Pre-Processing-Free Gear Fault Diagnosis Using Small Datasets with Deep Convolutional Neural Network-Based Transfer Learning

📝 Abstract

Early fault diagnosis in complex mechanical systems such as gearbox has always been a great challenge, even with the recent development in deep neural networks. The performance of a classic fault diagnosis system predominantly depends on the features extracted and the classifier subsequently applied. Although a large number of attempts have been made regarding feature extraction techniques, the methods require great human involvements are heavily depend on domain expertise and may thus be non-representative and biased from application to application. On the other hand, while the deep neural networks based approaches feature adaptive feature extractions and inherent classifications, they usually require a substantial set of training data and thus hinder their usage for engineering applications with limited training data such as gearbox fault diagnosis. This paper develops a deep convolutional neural network-based transfer learning approach that not only entertains pre-processing free adaptive feature extractions, but also requires only a small set of training data. The proposed approach performs gear fault diagnosis using pre-processing free raw accelerometer data and experiments with various sizes of training data were conducted. The superiority of the proposed approach is revealed by comparing the performance with other methods such as locally trained convolution neural network and angle-frequency analysis based support vector machine. The achieved accuracy indicates that the proposed approach is not only viable and robust, but also has the potential to be readily applicable to other fault diagnosis practices.

💡 Analysis

Early fault diagnosis in complex mechanical systems such as gearbox has always been a great challenge, even with the recent development in deep neural networks. The performance of a classic fault diagnosis system predominantly depends on the features extracted and the classifier subsequently applied. Although a large number of attempts have been made regarding feature extraction techniques, the methods require great human involvements are heavily depend on domain expertise and may thus be non-representative and biased from application to application. On the other hand, while the deep neural networks based approaches feature adaptive feature extractions and inherent classifications, they usually require a substantial set of training data and thus hinder their usage for engineering applications with limited training data such as gearbox fault diagnosis. This paper develops a deep convolutional neural network-based transfer learning approach that not only entertains pre-processing free adaptive feature extractions, but also requires only a small set of training data. The proposed approach performs gear fault diagnosis using pre-processing free raw accelerometer data and experiments with various sizes of training data were conducted. The superiority of the proposed approach is revealed by comparing the performance with other methods such as locally trained convolution neural network and angle-frequency analysis based support vector machine. The achieved accuracy indicates that the proposed approach is not only viable and robust, but also has the potential to be readily applicable to other fault diagnosis practices.

📄 Content

Pre-Processing-Free Gear Fault Diagnosis Using Small Datasets with Deep Convolutional Neural Network-Based Transfer Learning

Pei Cao Graduate Research Assistant and Ph.D. Candidate Department of Mechanical Engineering University of Connecticut 191 Auditorium Road, Unit 3139 Storrs, CT 06269 USA

Shengli Zhang Stanley Black & Decker, Global Tool & Storage Headquarter Towson, MD 21286 USA

J. Tang† Professor Department of Mechanical Engineering University of Connecticut 191 Auditorium Road, Unit 3139 Storrs, CT 06269 USA Phone: (860) 486-5911, Email: jtang@engr.uconn.edu

Submitted to

† Corresponding author

Pre-Processing-Free Gear Fault Diagnosis Using Small Datasets with Deep Convolutional Neural Network- Based Transfer Learning

1Pei Cao, 2Shengli Zhang and 1J. Tang* 1Department of Mechanical Engineering, University of Connecticut, Storrs, CT 06269, USA 2Stanley Black & Decker, Global Tool & Storage Headquarter, Towson, MD 21286

Abstract

Early fault diagnosis in complex mechanical systems such as gearbox has always been a great challenge, even with the recent development in deep neural networks. The performance of a classic fault diagnosis system predominantly depends on the features extracted and the classifier subsequently applied. Although a large number of attempts have been made regarding feature extraction techniques, the methods require great human involvements are heavily depend on domain expertise and may thus be non-representative and biased from application to application. On the other hand, while the deep neural networks based approaches feature adaptive feature extractions and inherent classifications, they usually require a substantial set of training data and thus hinder their usage for engineering applications with limited training data such as gearbox fault diagnosis. This paper develops a deep convolutional neural network-based transfer learning approach that not only entertains pre-processing free adaptive feature extractions, but also requires only a small set of training data. The proposed transfer learning architecture consists of two parts; the first part is constructed with a piece of a pre-trained deep neural network that serves to extract the features automatically from the input, the second part is a fully connected stage to classify the features that needs to be trained using gear fault experiment data. The proposed approach performs gear fault diagnosis using pre- processing free raw accelerometer data and experiments with various sizes of training data were conducted. The superiority of the proposed approach is revealed by comparing the performance with other methods such as locally trained convolution neural network and angle-frequency analysis based support vector machine. The achieved accuracy indicates that the proposed approach is not only viable and robust, but also has the potential to be readily applicable to other fault diagnosis practices.

Keywords: gear fault diagnosis, deep convolution neural network, transfer learning

1. Introduction In modern industry, the significance of condition monitoring and fault diagnosis has been ever-increasing due to the continuously raising standard for safety and quality. Gearbox, as one of the most common components used in contemporary machineries, is susceptible to failure under severe working conditions and thus requires the practice of fault diagnosis. Signal-based techniques have been shown to be an effective way to facilitate such practices (Kang et al, 2001; Randall, 2011; Marquez et al, 2012). As its name indicates, in a typical signal-based system, vibration signals are measured first, a feature extraction technique is then employed to characterize the fault-related features and a classifier is applied to predict fault occurrence in terms of type and severity. The accuracy of fault diagnosis is thus heavily correlated to the features extracted by the adopted technique. Last decades have seen diverse attempts to identify and extract useful features from measured signals for fault diagnosis, which fall into three main categories: time-domain analysis (Zhou et al, 2008; Parey and Pachori, 2012), frequency domain-analysis (Fakhfakh et al, 2005; Li et al, 2015; Wen et al, 2015) and time-frequency analysis (Tang et al, 2010; Chaari et al, 2012; Yan et al, 2014, Zhang and Tang, 2018). Although decent results of fault diagnosis tasks have been reported, most of these studies indeed exploit the features of preference. In other words, the types of feature extracted are mostly based on domain expertise and manual decision that may not be the most sensitive ones to reflect the health condition of the machinery. Moreover, expertise-based feature extraction techniques require vast human involvements and design efforts. In certain mechanical systems,

This content is AI-processed based on ArXiv data.