XAI FOR INDUSTRIAL PROCESS MONITORING: APPLYING SHAP TO DETECT ANOMALIES IN CHEMICAL MANUFACTURING DATA

Abstract

Industrial process monitoring is essential for maintaining operational safety, product quality, equipment reliability, and overall efficiency in chemical manufacturing environments. The growing adoption of Industry 4.0 technologies has enabled the collection of large volumes of real-time process data, creating opportunities for machine learning (ML)-based anomaly detection. While advanced ML models can accurately identify abnormal operating conditions, their black-box nature often limits transparency and trust among engineers and plant operators. To address this challenge, this research proposes an Explainable Artificial Intelligence (XAI)-driven anomaly detection framework that integrates machine learning with the Shapely Additive explanations (SHAP) method. The proposed framework uses SHAP values to explain model predictions by quantifying the contribution of individual process variables to detected anomalies. Experimental evaluation is performed using a benchmark chemical manufacturing dataset, where data preprocessing, feature engineering, and supervised learning techniques are employed to develop a robust anomaly detection model. Results demonstrate high detection performance in terms of accuracy, precision, recall, and F1-score while simultaneously providing meaningful explanations for abnormal process behavior. SHAP analysis identifies critical variables such as reactor temperature, reactor pressure, feed flow rate, and cooling system parameters as major contributors to anomaly predictions. Furthermore, the generated explanations support effective root-cause analysis, enabling engineers to diagnose faults more efficiently and make informed operational decisions. The findings indicate that integrating SHAP with machine learning significantly enhances model transparency, interpretability, and user trust without compromising predictive performance. Therefore, the proposed framework offers a practical and scalable solution for explainable industrial process monitoring and contributes to the adoption of trustworthy AI in chemical manufacturing systems.