A Fault Diagnosis of Sound and Vibration Signals Using Statistical Features and Machine Learning Algorithm

The sounds and vibrations that are given off by a rotating machine can provide insight into a wide variety of distinct problem situations. By conducting statistical analysis on the recorded sound and vibration signals and subjecting them to a variety of fault conditions, it is possible to generate statistical parameters. In order to obtain information regarding the numerous flaws, the signals in the temporal domain are analyzed using these characteristics. In order to locate flaws in the system, we make use of these values, also known as “statistical features.” In this work, dimensionality reduction was achieved through the utilisation of a decision tree, principal component analysis (PCA), and independent component analysis (ICA). In order to classify the various fault states, the various classifiers were given these streamlined features to work on. During this inquiry, different kinds of classifiers were utilized, such as decision trees, support vector machines, clonal selection classification algorithms, and proximal support vector machines. This paper presents the findings of two sets of research involving statistical features extracted from sound and vibration signals and used in conjunction with the aforementioned feature reduction techniques and classifiers to detect twelve different fault conditions involving shaft, rotor, and bearing failures. The research was conducted by using sound and vibration signals as the primary source of data. According to this Proposed Approach, the Mean Classification Efficiency% of Vibration Signals for 24 Classes is 94.27%, while the Mean Classification Efficiency”% of Sound Signals for 24 Classes is 95.41%.