基于数字孪生模型的机载光电探测系统性能退化预测理论与方法

The project aims at establishing a degradation model for Electro-Optical Detection System on the basis of the concept of Digital Twin. Based on this model, a corresponding prediction algorithm to predict the degrading system state in dynamic and uncertain environment is proposed. Examples combined with historical data and simulation data are supposed to demonstrate the research results...In this model, four layers are proposed. The influence of sub-system failure and degradation is firstly uniformly mapped into an energy domain with the help of Modulation Transfer Function (MTF), which is able to solve the problem of multi-field coupling existed in the complex Electro-Optical Detection System. Secondly, the Dynamic Bayesian Network (DBN) is used to model the common rule of the degradation, which describes the dynamic degradation process and quantifies uncertain factors. Meanwhile, degradation modes are extracted through historical data. Thirdly, a new structure of non-parametric DBN model is proposed to autonomously update the model structure and parameters according to individual’s situation. Finally, the established non-parametric DBN model is mapped to an illustrative DBN model in order for the online prediction of degrading system state. Quantized particle filtering and just in-time learning are combined in the prediction algorithm to reduce the computational burden and uncertainties...This project can provide theoretical support for the degradation prediction of the Electro-Optical Detection System based on outfield data, and is of great significance for improving the system safety and reducing the maintenance cost.

本项目依据数字孪生模型的思想，建立一套描述光电探测系统性能退化的模型体系，并在此基础上提出动态不确定环境下的系统整体性能预测算法，最后结合历史数据与仿真数据验证研究成果。借助调制传递函数将子系统故障和退化对性能的影响统一映射到能量域，解决系统性能退化多场耦合的问题。在能量域模型基础上，基于动态贝叶斯网络（DBN）对系统性能退化共有规律建模，描述系统退化的动态过程和量化不确定性因素，同时通过历史数据挖掘系统的退化模式。在此基础上进一步建立非参数DBN模型，根据特定个体情况自主学习模型结构和参数，增加模型对特定个体的描述能力。最后，针对所提出的模型体系，提出一种基于软分类、量化粒子滤波与即时学习相结合的性能退化预测算法框架，实现性能退化的预测并减小预测过程中的不确定性。项目研究能够为基于外场数据的光电探测系统性能退化预测提供理论算法支持，对降低光电探测系统维护成本及提高设备安全性具有重要意义。

机载光电探测系统是目前先进作战飞机广泛装备的光机电一体化火控武器系统，主要用于对目标被动搜索、精确跟踪及武器制导，是能够对抗新一代雷达隐身飞机的主要利器，其性能优劣将直接关系到战机的作战效能。本项目完成了拟定的研究目标。首先，实现了机载光电探测系统性能退化能量域统一描述，突破了复杂交联多场耦合性能退化建模难题；其次，提出一套复杂系统性能退化建模体系，应用机载光电探测系统性能退化预测理论与方法，实现了复杂系统退化能量表征先验知识、性能退化共有规律、特定个体个性化知识的混合集成；其三，面向机载光电探测系统建立了具有动态演化、自主学习能力的性能退化数字孪生模型，能够根据特定个体展开模型结构和参数的自学习；最后，设计了高效的非参数DBN模型近似推理算法，采用多重不确定性管理与控制策略，实现了动态运行不确定环境下复杂系统性能退化的高效推理、准确预测。不仅发表了多篇高水平学术论文、申请和授权了多项国家发明专利与软件著作权，且通过了实际机载光电探测系统的测试验证，能够满足部队复杂武器装备的视情维修和自主保障需求，提高了飞机、舰船等复杂武器装备的自主健康管理能力和任务执行效能，具有重要的国防应用价值。