Incremental supervised locally linear embedding for machinery fault diagnosis

Abstract

Locally linear embedding (LLE) is a promising algorithm for machinery fault diagnosis, but LLE operates in a batch mode and lacks discriminant information, which lead to be negative for fault diagnosis. In this paper, incremental supervised LLE (I-SLLE) is investigated for submersible plunger pump fault diagnosis. In the I-SLLE algorithm, block matrix decomposition technology is introduced to deal with out-of-sample data, while a part of old low-dimensional coordinates is also updated, upon which an iterative method is presented to update all the data for refining the accuracy. Furthermore, in order to improve the classification capability of LLE, discriminant information is assembled in the cost function of LLE. Based on I-SLLE, a new machinery fault diagnosis method is proposed. At first, I-SLLE is utilized to extract the feature of an original dataset, and then support vector machine (SVM) is employed to classify the test data in the feature space. Experiments on synthetic datasets as well as real world datasets are performed, demonstrating the efficiency and the accuracy of the proposed algorithm.