Degradation Prediction and Maintenance of Three-phase Inverters Under Fluctuating Temperature Conditions

Abstract

Three-phase inverters, as core power conversion units in modern energy systems, have an operational reliability that is paramount to system safety and stability. Under fluctuating thermal stress arising from both external environments and internal operations, the accurate degradation prediction and proactive maintenance of these inverters face significant challenges. Current research reveals several deficiencies: First, the independent or coupled effects of different thermal stress levels on component degradation remain unclear, challenging the theoretical foundation and applicability of conventional cumulative degradation models. Second, the selection of temporal granularity for degradation accumulation lacks a scientific basis, diminishing the rationality and accuracy of prediction results. Furthermore, existing preventive maintenance strategies fail to effectively incorporate prediction uncertainties, leading to insufficient decision robustness. To address these issues, this dissertation conducts the following research:

First, this study investigates component cumulative degradation modeling methods considering path dependence under thermal stress fluctuations. Through dynamic stress degradation tests integrating mixed-effects models and non-parametric statistical tests, the degradation path dependence caused by coupled stress levels is identified, revealing the degradation characteristics of key components under fluctuating thermal stress. Based on this, a path adjustment factor is introduced to correct the degradation rate, and a latent variable iterative algorithm for its solution is developed to accurately describe path-dependent features. The model's accuracy and superiority are validated through application in typical scenarios and comparison with experimental data.

Second, this study explores modeling and thermal stress calculation methods for three-phase inverters considering multi-scale fluctuating environmental temperature. To tackle the challenges posed by the complexity of ambient temperature profiles and the high computational cost of high-fidelity simulations, Seasonal and Trend decomposition techniques are introduced for multi-scale feature extraction, and an equivalent representation of complex profiles is achieved through a stress level discretization strategy. Subsequently, a high-fidelity digital model of the inverter is built and validated. A physics-informed neural network-based surrogate model is further developed, establishing an efficient and precise computational pathway for calculating thermal stresses on internal key components from external environmental conditions.

Third, this study develops a system degradation prediction method based on temporal granularity convergence and its verification method under small-sample constraints. At the prediction theory level, the impact of cumulative calculation temporal granularity on the numerical solution of nonlinear degradation processes is analyzed, and a selection criterion based on convergence assessment is proposed. For experimental verification, a theoretical framework for verification test design and error evaluation is constructed, integrating a priori prediction information with Bootstrap resampling techniques to address small-sample constraints. The mechanisms by which sample size and significance level affect the effectiveness of the prediction error evaluation are revealed, and the effectiveness of the proposed prediction and verification framework is confirmed through experiments.

Finally, this study investigates a robust optimization method for preventive maintenance of three-phase inverters considering prediction uncertainty. Through a comparative analysis of different decision-making paradigms, robust optimization models are established for two typical scenarios: high-reliability and economy-priority applications. To address the black-box and non-convex characteristics of the established models, a metaheuristic solving algorithm framework is developed to systematically balance maintenance costs and failure risks, thereby formulating robust maintenance strategies that accommodate both reliability and economy. The superiority of the proposed method is validated through a specific case study of the three-phase inverter, in comparison with traditional deterministic methods.

Date of Award	5 Jan 2026
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Xuerong YE (External Supervisor) & Min XIE (Supervisor)