Webinar recap: functional safety and artificial intelligence

This seminar, on “Functional safety and artificial intelligence” was given by Audrey Canning of Virkonnen Ltd, who has more than 35 years’ experience in Functional Safety (and before that some 6 years’ experience in development of digital systems and 2 years’ experience in development of AI-based systems).

Audrey’s interesting presentation drew on her experience as the Convener of the software engineering aspects of IEC 61508, the worldwide representative for functional safety on the IEC ACOS committee reporting to the IEC Standards Management Board and the IEC/SC65A liaison member to the Joint Working Group between IEC and ISO, JTC1/SC42.

The seminar noted that the current edition of IEC 61508 (more than ten years old) says that use of ‘Artificial Intelligence’ (AI) and ‘Dynamic Reconfiguration’ is ‘Not Recommended’ for any significant level of application safety.

Since the last publication of IEC 61508, the ability to derive effective algorithms through the use of AI techniques has increased dramatically, to the extent that it is routinely used in financial analysis, social media, image recognition and other data-intensive applications.

Inevitably, it is starting to be used to optimise control/health monitoring of safety-critical applications, such as manufacturing, machine health monitoring and autonomous vehicles.

Recognising the inevitability of the use of AI-based technology in safety systems, in April 2020 the IEC Committee SC65A and the ISO Committee SC42 agreed to collaborate on developing a Technical Report on AI and Functional Safety.

Audrey outlined some of the challenges of applying AI to safety systems and then gave an overview of the current state of the ISO/IEC’s Joint Technical Report.

This draft report contains a Classification Framework covering different AI technologies and usages and also includes three theoretical case studies where AI is used in the Safety Loop, used to control equipment but protected by a non-AI function and finally used for machine health monitoring.

The seminar generated a wide range of questions from the audience that explored the implications of the use of AI in safety systems.

The seminar and discussion, and feedback from the webinar participants elicited through the associated questionnaire, will inform the current work of the ISA Working Group on what represents sufficient rigour in safety cases and other, future, guidance material.