Novel L2-Discriminant Locality Preserving Projection Integrated with Adaboost and Its Application to Fault Diagnosis

Nowadays, the safety of industrial production processes has been paid more and more attention. Fault diagnosis methods based on data-driven techniques have been widely discussed in recent years. For handling the high dimensional data generated by large and complex systems, reducing the data dimension to extract fault information has been widely studied. However, the performance of traditional dimension reduction methods is limited due to the high complexity and integration of process data. As a result, the accuracy of fault diagnosis is unacceptable. In order to handle this limitation, a novel fault diagnosis model integrating L2-Discriminant Locality Preserving Projection with AdaBoost is proposed in this paper. Discriminant Locality Preserving Projection (DLPP) as a kind of manifold learning methods is good at extracting information from high-dimensional data but its performance is subject to the singular matrix decomposition problem. So, the L2 regularization term is incorporated into the objective function of DLPP to solve the singular matrix decomposition problem. After extracting fault information using the proposed L2-DLPP method, AdaBoost is adopted to classify faults. In order to verify the performance of the proposed fault diagnosis methodology, a case study using the Tennessee Eastman process (TEP) is carried out. The effectiveness of the proposed fault diagnosis methodology is confirmed by simulation results.