XAI for industrial coating processes in the era of Industry 5.0

Coating processes like Air Plasma Spray (APS) and High Velocity Oxi Fuel (HVOF) are increasingly used to protect parts from wear, high temperatures and corrosion. Determining the values of coating parameters (the recipe) to obtain products that comply with customers’ requirements is difficult, time and resources intensive. In this work we study an optimization technique to find optimal parameters of coating processes based on Bayesian Networks (BN). We employ Bayesian Networks to perform coating recipe optimization because they can integrate human experts’ knowledge with data into Artificial Intelligence algorithms. The probabilistic graphical model of BN can explain the results of optimization to users resulting in an explainable artificial intelligence (XAI) approach. XAI can be very useful in manufacturing processes optimization (like coating) for knowledge transfer/enhancing and mitigating the impact of experienced workers quitting the job. We perform several experiments involving spraying samples with different values of coating parameters. We use data gathered from these experiments to make the dataset for BN structure and parameters learning. To enhance the impact of this optimization technique on production processes we build a support decision system that gather data from coating processes, store in a data layer, perform the optimization and visualize results in an explanation interface. The approach used in this work enables a human-machine cooperation that contributes to processes and workers resilience, efficiency and skills improvements as stated in the human-centric pillar of Industry 5.0.