Abstract: 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：① only 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; ② the 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; ③ optimized 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; ④ generalized 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.

Key words: generalized health indices, spectral amplitude fusion, degradation properties, sparsity properties, machine degradation assessment, convex optimization