LAUSR

Complex and changeable working conditions are important factors affecting the accuracy and robustness of diesel engine fault diagnosis models. Working condition identification can provide a basic reference for the unit operation state, which is of great significance for fault diagnosis. At present, most working condition identification models take power as the identification parameter, divide the power parameter into several discrete intervals, and obtain the power interval of the current state through the classification model. However, describing the working condition only by power will lead to the coupling of speed and load parameters, and the working condition parameters should be continuous variables. In this paper, a continuous working condition model decoupled by speed and load parameters is proposed, and the working condition identification model is established based on a graph self-attention network (GSAN). A large number of experimental data of 32 working conditions under normal and typical fault simulations is obtained on a diesel engine experimental bench, which is used for training and testing models. Under the condition of untrained working conditions, the $R_{adj}^{2}$ coefficients of the proposed method are 96.70% and 97.27% for normal and typical fault experimental data respectively, demonstrating the efficiency of the proposed approach.