【作者】 孙海涛；

【导师】 汪洪波； 魏振亚；

【作者基本信息】 合肥工业大学 ， 车辆工程（专业学位）， 2025， 硕士

【摘要】 随着汽车智能化发展,悬架系统在保障驾乘安全性和舒适性方面的重要性日益突出。但随着悬架系统长时间工作,其零部件不可避免会出现老化或导致传感器、执行器等出现故障。这些故障轻则导致悬架系统性能下降,影响车辆平顺性和操纵稳定性,严重时可能会引发安全事故,威胁驾乘人员生命安全。尤其是在一些场景下,例如盲区场景中,悬架系统故障会进一步加剧恶化行驶安全。因此,悬架系统容错控制的研究格外重要。本文针对悬架系统出现多重故障时多目标容错控制展开研究,论文的主要工作内容如下:(1)针对盲区场景下车辆悬架同时出现作动器与传感器故障情形,设计安全速度并基于观测器设计作动器及传感器同时故障的容错控制策略。由于实际悬架故障是动态变化的,将执行器/传感器故障模型与马尔可夫过程理论相结合,构建具有跳变特征的马尔可夫型悬架故障模型。通过故障观测器实现多故障的实时估计,进而采用信号重构与动态补偿相结合设计主动容错控制方法。在进行主动容错的基础上,探究不同容错程度下容错效果,获得不同容错程度下的最佳容错。运用MATLAB/Simulink软件对所提出的容错控制策略进行仿真验证。仿真结果表明,基于故障诊断观测器的方法能准确估计悬架作动器和传感器的故障大小。通过故障补偿和信号重构,悬架性能可恢复至正常水平。容错效果分析表明,探究获得的不同容错程度下的最佳容错能使故障状态下的悬架性能接近正常水平。(2)当前悬架执行器故障容错控制研究,主要采用基于观测器的故障估计方法。虽然这种方法能够实现较好的容错性能,但其观测器设计过程复杂,参数调试难度较大。针对这一问题,提出一种基于柔性执行者-评价者(Soft Actor-Critic,SAC)算法的容错控制策略。该策略通过智能调节控制器参数并引入额外的补偿力,来改变悬架输出力。具体而言,首先基于车辆动力学模型设计输出反馈控制器,然后利用SAC智能体实现控制器参数的优化,从而构建完整的悬架容错控制系统。为优化系统性能,设计综合考虑平顺性和安全性复合奖励函数,并引入额外奖励机制。特别地,针对智能车辆在不同行驶工况下悬架故障时的多目标协同控制,采用可拓控制理论动态调整奖励函数中安全性与平顺性的权重系数,实现悬架控制系统输出特性的调节。基于MATLAB/Simulink软件对基于观测器的容错策略进行仿真,仿真结果可知,该策略能在悬架多执行器故障情况下有效维持系统性能。通过设计多梯度奖励函数,提升强化学习的训练效率和收敛速度。此外,基于三维可拓关联函数动态调整奖励权重,进一步增强悬架在不同路况和故障条件下的综合容错控制能力。(3)为了进一步验证提出的基于观测器和强化学习的多故障的容错策略,在盲区场景和不同路面工况下进行实车试验。结果表明,提出的基于观测器的容错控制策略能够有效提高故障悬架系统的性能,使其关键性能指标接近正常悬架系统的水平。基于三维可拓调节奖励函数权重的方法能提升SAC容错控制策略的适应能力,该策略能够适应不同等级的路面激励和多执行器故障模式的复杂工况,实现更优的综合容错性能,降低各类故障对悬架系统的影响。更多还原

【Abstract】 The growing intelligence of automobiles has heightened the significance of suspension systems in maintaining both safety and ride comfort.However,prolonged operation inevitably leads to component aging or faults in sensors and actuators.These faults may degrade suspension performance,compromising ride comfort and handling stability,or even cause safety hazards endangering occupants-especially in scenarios like blind scenes,where faults can exacerbate risks.Therefore,fault-tolerant control(FTC)for suspension systems is of paramount importance.This thesis focuses on multi-objective fault-tolerant control under multiple faults in suspension systems.The main contents are summarized as follows:(1)For the scenarios involving simultaneous actuator and sensor faults in blind scenes,a safe speed is designed,and an observer-based FTC strategy is proposed to handle concurrent actuator-sensor faults.Given the dynamic nature of real-world suspension faults,the actuator/sensor fault models are integrated with Markov process theory to construct a Markovian suspension fault model with jump characteristics.High-precision fault observers are employed for real-time multi-fault estimation,followed by an active FTC approach combining signal reconstruction and dynamic compensation.The effectiveness of varying fault tolerance levels is investigated to identify optimal compensation under different conditions.MATLAB/Simulink simulation results demonstrate that the proposed observer-based method accurately estimates the actuator and sensor faults.Through fault compensation and signal reconstruction,suspension performance is restored to normal levels.The fault-tolerance effectiveness analysis demonstrates that the optimal fault tolerance under different fault tolerance levels can make the suspension performance under fault conditions approach that under normal levels.(2)Current FTC study for suspension actuator faults primarily relies on the observer-based fault estimation.Although effective,this method involves complex observer design and challenging parameter tuning.To address this,a Soft Actor-Critic(SAC)algorithm-based FTC strategy is proposed.By intelligently adjusting controller parameters and introducing compensatory forces,suspension output is modified.Specifically,an output feedback controller is first designed based on vehicle dynamics,and the controller parameters are optimized by the SAC agent to form a complete FTC system.A compound reward function balancing ride comfort and safety is designed,supplemented by an additional reward mechanism.Notably,for multi-objective cooperative control under varying driving conditions,the weight coefficients of safety and comfort in the reward function are dynamically adjusted by extension theory,enabling adaptive output regulation.MATLAB/Simulink simulation results demonstrate that the observer-free SAC strategy maintains suspension performance under multi-actuator faults.A multi-gradient reward function enhances training efficiency and convergence,while three-dimensional extension correlation function dynamically adjusts reward weights,and the comprehensive fault-tolerant control capability of the suspension under various road conditions and fault scenarios is further enhanced.(3)To validate the proposed multi-fault FTC strategies(observer-based and reinforcement learning-based),real-vehicle tests are conducted in blind scenes and under different road conditions.The test results demonstrate that the proposed observer-based fault-tolerant control strategy can effectively enhance the performance of the faulty suspension system,bringing its key performance indicators close to the level of a normal suspension system.The method based on a three-dimensional extension-adjustable reward function weight improves the adaptability of the SAC fault-tolerant control strategy.This strategy can adapt complex working conditions involving varying levels of road excitation and multiple actuator fault modes,achieving superior comprehensive fault-tolerant performance and mitigating the impact of various faults on the suspension system.更多还原