Optimized deep intelligent features for root cause analysis in industrial applications

In industrial systems, operational effectiveness and machine availability are improved through root cause analysis (RCA). Typically, RCA techniques encounter multiple challenges when dealing with complex, high-dimensional sensor and industrial control system data, which often struggle with scalability, interpretability, and real-time performance. To enhance the accuracy and efficiency of RCA, this research proposes a new system that incorporates improved deep, intelligent features derived from lean CNNs along with attentional architectures. Specifically, the proposed system extracts and ranks significant features in real time through an enhanced Armadillo MobileNet Prediction Framework (AMPF) The proposed approach is superior to popular methods, such as Genetic Algorithms, Gradient Boosting Machines, and Bayesian Networks, in terms of accuracy (96.21%), error rate (0.0379%), and F-score while maintaining resource efficiency on closed devices. Results indicate that smart feature optimization enhances the predictive ability of smart manufacturing systems and improves the clarity of the fault diagnosis system for industrial tools.