A Novel Physics-Informed Hybrid Deep Learning Framework for Fault Detection and Diagnosis in Cyber Physical Systems

Cyber-physical systems (CPS) integrate computational algorithms and physical systems, facilitating coherent communication between the digital and physical world. Ensuring the reliability and efficiency of CPS is crucial across various domains, including space missions, manufacturing, transportation, and energy systems. One critical aspect is fault detection and diagnosis (FDD), which involves promptly identifying and addressing abnormalities or faults in the system. Fault detection in complex electromechanical systems remains a significant challenge due to non linear dynamics, diverse operating conditions and limited labeled data. Traditional model based methods, such as Bond Graph approaches, provides physics-informed insights but struggle with scalability, while deep learning based methods often lacks interpretability and robustness. To address these challenges we proposes a robust Hybrid Deep Learning Approach (HyDeLA), a hybrid architecture that integrate Convolutional Neural Networks (CNN), Long Short Term Memory (LSTM) units and an attention mechanism, combined with residual signals derived from bond graph modeling. The methodology employs residual signal augmentation, sliding window segmentation, Keras tuner-based hyperparameter optimization to adapt the model to specific datasets. Experimentation validation on two different CPS - DC Motor and Proton Exchange Membrane (PEM) Electrolyzer- demonstrates that HyDeLA achieves superior fault detection accuracy of 99.47% and 95.42% respectively, outperforming recent CNN-LSTM, VAE-GAN, Bond Graph CNN methods, even with a minimum number of labeled fault samples. In addition to this, the proposed model provides improved temporal-spatial feature capture and robustness under varying conditions. These results highlight HyDeLA’s capability to bridge physics-based modeling and deep learning, offering a reliable and generalizable framework for fault detection in safety critical cyber-physical systems.