The extraction of composite fault features in the fault diagnosis process of agricultural bearings has been challenging due to the influence of coupled propagation paths and strong background noise, an improved swarm decomposition （ISWD） method based on envelope spectrum correlation kurtosis is proposed to overcome this issue and achieve adaptive extraction of composite fault features for agricultural bearings. Firstly, the envelope spectrum correlation kurtosis, which is sensitive to periodic impacts, is employed as the fitness function to enhance the ability of SWD in extracting weak fault features. Secondly, an improved gray wolf optimization algorithm （GWO） is used to optimize the key thresholds Pth and Tth of SWD. Finally, the oscillatory component （OC） obtained by ISWD is subjected to envelope demodulation to highlight the fault feature frequencies and extract composite fault features of the bearings. Simulation and experimental analysis demonstrate that the proposed method can efficiently extract the features of composite faults in agricultural gearboxes. Compared with traditional Variational Mode Decomposition (VMD), the proposed method reduces the proportion of redundant components by 27% and increases the number of effective inner race features by 100%. Compared with SWD, not only does the proposed method increase the number of effective inner race features by 100%, but it also increases the number of effective outer race features by 25%. It provides a reference for the development of intelligent fault diagnosis methods for agricultural bearing faults.

针对农用轴承故障诊断过程中受到传播路径耦合与强烈背景噪声的影响，复合故障特征较难提取的问题，提出基于包络谱相关峭度的改进群分解（ISWD）方法，实现农用轴承复合故障特征的自适应提取。首先，利用对周期性冲击较为敏感的包络谱相关峭度为适应度函数，提升SWD对微弱故障特征的提取能力；其次，利用改进灰狼算法，实现SWD关键阈值Pth和Tth的寻优；最后，对ISWD分解出的振荡分量（OC）做包络解调处理，凸显故障特征频率，实现轴承复合故障特征的提取。仿真分析和试验分析表明，该方法能够高效提取农用齿轮箱复合故障的特征，相比于传统的变分模态分解(VMD)，减少27%的冗余分量占比，提高100%的内圈有效特征数量；与SWD相比，不仅内圈有效特征数量提升100%，外圈有效特征数量也提高25%，为农用轴承故障智能诊断方法的开发提供参考。