基于多SAE的机械臂运行可靠性深度特征提取与融合

Reliability Depth Feature Extraction and Fusion of Robotic Arm Based on Multi-SAE

在对机械臂系统运行可靠性进行综合评估时,针对现有工程应用中特征提取的层次结构和评估指标较为单一的问题,提出一种基于多稀疏自编码器(SAE)的深度融合特征构建方法。首先在多维统计特征提取的基础上,引入变分模态分解下各模态分量的样本熵特征,然后采用SAE对统计特征进行多层次的编码与解码,并以设备退化性能的标签值对整个SAE结构模型参数进行反向微调,从而将系统各关键部位的退化信息融入到SAE模型中,最后采用深度神经网络模型对系统运行可靠性进行评估。试验结果表明,提出的基于多SAE模型可自适应地提取出更能表征机械臂运行可靠性的深层融合特征,能有效提高后续评估模型的准确性和鲁棒性。

In the comprehensive assessment of the operational reliability of robotic arm systems,a deep fusion feature construction method based on stacked auto-encoder(SAE)is proposed to address the problem that the hierarchical structure of feature extraction and evaluation indexes are relatively single in existing engineering applications.Firstly,on the basis of multi-dimensional statistical feature extraction,the sample entropy features of each modal component under variable modal decomposition are introduced,and then SAE is used to encode and decode the statistical features at multiple levels,and the parameters of the whole SAE structure model are fine tuned in reverse with the labeled values of device degradation performance,so that the degradation information of each key part of the system can be incorporated into the SAE model,and finally the system operational reliability is evaluated by using a deep neural network model.The experimental results show that the multi-SAE-based model proposed in this paper can adaptively extract deep fusion features that can better characterize the operational reliability of the robotic arm,which can effectively improve the accuracy and robustness of the subsequent evaluation model.