FAULT TREE GENERATION FROM P&IDS USING DEEP LEARNING

Fault tree analysis is an effective approach for evaluating and improving the reliability and safety of industrial systems. However, building fault trees for modern industrial systems is challenging because of the complexity of the target system and the requirement high-quality experts be involved in the fault tree construction. This research proposes a computer-aided automation approach for fault tree generation and analysis based on Piping and Instrumentation diagrams (P&IDs). The proposed approach utilizes system information (e.g., system structures, component types, and failure modes) and qualitative physical laws (i.e., the trends of the impacts caused by faults) to infer the causalities of events during system operation automatically and to further build the fault tree for reliability and safety analysis. This research initially leverages the power of object detection algorithms applied to P&IDs to identify the components and their relationships. The relationships provide necessary information about the topology of the target system or its subsystems (series, parallel, feedforward loops, feedback loops) and the type of system components (pipes, valves, etc.). Data structures such as digraphs to model this information and the related process variables make Fault Tree generation possible. Digraph-based Fault Tree generating algorithms are deployed, which iteratively consider the component and related process variables to account for the top event occurrence. The fault tree generation is based on the approach where information about the components, the effect on the process variables, and failure modes is stored in the form of a table named equipment models. The digraph is then traced from the top event to generate the fault tree. A consistency checking algorithm is developed to avoid contradictions that may occur during the fault tree construction. Also, the generated fault tree is simplified by applying Boolean logic laws to the fault tree structure. In this paper, a simplified feedwater control system of the generic pressurized water reactor (GPWR) is utilized as the case study to verify the proposed method.