基于高维异构数据的电动汽车变速器早期故障识别方法研究

Electric vehicle is the key area of national development, transmission is the core component of electric vehicle, and its early fault identification is related to the operation safety of electric vehicle. Because the transmission of electric vehicle runs in enclosed space, the fault signal is reflected and refracted seriously, so it is difficult to get accurate and non-interference monitoring signal; during the operation of electric vehicle, many components outside the transmission produce resonance, the early fault signal is weak, and it is difficult to extract effective features under strong noise; the early fault signal of transmission has the characteristics of high-dimensional heterogeneity and large amount of data, and its fault types. The nonlinearity between the signal and the monitoring signal is strong and the complexity is high. Focusing on the difficult and scientific problems of early fault identification, the project analyses the characteristics of the monitoring data of transmission operation of electric vehicles, studies the transmission mechanism of fault signals in enclosed space, designs a high-dimensional fault feature representation method of dynamic depth neural network, and constructs a multi-source and heterogeneous fault signal identification method of multi-scale depth neural network, which is parallel with theoretical experiments and engineering verification. Improve the accuracy of early fault identification. In order to break through the difficult problems of electric vehicle fault identification and improve the theory and method of early fault identification, it is helpful to ensure the operation safety of electric vehicle.

电动汽车是国家发展重点领域，变速器是电动汽车核心部件，其早期故障识别关系电动汽车运行安全。由于电动汽车变速器运行在封闭空间，故障信号反射折射严重，难得到准确、无干扰的监测信号；电动汽车运行过程中变速器以外众多部件产生共振，早期故障征兆信号弱，强噪声下难提取有效特征；变速器早期故障信号具有高维异构和数据量大的特点，其故障类别与监测信号之间非线性强和复杂度高。项目围绕早期故障识别存在的难点科学问题，分析电动汽车变速器运行监测数据的特性，研究封闭空间内故障信号传输机理，设计动态深度神经网络的高维故障特征表征方法，构建多尺度深度神经网络的多源异构故障信号识别方法，采用理论实验与工程验证并行，提高早期故障识别精度。以实现电动汽车故障识别难点问题的突破，完善早期故障识别的理论和方法，有助于保障电动汽车运行安全。

变速器是电动汽车核心部件，其早期故障识别关系电动汽车运行安全。由于电动汽车变速器运行在封闭空间，故障信号反射折射严重，难得到准确、无干扰的监测信号；电动汽车运行过程中变速器以外众多部件产生共振，早期故障征兆信号弱，强噪声下难提取有效特征；变速器早期故障信号具有高维异构和数据量大的特点，其故障类别与监测信号之间非线性强和复杂度高。本项目围绕早期故障识别存在的难点科学问题，得到以下理论成果：分析了电动汽车变速器运行监测数据的特性，建立了基于对抗生成网络的变速器振动信号特性分析，得到了影响变速器54个变量中前12个变量的高斯分布；设计了基于边界约束生成对抗网络的变速器早期故障识别模型；构建了变速器振动信号动态神经网络模型的高维故障数据的表征模型。本课题采用理论实验与工程验证并行，提高早期故障识别精度。以实现电动汽车故障识别难点问题的突破，完善早期故障识别的理论和方法，有助于保障电动汽车运行安全。 ..本课题研究成果应用于重庆青山工业有限公司和泸州容大智能变速器有限公司的变速器试验台架和变速器设计中，并以排名第一获得重庆市科技进步奖三等奖1项“MF620汽车变速器关键技术及产业化”。同时培养研究生2名，青年教师2名，博士后出站1名。发表SCI期刊论文6篇，申请专利3项。..研究成果在IICSPI2020国际会议（项目负责人担任程序委员会主席）、第一届蓝瑙国际创新合作论坛（中国-奥地利）、长安汽车组织的AILAB交流会上进行展示，并作大会报告，广受好评。