基于多性能退化参数的装备关键系统实时可靠性评估与预测方法研究

For the long life, small samples and non-failure equipment, the reliability assessment is very difficult，which is limited by the number of samples data. Considering the characteristics of multi-parameters and real-time data for performance degradation, the project studies the methods of real-time reliability assessment and prediction based on multi-parameter performance degradation. For the multi-parameters characteristics, the project proposes the data fusion technique based on the algorithm of support vector data description(SVDD) to form a comprehensive performance parameter, and then forecasts the trend of the comprehensive performance parameter based on the algorithm of support vector regression (SVR); for the real-time characteristics, the Bayes methods is proposed and the real-time updating model is constructed based on the SVDD-SVR-Bayes techniques for comprehensive performance parameter, which integers the historical information and real-time data and accurately reflect the actual performance of the equipment. The project will address the real-time reliability assessment for rare failure of equipment, further enrich and perfect the existing reliability theory and methods, and provide technical support for the prediction and health management of critical systems and the application on reliability assessment in engineering.

本项目面向装备长寿命、小样本和无失效的可靠性评估问题，从装备的性能退化数据入手，针对性能退化数据的多参数与实时性等特点，研究基于多性能退化参数的装备实时可靠性评估与预测方法。针对多参数特点，提出基于支持向量数据描述（Support Vector Data Description, SVDD）方法的多参数信息融合技术与基于支持向量回归分析（Support Vector Regression, SVR）算法的综合性能指标可靠性变化趋势预测；针对实时性问题，提出并构造了基于SVDD-SVR-Bayes（贝叶斯）技术的综合性能指标分布参数的实时更新模型，融合了历史信息和实时数据信息，准确反映了装备的实际性能。本项目的研究将解决极少失效下装备的实时可靠性评估问题，进一步丰富和完善现有的可靠性理论和方法，为装备关键系统的故障预测及健康管理提供技术支撑，为可靠性评估在工程上的应用提供技术解决方案

对于高可靠性、长寿命以及试验经费高，周期长的设备而言，基于性能退化的可靠性评估相比传统的基于二元失效数据（正常和故障）的可靠性理论具有更大的实际意义。本项目系统深入地研究了基于性能退化数据的可靠性评估方法，利用装备性能退化数据来识别部件性能退化过程，综合考虑了多参数性能退化数据的信息融合方法和可靠性指标趋势预估方法，提出了基于支持向量数据描述（Support Vector Data Description, SVDD）的多性能参数退化数据的信息融合方法，并利用了支持向量回归（Support Vector Regression, SVR）算法预测SVDD距离的变化趋势，建立了装备失效与性能退化之间的数学模型，融合了装备历史信息和实时数据信息，构造了基于SVDD-SVR-Bayes（贝叶斯）技术的综合性能指标分布参数的实时更新模型，最后通过实车数据实现了某型装甲车辆机械系统的可靠性预估。本项目的研究解决了极少失效下装备的实时可靠性评估问题，为装备关键系统的故障预测及健康管理提供了技术支撑。