In this paper, a novel parametric model-based decomposition method is proposed for structural health monitoring of time-varying structures. For this purpose, the advanced Functional-Series Time-dependent Auto Regressive Moving Average (FS-TARMA) technique is used to estimate the parameters and innovation variance used in the parametric signal decomposition scheme. Additionally, a unique feature extraction and reduction method based on the decomposed signals, known as Latent Components (LCs), is proposed. To evaluate the effi ciency of the proposed method, numerical simulation and an experimental study in the laboratory were conducted on a time-varying structure, where various types of damage were introduced. The Fuzzy Expert System (FES) was used as a classifi cation tool to demonstrate that the proposed method successfully identif ies different structural conditions when compared with other methods based on non-reduced and ordinary feature extraction. In this paper, a novel parametric model-based decomposition method is proposed for structural health monitoring of time-varying structures. For this purpose, the advanced Functional-Series Time-dependent Auto Regressive Moving Average (FS-TARMA) technique is used to estimate the parameters and innovation variance used in the parametric signal decomposition scheme. A unique feature extraction and reduction method based on the decomposed signals, known as Latent Components (LCs), is proposed. To evaluate the effi ciency of the proposed method, numerical simulation and an experimental study in the laboratory were conducted on a time-varying structure, where various types of damage were introduced. The Fuzzy Expert System (FES) was used as a classifi cation tool to demonstrate that the proposed method successfully found ies different structural conditions when compared with other methods based on non-reduced and ordinary feature extraction.