Fault diagnosis and prediction algorithm of automobile engine based on digital twin model
Based on digital twin technology, this paper constructs a three-level coupled twin computing model of component-subsystem-whole machine, focusing on the health management requirements of automotive engines under complex working conditions. By integrating the spatio-temporal consistency modeling of multi-source sensor data, state variable encoding and dynamic update, it achieves fine reconstruction and evolution characterization of key working condition quantities. On this basis, a self-correction mechanism driven by virtual-real closed-loop residuals, multiscale health index representation, and an anomaly scoring model constrained by Mahalanobis distance are introduced. Combined with working condition adaptive thresholds and lightweight fault discrimination networks, a complete algorithm framework covering perception, diagnosis, and prediction is formed. Experimental results show that the proposed method maintains high diagnostic stability under typical working conditions such as steady state, high speed, cold start, and frequent acceleration and deceleration, with an average detection delay controlled within 20ms and an overall fault recognition accuracy rate exceeding 94%. It is suitable for online fault early warning and predictive maintenance scenarios of engines.