Fault Diagnosis Method and Application of Direction Axis Kernel Partial Least Squares

In the actual oil extraction process of the pumping unit, the data showed multiple features, repeatability and other problems, so that hidden information of variables cannot be completely expression, obtain low fault diagnosis accuracy, so a fault diagnosis methods based on the direction axis kernel partial least squares (DAKPLS) was proposed. The main contributions are as follows: (1) Constructing the maximum covariance set as direction axis for projection to maximize the degree of heterogeneity between data. (2) Construct a contribution map method to identify the source of failure variables. Compared with KPLS, obtain more latent variables between input and output variables. Applying the DAKPLS method to the process fault diagnosis of the pumping unit can accurately detect the fault and reduce the false alarm and false alarm rate, which shows the effectiveness of the proposed DAKPLS method.