noise reduction of vibrational signals based on singular spectrum analysis and fault diagnosis of ball bearing using artificial neural network

Document Type : full paper

Authors

• Mehdi Salehi
• Mehdi Hassanian

Department of Mechanical Engineering, Najafabad Branch, Islamic Azad University, Najafabad, Esfahan, Iran

Abstract

Fault detection and determination of rolling-element bearings is the highest priority in vibration analysis. Various methods have been proposed for fault detection in rolling-element bearings in frequency and time-frequency domains. In this work, a novel and effective method for fault detection in ball bearings in time domain is proposed. Most time domain methods are complex with high noise concentration in extracted signals. Single spectrum analysis (SSA) is an effective and simple to implement technique for noise removal in time domain series. In this method, the vibration signals for measuring each of ball bearing faults are decomposed into their main components and after selecting some of these components for reconstruction of vibration signal, the statistical characteristics of time domain are extracted from the reconstructed signal. These characteristics are used as inputs of Artificial Neural Networks (ANN) for fault detection and classification of ball bearings. The outputs of the ANN are the faults of ball bearings and determining the suitable number of hidden neurons (middle layer) will maximize the accuracy of fault detection. Adding a Gaussian white noise to signals, performance of SSA method and the accuracy of fault detection using ANN were also investigated. The results show the successful and effective implementation of SSA in fault detection of ball bearings with minimum error

Keywords

• Singular spectrum analysis
• Artificial Neural Network
• Fault diagnosis
• Roller element bearing

---