Methodology for Assessing Learning from Incidents - a Process Industry Perspective

Learning from incidents (from minor disturbances to major accidents) is considered one of the most important ways in the process industry to learn from experiences for improving safety performance. Most companies have a formal incident learning system and considerable resources are used for reporting incidents and for utilising these experiences to prevent future incidents. However, there are many indications that only a portion of the entire potential for learning from reported incidents is actually utilised. In order to know whether this apprehension is correct and to be able to manage and improve the situation, if necessary, one would need to have a methodology for assessing the learning in such systems. Therefore, the author has undertaken to develop a methodology for this, which is presented in the thesis. The methodology could be used on a wide variety of incident learning systems and by people in the process industry, in authorities and by researchers, with the general aim to improve learning from incidents. Several aspects of learning need to be included in a methodology for a comprehensive assessment of how effectively the learning from incidents works. One has to be able to address the following types of issues: 1. Do we handle the incidents reported in our incident learning system properly? Do the various steps in the learning cycle work effectively? 2. How much do we learn from the incidents which are reported? How does this learning compare with what could potentially have been extracted? What level of learning are we at and what level could we have achieved? 3. Do we report the incidents that are worth reporting (that have a learning potential)? What is the threshold for reporting? How big is the number of unreported cases, the “hidden number”? In order to address issue 1, a method has been developed which assesses the effectiveness of learning in every step of the learning cycle (Reporting – Analysis – Decision – Implementation – Follow-up) for each incident, and of the aggregated material of many incidents. The method contains a tool for each step, built on a number of dimensions which in turn contain a number of aspects. By using a rating system including a scale with formulated requirements for some levels, the effectiveness of each step can be assessed numerically for each individual incident. For issue 2, a method has also been developed that builds on classifying the learning product, the measures taken, in different levels depending on how well the experiences from an incident are handled. The basis for classifying an incident is the geographical application, the degree of organisational learning, and the duration of the measures taken. Incidents are classified both in actual levels of learning based on the measures taken, and in potential levels of learning, indicating the level that could have been achieved if all the potential for learning had been utilised. The relation (the ratio) between actual and potential levels of learning is a measure of the effectiveness of the learning. A specific method for evaluating the underlying causation has been developed to draw conclusions about the potential learning. For issue 3, a tool has been developed for assessing the threshold for reporting as well as guidelines for what can be considered reasonable frequencies of incident reporting in the process industry. In addition to providing information about how efficient the reporting of incidents is, this will also provide input to the method for issue 2. Together, the methods with their tools and guidelines constitute a methodology, which allows the user to make a total assessment of the effectiveness of the learning from incidents in process industry companies. The empirical material for the research was taken from the incident databases of six Swedish process industry companies, and from the EC MARS database for major accidents in enterprises which fall under the Seveso legislation. The results from the application of the methods have proven that learning from incidents is often limited, especially in relation to what would have been possible to achieve. Effectiveness in the learning cycle is often relatively poor, especially in the analysis and the follow-up steps. The results from the assessment of the learning effectiveness combined with the results from safety audits often offer valuable insight into the decisive factors for good learning