Fault diagnosis of bearings using multifractal detrended cross-correlation analysis and uncertain support vector machines

The effectiveness of data-driven fault diagnostic models is heavily reliant on the quality of the features. However, the current feature extraction methods primarily rely on probability theory and must adhere to the requirement of the “law of large numbers”. Consequently, if the number of accessible samples is insufficient, there will be a significant level of uncertainty in the resulting features. In order to tackle the problem, we have put out a novel approach for defect diagnosis called empirical mode decomposition, multifractal detrended cross-correlation analysis, and uncertain support vector machines (EMD-MFDCCA-UnSVM). Initially, the EMD technique is utilized to break down the vibration signal of the bear into several intrinsic mode functions (IMFs). Next, Multifractal Detrended Cross-Correlation Analysis (MFDCCA) is used to the primary Intrinsic Mode Functions (IMFs) in order to extract multifractal characteristics. These characteristics can then be utilized to describe and differentiate the various failure modes and levels of severity in the bearings. Furthermore, drawing inspiration from the uncertain theory, we treat each uncertain feature as an uncertain distribution and construct a Support Vector Machine (SVM) model with a chance restriction on uncertainty. Ultimately, the trained U nSVM is employed to diagnose the signal features that have been retrieved by the EMD-MFDCCA. The tests demonstrate that EMD-MFDCCA-U nSVM attains noteworthy outcomes on validation datasets.