The KPCA-based fault diagnosis algorithm is limited by the selection of kernel function parameters and the screening of non-faulty modeling data. In order to solve this problem, the DBSCAN clustering algorithm is deeply integrated with the KPCA algorithm, and the method of parameter optimization is proposed from the perspective of data mining. By analyzing the distribution of SPE values of core points, boundary points and outlier points obtained by DBSCAN algorithm under different kernel width parameters of KPCA algorithm, the objective function of kernel width and SPE control limit are proposed respectively, which transform the problem of parameter selection into the problem of maximum value optimization. Inheriting the advantages of both algorithms the fused algorithm can remove the fault data from the actual monitoring sensors for offline modeling and find the optimal kernel width and SPE control limit according to the data shape automatically. The proposed algorithm is simulated with synthetic crescent data and aerial rolling bearing data and the results are evaluated by ROC curve, which proves the practicability and effectiveness of the algorithm.
DBSCAN Based Parameter Optimization of KPCA for Fault Diagnosis
Lect. Notes Electrical Eng.
30.10.2021
20 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
DBSCAN Based Parameter Optimization of KPCA for Fault Diagnosis
TIBKAT | 2022
|An Integrated Fault Identification Method Based on KPCA-PNN for Complex Industrial Process
British Library Online Contents | 2009
|