A New Rotation Machinery Fault Diagnosis Method Based on Deep Structure and Sparse Least Squares Support Vector Machine

Abstract

In this paper, a fault diagnosis method that is based on the deep structure and the sparse least squares support vector machine (SLSSVM) is proposed. This method constructs the structure of a multi-layer support vector machine (SVM). First, the SVM on the first layer is trained by using the training samples, and it learns the shallow features of the data. Then, the ''feature extraction formula'' is used to generate a new expression of the sample, which is used as input of the next layer. The new layer of the SVM trains on the new sample, and it extracts and learns the deep features of the signal layer by layer; eventually, after multiple feature mapping, it outputs the diagnostic results on the last layer. Because of the deep structure, the algorithm complexity and operation time increase. Therefore, in this paper, the least squares support vector machine (LSSVM) is combined with the sparse theory. By constructing the approximate maximal linearly independent vector set in the feature space, we conduct the sparse expression of samples and obtain the discriminant function for classification, which effectively solves the problem of sparsity deficiency for the LSSVM. Last, the method is used to diagnose centrifugal pump faults and rolling bearing faults and compares with the several methods of the SVM, the SLSSVM, deep SVM, and convolutional neural networks. The diagnostic results indicate that the method in this paper has good performance. INDEX TERMS Fault diagnosis, deep structure, support vector machine, sparsity.

Quantitative Evaluation on Valve Leakage of Reciprocating Compressor Using System Characteristic Diagnosis Method

“…After extracting CIs, they need to be classified to identify different fault modes. SVM is a powerful classification technology that is based on statistical learning theory [47,48]. SVM was originally introduced as a generalized linear classifier of binary classification.…”

“…Function φ(x i ) is a feature mapping rule, which is determined by the kernel function. The most widely used radial basis function (RBF) is selected as the kernel function of the SVM algorithm in this study [47]. The expression of the function is shown as in Equation (15).…”

“…By using comb filter and setting appropriate basic frequency and bandwidth, the energy spectrum values of RC rotational frequency component and its two-fold and three-fold frequency components contained in vibration signal were extracted as the CIs of the response. SVM has unique advantages for solving small samples, nonlinearity, and high dimension classification problems [47,48], and it has good robustness and generalization ability [49,50]. Based on the CIs of excitation, response, and system, the SVM was used to conduct the contrast experiment of different fault pattern.…”

Permanent Synchron Magnet Motor Speed Observer Based on Least Squares Support Vector Machine Regression

“…The adaptation mechanism of observer MRAS is proposed using LSSVMR, as shown in Figure 4.The LSSVMR input is the reference current error and the estimated current from the adaptive model, while the output is the estimated back voltage emf which will determine the estimated PMSM speed based on Equation (7) and (8). The estimated current in the adaptive model is determined by the value of the motor voltage in the reference model and the value of the back voltage emf from the LSSVMR output.…”

“…LSSVMR is one type of machine learning with supervised training [6]. LSSVMR was chosen because it is global optimal in training data [7] - [9], so that it can increase validity and accelerate the work of the LSSVMR algorithm in identifying PMSM speed and rotor position based on the back emf voltage parameters.…”