An Explainable Hybrid Deep Learning Framework for Real-Time Signal Classification and Predictive Analytics in Intelligent Industrial System

Abstract

Intelligent industrial systems require reliable real-time monitoring to detect machine failures and reduce unexpected downtime. Industrial Internet of Things sensors generate continuous signals that can support fault diagnosis and predictive maintenance. This study proposes an explainable hybrid deep learning framework for real-time signal classification and predictive analytics in intelligent industrial systems. A simulated IIoT dataset containing 1,000 machine sensor observations was used, including 15 industrial signal features such as temperature, vibration, pressure, load, motor temperature, sound level, fan speed, voltage, current, and power-related variables. The target variable was machine failure, classified as normal or failure. The study involved data preprocessing, class imbalance handling, threshold calibration, hybrid model training, performance evaluation, and explainability analysis. The proposed framework achieved an accuracy of 0.90, precision of 0.87, recall of 0.78, F1-score of 0.82, and ROC-AUC of 0.91 at an optimized threshold of 0.62. Explainability analysis identified vibration, temperature, load, motor temperature, and pressure as the most influential features for machine failure prediction. The results show that the proposed framework effectively supports real-time machine condition classification, early fault detection, and predictive maintenance, while improving model transparency for intelligent industrial decision-making.