视情维修因能很大程度上避免“欠维修”和“过维修”的发生，为企业降低成本和风险故尤为重要。现有的视情维修理论中还存在以下不足：一是滤波去噪类故障检测方法有时会将微弱故障信号滤掉，从而造成故障漏检；二是现有的设备衰退预测模型大多以一定量的失效样本或截尾样本为建模基础，当样本受限且数据集不平衡时难以建立预测模型；三是大多维修决策模型忽略了设备衰退演化实时预测结果和维修备件可用性的影响。针对上述不足，本项目提出了基于多稳态随机共振模型的微弱故障检测方法，提高了对单一微弱故障和多微弱故障同时检测的效率；提出了基于深度学习理论和随机过程的设备衰退演化预测模型，解决在不同受限程度及数据集不平衡情况下的设备衰退预测问题，揭示了维修行为对可修设备退化的影响；考虑不同维修备件的可用性，结合设备衰退演化实时预测结果，建立了设备视情维修和备件订购集成优化模型。本项目的研究将完善视情维修理论，提高其工程实践性。

Predictive maintenance is important because it can avoid “over maintenance” and “less maintenance” to a great extent. Consequently, cost and risk can be declined for a company. The current predictive maintenance is deficient. First weak fault signal may be filtered and fault may be missed detected if denoising or filtering methods are used for weak fault detection. Second, the current equipment deterioration evolution prognosis methods are based on amount of failure samples or suspension samples which datasets are balanced. Prognosis models are hard to be built if samples are limited and unbalanced. Third, most of the maintenance decision models ignore the real-time prognosis results of equipment deterioration evolution and spare availability. To solve the problems above, a weak fault detection method is proposed based on multi-stable stochastic resonance. The fault detection efficiency is improved for a single weak fault and multiple faults. New equipment deterioration evolution prognosis methods are developed based on deep learning and stochastic processes. The problems of equipment deterioration evolution prognosis are solved considering the limited extent and imbalance. Considering the availability of different spares and real-time prognosis results of equipment deterioration evolution, a joint optimization model of equipment predictive maintenance and spare ordering. This study with improve the theory and application of predictive maintenance.