【作者】 刘洋；

【导师】 帅石金；

【作者基本信息】 清华大学 ， 动力工程及工程热物理， 2019， 硕士

【摘要】 随着轻型汽油车国6排放法规的即将实施,汽油车采用三效催化转化器(TWC)和汽油机颗粒捕集器(GPF)集成后处理系统降低有害物排放已是主流趋势。车载在线诊断(OBD)是发动机管理系统的必要组成模块,它的基本功能包括对排放相关部件进行监测和诊断;通过故障指示灯提示故障的发生;存储故障相关信息;与诊断软件进行通讯等。国6后处理系统复杂程度的提高为OBD开发工作带来了新的挑战。论文结合国内外OBD研究经验,对OBD软硬件要求进行了深入需求分析,基于快速原型的思想开发了一套轻型汽油车国6后处理OBD系统。系统硬件由NI PXI平台定制而成,根据后处理部件的测量需求选配硬件,实现了对集成后处理系统传感器信号和发动机工况参数的采集和通讯。OBD系统定义了GPF移除、泄漏、过载和堵塞四种故障类型以及TWC老化和临界两种状态,制备故障样件并根据排放结果建立了标准数据集。在MATLAB中基于神经网络进行了故障诊断算法的研究,分别选择反向传播神经网络(BPNN)和深度置信网络(DBN)作为GPF和TWC的模型架构,设置合理的训练参数,经过大量离线测试,证明算法具有良好的诊断精度和泛化性能。在Simulink中基于离线算法进行OBD系统模型的整体搭建,包括诊断条件判断、神经网络诊断和故障决策三个子模型,对各种系统接口和变量进行了定义,形成完整的诊断逻辑,完善了OBD系统的整体功能。基于Veri Stand设计了OBD系统的开发平台,实现了系统硬件配置、模型代码生成与下载以及ECU通讯功能,通过交互界面能够对系统进行实时的状态监控和参数标定。通过台架试验,对后处理OBD系统的基本诊断功能进行了验证。试验结果表明,OBD系统能够在未知稳态工况下,正确诊断出后处理部件的相应故障,各种故障模式的区分度良好。通过整车转鼓试验,对后处理OBD系统的诊断条件判断模块和故障决策模块进行了验证。试验结果表明,OBD系统能够在实际驾驶循环中,正确识别诊断工况,实时计算故障发生概率并进行合理的故障决策,满足设计要求。更多还原

【Abstract】 With the implementation of the China 6 emission regulations for light duty gasoline vehicles,the use of Three-Way Catalytic Converter(TWC)and Gasoline Particulate Filter(GPF)integrated aftertreatment system to reduce harmful emissions has become a mainstream trend.On-Board Diagnostics(OBD)is an essential component of the engine management system.Its basic functions include monitoring and diagnosing emissionsrelated components;indicating faults through Malfunction Indicator Lamp;storing faultrelated information;communicating with diagnostic software.The increased complexity of the aftertreatment system brings new challenges to OBD development work.Combining the domestic and international OBD research experience,this thesis analyzed the requirements of OBD software and hardware,and developed a set of aftertreatment OBD system for light duty gasoline vehicles meeting China 6 based on the idea of Rapid Control Prototyping.The system hardware was customized by the NI PXI platform.The hardware was selected according to the measurement requirements of the aftertreatment components,and the acquisition and communication of the sensor signals and engine operating parameters were realized.The system defined four fault types of GPF: removal,leakage,overload and blockage,as well as two state of TWC: aging and criticality.Fault samples were prepared and standard data sets were established based on emission results.The thesis researched fault diagnosis algorithms based on neural network in MATLAB,selecting Backpropagation Neural Network(BPNN)and Deep Belief Network(DBN)as the model structure of GPF and TWC respectively,and selecting reasonable training parameters.After a large number of offline tests,it was proved that the algorithm had good diagnostic accuracy and generalization performance.The overall construction of the OBD system model was completed in Simulink based on the offline algorithm,including three sub-models: diagnostic condition judgment,neural network diagnosis and fault decision.Various system interfaces and variables were defined to form complete diagnostic logic and the overall function of the OBD system were improved.Based on Veri Stand,the development platform of OBD system was developed,realizing functions including hardware configuration,model code generation and downloading,and ECU communication.The designed interactive interface enabled real-time status monitoring and parameter calibration of the system.The basic diagnostic function of the aftertreatment OBD system was verified by bench test.The test results showed that the OBD system could correctly diagnose the corresponding faults of the aftertreament components under unknown steady-state conditions,and the discrimination of various fault modes was good.The diagnostic condition judgment module and the fault decision module of the aftertreatment OBD system were verified by drum test.The test results showed that the OBD system could correctly identify the diagnostic conditions,calculate the real time probability of failure and make reasonable fault decisions in the actual driving cycle,which meets the design requirements.更多还原