【作者】 方煜坤； 程超轶； 闵海根； 赵祥模；

【Author】 Fang Yukun;Cheng Chaoyi;Min Haigen;Zhao Xiangmo;Chang’an University;

【机构】 长安大学；

【摘要】 本文提出了一种基于几种数据驱动方法并结合模糊比例-积分-微分(PID)控制的智能汽车故障诊断框架,该框架由传感器状态监测模块、异常侦测模块和执行器错误测试模块组成。其中,离散小波变换(DWT)被用于构建传感器状态监测器,实现对来自传感器数据的降噪和特征提取;基于超限学习机的自动编码器则被用来构建异常侦测模块,实现对车辆异常状态的侦测;进一步地,每一个关键的执行器都在健康状态下用一个神经网络对其进行系统逼近,并在车辆工作时将之作为参考系统,而模糊PID控制器的输出将同时送入实际系统和参考系统并对比二者输出的差异。本文的主要贡献包括:1)考虑到传感器数据序贯到达的特性,在对数据进行DWT操作时加入了滑窗机制;2)将神经网络和模糊PID控制结合对关键执行器进行错误测试,尝试了从控制角度解决执行器错误的定位问题。基于实际智能汽车平台的一系列实验和仿真验证了该故障诊断框架下各种用于方法的有效性。更多还原

【Abstract】 This paper has proposed a fault diagnosis framework which combines several datadriven methods and fuzzy Proportional Integral Derivative(PID) control. This framework is consisted of sensor monitor cluster, novel anomaly detector and actuator fault testing cluster. The Discrete Wavelet Transform(DWT) are used for constructing the sensor monitors to denoise and extract the features from sensors and extreme learning machine based autoencoder are applied for novel anomaly detection. Further, each crucial actuator is approximated under the healthy conditions via a neural network and these networks are used as reference systems. The output of the fuzzy PID controller will simultaneously send to the real actuator and the reference system to compare the difference between their outputs. Main contributions of this paper are as follow: 1) Considering the sequential arrival characteristic of the sensor data, an algorithm using DWT with slide window is proposed for fatal sensor fault detection; 2) Neural networks and fuzzy PID control are combined for crucial actuator fault testing, which tries to solves the issue of fault location from the perspective of control. Experiments on the real intelligent vehicle platform and related simulations validate the effectiveness of the proposed approaches in this fault diagnosis framework.更多还原