Industrial Machine Fault Detection Using Hybrid CNN-LSTM Model with Real- Time SCADA Integration for Mechanical Fault Analysis

This paper presents a comprehensive intelligent industrial machine fault detection system based on a hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) architecture integrated with a real-time SCADA-style monitoring platform. The proposed system analyzes multivariate sensor data—vibration, temperature, pressure, current, and ambient air quality—to identify five categories of abnormal machine behavior with high precision. A Flask-based web dashboard with WebSocket communication provides real-time visualization of machine health status, live waveform plots, and a fault event log with confidence scores. A NodeMCU ESP8266 hardware controller activates a piezoelectric buzzer upon fault detection and displays status locally on an LCD module. The integration of deep learning inference with real-time hardware response enables proactive predictive maintenance, reducing unplanned downtime and improving industrial safety standards. The system is exhaustively validated on a three-phase induction motor testbed instrumented with six sensor modalities under five fault conditions: bearing failure, overheating, pressure anomaly, vibration excess, and electrical fault. The hybrid CNN-LSTM model achieves 97.4% overall classification accuracy, precision and recall exceeding 96% across all fault classes, a 1.8% false positive rate, and an end-to-end system latency of 87 milliseconds, significantly outperforming conventional threshold-based SCADA methods, standalone CNN, standalone LSTM, SVM with FFT features, Random Forest, and other contemporary baselines. This paper provides detailed treatment of the system architecture, dataset construction methodology, model design and training procedure, hardware integration, experimental results, and future research directions.