Shen Weijie, Xiao Maohua, Wang Zhenyu, Song Xinmin
Zhejiang Technical Institute of Economics, Hangzhou 310018, China.
College of Engineering, Nanjing Agricultural University, Nanjing 210031, China.
Sensors (Basel). 2023 Jul 24;23(14):6645. doi: 10.3390/s23146645.
This study targets the low accuracy and efficiency of the support vector machine (SVM) algorithm in rolling bearing fault diagnosis. An improved grey wolf optimizer (IGWO) algorithm was proposed based on deep learning and a swarm intelligence optimization algorithm to optimize the structural parameters of SVM and improve the rolling bearing fault diagnosis. A nonlinear contraction factor update strategy was also proposed. The variable coefficient changes with the shrinkage factor . Thus, the search ability was balanced at different early and late stages by controlling the dynamic changes of the variable coefficient. In the early stages of optimization, its speed is low to avoid falling into local optimization. In the later stages of optimization, the speed is higher, and finding the optimal solution is easier, balancing the two different global and local optimization capabilities to complete efficient convergence. The dynamic weight update strategy was adopted to perform position updates based on adaptive dynamic weights. First, the dataset of Case Western Reserve University was used for simulation, and the results showed that the diagnosis accuracy of IGWO-SVM was 98.75%. Then, the IGWO-SVM model was trained and tested using data obtained from the full-life-cycle test platform of mechanical transmission bearings independently researched and developed by Nanjing Agricultural University. The fault diagnosis accuracy and convergence value of the adaptation curve were compared with those of PSO-SVM (particle swarm optimization) and GWO-SVM diagnosis models. Results showed that the IGWO-SVM model had the highest rolling bearing fault diagnosis accuracy and the best diagnosis convergence.
本研究针对支持向量机（SVM）算法在滚动轴承故障诊断中精度和效率较低的问题。基于深度学习和群体智能优化算法，提出了一种改进的灰狼优化器（IGWO）算法，用于优化SVM的结构参数，提高滚动轴承故障诊断能力。还提出了一种非线性收缩因子更新策略。可变系数随收缩因子变化。因此，通过控制可变系数的动态变化，在不同的早期和晚期阶段平衡搜索能力。在优化的早期阶段，其速度较低，以避免陷入局部优化。在优化的后期阶段，速度较高，更容易找到最优解，平衡两种不同的全局和局部优化能力以完成高效收敛。采用动态权重更新策略，基于自适应动态权重进行位置更新。首先，使用美国凯斯西储大学的数据集进行仿真，结果表明IGWO-SVM的诊断准确率为98.75%。然后，使用南京农业大学自主研发的机械传动轴承全生命周期试验平台获得的数据对IGWO-SVM模型进行训练和测试。将适应曲线的故障诊断准确率和收敛值与PSO-SVM（粒子群优化）和GWO-SVM诊断模型进行比较。结果表明，IGWO-SVM模型具有最高的滚动轴承故障诊断准确率和最佳的诊断收敛性。
