Authors: S.Radha Krishnan, Assistant Professor,M.Aishwarya, Dr.R.Natarajan
Abstract: Fault detection in industrial systems is crucial for ensuring operational safety, minimizing downtime, and reducing maintenance costs. This work proposes a hybrid deep learning model combining Convolutional Neural Networks (CNN) and Gated Recurrent Units (GRU) to detect and classify machine faults from time-series data. The CNN layers extract spatial features, while GRU layers model temporal depen,dencies in the data. The architecture incorporates residual connections to enhance gradient flow and improve learning efficiency. The model is evaluated on multi class fault detection datasets, achieving robust performance with high accuracy, precision, recall, and F1-score. Advanced metrics, including ROC-AUC, logarithmic loss, Cohen's Kappa, and Matthews Correlation Coefficient, demonstrate the model's reliability. Visualization of confusion matrices and detailed performance metrics validates its effectiveness in detecting anomalies and classifying fault types. This approach can be generalized for real-time monitoring systems in various industrial applications, ensuring predictive maintenance and operational excellence.
