Application of an intelligent classification method to mechanical fault diagnosis

Abstract

A new method for intelligent fault diagnosis of rotating machinery based on wavelet packet transform (WPT), empirical mode decomposition (EMD), dimensionless parameters, a distance evaluation technique and radial basis function (RBF) network is proposed in this paper. In this method, WPT and EMD are, respectively, used to preprocess vibration signals to mine fault characteristic information more accurately. Then, dimensionless parameters in time domain are extracted from each of the original vibration signals and preprocessed signals to form a combined feature set. Moreover, the distance evaluation technique is utilised to calculate evaluation factors of the combined feature set. Finally, according to the evaluation factors, the corresponding sensitive features are selected and input into the RBF network to automatically identify different machine operation conditions. An experiment of rolling element bearings is carried out to test the performance of the proposed method. The experimental result demonstrates that the method combining WPT, EMD, the distance evaluation technique and the RBF network may accurately extract fault information and select sensitive features, and therefore it may correctly diagnose the different fault categories occurring in the bearings. Furthermore, this method is applied to slight rub fault diagnosis of a heavy oil catalytic cracking unit, the actual result shows the method may be applied to fault diagnosis of rotating machinery effectively.

Introduction

Rotating machinery covers a broad range of machines and plays an important role in industrial applications. Typical applications are in aeronautical, naval and automotive industries. With the increase in production capabilities of modern manufacturing systems, machines are expected to run continuously for extended hours. As a result, unexpected downtime due to machine failure has become more costly than ever before. Therefore, fault diagnosis of rotating machinery is of great practical significance in industries to meet the need to increase machine availability and avoid personal casualties and economical loss (Lei et al., 2008, Saxena and Saad, 2007). Fault diagnosis of rotating machinery can be treated as a problem of pattern recognition. It consists of three steps: data acquisition, feature extraction and selection, and final condition identification. Feature extraction and selection is the key of fault diagnosis, and final condition identification is the core issue of fault diagnosis (Lei et al., 2008, Qiao et al., 2004).

The purpose of feature extraction is to extract parameters representing the machine operation conditions to be used for machine condition identification. Generally, in order to acquire accurate fault characteristic information from raw vibration signals at the fault early stage when the fault characteristics are usually immersed in the heavy noise, some preprocessing techniques based on advanced signal analysis are necessary to perform on the vibration signals before feature extraction. Wavelet transform (WT) (Peng & Chu, 2003) and empirical mode decomposition (EMD) (Huang, Shen, & Long, 1998), as two different kinds of advance signal process techniques, are proved to be highly powerful for machine signal analysis by many applications to machine fault diagnosis (Chow and Hai, 2004, Fan et al., 2006, Gai, 2006, Gao et al., 2008, Li et al., 2006, Liu et al., 2006, Ocak et al., 2007, Sanz et al., 2007, Tse et al., 2004, Wu and Qu, 2008). Wavelet packet transform (WPT) is an extension of WT and overcomes the shortcoming that WT does not split the high frequency band where the modulation information of machine fault always exists. Thus, both WPT and EMD are adopted to preprocess the vibration signals to highlight the fault characteristics in this paper.

Dimensionless parameters in time domain, such as skewness, kurtosis, crest indicator, etc., are effective and practical in fault diagnosis of rotating machinery due to their relative sensitivity to early faults, and robustness to various loads and speeds. Here, dimensionless parameters are extracted from not only the raw vibration signals but also the preprocessed signals of WPT and EMD. Unfortunately, features extracted from the raw and preprocessed signals may have large dimensionality, which may increase the computational burden of a subsequent classifier, and degrade the generalization capability of the classifier. Thus, to overcome these shortcomings, a few features which obviously characterise the machine operation conditions need to be selected from all the features. There are some feature selection methods such as conditional entropy (Lehrman, Rechester, & White, 1997), genetic algorithm (Jack et al., 2002, Samanta, 2004), distance evaluation technique (Widodo et al., 2007, Yang and Kim, 2006), etc. Because of the simpleness and reliability of the distance evaluation technique, it is generally adopted in fault diagnosis.

Final condition identification is another task in fault diagnosis of rotating machinery. Machine condition identification via artificial intelligence techniques can provide an automated fault diagnosis procedure. An artificial neural network (ANN) is an information processing paradigm that is inspired by the way the human brain processes information and has been established as a powerful intelligence technique in the condition identification of rotating machinery (Chen & Wang, 2002). Condition identification using ANN classifiers may largely increase the reliability of fault diagnosis methods. The radial basis function (RBF) network provides good transfer functions to approximate non-linear inputs and is fast in convergence. Also, it is able to automatically determine the number of neurons in the hidden layer during training. Hence, an optimized architecture for the RBF network can be obtained (Ham & Kostanic, 2001).

In view of the above analysis, a new intelligent fault diagnosis method is developed in this paper using WPT, EMD, dimensionless parameters, a distance evaluation technique and the RBF network. WPT and EMD are performed on raw vibration signals captured from rotating machinery to acquire more faulty characteristics. The dimensionless parameters are extracted from both the raw and preprocessed signals to obtain a combined feature set with rich faulty information. The distance evaluation technique is adopted to evaluate all features in the feature set and the a few sensitive features are selected. The RBF network is used to identify the machine operation conditions with the input of the selected sensitive features.

The proposed method is tested through identifying the different fault categories of rolling element bearings. The result demonstrates its effectiveness. Furthermore, the different fault severities of a heavy oil catalytic cracking unit are also diagnosed with a high accuracy. This result validates the good generalization ability of the presented intelligent method.