基于谱幅融合广义健康指数的可解释装备退化评估优化模型研究进展

Advances in Interpretable Machine Degradation Assessment Optimization Models Based on Spectral Amplitude Fusion Aided Generalized Health Indices

针对装备全生命周期退化评估所需的健康指数存在构建过程繁琐且未考虑退化特性、构建模型和学习权重物理难解释、退化趋势可分性和单调性不显著以及难以同步实现装备状态监测、早期故障诊断和退化评估等问题,总结近年来基于谱幅融合广义健康指数的可解释装备退化评估优化模型研究成果,探讨基于退化特性和故障特征稀疏性的两类典型可解释广义健康指数权重优化模型及其应用效果。该类优化模型核心思想是将谱幅(如频域幅值或包络谱域幅值)加权和定义为广义健康指数,然后基于可分性、单调性等退化特性和故障特征稀疏特性推导出相关的多种广义健康指数权重凸优化模型。与现有模型相比,所提出的基于谱幅融合广义健康指数的可解释装备性能退化评估优化模型具有以下特点:①仅需工程常用的傅里叶变换或希尔伯特变换作为原始数据预处理方法,避免使用复杂信号处理算法及其参数优化过程;②广义健康指数权重凸优化模型构造来源于退化特性和故障特征稀疏性的凸优化描述,因此权重优化模型具有可解释性和全局最优解;③优化的谱幅权重具备频域故障物理可解释性,谱幅权重可定位信息频率带并揭示故障特征频率来支撑广义健康指数退化评估合理性;④基于谱幅加权的广义健康指数可同时实现装备状态监测、故障诊断和退化评估三重目的,利于高效实施装备智能运维。最后,对基于谱幅融合广义健康指数的未来研究方向进行充分探讨。

Since health indices required for machine full lifecycle degradation assessment have problems such as cumbersome construction process,difficulty in explaining model construction and learning weights physically,insignificant separability and monotonicity of degradation trends,and difficulty in synchronously implementing machine condition monitoring,early fault diagnosis,and degradation assessment,research progresses of interpretable machine degradation assessment optimization models based on spectral amplitude fusion aided generalized health indices in recent years are summarized and two kinds of typical interpretable generalized health index weight optimization models based on degradation properties and fault feature sparsity and their application effects are explored.The core idea of this type of optimization models is to define the weighted sum of spectral amplitudes(such as amplitudes in the frequency domain or the envelope spectral domain)as a generalized health index,and then derive various generalized health index weight convex optimization models based on the degradation properties such as separability and monotonicity,as well as the sparsity properties of fault features.Compared with existing models,the proposed interpretable machine performance degradation assessment optimization models based on spectral amplitude fusion aided generalized health indices has the following characteristics:①only the commonly used Fourier transform or Hilbert transform in engineering are required as raw data preprocessing methods,avoiding the use of complex signal processing algorithms and parameter optimization processes;②the construction of generalized health index weight convex optimization models comes from the convex optimization description of degradation properties and the sparsity description of fault features,so various weight optimization models have interpretability and unique global optimal solutions;③optimized spectral amplitude weights have physical fault interpretability in the frequency domain,which can locate informative frequency bands and reveal fault feature frequencies to support the rationality of generalized health indices for degradation assessment;④generalized health indices based on spectral amplitude weighting can simultaneously achieve the triple goals of machine condition monitoring,fault diagnosis,and degradation assessment,which is conducive to efficient implementation of intelligent machine operation and maintenance.Finally,some future research directions of generalized health indices based on spectral amplitude fusion are fully explored.