Fault identification model of diesel engine based on mixed attention: Single-cylinder fault data driven whole-cylinder diagnosis

In the industrial and transportation sectors, fault diagnosis of diesel engines is critical to ensure their efficient and reliable operation. Current data-driven diesel engine fault diagnosis methods primarily rely on single-cylinder respective fault data to train diagnostic models for individual cylinders which face severe limitations regarding data availability. The limitation makes realizing all cylinder fault diagnoses with traditional models in practical engineering applications difficult. For this problem, the study proposes a whole machine fault diagnosis model for diesel engines based on mixed attention fusion multi-cylinder sensor data. The model not only incorporates single-cylinder sensor data through self-attention but also effectively fuses each cylinder's signal features using the mutual attention mechanism. Simultaneously considering the mechanism knowledge of cylinder structural consistency and signal time delay similarity, this approach innovative utilizes single-cylinder fault data to develop a comprehensive fault recognition model for all cylinders. This enhances the accuracy and generalization capability of fault diagnosis at the whole-engine level, overcoming the limitations of conventional methods. In a simulated misfire experiment on the 12-cylinder V-type diesel engine, the proposed model demonstrated outstanding performance, achieving a diagnostic accuracy of 98.17%. This result validates the model's ability to diagnose faults across all cylinders when only single-cylinder fault data is available. The proposed method holds significant importance for the application of diesel engine fault diagnosis technology in engineering practice.