Gray-box acceleration modeling of an electro hydraulic servo shaking table with Neural Network

An effective way for the testing of a large number of systems is using single and multi-axis shaking tables. Among the possible applications, the civil engineering field stands out for the testing of structures, or part of them, to high natural dynamic forces. However, due to nonlinearities and structured and unstructured uncertainties of the hydraulic systems, the acceleration signal of these systems with respect to sine input are distorted. This paper is focused on identification of a graybox acceleration model of a uni-axial servo hydraulic shaking table with respect to sine input signal with different frequencies and amplitude. First, a full system model of servo hydraulic system is developed based on fluid mechanical expressions and steady state friction. Second, for each input frequencies the unknown parameters of the friction, bulk modulus and leakage of the system are identified based on nonlinear least square method. Then, the identified parameters are used to develop a dynamic feed forward neural network model. The graybox model uses online short fast Fourier transform harmonic identification method to identify the harmonic and amplitude of input signal and the neural network model produces parameters of the system. Finally, the comparison between the experimental results of the acceleration signal and the simulation one demonstrates the accuracy of the model.