As one of the most vital energy conversation systems, the safe operation of wind turbines is very important; however, weak fault and time-varying speed may challenge the conventional monitoring strategies. Thus, an entropy-aided meshing-order modulation method is proposed for detecting the optimal frequency band, which contains the weak fault-related information. Specifically, the variable rotational frequency trend is first identified and extracted based on the time–frequency representation of the raw signal by constructing a novel scaling-basis local reassigning chirplet transform (SLRCT). A new entropy-aided meshing-order modulation (EMOM) indicator is then constructed to locate the most sensitive modulation frequency area according to the extracted fine speed trend with the help of order tracking technique. Finally, the raw vibration signal is bandpass filtered via the corresponding optimal frequency band with the highest EMOM indicator. The order components resulting from the weak fault can be highlighted to accomplish weak fault detection. The effectiveness of the proposed EMOM analysis-based method has been tested using the experimental data of three different gear fault types of different fault levels from a planetary test rig.

Wind turbine blades are key wind turbine components, and owing to constraints in their operating environments and year-round high-speed rotation, wind turbines may suffer from many types of damage. However, traditional defect detection algorithms for wind turbine blades have limitations, such as low accuracy for detecting tiny defects and a high number of parameters, leading to model deployment difficulties. Therefore, this study proposes a wind turbine blade detection algorithm MES-YOLOv8n (M_C2f-ECA-SBiFPN-YOLOv8n) based on YOLOv8n design to address these challenges. In the proposed algorithm, a multi-kernel, grouped convolution-based C2f module (M_C2f) is employed in the backbone network to enhance the feature extraction capability of the backbone network. In addition, an efficient channel attention mechanism (ECA) is introduced in the backbone network, which can accurately capture channel-level information between network layers. Finally, the feature fusion network adopts a structure that combines a small target detection head and a bidirectional feature pyramid network (SBiFPN) to capture the multi-scale information of defective blades, enhance the feature fusion capability of the model, and reduce the leakage rate and model parameters of small target defects of wind turbine blades. Experimental results demonstrate that the proposed MES-YOLOv8n algorithm has a mean average precision (mAP@0.5) of 85.5%, which is 1.7% higher than YOLOv8n, and the number of parameters is reduced by 30%. Thus, the algorithm can accurately detect small target defects on wind turbine blades and is sufficiently lightweight for practical use cases.

Under diverse conditions, the vibration signals of complex rotating machinery exhibit non-stationary behavior, multi-component characteristics, closely spaced frequencies, and non-proportionality, posing challenges to conventional time-frequency analysis (TFA) methods. These limitations hinder accurate instantaneous frequency (IF) estimation and time-frequency representation (TFR) construction, directly impacting machinery fault diagnosis. As such, we propose the Local Entropy Selection Scaling-Extracting Chirplet Transform (LESSECT), which optimizes entropy-based chirp rate (CR) selection to match non-proportional fundamental frequencies. By adaptively selecting multiple CRs at the same time center, LESSECT enhances TFR resolution and energy concentration, leading accurate IF identification. Experimental validation on bat echolocation, bearing fault, and planetary gearbox signals shows its superior performance in resolving non-proportional, closely spaced IFs. This significantly improves state estimation and enhances machinery diagnostics, contributing to greater system reliability.

Under time-varying operating conditions, the instantaneous frequency (IF) of the vibration signal of the planetary gearbox exhibits non-stationary time-varying closely spaced characteristics as well as non-proportional and non-synchronous characteristics. This in turn leads to challenges for traditional time-frequency analysis (TFA) methods to accurately identify its fault features and obtain accurate time-frequency representations (TFR). To address this challenge, in this study, a TFA method is proposed based on Synchronous Decomposition Match-Reassigning Transform (SDMRT). First, the successive variational mode decomposition (SVMD) is used to adaptively separate non-synchronous frequency components in the original vibration signal. Second, the chirp rate (CR) describing the signal harmonic structure is used to synchronously match each frequency component, obtaining the TFR corresponding to each component. Third, all TFRs are merged to obtain a complete TFA result. Finally, the energy spread is re-aggregated to the ridge of the IF using the reassignment principle, resulting in a new frequency estimation operator termed the Synchronous Decomposition Match-Reassigning Operator (SDMRO). SDMRT can adaptively separate non-proportional and non-synchronous IFs and realize precise matching of time-varying closely spaced IFs, thereby completing the TFR of the vibration signal generated by the planetary gearbox. By analyzing simulated signals and measured vibration signals from planetary gearboxes, Rényi entropy and IF estimation errors of SDMRT correspond to 12.14% and 0.59%, respectively. The energy concentration degree and IF estimation results are significantly better than other TFA methods, which proves the effectiveness and superiority of SDMRT.