Guidelines on Assessment of AI Reliability in the Field of Plant Safety
These guidelines present a systematic methodology for assessing and managing the reliability of machine learning components used in the field of plant safety. They adopt a hierarchical quality assurance framework comprising quality in use, external quality, and internal quality, and provide specific use cases, evaluation procedures, and implementation checklists for plant owner and vendor companies.
Summary and content fields are drawn from the document text and are pending editorial review.
Who issued it
Who it governs
Plant owner companies, AI vendors, and safety engineers should read this to properly assess, implement, and account for machine learning reliability in industrial plant environments.
What the document contains
The document addresses the assessment and management of machine learning reliability in petroleum and chemical plants, defining quality hierarchies, risk avoidance axes, performance axes, internal quality checklists, and five operational use cases.
The guidelines do not cover general hardware reliability, software reliability outside of machine learning components, legal or privacy liabilities, or cybersecurity and external cyber attacks.
- The guidelines establish a hierarchical reliability assessment framework for machine learning components in plant safety, dividing quality into use, external, and internal levels.
- Safety in plant operations must be maintained not solely through AI components, but through a combination of AI, external safety mechanisms, independent safety systems, and human oversight.
- The document provides five concrete use cases for maintenance and operation, including pipe wall thickness prediction, pipeline image diagnosis, equipment deterioration diagnosis, anomaly detection, and operation optimization.
- Internal quality requirements span data quality design, data collection models, and operation methods, tailored with specific perspectives for plant safety.
· 1. Introduction · 2. Reliability Assessment Structure of Machine Learning in the Field of Plant Safety · 3. Use Cases of Machine Learning in the Field of Plant Safety · 4. Flow of utilizing the Guidelines
How it scores
Each score is backed by a quote verified against the document text. Editorial review has not yet confirmed them.
Accountability
4/5Clear responsibility for AI outcomes
The responsibility of explaining to the regulatory authorities whether ML compoment development process is compliant to the Guidelines lies in the plant owner companies, should such explanation be necessary.
1.3 Scope of Application
screening score · 2026-10-23
Fairness
0/5Non-discrimination and equitable treatment
screening score · 2026-10-23
Human Dignity
0/5Respect for the inherent worth of persons
screening score · 2026-10-23
Privacy
0/5Protection of personal data and private life
screening score · 2026-10-23
Transparency
3/5Openness about how AI systems work and are used
A vendor company that develops and delivers AI can more easily explain the reliability of AI to the plant owner company because the Guidelines clarify how to build AI with sufficient reliability.
Executive Summary
screening score · 2026-10-23
Enforcement
1/5Consequences and compliance mechanisms with real force
By recording the results of assessment according to the template, evidence to which the Guidelines are applied can be kept and used for explanations within and without the company.
Executive Summary
screening score · 2026-10-23
Impact Assessment
4/5Required assessment of an AI system's risks/impacts before or during use
The basic procedure for reliability assessment is as follows (Figure 1). (1) Set the "quality in use" (See ch. 2.2.1). (2) Set the "external quality" required to achieve "quality in use" and determine the level of achievement
Executive Summary
screening score · 2026-10-23