Condition monitoring and fault diagnosis of bearings are essential for ensuring the reliability of rotating machinery. Due to their superior performance in quantifying the irregularity and complexity of time series, entropy-based methods have demonstrated significant potential for application in rolling bearing health monitoring and fault diagnosis. However, in the design process of most existing entropy-based methods, the focus is often placed on signal complexity, while the sensitivity to capturing the dynamic trends of impact responses present in vibration signals generated by bearing damage remains relatively weak. This limits the ability of the signal features extracted by entropy-based methods to fully represent the physical characteristics of rolling bearing faults, resulting in unsatisfactory health monitoring and fault diagnosis results. To overcome this issue, a novel entropy-based method called trend entropy (TE) is proposed in this paper. In the proposed TE, entropy is derived by calculating the trend transformations associated with four specific patterns of local extrema in the signal. By quantifying the variation trends of these specific patterns, the dynamic behavior of impact responses in bearing vibration signals is captured. Therefore, such designed TE can be used to improve the ability of entropy-based method to characterize the physical characteristics of rolling bearing faults, thus enabling better health monitoring and fault diagnosis. In this paper, the process of vibration signal processing based on the proposed TE entropy is designed and developed. Through two numerical simulations and two experimental examples, the robustness and performance of the proposed TE-based method for rolling bearing health monitoring and fault diagnosis are validated.
30 April 2026 in Mechanical Systems and Signal Processing