Fault Residual Threshold Over-Range Determination of Manipulator Under Deep Cascade CNN
The structure of the fully driven manipulator adopts PCas the upper computer.A large number of series structures lead to great difficulty in accurate transmission and poor load adaptability.In contrast,the underactuated manipulator has a simpler structure and can achieve more degrees of freedom than the control input.However,when the link type driven manipulator fails,it will produce natural vibration,and it is difficult to obtain the characteristic points,and the signal residual threshold cannot be ac-curately determined beyond the range.Therefore,a method to determine the fault residual error threshold of link type underactuat-ed manipulator is studied.The deep cascade convolution neural network is constructed to extract the deep meaning features of each group of fault signals and calculate the residual value of fault signals.The signal is normalized by the dimension reduction conver-sion of the connection layer.The shape and point level of the manipulator are extracted through the interpolation layer,and the fault signal prediction label is output.A deep cascade convolutional neural network signal observer is designed to update the data in real time following the operation of the manipulator,and the residual threshold of the feature point is set to determine whether the feature point is a fault point.The experimental results show that the research method can accurately judge the fault position of the link driven manipulator according to the output of the fault residual signal,and the convergence speed is fast and the error is small.
Deep Cascade Convolutional Neural NetworkManipulatorFault DiagnosisResidual ThresholdOb-serverCharacteristic Points
NETL
NSTL
万方数据
维普
