Explainable artificial intelligence to improve quality control in continuous industrial processes

This work presents a data-driven methodology to enhance quality control in continuous industrial processes by integrating Machine Learning (ML) models with Explainable Artificial Intelligence (XAI) techniques. The approach addresses the challenges of Industry 4.0 in environments where physical models are not available, and quality indices cannot be measured in real time. The proposal elaborates on the CRoss Industry Standard Process for Data Mining (CRISP-DM) methodology to provide a domain independent method to develop predictive models, which can be used to provide estimates for those indices in real time. The approach implements anomaly detection through residual analysis, and SHapley Additive exPlanations (SHAP) based local explanations are used to identify the most influential variables behind those predictions, that can be used to support the decision process performed by plant managers. The proposal has been tested using real data from a Medium-Density Fiberboard (MDF) factory, building predictive models to estimate thickness swelling from process variables. Several ML algorithms were evaluated, being eXtreme Gradient Boosting (XGBoost) the most successful one. The proposed system was validated in a real production setting, demonstrating its ability to anticipate quality issues and support decision-making obtaining an accuracy of 96.7%, with 75% precision, 1.8% False Positive Rate, and 25% False Negative Rate.