Digital Twins and Artificial Intelligence for HAZOP Enhancement in Process Safety: A Critical Literature Review

Conventional hazard and operability (HAZOP) studies remain central to process safety management, but their periodic, document-centered implementation and dependence on expert judgment limit their ability to track dynamic operational risk across the full plant lifecycle. This critical literature review, drawing on 113 sources, examines how digital twin (DT) and artificial intelligence (AI) technologies may augment HAZOP practice, advancing an augmentation principle: DT–AI should strengthen expert-led hazard analysis as a decision-support layer, not replace human judgment. The review maps conventional HAZOP limitations to four complementary technology pathways: AI-assisted knowledge capture and deviation reasoning; DT-based hazard monitoring and early warning; hybrid physics–data modeling for predictive safety; and explainable AI for operator trust and regulatory acceptance. Four representative industrial case studies illustrate current practice, and a structured barrier analysis identifies why many pilots stall before reaching sustained deployment. These pathways offer strong potential to improve HAZOP coverage, traceability, operational relevance, and lifecycle learning, but a study-level evidence assessment finds that no reviewed study demonstrates long-term, multi-site validation of an integrated DT–AI–HAZOP system; most implementations remain at conceptual, prototype, or limited pilot levels. Key failure modes, including model drift, sensor faults, large language model hallucination, and automation complacency, are examined alongside a thirteen-item implementation risk register spanning technical, organizational, economic, security, and scalability concerns. The review proposes a phased industrial implementation roadmap, a lifecycle-oriented safety-management framework, and a three-horizon research agenda for advancing toward reliable, validated HAZOP-informed digital twin systems.