Explainable AI for CPS-Based Manufacturing Workcell

This paper discusses the application of Explainable AI (XAI) for a Cyber-Physical System (CPS)-based manufacturing workcell. XAI involves providing interpretable outputs from deep learning models. CPSs are systems that integrate interacting analogue, physical and human inputs, and have applications in control optimisation, automation and process intelligence. Current state of the art and gaps in XAI and CPSs are introduced. The need for CPSs with in-built logic to correct and reduce dependency on conventional mathematical modelling of physical systems is identified, together with how the in-built logic must be explainable and understandable to build trust in the logic. This is especially true for applications where manufacturing accuracy and precision are critical. Whilst there have been advances in both XAI and CPSs in recent years, little progress has been made in integrating the two. We show that XAI can increase the efficiency and reliability of CPS-based manufacturing. XAI in CPS can ensure that manufacturing validation inputs made by deep learning networks within the CPS are understandable and correct. It thus ensures that the CPS outputs have actions that are explainable and logic dependent, with safeguards against malicious intent. This contrasts against conventional CPS which may have learning-based systems with ‘black box’ outputs that are assumed to be correct. Approaches towards XAI and how they can be integrated into a learning-based machine vision CPS manufacturing workcell are described. An analysis of the effectiveness of these approaches and their feasibility of implementation in such a CPS is performed. A recommendation of what is the most promising approach towards implementing XAI in the CPS manufacturing of a mass-market, specialised and personalised product, is made as a conclusion.