Tri-axial vibration information fusion model and its application to gear fault diagnosis in variable working conditions

Abstract

The differences between tri-axial vibration signals are often considered to be too small to affect the accuracy of gearbox fault diagnosis. Thus, the traditional approach of vibration signal processing based on one particular single axis is often used, which is often chosen by the accelerometer’s orientation relative to the fault location (vertical or horizontal) or the signal’s sensitivity to fault characteristics. This can cause some uncertainty and incompleteness in accurate diagnosis. To tackle this problem, we proposed a novel tri-axial signal information fusion model. In this model, measured vibration signals on three axes were analyzed at three levels to show their differences, and the visualized results were illustrated with histograms. Based on the comparison of the differences, the fusion model was used to analyze differences between the three different axes quantitatively based on the cross-approximate entropy, and such quantitative differences were utilized as adaptive weight coefficients to fuse three orthogonal axial signals. Through this process, the proposed model could transfer differences between different axial signals as adaptive fusion weights. To verify the superiority of the proposed strategy, a classical fuzzy C-means method was used to classify the fused signals from the model. With the measured gear fault signals under variable working conditions from the test bench, the reliability and superiority of the information fusion model proposed were proven through a comparison to fault diagnosis based on traditional signal axis selection.