“机主人辅”模式下智能汽车故障诊断、预测与容错控制研究

With the online perception and intelligent fusion of the “Machine, Human, Vehicle and Road” multi-source information, for the goal of active security of intelligent vehicles under the driving mode of “Human-assistance and Machine-leading”, two scientific issues are “Fault Diagnosis and Prediction for Intelligent Vehicles based on Human-machine Intelligent Collaboration” and “Security-guaranteed Fault-tolerant Control of Active Hedge for Intelligent Vehicles”. We focus on three aspects: (1) intensive studying on the theory of fault diagnosis, prediction and fault-tolerant control for intelligent vehicles, (2) making breakthroughs on the key technology of active hedge under the driving mode of “Human-assistance and Machine-leading”, and even “Completely Autonomous Driving” and (3) with the combination of theoretical methods and key technological innovation, developing a demo of autonomous driving system for sudden changes of driving environment of complex roads of cities, and the condition of the missing or mistaken perception information. Our main work includes: (1) active security with a hierarchical framework and an information fusion model of human-machine intelligent information (a hierarchical framework of information perception, analytical inference, comprehensive decision and driving-control integration and a information fusion model of human-machine intelligent information of autonomous driving), and (2) fault diagnosis and prediction of “Machine, Human, Vehicle and Road” multi-source information cooperatively online perception and intelligent collaboration (“Machine, Human, Vehicle and Road” multi-source information cooperatively online perception and prediction, diagnosis and prediction of small faults of multi-source sensors, and online human-machine interaction and machine learning with the knowledge of fault diagnosis of “Human-assistance” and prediction of underlying dangers), and (3) active hedge fault-tolerant control strategy for the sudden changes of driving environment and the missing or mistaken perception information (dynamic display of minimal security neighborhood scenes and hedge control decisions, fault-tolerant dynamic reconstruction of minimal security neighborhood scenes with the failures of a part of traffic sensors, and security-guaranteed fault tolerant control strategy with the failures of a part of vehicle sensors). ..Our project research results are devoted to, on the one hand, provide the theoretical and technological basis of the development of our national intelligent vehicle driving technology and the industrialization of the new generation of automobile active security system, and on the other hand, to develop a data-support and low-cost verification platform of our national vehicle autonomous driving system design.

基于“机、人、车、路”多源信息的在线感知和智能融合，针对“机主人辅”驾驶模式下智能汽车以主动安全为目的的两个科学问题，即“基于人机智能融合的故障诊断和预测”和“主动避险的保安全容错控制”，深入研究智能汽车故障诊断、预测和容错控制理论，突破智能汽车在“机主人辅”乃至“完全自主”驾驶模式下主动安全的核心关键技术；结合理论方法与关键技术创新，研制面向城市复杂道路环境的、能应对行驶环境突变和自身感知信息缺失或错误的智能汽车自主驾驶技术试验演示系统。具体研究内容包括：基于主动安全的人机智能信息融合分层架构及信息模型、基于“机、人、车、路”多源信息在线协同感知的故障诊断与预测、针对行驶环境突变与感知信息缺失或错误的主动安全容错控制策略。.项目研究成果将为我国车辆智能驾驶技术的发展和新一代汽车主动安全系统的产业化提供理论和技术基础，为我国车辆自主驾驶系统设计提供数据支撑和低成本验证平台。

本项目的研究目标是“提高智能汽车的主动安全性”，研究内容沿四条主线展开：智能汽车的容错感知；智能汽车的目标容错识别跟踪和容错运动规划；智能汽车的故障诊断；智能汽车的容错控制。研究成果主要包括：.“智能汽车的容错感知”方面：双目视觉中单目损坏时图像复原和深度估计、车辆运动下视觉信号的图像去模糊、视觉短暂“失明”时动态感知预测、雾霾下视觉图像的复原和增强、全天候低照度下可见光与长波红外图像融合的运动目标检测、车道线“残缺” 或局部被遮下的车道线增强检测、交通标志“缺陷”的容错识别、激光雷达失效时的点云补全或生成、激光雷达数据滤波降噪、毫米波雷达数据滤波降噪。. “智能汽车的目标容错识别跟踪和容错运动规划”方面：紧急避险的邻车运动轨迹预测、运动目标被遮挡或短暂消失下多目标容错跟踪、激光点云深度补全的运动目标检测、弱小运动目标的激光点云检测、激光点云受噪声污染下车辆目标的容错识别、三维车辆目标检测识别、自由换道行为态势评估与容错决策、具有容错安全监控的动态换道轨迹规划、具有防碰撞检测和轨迹最优重构的动态换道轨迹规划。. “智能汽车的故障诊断”方面：基于时间序列信号分析的故障诊断、故障样本数据不足下的故障诊断、ESP系统的故障诊断、汽车传感器故障检测与分类、电动汽车轮毂电机故障诊断、车云协同的故障诊断系统。. “智能汽车的容错控制”方面：自适应巡航系统主动容错控制、具有周边车辆驾驶意图识别的跨车道ACC系统容错控制、电动汽车分布式驱动协调容错控制、线控底盘车辆系统协同容错控制、ESP系统的主动容错控制。.以上成果的主要科学意义是，为智能汽车在非理想条件下的安全行驶提供了技术上的保障。.本项目共发表和录用标注论文68篇（含已发表SCI论文48篇），出版专著1部，共申请/公开/授权发明专利23项、软件著作权16项，共培养博士生15名、硕士生25名，部分成果已经在宇通客车、智能电动运载平台样车、无人驾驶道路清扫车样车上得到了实际应用。