End-to-end AI-assisted generation of HAZOP study reports in the process industries

This paper presents an end-to-end AI-assisted framework for the automated generation of hazard and operability (HAZOP) reports to address key challenges of current HAZOP practices, including time consumption, human bias, and knowledge loss. The framework consists of four modules, each employing a set of artificial intelligence (AI) models to support the generation of HAZOP notes, including deviations, causes, consequences, safeguards, and recommendations. The First module assesses the degree of similarity (expressed in 3 levels: high, medium, or low) between current and past piping and instrumentation diagram (P&ID) nodes using object, location, and functional similarity metrics. Object and location similarities are computed using the detected classes and bounding boxes coordinates obtained from YOLO models, while functional similarity is derived from node descriptions generated by large language models (LLMs). These metrics are then combined to compute the P&ID Node Similarity Index (PNSI), which determines the similarity level to guide subsequent analysis. For the high level, the associated HAZOP notes are directly retrieved and reused. For the medium level, the second module employs BERTopic to identify dominant topics from retrieved HAZOP notes. For the low level, the third module employs retrieval-augmented generation (RAG) and domain-adapted embeddings to enhance the reliability of generated HAZOP notes. The three modules, referred to above, are integrated into a Streamlit-based web interface to facilitate their respective applications to automatically generate HAZOP reports. The framework is tested primarily using a dataset of 510P&ID node images, 6,120 HAZOP notes and 1,140 incident investigation reports collected from diverse sources. The P&ID nodes were annotated using Roboflow , resulting in 13,780 labeled objects across 21 classes. The YOLOv8x model achieved 82.4% precision, 74.8% recall, and 79.2% mAP@50. Three P&ID nodes were used to illustrate how the modules interact under different similarity conditions. The results demonstrate the effectiveness to benefit from past HAZOP knowledge, thereby capturing the embedded experience and reducing the time required to perform HAZOP studies and prepare related reports.

Authors

Institutions

Publication Details

Journal
Safety Science
Published
2026-11-20
DOI
https://doi.org/10.1016/j.ssci.2026.107585
Primary Topic
Risk and Safety Analysis
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

End-to-end AI-assisted generation of HAZOP study reports in the process industries

Ehab Elhosary, Osama Moselhi
2 links
Safety Science
Risk and Safety Analysis
article

End-to-end AI-assisted generation of HAZOP study reports in the process industries

Ehab Elhosary, Osama Moselhi
article en

Abstract

This paper presents an end-to-end AI-assisted framework for the automated generation of hazard and operability (HAZOP) reports to address key challenges of current HAZOP practices, including time consumption, human bias, and knowledge loss. The framework consists of four modules, each employing a set of artificial intelligence (AI) models to support the generation of HAZOP notes, including deviations, causes, consequences, safeguards, and recommendations. The First module assesses the degree of similarity (expressed in 3 levels: high, medium, or low) between current and past piping and instrumentation diagram (P&ID) nodes using object, location, and functional similarity metrics. Object and location similarities are computed using the detected classes and bounding boxes coordinates obtained from YOLO models, while functional similarity is derived from node descriptions generated by large language models (LLMs). These metrics are then combined to compute the P&ID Node Similarity Index (PNSI), which determines the similarity level to guide subsequent analysis. For the high level, the associated HAZOP notes are directly retrieved and reused. For the medium level, the second module employs BERTopic to identify dominant topics from retrieved HAZOP notes. For the low level, the third module employs retrieval-augmented generation (RAG) and domain-adapted embeddings to enhance the reliability of generated HAZOP notes. The three modules, referred to above, are integrated into a Streamlit-based web interface to facilitate their respective applications to automatically generate HAZOP reports. The framework is tested primarily using a dataset of 510P&ID node images, 6,120 HAZOP notes and 1,140 incident investigation reports collected from diverse sources. The P&ID nodes were annotated using Roboflow , resulting in 13,780 labeled objects across 21 classes. The YOLOv8x model achieved 82.4% precision, 74.8% recall, and 79.2% mAP@50. Three P&ID nodes were used to illustrate how the modules interact under different similarity conditions. The results demonstrate the effectiveness to benefit from past HAZOP knowledge, thereby capturing the embedded experience and reducing the time required to perform HAZOP studies and prepare related reports.

Safety ScienceVol. 207
Tanta University (EG), Concordia University (CA)
Mitacs
Industry, innovation and infrastructure
Risk and Safety Analysis
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon