AI-Enabled Low-Cost Embedded Platform for Cascade Process Automation and Predictive Maintenance

Abstract

Cascade process control systems are extensively used in chemical plants, power generation, and water treatment, where fault detection stability is important to ensure operation stability and avoid economic and safety hazards. Common fault detection techniques, including threshold-based alarms and statistical process control, are not capable of describing nonlinear dynamics and do not offer much interpretability. This study introduces an AI powered, explainable fault detection system tailored for cascade control systems and optimized for execution on low-cost embedded hardware. The new approach combines engineered cascade-invariant and cascade-dependent features with light machine learning models, supplemented by Shapley value–based explainability to deliver transparent decision-making. Experimental tests run on artificially manipulated datasets show that the approach realizes an accuracy of 93.6% and an F1-score of 0.91, which are 10–25% higher than traditional methods. Outcomes from confusion matrices, ROC and precision–recall curves, feature importance analysis, and SHAP based explanations establish both the explainability and robustness of the method. Comparative analysis emphasizes its better accuracy-interpretability-computational efficiency trade-off. The results confirm that the envisaged framework offers an economically sound and scalable solution for predictive maintenance and fault identification in industrial cascade systems, tackling primary shortcomings of current methods.