基于CNN的主动悬架传感器故障诊断  

Fault Diagnosis of Active Suspension Sensor Based on CNN

作　　者：王艳平[1] 韩晓冰[1] WANG Yanping;HAN Xiaobing(College of Science,Liaoning University of Technology,Jinzhou 121001,China)

机构地区：[1]辽宁工业大学理学院,辽宁锦州121001

出　　处：《控制工程》2024年第12期2190-2195,共6页Control Engineering of China

摘　　要：汽车主动悬架系统中的传感器损坏后采集的信号会对系统的控制效果产生不良影响。因此,提出了基于卷积神经网络(convolutional neural network,CNN)的传感器故障诊断算法。根据高度传感器在3种不同损坏状态下所采集的数据,选取各周期内的训练样本数据,并保留一定数量的测试样本。结合基于CNN的传感器故障诊断算法,使用训练样本对神经网络进行训练,然后输入测试样本对神经网络的准确率进行测试,验证了卷积神经网络在高度传感器的故障诊断方面的准确性。并使用同样的数据对反向传播神经网络(back propagation neural network,BPNN)进行训练和测试。通过对诊断结果准确率的比较可知,CNN在汽车主动悬架高度传感器信号故障诊断方面具有明显优势,诊断准确率达到99.31%。The signal collected by a damaged sensor of active suspension system will adversely affect the control effect of the system.Therefore,a sensor fault diagnosis algorithm based on convolutional neural network(CNN)is proposed.Using the data collected by the height sensor under three different damage states,the data in each cycle are selected as training samples,and a number of test samples are retained.Combined with the sensor fault diagnosis algorithm based on CNN,the neural network is trained with training samples,and then the accuracy of the neural network is tested with test samples,which verifies the accuracy of convolution neural network in the fault diagnosis of height sensors.Besides,the same data is used to train and test the back propagation neural network(BPNN).After comparing the accuracy of diagnosis results,CNN has obvious advantages in the signal fault diagnosis of automotive active suspension height sensor,and the diagnosis accuracy reaches 99.31%.

关 键 词：卷积神经网络 BP神经网络 传感器 故障诊断 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]