Bearing fault diagnosis based on kernel independent component analysis and antlion optimization

Fault diagnosis of gearboxes based on vibration signal processing is challenging, as vibration signals collected by acceleration sensors are typically a nonlinear mixture of unknown signals. Furthermore, the number of source signals is usually larger than that of sensors because of the practical limitation on sensor positions. Hence, the fault characterization is actually a nonlinear underdetermined blind source separation (NUBSS) problem. In this paper, a novel NUBSS algorithm based on kernel independent component analysis (KICA) and antlion optimization (ALO) is proposed to address the technical challenge. The mathematical model demonstrates the nonlinear mixing of source signals in the underdetermined cases. Ensemble empirical mode decomposition is used as a preprocessing tool to decompose the observed signals into a set of intrinsic mode functions that suffers from the problem of redundant components. The correlation coefficient is utilized to eliminate the redundant components. An adaptive threshold singular value decomposition method is proposed to estimate the number of source signals. Then a whitening process is carried out to transform the overdetermined blind source separation (BSS) into determined BSS, which can be solved by the KICA method. However, the reasonable selection of parameters in KICA limits its application to some extent. Therefore, ALO and Fisher’s linear discriminant analysis are adopted to further enhance the accuracy of the KICA method. The separation performance of the proposed method is assessed through simulation. The numerical results show that the proposed method can accurately estimate the number of source signals and attains a higher separation quality in tackling nonlinear mixed signals when compared with the existing methods. Finally, the inner ring fault experiment is conducted to preliminarily validate the practicability of the proposed method in bearing fault diagnosis.

Improve the spectral clustering by integrating a new modularity similarity index and out-of-sample extension

Spectral clustering is one of the most important data processing methods which has been wildly applied to machine learning, computer vision, pattern recognition and image processing. However, one of the main drawbacks of spectral clustering is the fact that the clustering model is defined only for primal data without clear extension to out-of-sample data. To improve its efficiency, in this paper, we proposed a new modularity-based method for spectral clustering with out-of-sample extension. First, kernel independent component analysis is used to solve the demixing matrix on Stiefel manifold in order to extract high-order irrelevant data feature. Then, a new modularity similarity measure-based spectral mapping algorithm is proposed, which allows the clustering model to be directly extended to out-of-sample data. Based on above analysis, we present a spectral clustering algorithm with out-of-sample extension. Experimental results show our method has better performance compared with other related algorithms in different datasets.

A Fault Detection Method Based on CPSO-Improved KICA

In view of the randomness in the selection of kernel parameters in the traditional kernel independent component analysis (KICA) algorithm, this paper proposes a CPSO-KICA algorithm based on Chaotic Particle Swarm Optimization (CPSO) and KICA. In CPSO-KICA, the maximum entropy of the extracted independent component is first adopted as the fitness function of the PSO algorithm to determine the optimal kernel parameters, then the chaotic algorithm (CO) is used to avoid the local optimum existing in the traditional PSO algorithm. Finally, this proposed algorithm is compared with Weighted KICA (WKICA) and PSO-KICA with Tennessee Eastman Process (TEP) as the benchmark. Simulation results show that the proposed algorithm can determine the optimal kernel parameters and perform better in terms of false alarm rates (FAR), detection latency (DL) and fault detection rates (FDR).

Performance Evaluation of Kernel-Based Feature Extraction Techniques for Face Recognition System

Face recognition is considered to be one of the most reliable biometrics where security issues are of concerned. Feature extraction which is a functional block of a face recognition system becomes a critical problem when there is need to obtain the best feature with minimum classification error and low running time. Most existing face recognition systems have adopted different non-linear feature extraction techniques for face recognition but identification of the most suitable non-linear kernel variants for these systems remain an open problem. Hence, this research work analyzed the performance of three kernel feature extraction technique (Kernel Principal Component Analysis, Kernel Linear Discriminant Analysis and Kernel Independent Component Analysis) for face recognition system. A database of 360 face images was created by obtaining facial images from LAUTECH Biometric Research Group consisting of six facial expressions of 60 persons. Images were preprocessed (gray scaling, cropping and histogram equalization) and the kernel variants were used to extract distinctive features and reduce the dimensionality of each of the images from 600x800 pixels to four smaller dimensions: 50x50, 100x100, 150x150 and 200x200 pixels. Euclidean Distance similarity measure was used for classification. The performance of the three kernel variants was evaluated for face recognition system using 180 images for training and 180 images for testing using the following metrics: Recognition Accuracy (RA) and Recognition Time (RT). Empirical result indicate that KLDA performs best for face recognition system with an average accuracy of 94.52%. The larger image dimension also results in better recognition performance. We intend to experiment on other classifiers for face recognition system in our future work. Keywords— Biometrics, Face, Feature extraction, Kernel, KICA, KPCA, KLDA, Linear, Non-linear;