Fault diagnosis of ball bearings using machine learning methods

doi:10.1016/j.eswa.2010.07.119

Expert Systems with Applications

Volume 38, Issue 3, March 2011, Pages 1876-1886

https://doi.org/10.1016/j.eswa.2010.07.119 Get rights and content

Abstract

Ball bearings faults are one of the main causes of breakdown of rotating machines. Thus, detection and diagnosis of mechanical faults in ball bearings is very crucial for the reliable operation. This study is focused on fault diagnosis of ball bearings using artificial neural network (ANN) and support vector machine (SVM). A test rig of high speed rotor supported on rolling bearings is used. The vibration response are obtained and analyzed for the various defects of ball bearings. The specific defects are considered as crack in outer race, inner race with rough surface and corrosion pitting in balls. Statistical methods are used to extract features and to reduce the dimensionality of original vibration features. A comparative experimental study of the effectiveness of ANN and SVM is carried out. The results show that the machine learning algorithms mentioned above can be used for automated diagnosis of bearing faults. It is also observed that the severe (chaotic) vibrations occur under bearings with rough inner race surface and ball with corrosion pitting.

Introduction

Condition monitoring of rotating machinery helps in early detection of faults and anticipation of problems in time, so as to prevent complete failure. Bearing vibration can generate noise and degrade the quality of a product line. Severe vibrations of bearings can even cause the entire system to function incorrectly and that results in downtime for the system and economic loss to the customer. Rolling bearings defects may be categorized as point or local defects and distributed defects. The vibrations are generated by geometrical imperfections on the individual bearing components and these imperfections are caused by irregularities during the manufacturing process as well as wear and tear. The various distributed defects are surface roughness, waviness, misaligned races, and off-size rolling elements. The local defects include cracks, corrosion pitting, brinnelling and spalls on the rolling surfaces. McFadden and Smith, 1985, McFadden and Smith, 1984 have developed the models for vibration produced by a single and multiple point defects on the inner race of the rolling element bearing under radial load based on high-frequency resonance technique. Prabhakar, Mohanty, and Sekhar (2002) have considered single and multiple point defects on inner race, outer race and the combination faults and used discrete wavelet transform (DWT) to detect these faults on bearings. Kankar, Harsha, Pradeep, and Sharma Satish (2009) have applied response surface methodology (RSM), to investigate the effects of various defects on the non-linear vibrations of rotor bearing system.

Various artificial intelligent (AI) techniques such as hidden Markov models (HMM) (Li, Wu, He, & Fulei, 2005), artificial neural networks (ANN) (Vyas & Satishkumar, 2001) and support vector machines (SVM) (Widodo and Yang, 2007, Yuan and Chu, 2007) have been used in the fault diagnosis of machines. Zhitong, Jiazhong, Hongpingn, Guoguang, and Ritchie (2003) have carried out fault detection of induction motor using SVM technique for detecting broken rotor bars. In their experiment, induction motor was experimented with no fault, one broken bar, two broken bars and three broken bars. They used stator current to obtain the signal and calculated the frequency spectrum for fault detection. Samanta (2004) has compared the performance of gear fault detection using ANN and SVM. The time-domain vibration signal of a rotating machine with normal and defective gears were processed for feature extraction. The results compare the effectiveness of both types of classifiers without and with GA-based selection of features and the classifier parameters. The main difference between ANNs and SVMs is in the principle of risk minimization (RM) (Gunn, 1998). In case of SVMs, structural risk minimization (SRM) principle is used for minimizing an upper bound on the expected risk whereas in ANNs, traditional empirical risk minimization (ERM) is used for minimizing the error on the training data.

This paper is mainly focused on bearing fault classification using two machine learning methods ANN and SVM as both can work with non-linear classifications. Vibration data are collected by piezoelectric accelerometers as a time domain signals for the healthy bearing and bearing with different faults. Defects are considered as crack in outer race, inner race with rough surface and corrosion pitting in balls. The signals obtained are processed for machine condition diagnosis as shown in the flow chart Fig. 1. Features are extracted from the time domain signal by statistical method. These features are fed to a supervised attribute filter that can be used to select features. Selected features with the known output are used for training and testing of ANN and SVM.

Section snippets

Support vector machine

Support vector machine (SVM) is a supervised machine learning method based on the statistical learning theory. It is a useful method for classification and regression in small-sample cases such as fault diagnosis. Pattern recognition and classification using SVM is described in brief (Cristianini & Shawe-Taylor, 2000).

A simple case of two classes is considered, which can be separated by a linear classifier. Fig. 2 shows triangles and squares stand for these two classes of sample points. Hyper

Artificial neural network

Artificial neural network (ANN) is an interconnected group of artificial neurons. These neurons use a mathematical or computational model for information processing. ANN is an adaptive system that changes its structure based on information that flows through the network (Zurada, 1999).

A single neuron consists of synapses, adder and activation function. Bias is an external parameter of neural network. Model of a neuron shown in Fig. 3 can be represented by following mathematical model $y_{k} = ϕ (\sum_{i = 1}^{p} w_{ki})$

Experimental setup and data acquisition

Experimental tests are carried on a test rig to generate training and test data. The rig is connected to a data acquisition system through proper instrumentation. A variety of faults are simulated on the bearing at various speeds. Various parameters of bearing used for the study are listed in Table 1. Accelerometers are used for picking up the vibration signals from various stations on the rig. Signatures for healthy bearings operation establish the baseline data. This baseline data can then be

Feature extraction and selection

A wide set of features are calculated from the vibration signals using statistics. These statistical features are explained below.

(a)
Range: Range refers to the difference between maximum and minimum value of a signal.
(b)
Mean value: Average value of a signal is termed as mean value.
(c)
Standard deviation: Standard deviation is measure of energy content in the vibration signal $Standard deviation = \sqrt{\frac{n \sum x^{2} - (\sum x)^{2}}{n (n - 1)}}$
(d)
Skewness: Skewness is a measure of symmetry, or more precisely, the lack of symmetry $Skewness = \frac{n}{(n - 1}$

Results

ANN/SVM training and classification of faults is carried out using WEKA software. Training vectors are already compiled and are put as an input. The defects considered in the study are classified using ANN/SVM as shown in Table 3. Total 73 cases are considered for testing, which have 15, 14, 15, 14, 15 cases of ball with corrosion pitting, bearing with rough inner race surface, combined bearing component defects, healthy bearings and bearing with outer race crack respectively. From Table 3, we

Conclusions

This study presents a procedure for detection of bearing fault by classifying them using two machine learning methods, namely, ANNs and SVMs. Features are extracted from time-domain vibration signals using statistical techniques. Procedure incorporates most appropriate features selection by a filtering algorithm, which uses a density-based cluster to generate cluster membership values. The roles of different vibration signals, obtained with or without loader and at various speeds, are

References (15)

P.K. Kankar et al.
Fault diagnosis of a rotor bearing system using response surface method
European Journal of Mechanics A/Solids
(2009)
P.D. McFadden et al.
Model for the vibration produced by a single point defect in a rolling element bearing
Journal of Sound and Vibration
(1984)
P.D. McFadden et al.
The vibration produced by multiple point defects in a rolling element bearing
Journal of Sound and Vibration
(1985)
S. Prabhakar et al.
Application of discrete wavelet transform for detection of ball bearing race faults
Tribology International
(2002)
B. Samanta
Gear fault detection using artificial neural networks and support vector machines with genetic algorithms
Mechanical Systems and Signal Processing
(2004)
N.S. Vyas et al.
Artificial neural network design for fault identification in a rotor bearing system
Mechanism and Machine Theory
(2001)
A. Widodo et al.
Review on support vector machine in machine condition monitoring and fault diagnosis
Mechanical Systems and Signal Processing
(2007)

There are more references available in the full text version of this article.

Cited by (381)

Manifold assistant multi-modal multi-objective differential evolution algorithm and its application in actual rolling bearing fault diagnosis
2024, Engineering Applications of Artificial Intelligence
The fault diagnosis of rotating mechanical systems is crucial for ensuring the reliability of industrial equipment. In actual industrial scenarios, different combinations of fault features may yield the same diagnostic accuracy, but the difficulty and cost of obtaining different features generally vary. Therefore, in the feature extraction process, if multiple optimal feature combinations can be provided to decision-makers, it would be beneficial to select the feature subsets with lower computational cost or measurement difficulty that achieve the same fault diagnosis performance. Therefore, we model the fault feature selection task as a multi-modal multi-objective optimization problem (MMOP). To better solve MMOPs, a manifold assistant multi-modal multi-objective differential evolution (MA-MMODE) algorithm is proposed. It can establish a regularity model based on the distribution of the population to generate offspring and approximate the true Pareto set. Subsequently, this paper presents a preselection strategy that considers the distributions of both the decision space and objective space. The proposed preselection strategy can ensure that individuals with more promising solutions enter the mating pool to generate new offspring, thus exploring the search space more comprehensively. Using the CEC 2019 test set, the proposed MA-MMODE is compared with state-of-the-art algorithms, and the experimental results illustrate the effectiveness of the proposed algorithm. Moreover, based on two real rolling bearing testbeds, the proposed MA-MMODE is applied for feature selection in actual fault diagnosis. The experimental results show that the proposed MA-MMODE can provide multiple equivalent feature subsets to decision-makers while effectively improving the fault diagnosis accuracy for rolling bearings.
A generalized method for diagnosing multi-faults in rotating machines using imbalance datasets of different sensor modalities
2024, Engineering Applications of Artificial Intelligence
Fault diagnosis of rotating machines is essential for the safe and efficient operation of maritime vessels. It prevents potential failures in rotating machines in maritime systems. Simultaneously developed faults severely damage the machines, leading to unprecedented incidents and higher maintenance costs. Data availability for developing multi-fault diagnosis solutions is limited, creating maintainability problems. The unavailability of data also causes further obstacles in the accurate diagnosis of machines. The efficiency of maritime machinery can be improved with intelligent diagnosis by advanced machine learning algorithms. This paper proposes a methodology to diagnose multi-faults with balanced and imbalanced acoustic, vibration and current signals datasets. After acquiring the data from an experimental setup under various speeds, a time-frequency-based Stepping Frame Synchrosqueezed Fourier Transform (SFSFT) is deployed in each segmented signal prior to feature extraction. The extracted features are used for training the Radial Basis Function Neural Network (RBFNN). Different balanced and imbalanced datasets are created to check the multi-fault classification ability of the proposed methodology. The results show that SFSFT-RBFNN is a generalised methodology, which performed tremendously well with a maximum classification accuracy of 100 % for balanced and imbalanced datasets in all three modality datasets.
Prediction of friction coefficient of polymer surface using variational mode decomposition and machine learning algorithm based on noise features
2024, Tribology International
Establishing a correlation between the friction coefficient and friction noise of metal-friction polymer interfaces is a challenging task in various environments. To address this issue, our study utilizes machine learning algorithms to construct a friction data-based model, elucidating the relationship between noise and friction coefficient. We propose the variational mode decomposition (VMD) along with five machine learning algorithms, each capturing unique data characteristics. Algorithm optimization is achieved through the implementation of L1 and L2 regularization methods. By comparing the regression results, we develop a tribological model that offers a novel approach for constructing a nonlinear model, enabling precise monitoring of the friction coefficient by leveraging noise signals.
Deep learning algorithm to predict friction coefficient of matching pairs at different temperature domains based on friction sound
2023, Tribology International
The tribological properties of matching pairs at different temperature domain is predicted through friction sound. Two deep learning algorithms respectively by virtue of the Nonlinear Auto-Regressive models with Exogenous Inputs and long short-term memory neural network are adopted to analyze the acoustic features of friction sound to predict tribological properties of polymer surface at five temperatures within a large temperature domain under different working conditions. The deep learning algorithm precisely fits the friction coefficient in accordance with the performance analysis.
svm_friction: A Python based software for calculating, data analysis and modeling the coefficient of friction of aluminum metal matrix composites using support vector regression[Formula presented]
2023, Software Impacts
The Python software performs data analysis and modeling on the coefficient of friction (COF) of open-cell AlSi10Mg–SiC composites, which are tested by pin-on-disk method under dry sliding conditions. The software reads the experimental data from Excel files, calculates the average COF and time, saves the results to new files, prints and saves some descriptive statistics, merges the data into one DataFrame, and splits the data into training, validation, and testing sets for a support vector regression (SVR) model. The software also performs grid search to find the best hyperparameters for the SVR model, predicts the COF for the test and validation sets, calculates and saves some performance metrics, and plots and saves a histogram and a boxplot of the COF for the material. The software provides a simple, versatile, and extensible platform for other materials and models. The software is available at https://github.com/mihail-15/svm_friction.
Intelligent Diagnosis of Faults in Dynamic Equipment Through Fourier Transform, Wavelet Transform and Statistical Data
2024, SSRN

View all citing articles on Scopus

View full text