Fault diagnosis based on deep learning subject to missing data

With the development of modern industry and computer technology, deep learning is widely used in the field of fault diagnosis, but it still faces many challenges, such as data missing due to different sampling rate of different sensor or other data packet of dropout of control system network. Observation data with missing values will seriously affect the result of fault diagnosis. Imputation methods such as regression imputation method intend to solve the problem of data missing to a certain extent. However, as the missing rate increases and the cross-correlation coefficient between observation variable-decreases, the traditional imputation method will fail in extracting the potential feature involved in the missing data. In this paper, a neural network imputation method is proposed to estimate the missed observation. Once online observation data with missing value is available, neural network imputation method is first used to get a structural complete observation sample. Then DNN trained by structural complete data can be effectively classify different fault with high accuracy. Experiment analysis shows the efficiency of the proposed method.