Physics-informed probabilistic neural network for reliable condition assessment and failure prognosis of substructures in floating offshore wind turbines

Xu, Z ORCID: 0000-0003-2661-517X (2026) Physics-informed probabilistic neural network for reliable condition assessment and failure prognosis of substructures in floating offshore wind turbines. Engineering Structures, 366. ISSN 0141-0296

Preview

Text Physics-informed probabilistic neural network for reliable condition assessment and failure prognosis of substructures in floating offshore wind turbines.pdf - Published Version Available under License Creative Commons Attribution. Download (10MB) | Preview

Abstract

The structural integrity of substructures in Floating Offshore Wind Turbine (FOWTs) is critical importance and efficient and reliable health condition assessment methods are essential to ensure their long-term reliability and safety. Modern intelligent prognostic approaches based on deep neural networks are generally developed by learning the mapping between theoretical health condition with the monitoring observations. However, their performance can deteriorate significantly when failure data are limited or unavailable. To address these challenges, this study proposes a reliable intelligent prognostic framework capable of predicting failures even in the absence of direct failure observations. The proposed framework consists of three main modules: (i) a robust health indicator construction module for extracting representative degradation information from monitoring data; (ii) a Physics-Informed Dual Decoder prognostic module, which incorporates physics-informed inductive bias to constrain latent fatigue-evolution dynamics and predict future health states; and (iii) a reliability assessment module for quantifying structural reliability based on the prognostic results. The framework's effectiveness is validated through fatigue degradation analysis on three typical FOWT substructures. The results show that, as structural fatigue degradation progresses, degradation-related dynamic response changes can be effectively captured by the proposed TDA-VMD-based health indicator construction method, leading to more robust health assessment features. Furthermore, by introducing physics-informed inductive bias to constrain the latent dynamics of fatigue evolution, the proposed PIDualDecoder model enables more physically consistent and reliable fatigue prognosis, particularly for long-term prediction. These improved prognosis results further provide more confident support for reliability assessment, thereby contributing to safer and more cost-effective operation and maintenance of FOWTs.

Item Type:	Article
Uncontrolled Keywords:	Prognostics and health management; Floating offshore wind turbine; Uncertainty quantification; Physics-informed Machine Learning; Lifecycle Reliability; 4005 Civil Engineering; 40 Engineering; Machine Learning and Artificial Intelligence; 7 Affordable and Clean Energy; 0905 Civil Engineering; 0912 Materials Engineering; 0915 Interdisciplinary Engineering; Civil Engineering; 4005 Civil engineering; 4016 Materials engineering
Subjects:	Q Science > QA Mathematics > QA76 Computer software T Technology > TA Engineering (General). Civil engineering (General)
Divisions:	Engineering and Built Environment
Publisher:	Elsevier
Date of acceptance:	25 June 2026
Date of first compliant Open Access:	1 September 2026
Date Deposited:	01 Sep 2026 14:20
Last Modified:	01 Sep 2026 14:20
DOI or ID number:	10.1016/j.engstruct.2026.123282
URI:	https://researchonline.ljmu.ac.uk/id/eprint/29253

View Item