Metaevaluation: a comprehensive evaluation of health indicator on real options-based maintenance scheduling and health prognostics of bearing degradation
研究了如何通过传感器数据构建健康指标来评估轴承退化，并提出“元评估”方法（含五个指标）比较不同健康指标，结合实物期权框架优化维护调度决策。
This study focuses on improving maintenance strategies in the competitive manufacturing sector, emphasising the role of predictive maintenance (PdM) through sensor data and the creation of robust health indicators for proactive planning. The study analyzes high-frequency vibration data from aging bearings. By extracting and selecting key features, it identifies crucial time-domain and frequency-domain characteristics to develop health indicators which are used to evaluate the machinery health. To account for the linear, non-linear patterns and uncertainties in the degradation process, we propose ‘metaevaluation’ (i.e., evaluation about evaluation), a comprehensive evaluation to evaluate the health indicator, with five metrics including Monotonicity, Prognosability, Trendability, Reliability and Internal Robustness. We build metaevaluation for the time-dependent and noise-resilient health indicators, including Principal Component Analysis (PCA), Independent Component Analyzes (ICA), and Convolutional Variational Autoencoder (CVAE). The study employs the Exponential Wiener Process (EWP) to address the complexities and uncertainties of aging equipment. Though real options framework and Monte Carlo deterioration simulation, we are able to extended operation and establish a maintenance decision considering multiple operational criteria. This method is key in pinpointing optimal maintenance time from premature scheduling based on fixed thresholds.