Hazard and operability (HAZOP) analysis is a systematic way of identifying every possible process deviation, propagational, and adverse hazardous consequences in a chemical plant. This is a complex, time-consuming, and labour-intensive process with multiple levels of brainstorming that can benefit from automating and AI assistance. Though the idea of automating HAZOP analysis is not new [1,2,3], we propose a novel simulator framework that combines the ontology-based symbolic knowledge and the data-based attention approach to create a hybrid HAZOP graph. Data-based brute-force techniques have a significant downfall regarding explainability and trustworthiness, making these hybrid models work and reason better for precious data problems in chemical engineering. Using an ontology with a well-defined knowledge base with process-specific and process-generic frameworks for failures, causes, and propagation, a HAZOP graph with a property vector for each node corresponding to its failures is introduced. A complete simulation is performed by disturbing these property vectors while the attention mechanism predicts the hazard in the subsequent nodes, mimicking the hazard flow as a continuous and discrete process. Using the simulation results, the likelihoods of failures are quantified and ranked for better analysis.
Venkatasubramanian, V. and Vaidhyanathan, R., “A Knowledge-Based Framework for Automating HAZOP Analysis”, AIChE Journal, 40 (3), 1994, pp. 496 – 505.
Srinivasan, R. and Venkatasubramanian, V., “Automating HAZOP Analysis of Batch Chemical Plants: Part I. The Knowledge Representation Framework”, Computers and Chemical Engineering, 22 (9), 1998, pp. 1345 – 1355.
Zhao, C., Bushan, M. and Venkatasubramanian, V., “PHASuite: An Automated HAZOP Analysis Tool for Chemical Processes Part I Knowledge Engineering Framework”, Process Safety and Environmental Protection, Trans IChemE Part B, 83(B6), 2005.
