April 2000 Lead Story

Failure and Progress

by Peter Carr

I trained as a structural engineer. I have always been interested in structural failures. Reading about past structural disasters, I noticed that many had been associated with innovation and inability to appreciate the significance of new failure modes. Famous examples include the Tay Bridge (1879), which collapsed because it had not been designed for wind load, the Tacoma Narrows Bridge (1940) which was destroyed by wind-induced vibration and the Ronan Point Tower Block (1968) which broke up following a gas explosion due to lack of structural continuity. After each failure, revisions were made to engineering methods to prevent repetition and the techniques that had been innovatory became conventional. After years of safe construction, there would be other technological developments followed by other failures. This seemed an undesirable way to proceed. Was it impossible for engineers to foresee failures and the causes of failure? Or, could we start anticipating failures and taking measures to prevent them?

In the course of my working life, I have watched the slow emergence of probabilistic methods and of risk analysis. These subjects were never mentioned by teachers during my undergraduate or postgraduate studies in the late 60's and early 70's. When I was first aware of them, they seemed topics of interest only to a handful of academics.

My first encounter with probabilistic methods came in the mid 70's, when I came across the statistical methods of quality control that were being used in the more advanced manufacturing industries. I was impressed by the use of mathematically rigorous procedures to quantify and balance supplier's risk and customer's risk. I learnt that probabilistic methods dealt in trade-offs, the balancing of conflicting objectives or conflicting interests. Subsequently, I became aware of increasing numbers of other problems also involving trade-offs.

I realised that decision-making could be assisted by mathematical methods.

My first opportunity to apply probabilistic methods to a major problem did not come until 1984. An operator of gas production platforms in the North Sea was faced with large annual maintenance bills for the inspection of subsea welded joints in the platform structures. These inspections were performed because of the possibility of fatigue cracking - the cracking that results because the platforms are constantly moving in response to wave, current and wind actions. Small initial imperfections in the welded steel joints can grow to become large cracks and ultimately joints can become completely severed and threaten the integrity of the whole platform. The task was to develop a rationale for which joints should be inspected, how often and by what methods. It was also required to identify the appropriate responses to defects detected during inspection.

Attempts were initially made to produce the required rationale by gathering a number of engineers together to brainstorm the issue. These attempts failed, with different people expressing incompatible views that could not be reconciled. I had just acquired my first personal computer and had become aware of the technique of Monte Carlo simulation, a method for the numerical analysis of probabilistic questions that is both rigorous and easy to program. A Monte Carlo simulation is essentially a process of conducting experiments on a mathematical model of the situation that is under investigation. The computer's random number generator is used to introduce a degree of uncertainty into the mathematical model, so that no two analyses will give exactly the same outcome. The model is analyzed over and over again (maybe many millions of times) and the results are studied and post-processed just as if they were results from physical experiments. Therefore in the hope of developing a less subjective soultion to the client's problem, I developed a Monte Carlo simulation of the whole process of fatigue cracking, diver inspection, the probability that defects were found or not found, the remedial responses that might be made when cracks were detected, and the significance of cracks to overall platform integrity.

The simulation showed that there were a few joints in the platforms being studied that needed inspection by sensitive and costly techniques, but that the majority of the joints were less critical, and an annual diver swimround inspection was sufficient for these less critical joints. The simulation also embodied a procedure for using the results of each inspection to reduce the uncertainties in the model, with the result that a rational basis could be given for relaxing inspection requirements as time went on provided that all inspection results continued to be favorable. The simulation identified the joints that should be inspected, the appropriate inspection methods and the optimum inspection intervals. The benefits to the operator were that it was found possible to reduce the inspection effort but to better target the inspections to the most critical joints so that the predicted safety levels were actually improved.

Shortly afterwards, I was involved with the design of a number of new platforms in the North Sea. For these platforms, explicit consideration was given to providing structural redundancy so that no joint would be critical to overall integrity of the platform. In addition, joints were designed to achieve low cyclic stress levels and hence low rates of fatigue crack growth. The Monte Carlo simulation model was then applied to show that the new platforms needed very modest inspection requirements, thus reducing the lifetime cost of ownership.

Later on I used Monte Carlo simulation techniques time and again to help understand behavior of engineering systems and to assist decision-making.

In 1988 came the Piper Alpha disaster. An initial explosion led to an escalating fire that destroyed the platform with the loss of 165 lives. This single accident led to a revolution in offshore safety in the UK and across the world.
Within days of the Piper Alpha accident, every operator in the North Sea had set about reviewing the hazards to their own platforms and the Government had appointed Lord Cullen to conduct a public inquiry.

During the two years after Piper, I was one of the many people evaluating risks to other North Sea platforms.

Lord Cullen reported his findings in 1990. The platform had design deficiencies and the operator, Occidental, had had no effective safety management system. Cullen was also critical of the regulatory regime which was highly prescriptive and tended to impose 'solutions' rather than 'objectives'. As a result, compliance with the letter of the law had taken precedence over wider safety considerations. During the 1990's, all the UK offshore regulations were progressively revoked and replaced with new goal-setting regulations that specified only objectives and not methods.

The three key features of the new approach were:

• hazard identification;
• risk analysis;
• formal demonstration that major risks had been reduced 'to as low as is reasonably practicable' (ALARP).

The point at which risks have been reduced to ALARP is essentially to be determined on cost-benefit considerations using a principle that had been defined by a UK court as long ago as 1949 in the case of Edwards Vs National Coal Board. The following summary of this case is taken from the Health & Safety Executive's discussion document 'Reducing Risks, Protecting People' issued in 1999:

"This case established that a computation must be made in which the quantum of risk is placed on one scale and the sacrifice, whether in money, time or trouble, involved in the measures necessary to avert the risk is placed in the other; and that, if it be shown that there is a gross disproportion between them, the risk being insignificant in relation to the sacrifice, the person on whom the duty is laid discharges the burden of proving that compliance was not reasonably practicable."

Thus, risks are ALARP when the cost of reducing them further is grossly disproportionate to the further benefits that can be achieved.

During the next few months, it is planned to add articles to this website to explore in details the topics of risk and risk management. While some of these articles will be for the general reader, it is also intended to present useful data, algorithms and downloadable software for the risk specialist.