【作者】 魏星宇；

【导师】 李端玲；

【作者基本信息】 北京邮电大学 ， 机械工程， 2023， 硕士

【摘要】 在新能源车辆中,电机系统是车辆动力系统的关键组成部分,驱动电机出现故障可能会造成严重的后果。针对目前新能源车辆电机检修主要依靠人为检修、电机正常和故障数据获取较难且数量不平衡等问题,提出了结合数字孪生技术和复杂系统临界相变理论的故障预测方法;针对工业时间序列数据含有噪声难以利用等问题,提出了基于GRU-卷积降噪自编码器模型来进行电机系统的故障诊断方法。解决了少量样本时如何实现新能源车辆电机系统故障预测和诊断的问题,为新能源车辆智能运维开辟了新的思路。首先,基于本课题研究要求,设计出简化的数字孪生架构。分析新能源车辆电机系统在运行时的物理机理,建立正常电机系统机理孪生模型。进而具体研究故障发生机理,建立故障电机系统机理孪生模型,得到数字孪生数据在后续研究中备用。其次,结合复杂系统临界相变理论进行新能源车辆电机系统的故障预测研究。通过数字孪生模型得到电机运行时的随机波动信号,从中提取早期预警特征,通过多种检验方法检验了系统故障和复杂系统临界相变具有一致性,最终根据早期预警特征的关键跳变实现潜在故障的预测。最后,由于新能源车辆电机系统的各类故障数据都属于工业时间序列数据,通常故障诊断方法无法有效学习其时序特征,并且采集到的数据会掺入一定的噪声,因此提出了一种基于GRU-卷积降噪自编码器的故障诊断方法,实现了对电机各类故障的诊断,便于进行检修和维护。通过数字孪生数据和多组项目中采集的真实电机数据进行验证,结果表明,在两类数据上,提出的故障预测和诊断方法能在样本较少的情况下实现故障预测、对时间序列数据的故障诊断,且拥有较高的诊断准确率,为研究新能源车辆电机故障预测和诊断通用方法开辟了新思路。更多还原

【Abstract】 In new energy vehicles,the motor system is a key component of the vehicle power system,and the failure of the drive motor may cause serious consequences.In view of the current problem that motor overhaul in new energy vehicles mainly relies on human overhaul,and it is difficult and unbalanced to obtain normal and fault data of motors,a fault prediction method combining digital twin technology and the theory of critical phase change of complex systems is proposed;in view of the problem that industrial time series data contains noise and is difficult to use,a fault diagnosis method based on GRU-convolutional noise reduction selfencoder model is proposed for motor systems.The problem of fault prediction and diagnosis of new energy vehicle motor systems with a small number of samples is solved,filling a gap in the research on intelligent operation and maintenance of new energy vehicle motor systems.Firstly,a simplified digital twin architecture is designed based on the specific needs of this research.The physical mechanism of the motor system of the new energy vehicle is analysed during operation,a mechanistic twin model of the normal motor system is established,and then the failure occurrence mechanism is specifically investigated,a mechanistic twin model of the faulty motor system is established,and the digital twin data is obtained for backup in subsequent studies.Secondly,the fault prediction study of the motor system of new energy vehicles is carried out by combining the theory of the critical phase change of complex systems.The early warning features are extracted from the random fluctuation signals of the motor operation by means of a digital twin model,and the consistency between the system fault and the critical phase transition of the complex system is checked by various tests.Finally,a fault diagnosis method based on GRU-convolutional noise reduction self-encoder is proposed to solve the problem that all kinds of fault data of the motor system of new energy vehicles belong to industrial time series data,and the usual fault diagnosis methods cannot effectively learn its time series characteristics,and the collected data will be mixed with certain noise,etc.A fault diagnosis method based on GRU Convolution Denoising Auto-Encoder is proposed to realize the fault diagnosis of various kinds of motor faults,which is convenient for overhaul and maintenance.Validation is carried out with digital twin data and real motor data collected in multiple groups of projects.The results show that the proposed fault prediction and diagnosis method can be achieved fault prediction and fault diagnosis by time series data with fewer samples on both types of data.It possesses a high diagnostic accuracy rate.It opens up new ideas for the study of a general method for fault prediction and diagnosis of new energy vehicle motors.更多还原