Acceleration Harmonic Identification For An Electro-Hydraulic Shaking Table Based On Bp Network

There are nonlinearities in an electro-hydraulic shaking table, causing harmonic distortion when it corresponds to a sinusoidal excitation, due to higher harmonics in the acceleration response. This paper aims to develop a harmonic identification method for a hydraulic shaking table by using Fourier series based Back Propagation (BP) network. The learning problem of hidden layer is solved by selecting appropriate BP network model, which has a single input, a single output, and two hidden layers for each harmonic. The hidden layer is excited by trigonometric function, and the weights are adjusted by the gradient descent method. The Lyapunov theorem is adopted to limit the network's the learning rate in order to guarantee the convergence of the algorithm. The network's weights are adjusted by the identification error between the sinusoidal response and the identified signal. Each harmonic's amplitude and phase are directly computed from the trained weights. Experimental results show that the proposed scheme can effectively identify each harmonic's amplitude and phase with high precision.