Decoupling Identification Method of Continuous Working Conditions of Diesel Engines Based on a Graph Self-Attention Network

For diesel engine malfunction detection, machine learning-based intelligent detection approaches have made great strides, but some performance deterioration is also observed due to the significant ambient noise and the change in operating circumstances in the actual application situations. Diesel engine fault diagnostic models can be negatively impacted by complex and erratic working circumstances. Identifying the working condition can provide as a baseline for the current unit operating state, which is crucial information when trying to pinpoint the source of an issue. Many existing techniques for identifying operational states use power as an identifier, segmenting it into discrete intervals from which the current state's power may be derived using a classification model. However, the working condition characteristics should be constant, and defining it exclusively in terms of power would lead to the connection of speed and load elements. In this study, we offer a regular working situation model that is independent of speed and load characteristics, and we use a graph self-attention network to construct a model for identifying the working condition. On a diesel engine research bench, a vast amount of experimental data is acquired for training and testing models, including 32 different operating situations under normal and typical fault scenarios. The R2 adj values of 99.70% and 99.27% for normal and typical defect experimental data, correspondingly, demonstrate the efficacy of the suggested technique under the circumstance of uninformed nnerating situations.