Hybrid physics-infused 1D-CNN based deep learning framework for diesel engine fault diagnostics

Fault diagnosis is required to ensure the safe operation of various equipment and enables real-time monitoring of associated components. As a result, the demand for new cognitive fault diagnosis algorithms is the need of the hour. Existing deep learning algorithms can detect faults but do not incorporate the system’s underlying physics into the prediction and model training processes. Therefore, the results generated by this class of fault-detecting algorithms sometimes do not make sense and fail to deliver when put to the test in actual operating conditions. We propose an end-to-end, autonomous hybrid physics-infused deep learning framework that consists of a low-fidelity physics model combined with a 1 Dimensional Convolutional Neural Network (1D CNN) to address the aforementioned issues. The application system under consideration is a 6-cylinder, 4-stroke, 7.6 L Navistar diesel engine. The physics model in the hybrid framework ensures that the predictions made by the framework are in coherence with the actual dynamics of the engine. In contrast, the deep learning component of the hybrid framework makes up for the simplifications involved during the development of the physics model of the engine, where the 1D CNN module enables robust Spatiotemporal feature extraction. Using empirical results, we demonstrate that our proposed hybrid fault diagnostics framework is autonomous and efficient for fault detection and isolation. The robustness of this framework is put to the test against the data obtained by the engine when subjected to different operating conditions, such as varying speed, changing injection pressure, and injection duration. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2024.