Bearing Fault Analyses through the Application of ANFIS and Vector Array Indicators Based on Statistical Parameters of Wavelet Transformation Components

The fault diagnosis of machinery components, such as bearings, through their vibration signals is one of the valuable tools for failure prevention in industrial operations. Recent developments in this area have produced various methods for vibration signal processing. Some methods combine several signal processing techniques for fault feature extraction of bearing signals. One of these combinations which has been the focus of research is the application of the Wavelet transform, Statistical analysis and Adaptive Neural Fuzzy Inference System (ANFIS). This paper will provide an explanation and survey of the application of the Wavelet method. The advantages and disadvantages of its application in relation to vibration signal analysis will also be evaluated. The paper then reports on the pre‐processing of vibration signals using the Wavelet transform and statistical analyses to generate suitable inputs for ANFIS. For the purposes of training and testing ANFIS, there is an investigation of the generation of vector arrays which will be used as inputs for ANFIS. This vector generation scheme is based on several statistical parameters such as mean, standard deviation and vector distance which are calculated from the components of the wavelet transformation results. Prior to vector generation, the vibration data is pre‐processed using wavelet decomposition and the data is separated into two sections. The first half section is used as test data and the second half section is used as training data. A scheme of data normalization is also investigated which aims to scale the magnitude of each data row. The final results of this investigation are an array of vectors which can be used as additional inputs to ANFIS for use in the next stage of bearing vibration signal analysis.