一类复杂系统精细建模与多级多粒度故障诊断方法研究

Permanent magnet synchronous motor is a typical nonlinear complex system. With its high torque density, high efficiency and high reliability, has become the mainstream motor in the field of active aircraft, electric vehicles, industrial servo drives, etc. However, once the motor fault occurs, the output torque is not controlled and will affect its stability, thus endangering the user's life safety. The existing method of fault diagnosis based on an integer-order motor model has not considered the fractional-order characteristics contained in the electromagnetic coupling and friction in the motor system, resulting in poor results, and is only based on deviations or residual signals of the system's physical numerical output, thus the description of faults and essential changes, as well as the uniform and efficient diagnosis of multiple faults are difficult. Therefore, based on fractional calculus modeling and multi-granularity measure of residual sequence, a new fault diagnosis method for fractional-order complex system is established: multi-level fine modeling based on fractional calculus representation; multi-stage observer and multi-granularity (Euclidean distance, relative change rate, gap metric) residual generator, classifier design, for the purpose of fault detection, classification and severity assessment, and experimental verification. Focus on solving the key scientific issues encountered in the study: multi-parameter identification for the fractional transfer function model, design of the multi-level multi-granularity residual generator and classifier.

永磁同步电机是一类典型的非线性复杂系统，凭借其高转矩密度、高效率和高可靠性等优良性能，成为现役飞机、电动汽车和工业伺服驱动等领域主流电机；然而，一旦电机发生故障，其不受控制的输出力矩会影响自身及系统稳定性，从而危及使用者的生命安全。现有基于电机整数阶模型的故障诊断，因未考虑描述电机系统中电磁耦合和摩擦力等包含的分数阶特性，导致效果欠佳，且仅基于系统物理数值输出的偏差或残差信号，难以呈现对故障的刻画和本质变化，以及多种故障统一高效诊断。为此，基于分数阶微积分建模和残差序列多粒度度量，建立一类具有分数阶特性复杂系统的故障诊断新方法：基于分数阶微积分表示的多级精细建模；多级观测器和多粒度（欧氏距离、相对变化率、间隙度量）残差生成器、分类器设计，实现故障检测、分类和严重程度评估，并进行实验验证。重点解决研究中所遇到的关键科学问题：分数阶传递函数模型多参数辨识，多级多粒度残差生成器和分类器设计等。

永磁同步电机是一类典型的非线性复杂系统，凭借其高转矩密度、高效率和高可靠性等优良性能，成为现役飞机、电动汽车和工业伺服驱动等领域主流电机；然而，一旦电机发生故障，其不受控制的输出力矩会影响自身及系统稳定性，从而危及使用者的生命安全。现有基于电机整数阶模型的故障诊断，因未考虑描述电机系统中电磁耦合和摩擦力等包含的分数阶特性，导致效果欠佳，且仅基于系统物理数值输出的偏差或残差信号，难以呈现对故障的刻画和本质变化，以及多种故障统一高效诊断。为此，项目具体实现了：（1）采用改进差分进化算法用于辨识分数阶模型参数，以增加种群的多样性并提高搜索能力。在降低计算量的同时，兼顾了算法的辨识精度和动态收敛速度，具有更优的全局搜索能力。（2）在传统整数阶和分数阶模型状态空间描述基础上，利用代数方法解耦建模误差、扰动及噪声，实现多级精细模型确定性状态空间描述，基于建立的多级模型和观测器，在每个级别上用基于状态输出变量和观测输出变量，分别采用欧氏距离、相对变化率、间隙度量等度量方式，生成多粒度下的残差序列及多粒度区间序列。（3）基于残差序列实时数据，利用设计好的多级多粒度分类器，实现故障分类。基于残差序列实时数据，在不同粒度度量下建立度量故障大小（严重程度）的阈值序列比较，评估故障严重程度。共撰写了研究论文16篇，已发表学术论文12篇（SCI检索论文5篇，EI检索发表论文5篇，已授权发明专利4项）。