基于遗传神经网络的履带行驶系统载荷识别方法

针对煤矿掘进机器人履带行驶系统工作环境恶劣,载荷无法直接有效获取这一工程实际问题,提出了基于遗传神经网络的振动信号载荷识别方法。构建了遗传算法(GA)优化BP(back propagation)神经网络载荷识别模型,采用路试法试验采集了履带小车的5组振动加速度数据和单组应力载荷数据,探讨路面不平度频率和驱动轮啮频等对履带车振动和应力载荷的影响规律;借助快速傅里叶变换(FFT)对原始应力载荷数据进行去噪处理,依据履带小车行驶平顺性指标,利用sym8小波函数对振动加速度信号进行5层特征提取以提高载荷识别的精度,然后将5组小波变换分解的加速度数据和滤波后的应力载荷数据分别作为GA-BP神经网络的输入和输出进行训练及验证,揭示了履带行驶系统运动过程中振动与应力载荷之间的关系。研究结果表明,路面不平度频率、驱动轮啮频及转频为小车振动的主要频率成分,路面不平度引起的振动频率为13.765 Hz,驱动轮啮频为68.25 Hz,转频为3.25 Hz。多组试验得到的BP神经网络最佳隐含层神经元数为63,GA-BP神经网络识别的应力载荷与期望应力载荷具有较高吻合度,相对误差为4.5%,验证了该方法的有效性...

Load identification method of track driving system based on genetic neural network

Here,aiming at practical engineering problems of bad working environment of track driving system of coal mine robot and load being unable to obtain directly and effectively,a vibration signal load identification method based on genetic neural network was proposed.The load identification model based on genetic algorithm(GA)-optimized and back propagation(BP)neural network was constructed.5 sets of vibration acceleration data and a set of stress load data of a track car were collected with the road test method.Effects of road roughness frequency and driving wheel meshing frequency on vibration and stress load of the track car were discussed.The original stress load data was denoised by means of fast Fourier transform(FFT).According to the ride comfort index of track car,Sym8 wavelet function was used to do 5 layer feature extraction for vibration acceleration signals,and improve the accuracy of load identification.Then,5 sets of vibration acceleration data after decomposed using wavelet transform and filtered stress load data were taken as the input and output of the GA-BP neural network to perform training and verification to reveal the relationship between vibration and stress load in motion of track driving system.The results showed that the road roughness frequency,driving wheel meshing frequency and rotating frequency are main frequency components of the track car’s vibration;the vibration frequency caused by road roughness is 13.765 Hz,the driving wheel meshing frequency is 68.25 Hz and the rotating frequency is 3.25 Hz;the optimal hidden layer neuron number of BP neural network obtained from multi-set tests is 63;the stress load identified with the GA-BP neural network has a higher degree of consistence to the expected stress load,and the relative error is 4.5%to verify the effectiveness of the proposed method and provide a good theoretical basis for studying reliability of track driving system in coal mine machinery.