面 向 高 可 信 机 械 故 障 诊 断 的 多 模 态 数 据 插 补与 融 合 方 法

Due to the issues of sensor data acquisition anomalies and data transmission errors under complex operating conditions, equipment monitoring data often contain missing values, making trustworthy machine fault diagnosis in scenarios of low-trustworthiness sensor data extremely challenging. Therefore, this paper proposed a novel multimodal data imputation and fusion method for highly trustworthy machine fault diagnosis. A generative adversarial imputation network (L2-TSGAIN) based on L2 regularization function and temporal-spatio feature extraction module (TSFEM) was proposed to comprehensively extract the features of data from the perspective of time and space, so as to achieve high-quality imputation of abnormal sensor data. A multi-input single-output auto encoder (MISO-AE) was designed to extract a universal representation of imputation data from different modalities and perform the feature fusion. The feature-fused data of different fault types were input into the convolutional neural network (CNN) classifier to realize fault diagnosis. The proposed method was experimentally validated on the planetary transmission system test bench, and the results showed that the performance of the imputation data generated by the proposed L2-TSGAIN method on quantitative metrics of MAE, MSE and RMSE are 0.0257, 0.0107 and 0.1031, respectively, verifying that the proposed method can realize high-quality multimodal data imputation. The comparative experiment results showed that the proposed method achieved the highest accuracy of 99.68% on the task of planetary transmission fault diagnosis, which demonstrated the excellent diagnostic performance of the proposed method.