6.1.1 General

Various statistics reflecting oil production from time to time on the Piper Alpha platform were relayed to the Flotta terminal and recorded there as well as being noted by the operatives working at the terminal. One of these records was produced and witnesses spoke to what they had observed. The implications of this material was commented on by the parties at some length. It was principally the defenders who sought to place importance on this material. Thus the defenders averred that "there was a sudden and material drop in the flow of oil from Piper Alpha at least 7 minutes before the initial explosion". The defenders sought to get support for this aspect of their case from what had been noted at Flotta at the time of the accident. However in argument they were content to modify the time span so as to argue that production at the platform had diminished substantially or ceased about 2 minutes before the explosion rather than seven. Their contentions in this respect are important since if there was a major process upset at the platform some minutes before the accident (particularly affecting the production of oil) this may afford an alternative explanation of the accident and indeed eliminate a leak of a relatively limited amount of hydrocarbon as being the cause of the explosion. In fact Senior Counsel for the pursuers accepted that if it were proved that there had been a sudden and material drop in oil production minutes before the explosion this could cause him difficulty. The pursuers accept that there was some loss of oil production minutes before the explosion but not the major upset suggested by the defenders and they say this is explicable as being caused by loss of production following upon the tripping of the condensate injection pump B. This brought about the need for the operators to unload and recycle the reciprocating compressors. This of course would cause a loss of gas lift which in turn would bring about a loss of production. Indeed in my view there is no doubt that a loss of gas lift would bring about a certain loss of oil production. Moreover if the reciprocating compressors were recycled gas which would otherwise have been available for gas lift would escape to flare. The operators had installed a system of alarms and process trips designed to alert the Control Room to any serious change in process conditions but in respect of most of these we were not told in evidence just at what levels of disturbance they had been set to respond. I think that the amount of hydrocarbon which could have possibly escaped from PSV 504 was restricted in relation to the very considerable quantities of condensate produced by the platform and therefore may not have had much effect on the systems for identifying process upsets (other than gas alarms). It is therefore unlikely that any phenomena at Flotta which indicated an upset equivalent to a shutdown could have been in consequence of the mere escape of condensate from a jagged condensate injection pump. On the other hand the process disturbance recorded at Flotta, had this been due to an escape of crude oil, must have been reflecting the loss of a substantial amount of oil. Certainly enough crude oil to have created a large explosion. The scale of the operations is illustrated by the fact that loss of gas lift alone could produce a production loss of about 25,000 barrels of oil per day and yet this loss would reflect nothing like a shut down. On the other hand one would have expected any escape of a large amount of oil sufficient to register a significant drop at Flotta (as distinct from a mere drop in production) would have triggered some of the various pressure and level drop alarms on the platform and the evidence was to the effect that this did not happen. The starting point for the pursuers’ submissions on this matter is that the reciprocating compressors were unloaded at least 10 minutes before the explosion and I agree that in fact that happened. Mr Vernon reported as much to Bollands the Control Room Operator. I further think that it has been established that the full effect of the loss of gas lift would have been noticed in about 10 minutes although some effect would be noticed after about 3 minutes.

6.1.2 The Spectra-Tek System

Details of the operators’ metering and telemetry system was given by the witness Bruce Lawson. Mr Lawson seemed to me to be a reasonably reliable witness. He was a senior metering engineer and had been employed by OPCAL in 1988 as a metering engineer. He had detailed knowledge of OPCAL’s metering and telemetry systems in their offshore installations. OPCAL has a telemetry system. Insofar as this was based on the Claymore platform it was the system known as the Spectra-Tek system and Mr Lawson had been the project engineer responsible for the installation of this. This system involved the automatic transmission of information between the pursuers’ platforms and also from the platforms to the Flotta Terminal. The communication network used microwaves and tropospheric systems. The microwave system operated on line of sight and the other did not. In relation to Piper Alpha the transmission dishes required by these systems were situated at the north face of the platform. As was explained to me a communications system is one which can ferry information from point A to point B. Telemetry is the information that is carried along the communication system. There were two functions of the operators’ telemetry system. The first of these is the offshore function. The Operators offshore are concerned primarily with the operations which are current and the information which at any point of time may affect these. This information can affect their productive capacity. Thus for example the pressure at Claymore can affect the other platforms’ ability to export products. The onshore interest on the other hand is not so focused on immediate data as on overall trends. The system on Piper Alpha was not as modern as that on Claymore and was known as the Solartron system. The advantage of the Spectra-Tek system on Claymore was that it had a trending facility. The system on Piper was only a real-time system. It is perhaps important to note that the Solartron system on Piper was situated in the Control Room there. It had an alarm panel which would annunciate among other matters serious interruptions of the process flow. It could also annunciate what were regarded as minor alarms and this would include indications of low or high pressure. This alarm had not annunciated prior to the accident. The telemetry systems would also relay from one platform to another or to Flotta alarms relating to a communications failure or a shutdown. However Piper Alpha was at the hub of the system so that any communication from another platform to Flotta had to be relayed through Piper. This means that if Piper were blown up the alarm system at Flotta would indicate not only a loss of communication at Piper but a loss of communications with the other platforms and alarms at Flotta would show this.

The pursuers argued strongly that if prior to the explosion a serious loss of oil pressure had occurred this would have registered on the alarms in the Piper Control Room, including the Solartron alarms. The defenders contended on the other hand that in the absence of evidence about the levels at which alarms were set there could be no conclusion arrived at as to what was required to set-off an alarm. In my view this can only be so within relatively fine limits. For example there was a substantial volume of hydrocarbon circulating through the process system at any point of time. It certainly cannot be assumed that the operators would want to be alerted to every variation in the process flow. An oil-well might be withdrawn from production causing some diminution in production but it may be that this would not give rise to a flurry of alarms. However one must pay some regard to the purpose in having an alarm system. It would be most surprising if the alarm system was not set to register any change in the process gross enough to require urgent action and in particular one which might signify the development of a serious danger. I think that it is clear that any change in the process approximating to an involuntary shutdown would be in that category and I find it difficult to believe that there could be a loss of flow equivalent to such a shutdown without alarms annunciating in the Control Room. If this could happen the elaborate alarm and control system would be relatively pointless.

The computer graphic number 56/2A of process illustrates the lay out of the Control Room on the Piper Alpha platform. A person sitting at the Control Room operator’s desk could see the alarm annunciator and in any event would hear the audible alarm. Number 42/1 of process illustrates the Control Room at Flotta. If there were to be a communication fault at Piper Alpha there would be flashing lights and an audible alarm at the control panel at Flotta and the alarm could be converted into a solid state light by operating the acceptance button.

The Spectra-Tek system on Claymore in the first instance recorded local data from the metering skid on Claymore itself and this data would include the flow from the Scapa Field which converged on Claymore. The said data is collected by what is known as a multi-drop loop. This means that to collect data from each of the nine microflow computers it takes 3 seconds. Thus to go round the whole loop takes 27 seconds. Therefore if a particular piece of data just misses its place on this chain it may take approximately 30 seconds before that data can be displayed. The computer is able to set out trends in the data it collects and indeed the data for the nine streams is amalgamated. The data collected for each stream would include readings for a number of matters such as pressure differentials between say Claymore and Flotta. The data used for this purpose is collected on a 1 minute basis. This means that if a particular value is shown on the output documentation at a specified time it is possible that such refers to data actually recorded 59 seconds earlier. The gathering of data by the Spectra-Tek system from sources external to Claymore, such as Piper, does not require to go through the multi-drop loop and thus the input is much faster, perhaps by a second or two. Only the trending data is recorded. In making comparisons between the different kinds of data allowance has to be made for the time skew caused by the multi-drop system. Eighteen different trends were configured. The trending data was also collected on a hard disk at Flotta. This meant that after the accident information was available of the trending records relating to Piper and in this respect two trends 1 and 4 were particularly relevant. The trending data is timed according to the system’s own clock and this is not synchronised with any other clock. After the accident the hard disk containing the Spectra-Tek data was brought to Mr Lewis, a software engineer, (and a witness in the case) and he produced the document setting out the data which is number 41/15 of process.

Trend 1 relates to the differential pressure between Piper and Flotta. Trend 4 relates to the flow rate from Piper to the Tee junction on the MOL where the Piper flow meets the Claymore flow. Thus the oil flow leaving Piper was relayed to Spectra-Tek and recorded there at minute intervals. Trend 5 was the flow rate from Claymore to the said junction. Piper had an indirect input into this trend because the data from the Tartan platform went through the telemetry system on Piper. Trend 18 relates to the main oil line pressure as measured at Claymore.

If there was a failure of the telemetry system on Piper certain trends would be lost and the data for any such trend would become invalid. For example because of the inter-relationship of the data affecting the various trends, trends 1, 2, 3, and 4 would all become invalid if telemetry at Piper were to fail. If the equipment in the Control Room at Piper were damaged by the effects of an explosion then failure of the telemetry system could result either by disconnection of cables or disconnection of energy supplies.

If there was a problem with the telemetry system on Piper then there would be a short delay before this registered on the Spectra-Tek system because of what the witness Lawson referred to as the "handshake". This is in effect attempts made by the stations under the system to establish contact with one another before an alarm is annunciated. This process could take as little as 20 seconds but could also take longer.

On the day of the accident in relation to the recording of trend 1 at 2200 hours there is a pressure value of 43.4 bar brought out but at 2202 there is a value of 42.30 and that value thereafter continues to be repeated. The freezing of values indicates that telemetry is no longer working. Although it is not impossible that the system would record the same value on successive readings this is unlikely. Thus subject to adjustments on the time scale the time when the telemetry on Piper failed can be worked out. In relation to trend 1 the telemetry at Piper failed no earlier than 2202 hours as printed although failure a minute or so later is just possible. However the trend information also shows that there was a gradual reduction in oil flow after about 2156 hours and this could be attributable to the loss of gas lift. Certainly in my view the trend information would not justify a finding of a sudden and material drop in oil production 7 minutes before the explosion. Trend 4 also shows a gradual loss of production from about 2158 hours followed by a constant value after 2202 hours. A particular value may in normal circumstances repeat 2 or 3 times but given that the values being discussed continued to repeat in my view it is likely that the beginning of this repetition marked the loss of telemetry. Indeed the values at 2201 hours could also have been a sign of the loss of telemetry since they show a reduction in pressure.

At the time of the accident there were about 15 wells in production and the effects of a loss of gas lift would have begun to be felt after about 3 to 5 minutes with the full effect happening after about 10 minutes. The effect of the loss of gas lift would vary from one well to another. The wells which required gas lift would have suffered a loss of production of about 50%. Thus given that the recycling and unloading of the reciprocating compressors occurred between say 2145 and 2150 hours there should have been a notable loss of pressure about 2200 hours or slightly earlier. It takes about 40 seconds for the pressure pulse to travel from Piper to Claymore so that the latter platform would see the change of pressure after that interval. A further important consideration is that it takes about 3 minutes for the pressure pulse to travel from Piper to Flotta so that a pressure drop at Piper would not be noted in the pressure measuring equipment at Flotta until about 3 minutes after it had occurred. The loss of pressure at Piper would result in a pressure drop in the MOL and this would affect the production at Claymore since the oil from that platform would meet a different pressure as it traversed the line. The production at Claymore was in fact recorded in trend 18. The pressure at Flotta would not drop below about 230 psi because there were pressure control valves (PSV 90 A and B) at Flotta designed to maintain the MOL pressure at that value. The graphs in number 41/16A and B of process show the recordings of trend 18 and a loss of Claymore pressure some minutes before 2200 hours to be followed by a sudden drop of pressure at 2202 hours. The advantage that trend 18 has over trend 2 is that the material in the latter contains a combination of locally and remotely derived data which causes a certain degree of time skewing. The graphs relating to trend 1 (the pressure differential between Piper and Flotta) also show the commencement of a gradual drop in pressure about 2156 hours to be followed by a drastic drop between about 2201 hours and 2202 hours. The loss of pressure due to the cessation of gas lift would be compounded slightly by the loss of condensate production when the condensate injection pump system failed.

6.1.3 Process at Flotta

The flow into Flotta first went through the feed preheater the function of which was to heat up the fluid. The flow of hot oil was controlled by temperature control valves (TCVs). If the flow coming into Flotta were to diminish the temperature of the oil flow would increase because the oil was itself heated by hot oil and the amount of this heating oil would not diminish. Thus a reduction of oil coming into Flotta would require process adjustment by an operator because otherwise the TCVs would automatically close if the pre-set pressures were exceeded. Such closure would cause alarms to annunciate in the Control Room. Alarms of this sort were not in themselves an unusual occurrence. After it was heated the oil flow went through a separator and then a desalter. The flow emerging from the desalter was split into 2 streams. One representing 40% of the fluid (called the cold stream) went directly to the stabiliser while the remaining 60% (called the hot stream) went through a vessel called the feed bottom exchanger before then going into the stabiliser. These 2 streams had first to pass flow elements which measured the flow going into the stabilisers and these measurements were recorded in a series of pen charts in the Control Room.

6.1.4 Evidence of Flotta Operators

Mr Kelly was the Process Shift Controller at Flotta on the evening of the accident and as such was responsible for the process side of the Flotta operation during his shift. The schematic number 42/1 of process shows his office within the Control Room where he was when the accident occurred. The Lead Operator that evening (working under Mr Kelly ) was Mr Stockan, and a Mr Slater was the Process Operator which is the equivalent of the Control Room Operator on Piper. In the Control Room (as shown in the said schematic) there were four stabiliser trains (A, B, C, and D) and these represented each of the trains in the process flow. The alarms for the TCVs were on the stabiliser train to which they related. There was a pen chart for each train and this also would be situated on the train to which it related. The pen charts were number of process 56/1. On these charts the red pen represented the vapour draw, the blue pen the hot feed and the green pen the cold feed. The rectangular boxes were intended each to represent the passage of an hour and the square subdivisions within the rectangles a period of 15 minutes. However I think it was agreed that no accurate timings could be taken from the charts. For some reason the instrumentation of train C was rather more sensitive than that on the other trains.

At a point in the evening approximate to 2200 hours Mr Stockan came to the window of Mr Kelly’s office and asked him to come out and have a look. He told Mr Kelly that telemetry alarms had come up offshore and that the feed had started to drop off. When Mr Kelly came out of his office and walked past the telemetry display units he could see that communication alarms had come up for every platform. He claimed that as he walked through the Control Room he had looked at the clock there and noted that the time was 2202 hours. It is not absolutely clear that such a precise recollection was not perhaps prompted by a degree of hindsight because it was at 2200 hours that the telemetry system signalled a clear loss of telemetry. Nevertheless he did testify that the Control Room clock did keep accurate time. Because all the telemetry signals were relayed through Piper a telemetry failure there would of course affect the telemetry relating to every platform. Mr Kelly then went to look at the pen charts. He reckoned that it took about 10 seconds from the time Mr Stockan approached him until he was consulting the charts and there is no reason to doubt the approximate accuracy of this recollection. When Mr Kelly looked at the chart relating to train A he saw at once that the blue and green lines on the chart had begun to drop markedly which indicates that there was a serious diminution of the flow. This drop continued as he watched. Before the point where the traces had begun to drop markedly they had been within normal limits. The movement in the traces from relative normality to clear abnormality as observed by Mr Kelly according to his evidence took from about thirty seconds to about a minute. The extent of the drop in terms of Mr Kelly’s experience was sufficiently marked to signal a total shutdown at Piper and I accept his judgment on this. When the traces on the various trains had dropped to their base levels he instructed Mr Stockan to contact Piper but the telephones were out of order.

The witnesses who gave evidence on the matter of what was observed at Flotta did not always agree on matters of timings. I did not form the impression that any of these witnesses were doing other that their best but given the compressed time scale of the relevant events it is only to be expected that there could be a degree of inaccuracy. Thus there may be some uncertainty as to just how long it took before Mr Kelly could be certain that he was observing a drop in flows equivalent to a shutdown at Piper but it certainly was relatively short.

It has to be observed that the pen charts are not related to the telemetric system but rather record the actual flow into Flotta and this would explain why the pen charts continued to show some flow because flow would be continuing from Claymore. It should also be noted that it was accepted by parties that the pressure pulse would take about 3 minutes to travel from Piper to Flotta and it would therefore take that time before the flow elements at Flotta would record pressure changes at Piper.

Some evidence about timings can be derived from the entries in the Control Room log book. This was in general kept by Mr Slater but on the relevant occasion because the operators were under pressure as the critical events occurred much of it was not written up until the morning following the accident. However the entries were made by reference to a scratch pad kept at the time of events and Mr Kelly, Mr Stockan, and Mr Slater consulted before the log was written up. The scratch pad was not produced and indeed no one was asked if it was still available. The log had an entry that at 2200 hours there was a communication fault for all rigs. On the other hand Mr Stockan gave evidence that there was a clock forming part of the telemetry system and that this clock froze at 2202 hours. He may well be right about this because there are other indications in the telemetry records that this was the time at which telemetry failed. If so the log would be inaccurate as to the time of the communication failure and this would not be surprising. On the other hand on one view Mr Slater would have had the best opportunity to note the time when the communication alarms came up. Mr Kelly said that he personally did not make any contribution to the entry relating to these alarms.

The loss of telemetry would not in itself have prompted Mr Stockan to report to Mr Kelly since this happened from time to time for a variety of relatively innocuous causes. What on the other hand did motivate Mr Stockan to report events was signs of an obvious drop off in production and inability to contact Piper.

Mr Slater asserted that the first thing out of the ordinary that he noticed was that the pen charts were showing a serious drop in production. This was a larger drop than might have been brought about by loss of gas lift. He had been in the vicinity of the stabiliser trains in the Control Room. Shortly afterwards alarms indicated that the TCVs had closed which also indicated a serious drop in the flow. Mr Slater’s evidence that he had not noticed the communication alarms is not too convincing since it was apparently he who made the log entry that such alarms had gone off at 2200 hours. However he claimed that his recollection was that he had noticed a drop in the flow before he heard any alarms - such alarms relating to the TCVs. It should be noted in relation to the log that in the morning after the accident there was a debate between Mr Kelly, Mr Slater, and Mr Stockan as to what time ought to be entered in the log in respect of the time when the feed rate began to drop. At the time Mr Stockan was taking the view that the drop in feed rate did not occur until after 2200 hours but when he gave evidence he claimed that the drop had occurred about 2155 hours - that is before the accident. He claimed that the low flow alarms had gone off about that time. Such alarms are likely to have indicated at least a 25% drop in feed rate although the precise figure would depend on the settings. However if this were so it is curious that he did not refer to Mr Kelly until as he said the communication alarms had gone off. Moreover when Mr Kelly came out of his office he noted that the pen charts were still dropping which might be odd if a significant drop had been noticed at least 5 minutes earlier. Moreover Mr Kelly noted that the pen charts were dropping rapidly. Nevertheless Mr Stockan explained that during the periods in question it was not at all unusual for low flow alarms to go off because the production processes on the platforms were somewhat unreliable. When pressed by Counsel Mr Stockan was prepared to admit that his timings may not have been accurate and given the whole circumstances I should have thought that this concession was not surprising. Mr Slater on the other hand claims that he did not notice the TCV alarms until after such time as he had observed a serious loss in gas flow. For some reason the sounding of the TCV alarms does not appear to have been logged. Number 94 of Process is a rather curious document. It is dated 7 July 1988 and Mr Stockan accepts that it is in his handwriting although he cannot remember how it originated. It would appear to state that all the communication alarms went off at 2201 hours. The next entry is that the feed to the plant started to drop off immediately and this is certainly not consistent with the evidence Mr Stockan gave about the sequence of events. Although Mr Stockan cannot remember writing this document it must have been written shortly after the accident and I think it places a question mark over his subsequent memory of detail. It is to be noted that there was a shutdown alarm on the telemetry panel on Flotta. This initiates an alarm if the combined flow rate from Piper falls below a certain level and the purpose of the alarm is to give a signal if Piper shuts down. It is important to note that none of the Flotta witnesses speak to this alarm annunciating before the break-down in telemetry at the time of the accident. The absence of such an alarm is difficult to reconcile with any suggestion that there was a major upset on Piper before the explosion.

6.1.5 Defenders’ submissions on Flotta

The defenders sought to argue that on a proper analysis of the Flotta evidence it was clear that there had been a major drop in production at Piper Alpha - indeed one equivalent to a shutdown - some minutes before the telemetry broke down. Since it must be assumed that the failure of the telemetry system came at the time of the explosion then there must have been a major upset of the production at Piper before the explosion. A leak from PSV 504 could not have caused such a response and therefore it must be inferred that some cause other that a leak from the said blind flange was responsible for the shutdown conditions and if this happened then such a breakdown would have been a pre-eminent source of the explosion. I would certainly agree that if conditions on Piper were such that a major general shutdown of production was signalled before the explosion then it would be difficult to reject the possibility that the shutdown was occasioned by a large escape of gas and this event could of course have caused the accident. The defenders contended that the conclusion they relied upon was supported not only by Mr Stockan but by the pursuers’ own witnesses. The defenders accepted that the computer clock at Claymore from which the Spectra-Tek records were taken was 38 seconds fast and since that data was recorded only at minute intervals a recording shown could fall anywhere within that period. Another important consideration is that the Spectra-Tek data might not freeze until about 20 seconds after the loss of telemetry. This is the result of what is described as the "handshake" between the computers as they tried to adjust to the signals indicating a loss of telemetry. The defenders also agreed with the pursuers that a fall in pressure at Piper would not be detected by the metering system for flow at Flotta until about 3 minutes after the event because of the time the pressure pulse would take to travel from Piper to Flotta. Subject to a possible time lag of about one and a half minutes the Spectra-Tek system is recording real-time data whereas the pen chart system at Flotta indicates a time lag of about 3 minutes because of the pressure pulse. Senior Counsel for the defenders argued that on the most favourable view for the pursuers the loss of Piper production pre-dated the explosion by 2 minutes. Mr Lawson gave evidence that a failure of the Spectra-Tek system could occur if the communication disks on the north face of the platform were damaged or if the communications between these disks and the communication room were damaged. Mr Lawson also explained that the system could also fail if the equipment in the platform Control Room was damaged or there was a disconnection of cables or supplies. There certainly was evidence that the explosion caused damage in the Control Room and this included damage to the computer VDUs. The defenders argued that the explosion was certainly not earlier than 2200 and on that point I agree with them since the preponderance of the evidence places it close to 2200 hours but not before that time. Number 41/17 of process was said by the pursuers’ witnesses to be a print-out from Claymore showing various readings from the Spectra-Tek system at about the time of the accident. The defenders founded on this document. This duplicates what would have been recorded at Piper. The document shows that there was a loss of telemetry from Piper between about 22O2:08 hours and 2202:50 hours. Moreover the trends shown in number 41/15 of process indicate that telemetry was lost just after 2200 hours. When the values on the trends freeze this is an indication that the telemetry has failed. Two or three constant values may occur in normal course but if these are followed by a string of the same numbers this in my view suggests strongly that the constant numbers have from the beginning indicated loss of telemetry. Trends 1 and 4 may be particularly significant because they do not include any locally derived data. The pressure control valve at Flotta seeks to maintain pressure in the flow from the MOLs at the set level which means that the pressure differential between Piper and Flotta should not vary much (although it may be influenced by flow rates from the other platforms). Looking to the whole evidence I have concluded that the loss of telemetry at Piper occurred at 2201 hours or, allowing for the intrinsic delays in the system, seconds before. However even in attempting to interpret the data care must be taken not to rely too much on the apparent precision since the time scales are very narrow. The defenders’ Senior Counsel was at pains to point out that the experts had conceded that a telemetry failure as late as say 2203 was within the bounds of possibility. However possible a slightly later time may be as I have indicated I think that it is unlikely that normally recurring constants would immediately precede the failure. This would be quite a coincidence. Mr Stockan testified that he had noticed that the computer clock in his control room had failed at 2202 hours but I doubt if his memory is sufficiently reliable to cope with fairly fine detail. The defenders argued that it can therefore be concluded that the explosion occurred at the time of the telemetry failure. I think this is stating matters rather too baldly. I prefer the way the pursuers put it which is to contend that the telemetry failed at the time of the explosion or shortly after. It is not obvious that the telemetry would fail instantaneously with the explosion. For example not only the damage to the basic equipment could have caused the telemetry to totally fail but also damage to the transmission disks or interruption of the energy supply. On the other hand it would appear unlikely on other grounds that the explosion occurred materially before 2200 hours. For example the witness Captain Morton remembered that just before the explosion the 10-o’clock news had started.

After the loss of telemetry and he had noticed some drop in pressure Mr Stockan brought Mr Kelly out of his office. Mr Kelly claimed that it was 2202 hours when he looked at the clock on his way to look at the pen charts. This was not a digital clock. The clock, he said, was always kept accurate. Mr Currie argued that the pursuers had not challenged Mr Kelly in respect of this time. This is true and it may be that the pursuers accepted that Mr Kelly was giving times to the best of his recollection. If so I think that they were right. Nevertheless I am faced with the task of deciding whether Mr Kelly’s precise timings can be reconciled with other acceptable evidence. Mr Kelly certainly accepted that there was a drop in the feed rates shown on the pen charts and expressed the view that this drop must have started about 30 seconds before he looked at the charts. I think there is little doubt that the pen charts are perhaps the most reliable record we are left with of the flow rate from Piper about the time of the explosion. Moreover I accept that they show that at or about the time of the explosion there was a rapid drop of pressure at Piper equivalent to a total loss of production. Unfortunately however the charts do not show timings that could be regarded as reliable. Relying on Mr Kelly’s timings the defenders argued strongly that if the major drop in production at Piper was noted to have occurred by about 2202 hours then allowing for the delay of 3 minutes required for the Piper pressure pulse to reach Flotta then the drop must have occurred 2 minutes or so before the explosion.

It should be noted that as Mr Kelly walked through to look at the pen charts he saw that the communication alarms were up in the control room. The Log number 42/3 of process is interesting. It shows that at 2200 hours communication faults were noted on all the rigs. Then there is an entry that at 2202 hours the feed rate was dropping off fast. The Log was partly made up the next morning from discussion between the operators and perhaps from entries on the scratch log. The significant thing is that at that time close to the incident the impression of those concerned was that the communication fault occurred 2 minutes before a pressure drop was noted. If the entry about the time of the pressure drop should have read 3 minutes rather than 2 then that would be reasonably consistent with the explosion having caused the pressure drop.

According to the evidence given by Mr Stockan the first thing he noticed were the TCV alarms which he thought had annunciated about 2155 hours and it was not until 2202 that the telemetry alarm went. It was only after that point that he communicated with Mr Kelly. Mr Stockan had noted a drop in pressure shortly after the TCV alarms. This programme would mean that the oil flow at Piper had reached a critical level of reduction about 10 minutes before the explosion. This evidence certainly does not square with what is written in the Log nor indeed is it possible to reconcile the timings with those of Mr Kelly. Mr Stockan accepts that his timings may be wrong and Mr Currie did not seem to be too keen to rely on them preferring those of Mr Kelly. Mr Stockan also accepted that it is possible that he is mistaken in attributing the drop in oil flow to a time preceding the alarms. On the other hand apart from the niceties of the times Mr Stockan has the pressure showing a pronounced drop before he summoned Mr Kelly.

Mr Wottge took data of oil production of the platforms from the Spectra-Tek records at Claymore and he averaged this out for the period just before the accident. Looking at the average position his analysis shows that there was a downward trend in production starting at about 2148 hours and becoming more significant about 2158. The drop is in percentage terms from about 102% to about 89%. Some of this drop was attributable to Piper and some to Claymore. However this loss of production was before the system shows frozen values and is likely to have been attributable to loss of gas lift on Piper and of course there is also a loss of condensate production from the time the pump failed. Defenders’ Senior Counsel argued that the drop noted by Mr Wottge at 2158 hours is consistent with his thesis that the pressure was dropping drastically before the explosion. However what the defenders seek to take from the pen charts is not a drop of a gradual character but a drop signifying something similar to a shutdown. Indeed Mr Wottge’s analysis of the Spectra-Tek data discloses what he calls a "drastic reduction" in the flow from Piper about 2202 hours. The pressures coming from all the platforms interact with one another so that fluctuations within normal ranges cannot necessarily be attributed to a particular platform.

6.1.6 Conclusion on Flotta recordings

The defenders urged me to conclude that several minutes before the explosion the production processes at Piper had effectively shutdown. It is certainly true that it is impossible to reconcile all the evidence spoken to by Flotta witnesses. Indeed it was in recognition of that fact that the defenders seemed ready to abandon the evidence of Mr Stockan. On the other hand the defenders’ case on this Chapter depends very much on the evidence of Mr Kelly. Given that the telemetry system clearly failed about 2201 or even 2202 then if this represents the time of the explosion and the production breakdown can be attributed to some minutes earlier then this would be difficult to reconcile with the pursuers’ hypothesis about the cause of the accident. A point made by the defenders, and it has a measure of force, was that the pursuers did not appear to challenge any of Mr Kelly’s evidence particularly as it related to timings. If there was evidence to support the view that there was a breakdown in production at Piper some time before the explosion then in that situation the evidence from Flotta may have given support to the defenders’ contentions. However there is a strong body of evidence that suggests that the inferences the defenders seek from the position at Flotta would be ill-founded. In respect of the position at the Piper control board at the time of the accident Mr Bollands, the Control Room Operator, gave his evidence about the state of the alarms at the time of the accident in a clear and convincing manner. His observation about the incidence of alarms was materially supported by Mr Clark, the Maintenance Lead Hand. Prior to the accident Mr Bollands did not observe any process alarms other than those relating to the tripping of the Compressors and the Condensate Injection Pump. In relation to oil flow the platform operators had an elaborate system of safety features and alarms. Thus anything untoward in the oil flow should have produced trips and alarms. It is true as the defenders argued that we were not given details of the settings for the various safety flow and level meters and alarms. However these features were highly important. Indeed several witnesses spoke to the fact that at about the period of the accident process failures were a frequent occurrence. The pen charts were indicating a sharp and rapid drop in the flow rate equivalent to a shutdown. In my view it is inconceivable that the safety features would be set in a way other than that they would respond in the event of conditions equal to a total failure in production. If the alarm and safety system did not respond to an accidental production shutdown it would have scarcely been worth having. Thus if the situation suggested by the defenders had arisen appropriate alarms would have been annunciating in the Piper Control Room and this did not happen. Moreover it is difficult to believe that the operators working in the Modules such as for example the Oil and Water Operators who had duties in Module B would not have been alerted to a process breakdown. It is of course possible that the module B operators were not in the module at the time but given that the reciprocating compressors were being re-cycled and unloading at the time it may have been surprising if they had not duties to perform there. Moreover I should have expected that any mass leakage of gas in Module B would have triggered at least some of the gas alarms in that module. There is of course evidence that might signify that some of the alarms in that module were not operative at the time but it would certainly be another coincidence if the absence of any alarms at all in the module could be explained in that way.

In my opinion the evidence supporting the view that there was no shutdown in process preceding the explosion is far stronger than the rather tendentious inference which the defenders sought to extract from the evidence concerning Flotta. In the circumstance it is not necessary for me to make findings as to the source of the weakness in the Flotta evidence. However there may be a variety of explanations for the apparent anomalies in the Flotta evidence. For example the defenders’ theory depends to a degree on Mr Kelly’s timings being very accurate. It is certainly the case that the pursuers did not challenge these but as I have said this may have been due to the realisation that Mr Kelly was giving his timings to the best of his recollection. The defenders for their part led Mr Stockan who in some respects differed from other witnesses in regard to timings. As the evidence in this part of the case indicated not surprisingly at least some witnesses had difficulty in being accurate about the very precise timings on which the defenders’ case depends. Some of the clocks either on the Spectra-Tek system or in the control room may unknown to witnesses have been inaccurate. The loss of telemetry may not have coincided exactly with the explosion but occurred shortly after it. Finally I think it could be very significant that in the Log compiled in the hours immediately following on the explosion the loss in production is thought to have occurred some minutes after the loss of telemetry.

6.2.1 General

The question of gas dispersion is central to the pursuers’ case as to the cause of the explosion. The question which they must answer is whether an escape through PSV 504 of such quantity of gas as the condensate pump system could have accumulated could have caused the explosion which took place. Moreover could such escaping gas have dispersed in such manner as to account for the gas alarms which were noted by Mr Bollands? Could the gas which may have thus accumulated have exploded so as to create the forces necessary to destroy the B/C and C/D firewalls and to have hurtled the projectiles into Module B which are the pursuers’ explanation for the damage to equipment in that module which damage they say released hydrocarbon?

6.2.2 Dr Davies

The pursuers’ main witness on this branch of their case was Dr Davies who gave evidence over a period of 4 weeks. The defenders led a witness, Dr Bruun, whose evidence was restricted to a discussion of certain aspects of Dr Davies’ modelling methodology and only lasted 2 days.

At the time of giving his evidence Dr Davies was 47 years old. He was the managing director of BMT Fluid Mechanics. Before it was privatised this organisation had been the National Maritime Institute and then it merged with the British Research Association. Dr Davies was also the Chairman of an associated company BMT Offshore Ltd based in Aberdeen. He held a BSc Honours degree in applied mathematics and had a Masters Degree in Aeronautics from Imperial College. That work was concerned with airflows as they relate to structures. He had worked for the British Aircraft Corporation as an Aerodynamicist. In 1971 he returned to Imperial College where he carried out research on the behaviour of flows round structures and thus earned a PhD. In 1974 he went to work at the National Physical Laboratory where he did work on the low of bodies around circular objects particularly in relation to turbulent flows. This work arose from the requirements of the offshore oil industry. Thereafter he had a continuing involvement with that industry. From about 1976 he was working on topside aerodynamics including fluid dynamics. This was concerned very much with the nature of airflow around platforms. Thereafter he had been involved for 18 years in consultancy work for the offshore industry. His experience included wind tunnel modelling. He had carried out advisory work in relation to the Piper and also the Claymore platforms. Some of his work was the subject of publications. In the 1980s he was considerably involved in heavy gas dispersion and this included the modelling associated with the work. He did work in connection with the British Gas Terminal at Canvey Island and in that work he was looking at the behaviour of heavy flammable clouds of gas. Thereafter he was engaged in a number of further major studies into gas behaviour. He was given a contract by the US Gas Research Institute to carry out a wide range of wind tunnel exercises. Among his other work was a study into high pressure releases for a consortium of European companies. He has been involved in consultancy work in relation to about 30 or 40 different oil platforms and this covered work into the dispersion of gases within Modules. Since the Piper Alpha disaster he had carried out ten studies into the kind of problems that arose there. Since 1972 he has been continuously involved in the setting up measurement systems. Since 1982 he has been a Chartered Engineer. He has been the author of a number of publications all within the field of fluid mechanics. One of these was in connection with Froude number scaling a problem that arose in his evidence. In my view there was no question about Dr Davies’ qualification to give the evidence he gave and indeed the defenders never doubted those.

Dr Bruun was also well qualified. He was 55 years old when he gave his evidence and held the post of lecturer in the Department of Mechanical and Manufacturing Engineering at the University of Bradford. He was awarded a first class degree in Mechanical Engineering in 1964 and a PhD in 1967. This latter degree was in the general area of fluid mechanics. After qualifying he had periods of research work at first Southhampton and then Cambridge Universities. He was a guest professor at Karlsruhe University before taking up his present post. He has a number of publications to his name. Some of these concern the behaviour of flows but he has done a considerable amount of specialist work into hot wire anemometry. Hot wire anemometry involves the use of a wire heated up by means of passing an electric current through it. The device can be used to measure quantities such as velocity and also concentration measurements. There is no doubt that Dr Bruun has a lot of experience of the use of these. On the other hand most of his experience in fluid mechanics has been in the academic field. Dr Bruun sought to challenge a certain amount of Dr Davies’ methodology particularly in regard to the use of hot wire anemometry in his modelling runs. However a lot of the detail in his evidence was not pursued by the defenders in their submissions no doubt because at the end of the day they were happy enough to adopt some of Dr Davies’ evidence.

In cross-examining Dr Davies, the defenders sought to attack his use of Froude numbers and Reynold numbers in his modelling. The problem was of considerable technical difficulty. At the end of the day the defenders did not support their criticisms by evidence - even from Dr Bruun who had relevant experience about such matters.

Dr Davies concluded that a release of flammable material within Module C which affected only the gas detectors within Module C at gas zones C2, C3, C4 and C5 would have had to be released at the eastern half of Module C. I did not understand the defenders, at the end of the day, to challenge that particular assertion. Moreover he opined that any release within Module B that could have passed to the east end of B would have had to initiate gas alarms within B to get there. This of course assumes that the alarms within B were working correctly. In 1989 Dr Davies had carried out certain wind tunnel work to investigate a variety of gas release scenarios in Module C. This work was done on the basis of a number of assumptions which had been put to him and in particular assumptions relating to the pattern of gas alarms which had annunciated at the accident (these latter taken from the evidence of Mr Bollands). He concluded from his 1989 work that releases from the vicinity of PSV 504 frequently produced the first alarm at zone C3. However the gap between the first alarm and the second alarm which had annunciated was about 2 minutes and this was never sufficiently long to explain the alarm pattern that had been given to him. Because of this he carried out further work in 1993 in relation to small releases. He also carried out in 1993 work on gas escapes within Module B. The pursuers contended that the evidence of Dr Davies confirmed that a pattern of alarms such as was experienced was consistent with an escape of gas in the general vicinity of PSV 504. Mr MacAulay submitted that the evidence of Dr Davies showed that the initial alarm at C3 could have been caused by a leak from the area of PSV 504 if the release had been a small continuous release or a curtailed release such as a puff type release. A second release from the same site, if more substantial could explain the remaining alarms. The pursuers argued that it would in normal circumstances be highly coincidental to have two separate gas leaks separated by about 2 minutes. However this is just what one would expect if the escape was due to a pump jagging operation. Certainly I think that it would be reasonable to conclude that there was some connection between the gas escape which caused the first alarm and that which caused the second. It would be rather a coincidence if there were two totally unrelated escapes within minutes of each other.

Dr Davies’s work showed that any gas released in Module B would dilute with air as the wind moved it through B to the east face. This process of dilution would continue at the east face of the platform even before gas could be ingested into C. Thus to transmit gas generating from B which could be absorbed into C and form a flammable mass there the gas released in B would have to constitute a relatively large mass. Such a mass would necessarily initiate any gas alarms in B which were working. Mr MacAulay emphasised that no alarms had gone off to indicate any process disturbance in B and this might have been expected if there had been a large release of gas there. Moreover Dr Davies could envisage no circumstances where it would be possible to ingest gas into Module C so as to trigger an alarm at zone C2 and the defenders did not lead any expert evidence adverse to this view.

Dr Davies did not consider that the precise sequence of the final flurry of low level alarms made any difference. In any event the alarms were so close together that it is difficult to suppose that their sequence accurately displayed the sequence of the arrival of critical quantities of gas at the detectors.

The natural wind over the sea is a turbulent wind with a speed that varies with height above the platform. When the wind encounters the platform it will be disturbed by the presence of the platform and this will cause it to flow around, over, under, and to a degree through the platform. The wind external to the platform is the primary flow and any wind internal to the platform is the secondary flow. It is called secondary because it is generally much smaller than the primary flow. The secondary flow is principally created by the pressure difference across the Modules and these differences are generated by the primary wind flow. The amount of secondary flow through the Module will depend on congestion within it and the resistance to flow caused by the congestion is measured as a pressure drop which can be expressed as a pressure drop coefficient. Principally the windward faces of the platform will have positive pressures and the leeward will have negative pressures. Thus on the occasion of the accident there would be pressure driven ventilation through Modules C and B from the positive pressure side (the west) to the negative pressure side (the east). However the equipment causing obstruction to the wind in each Module will restrict this flow. The summation of all the drag the wind flow experiences as it goes through the Module can be measured and expressed as the pressure drop coefficient. Of course large objects such as compressors and tanks (known as "bluff objects") disturb the wind significantly as compared with more streamlined objects. Since such objects will cause variations in the air flow through the Module they are in effect creating turbulence. Thus the flow through the Module becomes dominated by local effects. The ventilation flow through the Module can be expressed as a ventilation rate in cubic metres per second. It will not necessarily be constant throughout the module and indeed can vary by a factor of 2 or even 4. The secondary flow can also be expressed as the number of air changes per hour. The foregoing matters were spoken to by Dr Davies and I did not understand that the defenders disputed any of them. Of course in determining how an escape of hydrocarbon within a Module would behave account has to be taken of the secondary air flow which would have the effect both of moving the gas and providing a source of air for the formation of a gas/air mixture. Moreover assuming an escape of 100% gas this would dilute as it moved through the module and mixed with air. If the escape is slightly dense and into a strong air flow it will act much as a neutrally buoyant gas. The defenders argued that this would be the case if the gas that had escaped was predominantly composed of the lighter ends and this they say must be so. Air of course represents neutrality. If you have a very dense release into still air the gas will bear all the characteristics of its heaviness and slump. As such gas dilutes its heaviness is progressively reduced. Thus the characteristics of the cloud may change during the course of its travel. Moreover if a heavy gas was released in the west part of either Module B or C as it travelled through the module some of the gas would become lighter and rise and thus could trigger an alarm in a high location at the east end of the module. As the released gas impinges on the equipment within the module that would also affect its dispersal. If a cloud or jet of gas is released it will take a finite time to arrive at a particular point in the module and to build up the concentration of gas there. However if we are dealing with a continuous release then once the maximum concentration at the point in question is attained that concentration will remain steady for so long as the release continues. That plateau of concentration is known as the steady state value. If that value is not sufficient to trigger a gas alarm then such alarm will never be triggered at that point. Moreover the arrival at the steady state condition does not necessarily reflect the maximum volume of gas within the module since some gas could escape out the end of the module before the steady state condition is reached. Thus the steady state is not necessarily the maximum concentration. With a transient type of release if this is terminated abruptly the gas may never reach the steady state level. Moreover because the gas detectors have specific response time a situation could arise with a transient release that the steady state was reached for such a short period that the detector did not have the time to register what otherwise would have been a concentration sufficient to trigger the alarm. If a liquid jet is released and then flashes the flashing process itself will accentuate the dispersion.

Dr Davies looked at a number of hypothetical release sources in Module C in order to ascertain how a gas cloud might behave in that module and in that respect he had regard to the location of the gas detectors. In so doing he used a model that replicated in basic form the obstructions in the module as these were in fact portrayed in the model of the platform which was in Court. Because the obstructions in the module offer considerable resistance to the air flow particularly towards the east end the initial wind speed as the air first enters the module will be considerably less than the wind speed attained once turbulence develops. Thus the wind speed out of the east end of the module would be greater than the initial speed of the air flow within the module particularly because of the substantial obstruction represented by the centrifugal compressors.

The location of the gas detectors in Module C are shown in number 12/112 of process. The first alarm to annunciate at the time of the accident was a C3 alarm which puts the first concentration of gas observed in the south east quadrant of the module. Dr Davies pointed out that generally in relation to the problem facing him the difference between a release of a neutrally buoyant hydrocarbon and one that is heavier primarily relates to the behaviour of the gas at the early stage after the release. A neutrally buoyant release would tend to spread very rapidly in all directions. A heavy cloud would tend to slump at the early stage of the release dependent on how large it is in relation to the ventilation. A heavier cloud will tend to stratify. Thus the density and concentration will tend to be greater towards the bottom of the cloud. Not only will the cloud be lower but it will be more packed and not spread out so much sideways to the airflow. Essentially it will be more contained. This is because nearer the top of the cloud it will be better mixed with air. The cloud will eventually disperse widely but not so fast as a lighter cloud. If a cloud of gas be it neutrally buoyant or heavier had escaped from the west end of Module C then Dr Davies would have expected it to trigger other alarms before arriving at the east end and the defenders make much of this opinion. Therefore he discounts an escape of gas from the west end of Module C. The gas would have become dispersed as it moved the length of Module C and thus since it would have become lighter a response from the higher placed detectors in the Module would also have been expected.

Dr Davies considered a release from points just to the east of reciprocating compressor A. He assumes a release somewhere along the line just to the east of that compressor and also at mid-height. In relation to a neutrally buoyant gas if it was released at the point being considered then his view is that the detector G101/2 would be unlikely to avoid being triggered. The same applies to G101/3 another detector in the C2 area. Indeed most of the detectors at the east end of the module would see the gas cloud. The point is that a neutrally buoyant gas might be expected to spread out quickly in all directions. He considered that a similarly sized release of a heavier gas from the same point would spread at a lower level and because of the configuration of the module would be more likely to move towards the south east corner. He would expect G101/2 to alarm under these circumstance but G101/1 may not. On the other hand all the gas detectors located at low level would certainly see a low level gas alarm. His assumptions about the spread of different weights of gas is illustrated by his modelling

A release to the north and to the east of the reciprocating compressors of a neutrally buoyant gas was also analysed by Dr Davies. He assumed that such a release would result in a relatively strong flow because of the space contours. It would encounter the process skid quickly and thereafter disperse widely as it mixes with air. G101/2 would alarm but G101/1 would be unlikely to detect. The earliest alarms would probably be in the C5 area. Alarms in the south sector may also trigger depending on the size and spread of the cloud. The release of a heavier gas at the same location, if it was released in such a way that it was not spread at source, would tend to move in a more southerly direction round the process skid. However it would also go through the process skid and would mix to a degree that it would rise and trigger G101/2. It would also trigger low placed detectors in the C5 and C4 areas and possibly also in the C3 area.

Regarding a release of gas just to the east and south of the reciprocating compressors, then if the gas was neutrally buoyant Dr Davies would expect the flow to be highly turbulent. He would have expected the flow to move from west to east not encountering many obstacles. Spreading would be encouraged by the fact that the flow would be faster on the north side. G101/1 would be a strong candidate to respond. Because there is some air trapped in the detector itself even very strong concentration of gas would lead to an alarm response and the stronger the concentration of gas the faster the response. Moreover with the spreading with height G101/3 would also detect. All the detectors in the C3 zone would detect irrespective of height and G 101/2 would also almost certainly detect. Because of the distances involved the timings would be short. If the release was of a heavier gas the gas would have a relatively easy path to exit out of the east of the module. He would expect low placed detectors in C3 to respond and then progressively detectors would respond across C4 and C5 depending on the size of the cloud. G101/2 would be a likely candidate to respond but G101/1 may not. Senior Counsel for the pursuers accepted that in the light of the foregoing if a lighter gas had escaped from PSV 504 then one would have expected an alarm at G101/1. However he contended that the escape from PSV 504 had been of a heavier type gas in which case G101/1 may not have detected it. He maintained that on the evidence that detector may have been located at a height of about 20 feet in which case there was more possibility that the gas could have escaped detection by it. These considerations represent a vital difference between the pursuers and the defenders since the defenders claim that on their analysis of the evidence of Dr Richardson the first stage release upon jagging would in fact have been effectively a neutrally buoyant gas cloud. This is in contra-distinction to Dr Davies who modelled the alleged gas escape as a propane equivalent. It is clear that as Dr Davies stated the dispersion of a gas release in general terms will depend on its initial location in terms of how far it will travel in Module C and also the precise dispersive nature of the environment in which it starts from. It will also depend on the composition and temperature of the gas.

The defenders contended that for Dr Davies’ views to have validity then at best to explain the fact that the C2 detectors did not annunciate in relation to the first stage release one would require conditions where there was a release of a heavy gas at PSV 504 with the release pointed downwards. This they say is not what the evidence points to. Dr Davies said that if you can get about 20 to 23 kilograms of gas out either in one puff or a low continuous release then if it is a heavy gas that comes down and under the C2 gas detector that could cause the C3 alarm. This particular approach seems to be postulated on the basis of escape of a heavy gas. The evidence of Dr Davies was that even a much lower level release such as 2-4 kilograms per minute could trigger a C3 alarm if continuous or even terminated after a short time provided that a steady state is attained. It is a puff burst of gas at a low level rate of release that could not trigger a C3 alarm. For such a short release a much higher rate of release would be required.

In relation to the natural gas as the gas that he uses in some of his tests Dr Davies defines this as being in relation to his work a mixture of methane, ethane, and propane which was so mixed as to be neutrally buoyant. He seems in his work to use ‘neutrally buoyant gas’ and ‘natural gas’ interchangeably. It is perhaps a pity that we do not know that the natural gas he used contained components in the same proportions as condensate.

If gas is introduced into a still atmosphere the gas can spread by the process of diffusion. If on the other hand it is introduced to moving air there is only diffusion at the interface between hydrocarbon and air but the turbulence or ventilation can also accelerate the diffusion process. In any calculation of flammable mass of gas at a particular location ventilation and turbulence have to be taken into account. The continuity of the source of gas is also an important factor. As his approximate starting point Dr Davies considers that you need a continuous release at the leak source of about 100 kilograms per minute to develop a significant flammable mass at the east end of the module. Moreover if there was a release as low as 4 kilograms a minute there would be no alarms at all. The concentration of gas to air will vary as the gas gets more distant from the leak source and becomes more diluted. If the release is continuous then the gas cloud at any point will increase in concentration to reach a steady state and this should not change so long as the release continues at the same rate. On the other hand if the release is short lived a particular location may never see the steady state value of the gas because such gas as there is will move on too quickly. Of course at different points the steady state concentration will not be the same since the further away from the source the concentration is going to decay. The point that the defenders seek to make out of this is that if any detector is going to see a particular level of gas it is likely to be the one nearest to the leak source.

It should be noted that the above views are those which Dr Davies also expressed on the basis of his general experience so that they rest independent of his methodology in modelling. Moreover although he was asked to assume that the gas which escaped was a heavy gas he was highly experienced in the behaviour of gas and he never doubted that a leak of condensate could conform to the assumption given to him.

Dr Davies also expressed general views as to how he would expect gas to behave if it escaped in Module B. Number of process 41/1 shows the location of gas detectors in Module B. He is of the view that as Module B is a more open module it would be likely to have a higher ventilation rate than C. The flow would be largely a turbulent mixing flow. There are open corridors in B which would speed up the flow. He opined that if a gas cloud was released towards the west end of B then he would expect many of the detectors in the module to see gas concentrations above the alarm levels. If material proceed to the east end of B so as to permit ingestion into C at alarm levels then the original escape would have required to be well above these levels because the gas would have been substantially diluted as it moved along Module B. Thus a release in Module B that could result in serious ingestion into C would have required to be a major release that would have triggered gas alarms in B (assuming that these were operational). I have no difficulty in accepting that view. The gas would stratify as it progressed so that both low level and high level alarms would be triggered and the alarms triggered would include G99/1, G99/3 and G99/4. G19, G20, and G21 would also see alarm concentrations. He specifically considered what the gas behaviour might have been if a natural gas (that is a neutrally buoyant gas) had been released about the middle of Module B but towards the north. He would have expected such a cloud to have been seen by a number of detectors including G99/3 among others. The lower detectors in the module had been placed in a low position particularly to enable detection of the heavier hydrocarbons.

If a dense cloud of gas had come out of the east end of Module B it would tend to drop and Dr Davies would have expected detector G127 at the 68-foot level to have seen gas. However that detector is external. In the present case since there certainly were men working at the 68-foot level when the accident occurred one might have expected that they would have noticed any material intrusion of gas into the platform at that level particularly if it had exploded.

The quantitative work that Dr Davies did in 1989 is summarised in his report number 12/362A of process. He did not at that time perform his quantitative exercise in relation to Module B because as he stated he considered that the possibility of a release there not triggering alarms was extremely remote. Moreover because the wind pattern would have spread the gas along the east face quite widely he thought that the creation of a significant inventory of flammable gas within Module C by such a process most unlikely. However in the light of certain averments of the defenders in the present actions he did in 1993 carry out some quantitative tests relating to Module B. Dr Davies also approached his tests on the basis that between the first alarm and the second flurry of alarms noted by Mr Bollands a period of 2 or 3 minutes had lapsed. Mr Bollands’ evidence in this respect was indeed not challenged. In the view of Dr Davies the fact that there was a gap between alarms of longer than about a minute would be sufficient to require two separate events. Dr Davies also said that if there was a flurry of alarms that event would signify a largish release of about 100 to 200 kilograms per minute. Thus in the first instance the flurry of alarms noted by Mr Bollands might point to a gas release of that size.

Dr Davies assumed in his 1989 tests that he could regard a concentration of 0.75% of the neutrally buoyant streams as being sufficient to trigger a low-level alarm and a concentration of 3.75% to trigger a high-level alarm. These figures are based on the lower explosive limit for methane. Indeed the detectors were calibrated for methane. When he came to do further work in 1993 he used the figure of 3.3% as the lower explosive limit of the neutrally buoyant stream. This was taken from Dr Balfour and the change should not affect the implication of his results. In relation to heavier gases, such as one would find in the condensate streams Dr Davies, used a lower explosive limit of 2.15% again taken from Dr Balfour. He used 1:100 and 1:33 scale models for his work and these were said to be accurate representations of the salient aerodynamic features of the platform and Module C. Because the detectors were calibrated for methane calculations were required to adjust the explosive limits for the heavier gases. These calculations were done by Dr Balfour and appear in his report. Thus for a calibration set for the low level alarm setting, namely 15% of the LEL of methane, the equivalent value required to generate a low level alarm for condensate would be 23%. The high level value for the heavier gases was taken by Dr Balfour as 115% of the LEL of 2.15% to give him 2.5% as being the level at which a high level alarm would be triggered.

Regarding the response times of the gas alarms Dr Davies again used the figures in Dr Balfour’s report. Thus for example for the Sieger 910 detector the time to alarm for the first alarm set at 15% LEL would be 1.5 seconds and the time to the high level alarm would be about 14 seconds. The applicable response times of the detectors are set out at page 5 of Dr Davies’ report. It will be seen that the response times for a low level alarm are in practice very quick.

PSV 504 was located about 15 feet above the deck and to the south and east of the reciprocating compressors and Dr Davies worked on that assumption. To adjust for possible imprecision in the exact location of the valve, in his 1993 experiments Dr Davies changed the assumed position of the valve about a metre northwards from the position that he had used in 1989. Dr Davies also worked on the basis that the flammable mass at the initiation of the explosion was about 40 to 60 kilograms. By flammable mass I mean the amount of hydrocarbon needed to be within flammable limits for the purpose of the initial explosion. These values for the amount of gas in the explosion were taken from the evidence of Dr Mitcheson and Dr Cubbage. Of course the flammable mass may not represent all the gas that has escaped from a leak source since some gas might for example escape or not be absorbed in the gas and air mixture. In any event a proportion of the leaked gas will decay. Indeed Dr Davies said that if a flammable mass in the region of 40 to 60 kilograms is required you would need a hydrocarbon release of about 140 to 160 kilograms per minute over 30 seconds. However the defenders contend that although a release rate of 180 kilograms per minute is quite possible according to Dr Richardson this rate is only going to last for about 6 seconds because at that point the flow will revert from liquid to gas. Dr Richardson’s calculations are of course all related to total mass and it is Dr Davies who translates this into flammable mass. There is of course a practicable limit in the release rate that would be consistent with the alarm patterns observed. Thus if the release rate was much higher than say 200 kilograms more alarms would have annunciated. However Dr Davies considered that the alarms could be accommodated with a release rate of between 150 to 200 kilograms per minute over about 30 seconds. Of course these views of Dr Davies were based on tests with propane. It should also be noted that the flammable mass is considerably less in a natural cloud than in a propane cloud because of differing dispersion and LEL levels. These broad estimates were generally based on the size of a flammable cloud that at the time of the explosion would not have generated a flame or hot combustion products in the Control Room. The witnesses who had been in that location when the explosion occurred did not experience either of these phenomena and this gives some support that the gas cloud was restricted in mass. Dr Mitcheson had worked on the assumption that the flammable cloud which exploded after expansion would have occupied about three-quarters of the volume of the module (taking the equipment at the east end into account) and that the explosion would generate an expansion of 7.5 to 1. He assumes that the module had a capacity of about 4,500 cubic metres. The total volume of the module was about 5,175 cubic metres and Mr Wottge had estimated that about 15% of the module was occupied by piping and equipment. That would leave about 85% of the volume of the module available for the gas. Dr Mitcheson indicated that in his calculations he did not take into account any material that might have escaped but been outside the explosive limits. Essentially he was looking for the mass of the flammable cloud. He expressed the view that in order to generate the sort of over pressures which occurred he would have expected the minimum mass of the flammable cloud to be about 15 kilograms. He is of course talking about condensate gas. He thought the upper bound would be about 50 to 60 kilograms. Thus he arrives at what can be no more than a broad yardstick. Mr Cubbage agrees that Dr Mitcheson’s figures would give a good view of the mass of flammable gas likely to have caused the explosion. On this part of his evidence Dr Mitcheson was not challenged and the defenders led no evidence to contradict it.

Dr Davies had to take account of ingestion of gas into the centrifugal compressors. With the tripping of these pumps combustion gas would no longer be ingested but that may not have stopped the ingestion of air for ventilation purposes. Combustion air when taken into the compressors was taken in via intake hoods positioned on either side of each of the compressors. The situation is shown in the schematic which is number 12/126 of process and the plan number 12/106. The air for combustion goes into the turbine section where combustion occurs. Most of the air for the combustion process is drawn in outside the module from beneath the external grating but about 1o to 25% is drawn in from Module C itself. When the compressors were working there was a total requirement of combustion air of about 27,000 standard cubic feet per minute. If gas were to emerge from Module B whether or not it might be ingested into the combustion system would depend on the ambient wind conditions and its weight. But in any event such gas would have to drop down and then be drawn up. There was a gas detector, G102/2, sited within the air intake hood of compressor A. There would be another detector, G 102/1, at the south intake in a equivalent position. The other two compressors had similar arrangements. In addition to air for combustion purposes these compressors ingested air for ventilation purposes. The gas turbine ran at relatively high temperature producing radiant heat so that coolant air was required for both the turbine and compressor compartments of the compressor enclosures. The air intakes for ventilation purposes is shown in number of process 12/126 where this aspect of the system is illustrated in red. The air is coming in from the outside of Module C and at the same level. Gas emerging from Module C would flow directly into the paths of these intakes. The air is drawn into the intakes by fans. After being drawn in the air goes to the compressor and turbine compartments. In relation to Compressor A there is a gas detector, G26, positioned almost at the mouth of the intake. That would probably be the first detector to register any gas drawn into the compressor through the ventilation system and it is in the C3 zone. However five detectors were available in C3 which might have triggered the first alarm noted by Mr Bollands. The ventilation air drawn into the compressors totalled 8,000 standard cubic feet per minute and of this 75% went into the turbine compartments and 25% to the compressor compartments. The compartments were pressurised though the pressure difference between the turbine compartment and the module was greater than in the case of the compressor compartment. The ventilation fans would run for about 2 hours after a compressor trip in order to ventilate and cool the compartments. The detector G27 is situated on the outside wall of the turbine compartment and G28 within the compressor compartment. Since the alarm panel in the Control Room can only accept one C3 low level alarm at a time it is possible that what Mr Bollands experienced as the first low level alarm in fact indicated responses from more than one of the C3 detectors. If high gas alarms were to occur in G27 or G28 the compressors would trip but not the ventilation. On the other hand if there was a high concentration alarm perceived at G26 then not only would the compressor immediately shut down but the ventilation would shut down as well. Thus it would appear that the centrifugal compressors were not caused to trip by a high concentration of gas because they seem in the case of the accident to have tripped before a high level alarm annunciated. Ventilation from the turbine compartment in each compressor vented through ducting to the south of the compressor as shown in number 12/126 of process. Ventilation air for the compressor compartment was taken into Module C through louvres at each side of the west end of the compressor compartments. The fact that each compartment had 2 louvres can be deduced from the said schematic and by the video film number 41/ 21 of process. Dr Davies proceeded on the basis that the exhaust amounted to 2000 standard cubic feet per minute and that this would exhaust equally between the 2 louvres. Thus 6,000 scf would be exhausted by the turbine compartment exhaust system.

With regard to the combustion air intakes Dr Davies was of the view that they had little influence on the air flows within Module C but he did not consider what effect if any these may have had on gas detectors. In his 1993 tests Dr Davies could not find a mechanism, even assuming that all the centrifugal compressors were running and the compressor compartment louvres were emitting 100% gas, which would account for the annunciation of a C2 detector. Before anything approaching a high level of gas could be expelled from the louvres the G26 alarm at the ingestion point would have to register. There was of course a C2 detector above the central centrifugal compressor. The closest C2 detector to the east of Module C was G101/3 which was situated above and outside the C compressor and was about 15 to 20 feet above the deck level. There are three alarms altogether in the C2 zone. The other two are G101/2 and G101/1 which is more inboard and towards the south. Their positions are shown in number 12/112 of process. Another alarm that may be critical to the pursuers’ hypothesis and which Dr Davies required to consider was G103/1 which the pursuers suggested was the C3 detector most likely to have triggered the first gas alarm noted by Mr Bollands. That detector was located within the fuel gas valve compartment of the centrifugal compressor C and was located beneath the walkway grating. Dr Davies considered that if a quantity of gas did get into the Centrifugal Compressor this would probably dilute with time because of the air being ingested.

In modelling one has to adjust the results of the modelling to take account of the effect produced by the scale of the modelling. Certain pragmatically derived compensatory numbers can be applied to the result and these are known as Froude numbers and Reynolds numbers respectively. Dr Davies considered that in the case he was dealing with it was only necessary to take account of the Froude number. This was as he explained because Froude and Reynolds numbers are incompatible. He was tested at length as to why he had omitted the Reynolds number but although his reasoning was very complex in the event the defenders led no evidence about this matter and I have no difficulty in accepting the reliability of Dr Davies on the subject. In testing he used two kinds of wind tunnel. One was a blower tunnel which was used essentially to blow air into the model. The second tunnel was much larger and was know as the number 7 environmental tunnel. Dr Davies was able to work out a pressure drop coefficient for Module C by attaching a 1:33 model of the module to the blower tunnel. Having got that figure he measured the coefficient of Module C for the 1:100 scale model and then adjusted the model until he could get a similar pressure drop coefficient. Thus he had secured a coefficient value for the larger more reliable model. Having got a ventilation rate for the 1:100 model he replicated that in the 1:33 model which was then attached to the blower model for the gas dispersion tests. His value for ventilation rate was 46 cubic metres per second which is equivalent to a wind speed of about half a metre per second. This can be compared with the external wind speed of about 8 metres per second. Dr Davies considered that on a fair estimate there would be about 50 air changes an hour in Module C. The significance of such speed is that a particular cloud of gas will have a finite time to reach the end of the Module and escape into the atmosphere. The average rate of air movement can be calculated form the air changes. Thus Dr Davies considered that this would give a ventilation rate of about 54 metres per second which compares well with his measured value of 46 metres per second. As I shall consider later as hydrocarbon moves from the source of a leak there will be concentration decay. As the defenders argued if you have a release source 5 metres from a downwind detector and there is another detector 10 metres downwind then not only should the gas cloud hit the nearer detector sooner but it should also have a higher concentration of gas when this happens. The defenders contend that on this supposition there is no way that there could be an alarm at C3 without an alarm earlier having been triggered at C2. The defenders point out that the pursuers’ hypothesis would involve the gas cloud triggering a low level detector, somehow developing so that it later triggered an alarm at C2, then somehow increasing to trigger a high level alarm at the compressors. However it was accepted that this analysis may be more suspect if in fact the gas cloud is composed of heavier ends.

The defenders accepted that a puff type release would not necessarily attain the steady state concentration that would result from a continuous release.

For Module B his values would be higher because that module is not so congested. For Module B he would expect the air speed to be about one-tenth to one-twentieth of the external air speed. He was certainly cross-examined about the accuracy of his estimates in relation to the modules for the air flow was introduced straight-on to the model whereas in reality the wind was blowing at a slight angle to the platform. Dr Davies explained that unlike the situation at the very mouth of the modules within them the air flows would be largely dominated by the turbulence and pressure differentials rather than the external wind. Moreover it was suggested that the model used in the 1993 experiments may not have represented the modules with sufficient accuracy. However no evidence was led to support the cross-examination. For his 1993 experiments Dr Davies decided to adjust the 1:33 model to improve its salient dynamic features and it was this model that he used for his work on the interaction between Module B and Module C. It was this model that he placed in his wind tunnel. His approach was to apply to Module C the air flow that would reproduce the ventilation flow he had found was applicable to that module. With that assuming the model gives an accurate representation of the aerodynamic features of the modules then the experiments should give valid values for both modules. It was suggested to him that heated equipment within the modules could influence air flow patterns. Dr Davies was also cross-examined on the question of how convective heat may affect air flows. His response was by reference to scientific literature to the effect that convective currents would have to be exceptionally strong to overcome the dominance of pressure differential and turbulence. Dr Davies was very experienced in the use of models for air flow experiments and he also made reference to the literature on the questions now being considered to back his views. In the absence of contradictory evidence I see no reason not to accept his judgment on these matters.

In his experiments Dr Davies sought to simulate both natural gas and condensate. For natural gas he used a mixture of carbon dioxide and helium to give him a neutrally buoyant gas. This had a molecular weight of about 18 to 20 kgs. For condensate vapour he used a mixture of argon and freon to give him a heavier than air gas approximate to propane. This had a molecular weight of about 42 to 44 kgs. In relation to the probes so far as affects the heavier gases the calibration of these probes was challenged by the defenders both in cross-examination and through their witness, Dr Bruun. To carry out his tests Dr Davies had manufactured devices which would produce gas releases both of the circumferential type and partial circumferential type. The amount of gas released could be controlled. These release mechanisms were improved for the 1993 experiments. He had various release points in his representation of Module C. Position 1 was in the general area of PSV 504, position 2 being the equivalent position opposite near the north wall, position 3 which is generally within the centrifugal skid area, and position 4 which is in the west of Module C. His thermal conductivity aspirating probes were placed in positions representing the positions of gas detectors in the module (G probes). He also had a number of additional probes (designated as B probes) and they were intended to test additional points in the module. The probes operated on the hot wire system. That is as air or gas passed over them the loss of heat would produce an output by way of an electrical signal. These probes would respond to gas quickly and to that extent they differ from detectors on the field which have response factors. The electrical signals transmitted by the probes are converted into a number which reflects it size. Those numbers are the essential output and the number relates to voltages. As far as Dr Davies is concerned his methodology was established and well tried. His 1989 results are set out in reports 12/362 A and B. He divided his tests into different series. A series would set out the applicable scenario as for example the fact that there was a release of a particular gas from a certain location and simulating certain conditions of release. Thus for example he carried out tests for a variety of release conditions such as jet leaks or circumferential leaks. For each series there were a number of runs. Thus in effect a series was a group of runs at the same conditions. He also sought to model the ingestion and exhaust of gas from the centrifugal compressors. He tested the various possibilities for neutrally buoyant gas leaks and for heavy gas leaks. From his data Dr Davies worked out the consequential traces manually.

In respect of the runs at position 2 Dr Davies released simulations of propane (series 43) and natural gas (series 46) respectively. In the first case the supposed rate of release was 37 kilograms per minute and the sequence of low level alarms he discovered to be C2; C4 ; C3 and C5; whereas the only high level alarms which would have responded was in C2. The release was designed to simulate a partially circumferential leak. He sought to measure the steady state concentrations as a basis for his results. Thus Dr Davies was able to measure the time to particular detectors in particular conditions and from that to know the sequence of alarms. Dr Davies was able to conclude that neither of the position 2 scenarios resembled the alarm patterns reported by Mr Bollands. In fact he becomes rather more positive than that for he asserts that in the light of his general experience he would not expect any scenario constructed round the position 2 release point to provide the necessary alarm sequence.

Position 3 was intended to represent a point on the centrifugal compressor B skid - that is a point just to the west of the compressor itself. Again he concludes that a leak from that point would not fit Mr Bollands’ gas alarm pattern. He likewise found that any leak from position 4, at the west of Module C would produce a well dispersed cloud of gas which would affect a number of alarms within the module and not in the pattern observed at the accident. On the other hand in relation to position 1 which seeks to replicates the site of PSV 504 he is able to conclude that with this site generally C3 was the first zone to detect gas when compared with C2; C4; and C5. However it is necessary for his results that the leak is generally downward-pointing and involves a heavier than air material like condensate. This is the circumstance that the defenders say was not proved. The need for a downward-pointing leak is said to be the fact that only such a leak of a heavier material could avoid setting off the G 101/1 detector which would be in the path of a eastward moving cloud from position 1. Thus subject to his qualification Dr Davies tends to favour the site at position 1 as being a likelier source of the leak than the other positions he explored. He found from his tests that looking to a release rate of about 90 kilograms per minute he could conclude that longer release times like 17 seconds were too short and that a value such as 13 seconds was appropriate to set off an alarm at C3. He accepted that if the release rate were 180 kilograms per minute then a puff release of about 6 seconds could trigger the C3 alarm. However one could have a higher rate for a shorter period of time or a lower rate for a longer period of time. He makes the point that there is a definite correlation between rate and time. To get to his result he of course requires a shrunken cloud of heavier gas. Indeed the defenders did not challenge that the foregoing results were appropriate if dealing with a propane cloud and they accepted that much of the force of their challenge would depend on their hypothesis that the first stage release would not in fact behave like a propane cloud. Moreover Dr Davies does accept that a neutral buoyancy release such as he had been considering would not miss a C2 alarm. For his opinion Dr Davies relies not only on his quantitative material but on his general experience. As he declares "moving the release position to anywhere else never produced C3 as the first alarm". However he acknowledges that although his results fit in well with a first alarm generally there should be a shorter time than 2 minutes before the second series of alarms. Senior Counsel for the pursuers sought to urge me to accept the view that the problem highlighted by Dr Davies would fit in well with a scenario where there had been a two-stage release. Indeed his view was that the pattern of alarms which occurred could only be explained if the first release was a relatively small release. By that he means a release of 2 to 4 kilograms per minute were the escape to be continuous. As we shall see there are certain difficulties with that view. On the other hand a puff type release of a restricted quantity of gas could cause the same effect. The gas cloud would require to be just enough to set-off the C3 alarm. However Dr Davies could not envisage circumstances where a continuous release of gas could account for the presumed alarm pattern. In his 1993 report Dr Davies gives further consideration to the possibility that there was a puff type release. The essence of a puff is that there is a sudden release and then it is curtailed. He explained that the key features of a puff are the release rate when it rises to its maximum and the duration of the release. Thus 75 kilograms of gas a minute at 17 seconds may produce the same sort of effect as 90 kilograms a minute at 13 seconds. If the release rate is put up then to produce the sort of cloud that would explain the alarm pattern one has to shorten the duration. In fact it was possible for Dr Davies to simulate a single C3 alarm with a puff type release from the locality of PSV 504. He concludes that for a release time of 15 to 20 seconds no low alarm would be triggered if the release rate was less than about 70 kilograms per minute but that one alarm would generally register for releases greater than 100 kilograms per minute. Whatever the correlation between release rate and time it would be necessary to generate a mass of propane type gas of about 20 to 25 kilograms to set off the C3 alarm. With some rates of release and times the critical mass is never reached because the ventilation has removed some of the gas from the module before that mass is reached. To consist with the alarm pattern the release has to be relatively quick. A first release of the kind necessary to create one low alarm at C3 would not generate a sufficient gas cloud to have caused the explosion. In Dr Davies’ view the quantity of gas necessary to generate an explosion would result from the second stage release of a two stage release. He also accepts that the dynamics of the situations he has been modelling are complex so that some care is needed.

Because of the phenomenon of pressure drop any escape of gas would not be at a steady rate. Thus Dr Davies was asked to calculate what would be the escape if the initial release rate was 30 kilograms a minute and after the first 9 seconds the release rate halved and continued to do so thereafter at the same rate. This is to take account of the fact that escape rates drop with time. The calculation brought out a total discharge of hydrocarbon of 6.33 kilograms which the defenders said would not produce a single C3 alarm. Even if the leak is continued for 36 seconds the release rate drops to 1.8 kilograms per minute. To arrive at a discharge of 20 kilograms the escape would have to last 10 minutes and no gas leave the module. The calculation of Dr Davies is set out in number 68/1 of process. The calculation above represents an eventual revisal by Dr Davies of his original opinion.

It has to be noted that in terms of the ventilation rate it would take about 12 to 13 seconds from any gas release for that gas to reach the compressor detectors and then of course perhaps 15 seconds for a high alarm to respond. Perhaps another 5 seconds or so would pass before the explosion. This means that if there is about 2 minutes between the first C3 alarm and the explosion only about 30 seconds was occupied in the time from release to explosion. Thus there must have been at least about one and a half minutes between the final jagging and the one that preceded it. It should perhaps also be noted that Dr Davies used his general engineering judgment to determine the proportion of total gas mass that would be flammable. In Dr Davies’ tests it was generally 103/1 that was the forerunner of the C3 alarms.

In relation to the first stage release the defenders spent some time developing their thesis that Dr Davies was wrong to suppose that the gas escaping at this stage would behave as a propane gas. The defenders suggested that Mr Clark’s flow chart shows that the stream coming off the JT Flash drum had a molecular weight of about 22 and they asked why the stream leaking from the pump would be any different. The condensate that goes into the injection pump comes from the JT Flash Drum. However a larger proportion of the lighter gas and in particular the methane will have flashed off in the Drum so that the material proceeding to the injection pump should have a higher proportion of the heavier ends. The defenders maintained, and in this they were correct, that with heavy hydrocarbon such as the crude oil (with a molecular weight of 38.7) there is still a capacity to flash off lighter ends such as methane. However the gas that flashes off in such circumstances would be slightly heavier than air. The defenders’ point is that one would expect the gas flashing off the condensate entering the pump during the first stage of jagging to have a molecular weight falling between that of the flash drum and the oil flow. Dr Drysdale and Dr Balfour indicated that the proportions of components representing 80% of the condensate flow in stream 360 and in relation to the stream as a whole he finds that it contains 19.6 % methane, 18.3% ethane and 30.7 % propane. Dr Balfour further indicates that if the condensate sustains a substantial release of pressure about 50% of the propane will flash off giving a percentage of 15.3% of the total. Of course flashing into a restricted space such as a pump chest is different to a situation where there can be unrestricted flashing but the defenders point out that Dr Richardson said that the lighter ends will flash off first as the condensate flows into the largely empty pump. The reduction in temperature will inhibit the heavier ends from flashing. The defenders claimed that if there is more methane than propane the mixture will be marginally lighter than air. It seems clear that the flash fraction of the condensate is roughly 50%. Dr Drysdale calculated that if the pressure in the pump was 20 bar the density of the gas that would flash off would be 20 kilograms per cubic metre. The evidence submitted the defenders is that the gas that flashed off in the pump would be approximately neutrally buoyant. Defenders’ Counsel maintained that the average molecular weight of the flashing mixture can be looked at but I was not sure that the gases would not stratify and although there was some evidence that the mixture will have certain characteristics the question of stratification was not specifically explored. Attractive as the defenders’ argument may appear at first sight the experts seemed quite content to work on the supposition that the leaked gas would be similar as a propane gas and they were not really asked if this could not be the case. The defenders construct their hypothesis from fairly general evidence.

It should be noted that Dr Davies seemed to concede that his "natural gas" represented a mixture of propane, methane and ethane. Dr Richardson it was contended gave evidence that it was only after a pressure in excess of 43 bar that liquid would escape because before that point there would be a gas plug which would permit further flashing. I certainly consider that at 43 bar it is eminently possible that liquid would emerge from any leak since one is at least about the area of sufficient pressure to form condensate. On the other hand it is doubtful if you would be beyond the gas bubble threshold at 40 bar. The argument was that one needs liquid to escape before escapes at a rate of 100 kilograms per minute or more are possible. It has to be noted that the flashing as the condensate enters the pump takes place in a fraction of a millisecond. Certainly Dr Davies thinks that if the first stage leak is neutrally buoyant G101/1 would be "a very strong candidate" but that view is expressed "unless the source is not diffuse in the way I am imagining it". That last remark may indicate a somewhat tentative quality in Dr Davies’ opinion. He also expected G101/3 to detect and indeed all the C2 detectors. However it is fair to say that Dr Davies says that if the gas escape is neutrally buoyant his expectation would be that the cloud would rise and hit G101/1 first. I think the defenders can derive some comfort from the fact that the pursuers seemed to accept that it was central to their assertion that the first alarm to go off was caused by a leak during the jagging process resulting in the first gas which triggered the alarm being heavier than air.

Dr Davies deals with the suggested second stage release and he extended his work on this aspect of the case in 1993. He deals with the additional work in his report 14/53. He indicated that to trigger the multiple alarms said to have occurred at this stage a increased leakage rate was necessary. Moreover such an increased rate would have been needed to generate the mass of gas required to cause the type of explosion which occurred. Indeed he considers that leakage rates of 100 kilograms per minute may have been necessary to develop a flammable mass in the range from 40 to 60 kilograms. If the time scale of the second release was about 30 seconds then a release of perhaps 140 to 200 kilograms per minute would have been necessary to cause a flammable cloud with a mass of 40 to 60 kilograms. He thought that to avoid a second high alarm requires a release duration of less than 55 seconds but to ensure a first high level alarm requires a duration of approximately 25 seconds. He concludes that these are the time bounds of the second stage and nobody challenged him on this.

Generally Dr Davies was of the view that there was no mechanism that would have resulted in a high level alarm from gas ingested into Module C from the east end. The detectors G27, G30, and G33 were located beside the turbine compartments of the Centrifugal compressors and these were devised so as to detect gas within the turbine compartments by way of a tube which went from the detector into the compartment. A question was raised by the defenders as to whether the said detectors would have been able to detect gas within Module C because of differential pressure. However Dr Davies did not think that these alarms were a significant factor in affecting his conclusions. Senior Counsel contended that alarms such as 103/1, 102/1 and 102/2 could have responded to gas from Module C.

Dr Davies considered that on the basis of his modelling work the probability of a significant inventory of gas accumulating in Module B without triggering a gas alarm was very low. That of course assumes that the alarms were operational. Thus I consider that if the alarms were working properly then apart from other considerations it is unlikely that Module B was the source of the gas that caused the explosion. That means that the evidence about the effectiveness of the detectors in Module B at the time of the accident is central to any decision about the possible role of Module B in the accident. Dr Davies did some work in 1993 on the topic of Module B this being contained in his Report number 65/1 of process (and the appendices number 65/3). In his experiments he set-up a plane of probes at the east face of Module B to replicate gas detectors. To explore the interaction between the modules he set-up a similar plane at the east face of Module C. These planes each consisted of 25 probes to give extensive coverage of the gas concentrations at different points. He set-up probes to replicate detectors G21 and G99/4 on the basis that any gas leaving Module B from the east was likely to be felt by these detectors because they were the two eastmost. In his experiments he simulated gas releases near the MOLs at the west end of B and also at two points towards the east of the module. He tested for both a natural gas and a heavy gas. His view is that if the release had been towards the west end dispersion would have been fairly rapid so that most of the detectors would have seen gas. If the release was towards the east of the module from Position 5 then G99/3 and G20 would receive the gas cloud, whereas the position was less sure in relation to G19. The release rate of gas from Position 5 was based on a supposed rate of 100 kilograms per minute. The pursuers argued that if a cloud of H2S sufficient to poison the detectors in Module B had been at the east face of B then it is likely that some of this cloud would have been ingested into the centrifugal compressors and also poisoned the detectors there. The principal consideration in Dr Davies’ analysis is that any gas originating in B will be substantially diluted by the time it can enter C. Thus where a gas leaving the east end of Module B has an average concentration of 1.63 % then the east face of C would show an average of 0.3 % for the same run. Another run, 107, involved a release from Position 5 at a rate of 100 kilograms per minute and a sucking mechanism was operated to simulate ingestion into Module C by the Compressors’ intakes. The average concentration at the east face of C was only slightly higher at 0.52%. Thus Dr Davies did not consider that the effect of operating the compressor exhausts was material. Dr Davies maintained that as gas moves through air the concentration of gas can only decay. In fact he found that gas decays at approximately the square of the distance from its source. However within a module the amount of decay may be limited by the restricted availability of air. He asserts that the concentration limit of gas equals the release rate of gas over ventilation rate. He reckoned that if a neutrally buoyant gas was released at the west end of Module B at the rate of 100 kilograms of gas per minute then the dilution rate by the time it reached the east end would be about 2.6 %. If the release rate was much smaller then correspondingly the dilution would be greater and the gas would not trigger an alarm at the east end of the module. He also says that a heavier gas would also mix and separate rapidly to become dilute. The gas cannot be taken to a higher concentration of gas by concentrating the mixture at one point. The difference in concentration from a gas leaving B and what would enter C is about one-fourth or even higher. Thus if gas coming from B were to set-off a high level alarm in C the original cloud in B would be well above the explosive limit. Dr Davies found that even a release rate of 150 kilograms per minute in Module B would not create a flammable cloud in C.

He did further work to look specifically into ingestion into the centrifugal compressors. For this purpose he set up his equipment so as to have probes replicating the air ventilation detectors (G26, G29, and G32). He also had probes representing the detectors at the combustion air intakes and positioned a probe to represent detector G101/3 in the C2 area directly above the B compressor. He placed a plane of twelve detectors just to the west of the centrifugal compressors. He found that the average ingested material at the plane of detectors was about 10% to 20% of what had emerged from Module B. Releases of heavier gas and of natural gas at the rate of 200 kilograms per minute and from a point to the north east of Module B showed no concentration at all at the plane of probes in Module C (series 109 and 110). This conclusion was related to the assumed flow patterns at the eastern exterior of the modules and was what Dr Davies considered to be what he would expect. In other experiments he simulates the ventilation that would come into Module C by the louvres at the end of the compressor compartments even assuming 100% gas was being exhausted. He tests on the basis of a flow of gas at the rate of 1000 kilograms per minute and measures the effect of final dilution as it is expelled through the louvres into Module C. Moreover he tests in relation to a neutrally buoyant gas since he considers that this is what would be encountered in the situation he is envisaging. Even in this worst case scenario no alarms trigger at C2. The explanation is that the gas emerging from the louvres is rapidly swept downstream to the east by the strong airflows between the compressors. Thus Dr Davies finally arrives at a view that he could not envisage a scenario that would result in gas generated in Module B setting off an alarm in C2 zone in C. The defenders led no expert evidence to a different effect.

Dr Davies also did work to test the hypothesis that gas originating in Module B might have been ingested in the ventilation or compressor intakes of the centrifugal compressor and thus set off alarms. Series 115 and 119 relate to releases from Position 6 in Module B at the rate of 1000 kilograms per minute. The detectors at the combustion air intakes of the compressors would see gas which would trigger both high and low level alarms. On the other hand the detectors at the ventilation air intakes would not see flammable levels of gas. Of course a heavier gas would sink and therefore be more affected by the combustion air intakes which are lower than the ventilation air intakes. The combustion air intake route does not provide a mechanism for getting gas into Module C. Moreover to have gas in Module C at the C2 detectors such high rates of release in B would be required that high level alarms in the combustion air intake would inevitably be triggered before the gas encountered C2.

It has to be noted that Mr Cubbage and Dr Mitcheson thought that the characteristics of the explosion pointed to it having been caused by a heavier than air cloud.

6.2.3 Dr Bruun’s Criticism

Of course the accuracy of the work of Dr Davies was attacked by Dr Bruun who challenged his calibrations of the probes used. It has to be appreciated that the fundamental methodology used by Dr Davies in his general experimental work is what is being attacked.

The first issue raised was whether the probes which were being employed by Dr Davies would respond to the gases used in his modelling in a linear manner or not. Dr Davies for his part described the relationship between the concentration fed into the probe and the voltage output by the probe as a straight-line relationship particularly for the heavy gas releases. Dr Davies when asked why he had been assuming linearity for all his calibrations stated that this was an empirical decision based on the work his group had been performing for years. He claimed that at the very outset of his work he investigated the suitability of a linear approach with the authors of the British Gas Team paper (Birch, Brown, Dodson, and Swaffield). That assumption had been used in their joint work during the 1980s. Other studies supported the same techniques. The proposition put to Dr Davies in cross-examination was that for heavy gases linearity of concentration cannot be assumed between concentrations of 0 and 100%. Dr Davies disagrees with that on the basis of the empirical experience of himself and other scientists over an extensive period. Moreover his decision to use linearity was practical for it simplifies his experiments. Dr Bruun on the other hand thought that Dr Davies in his calibrations for heavy gas had used a most unusual procedure. He said that in doing experiments one should never just assume a priori a linear relationship. Moreover instead of considering concentrations between 1 and 100% Dr Davies should have investigated the range he was actually using which was 0 to 5%. The first of these points is scarcely fair because Dr Davies’ approach is not that of a person with no experience of the problem arriving at an a priori solution but rather a reliance on personal and general experience on the matter. 0n the other hand Dr Bruun asserts specifically that there are people who have carried out the calibration of what he describes as "certain gases" and that the relationship is not linear. However Dr Bruun does not himself appear to have carried out work to demonstrate that the relationship between concentration of an argon/freon mixture is non linear nor does he refer to any research with that combination of gases. Indeed he declares that he himself is not an expert in doing measurements in gas concentrations. He refers to research by others into freon showing that a curved relationship develops rather than a straight one but the mixture used by Dr Davies was 82% argon which seems to develop in a straight line. Dr Bruun’s evidence on this matter appears to be speculative and academic. Dr Davies and others have used the linear approach in their work for many years apparently without noticing any distortion in their results. Even if Dr Bruun were correct that the introduction of freon may cause a degree of curve there was no evidence to show that this would materially affect results. I do not consider that the issue of linearity affects the utility of Dr Davies’ results.

The second criticism made by Dr Bruun concerns what is generally referred to as drift. This arises from the tendency of the voltage from a probe to vary with time. The system of calibration employed was that readings of voltages were taken of a mixture with zero gas and then a mixture of 100% gas and from such readings the calibration co-efficients were derived by computer process. The matter is complicated by the fact that the signal from the probes goes to a digital converter which converts the electrical signal into numbers. The risk is that after the calibration is completed voltage drift will occur and that should invalidate the results. To allow for drift Dr Davies would do a number of zero runs before the actual test runs. It seemed to be agreed by the experts that at 100% concentration the significance of drift is negligible. Originally Dr Davies had one calculation in cross-examination that appeared to show that even at 100% concentration drift would be a significant factor. However he repeated the calculations and found that they had been incorrectly done in cross-examination. This was because he had done his calculation by reference only to the lower co-efficient whereas the proper approach is to regard the co-efficients as being interdependent. Doing the calculation correctly confirmed that the 100% concentration was not materially affected by drift. Dr Davies considered that the question of drift at zero concentration had been properly dealt with by the zero runs. The degree of drift established by these runs can be off-set. Dr Bruun gave evidence that there was a better equation for calculating the effect of drift than that used by Dr Davies but I am handicapped in accepting that evidence because the matter of the alternative equation was not put to Dr Davies. In any event Dr Bruun eventually accepted that his equation and that of Dr Davies were mathematically essentially the same.

Dr Davies’ calibration tests are recorded in number 65/5 of process. It can be seen that six readings are taken for each probe and averaged. The two co-efficients are then calculated on the basis of the calibration exercise and fed into the computer. In the said calculation it is necessary to look at the interrelationship of the readings at zero concentration and 100% concentration rather than seek to calculate each co-efficient individually. When the computer is fed a voltage for that probe it uses the co-efficients to determine what the voltage represents by way of a gas concentration. Dr Davies gave evidence that in relation to his experience with probes he would expect his methods to produce accurate results to within a tolerance of about 0.1 %. The suggestion derived from the calculations that were put to Dr Davies in cross-examination that the probes were susceptible to considerable error because of voltage drift can be explained by the fact (later corrected) that in his first attempt at the exercise there was an inadvertent failure to take account of the 100% co-efficient. This is perhaps explained by the fact that the calculations he was asked to do in Court are quite complex. The result of the recalculated figures was that the observed voltage variation produced a variation in concentration between .49% and .47%. Such a variation would be within the tolerances that Dr Davies would expect and be prepared to accept. At the end of the day when he gave his evidence Dr Bruun did not challenge Dr Davies’ revised way of doing his calculations. In any event it would only be the variation of voltage in relation to the lower co-efficient that would have to be watched since mathematically the upper coefficient is relatively insensitive to voltage changes. This is because the range of concentrations being tested were low values of concentration (about 0.5 %) so that changes at the upper end of the linear scale do no affect the results so drastically as those at the lower end of the scale.

Dr Bruun for his part claimed originally that a calibration that only took account of variations of the zero concentrations was deficient. He claimed that the equation set out in 96/101 of process was the proper and usual one to apply to hot wire calibrations. Dr Bruun’s equation begins by placing the emphasis on deriving a value for voltages whereas Dr Davies concentrates on deriving the gas concentration. However, as I have said, in cross-examination Dr Bruun seems to agree that the equations in mathematical terms are identical and thus should produce the same results albeit through different routes. A comparison can be made through numbers 44/145 and 96/101 of process. The equations both express the same relationship between concentration and voltage but one is the inverse of the other. But even Dr Bruun’s calculations show that the effect of drift towards the 100% values is negligible when the experimental situation is dealing with a low concentration of gas. Dr Bruun was probably more of a theoretician than Dr Davies but although his equation may have been theoretically more attractive than that of Dr Davies it emerged that after he had used it on the actual data of the test probes the results obtained were not materially different from those of Dr Davies. Dr Bruun’s final position was that although Dr Davies did not use the best method his methods were sufficient to produce results that can be taken as approximately correct. One difference between the two experts was that Dr Davies calculates his offset by reference to his original co-efficients whereas Dr Bruun declares that the preferable method is to calculate new co-efficients before introducing the off-set. I would have difficulty in deciding which of the two possible methods is theoretically preferable but ultimately Senior Counsel for the pursuers conceded that Dr Bruun’s method may in fact be more correct. On the other hand given that it seems to be conceded that any deficiency in Dr Davies’ method would produce a very small margin of error and that his method has been used without difficulty by a number of experimenters over a period of years then I consider that the difference in methodology on this matter is of little importance. This particular question of methodology was not put to Dr Davies in cross-examination and had that been done he may have been able to produce practical reasons for the adoption of his own rather less precise methods.

Dr Bruun also questioned the way that Dr Davies had used the analogue to digital converter. The system which Dr Davies used involved the electrical signal from the probes going to a converter where it was translated into digital numbers. The converter used had a total range of 10 volts that is from plus 5 volts to minus 5 volts. This allowed for a range of 4,096 voltage steps. Each of these steps was equivalent to an input voltage of 0.0024 volts. The output generated by the probes varied generally between minus 1 volt at 0% and minus 0.5 at 100%. Thus in terms of the potential range of the A to D converter Dr Davies was using output from the probes amounting to some 205 voltage steps. The concentration of 0.5 which Dr Davies was using was equivalent to 0.0025 volts which in turn is almost equivalent to one voltage step. It follows therefore that one voltage step almost represents the difference between no gas and a critical concentration. Thus it was suggested to Dr Davies that it was not possible to get proper measurement of low concentrations of gas. Dr Bruun suggested that the proper measurement approach was to apply gain - that is to magnify the signal being sent to the A to D converter - so that more refined measurements could thus be taken. Under Dr Davies’ system it was claimed there was a complete step from zero to 0.5 % concentration and nothing in between could be measured. However Dr Davies maintained that this problem can be overcome by what he referred to as the smoothing process. This was in fact a measurement averaging process. In the course of one second of model use 21 voltage readings are received by the A to D converter. Thus every twentieth of a second or so there is a digit reading. What Dr Davies did was have the computer look at the twenty or so digital readings per second and average them together. This process gave what he called a smoothed concentration value. Averages are obtainable because there are inevitably peaks and troughs in the signals. Thus the difference between the experts is that Dr Davies was content to take his readings on the basis of averages whereas Dr Bruun would seek to take actual readings. One justification of Dr Davies’ method is that looking to the detectors they would never see more than what in effect is the average because their response time would not permit measurement of the very short measurement times issued by the converter. The averaging that is obtained in the smoothing process is essentially a sliding average obtained on a continuous basis. It should be noted that with regard to the detectors one requires to have the critical concentration of gas present over the response period to get the alarm triggered. Dr Bruun says at one point that "when you do scientific experiments the whole objective is to have a true signal". However it is clear in my view that Dr Davies was not concerned with measurements of absolute scientific accuracy. Indeed that kind of accuracy was probably not available over the whole scheme of his tests which inevitably had to depend on a degree of judgment and simplification. The question really is did Dr Davies apply methods that would give reasonable results looking to the practical questions the experiments were designed to answer. Indeed Dr Bruun acknowledged that he was not an expert in the actual taking of measurements. Dr Davies was a practical modeller with years of experience in modelling and I have no reason to conclude that his methods were faulty in any practical sense.

As has been seen Dr Davies is basically concerned with the likely behaviour of gas clouds within Module C and to a lesser extent Module B. As he put it he starts after the gas is out. The nature and characteristics of any prospective leak from PSV 504 was dealt with in the evidence that pivoted rounds the evidence of Dr R