基于SVDD和PF的剩余使用寿命预测方法研究

Remaining Useful Life Prediction Method Based on SVDD and PF

通常情况下剩余使用寿命(Remaining Useful Life, RUL)预测的实现都是基于各种传感器监测得来的数据,如何从具有过程噪声的连续采集的大量数据中提取退化特征来监测系统的性能退化趋势,从而准确地预测出RUL是一项挑战。因此,提出了一个基于支持向量数据描述(Support Vector Data Description, SVDD)和粒子滤波(Particle Filter, PF)的RUL预测方法。首先,利用SVDD从大量历史数据中提取一个可以反映系统性能退化趋势的健康指标;其次,针对健康指标的退化趋势构建一个退化模型,同时可以相应地确定一个故障阈值;然后,基于PF算法和自动确定的故障阈值来实现RUL的准确预测;最后,利用航空发动机退化数据集验证了所提出方法的有效性和优越性。

In general, the implementation of remaining useful life(RUL) is based on data monitored by various sensors.However, it is a challenge to extract the degradation characteristics from the continuously collected large amount of data with process noise to monitor the performance degradation trend of the system, and then accurately predict RUL.Therefore, a RUL prediction method based on support vector data description(SVDD) and particle filter(PF) is proposed.Firstly, SVDD is used to extract a health indicator that can reflect the degradation trend of the system performance from a large amount of historical data.Secondly, a degradation model is constructed according to the degradation trend of health indicators, and a fault threshold can be determined accordingly.Then, the accurate prediction of RUL is realized based on PF algorithm and automatically determined fault threshold.Finally, the effectiveness and superiority of the proposed method are verified by using the aero-engine degradation data set.