Adaptive Neural Networks Control Of High Speed Ball Screw Drives

The ball screw drive system has been widely used in the feed system of machine tool because of its high rigidity and transmission efficiency.The tracking accuracy of machine tool has a direct impact on the quality of workpieces.During high-speed motion process,there are many nonlinear factors and time-varying uncertainty in ball screw drives,which are against the improvement of the control system performance.These elements include workpiece mass and position variations,friction,screw elastic force,torque and motor torque fluctuation.Neural network has infinite ability to approach any continuous function.Therefore,neural network is combined with PID,sliding mode control and backstepping control in order to realize the goal,which is improving the tracking precision of the ball screw drives.Traditional k-mean clustering method is sensitive to initial value and tends to cause local optimal solution.Therefore,improved neural network learning methods based on genetic algorithm and particle swarm optimization are respectively applied so as to determine the network initial values.Respectively using simulation and experiment samples to identify plant offline,the results show that the convergence speed and identification precision of neural network are higher based on PSO.Considering ball screw drives as a rigid body and using neural network to identify plant online.The results show that neural network identifier with PSO is more accurate.By using PSO-improved RBF neural network,the Jacobian information of ball screw drives system is identified so that three parameters of PID controller are self-calibrated.By using Hebb learning rules,a single neuron PID control method is designed.The two controllers conduct their parallel control for the ball screw drives.The simulation and experiment results show that compared with traditional PID controller,directing at ball screw drives with time-varying and nonlinear characteristics,paralle neural network PID controller can adjust parameters and lead to better tracking performance.High speed ball screw drives are influenced by the unknown disturbance.According to rigid body model of ball screw drives,using RBF neural network with PSO to identify unknown elements and disturbance.Meanwhile,the sliding mode controller is applied to realize real-time control of ball screw drives.Simulation and experiment results show that compared with parallel neural network PID controller,adaptive sliding mode controller based on improved neural network can effectively reduce the maximum tracking error because it can identify the unknown disturbance and essentially contains the two feed-forward controllers.Flexible ball screw drives model which contains time-varying uncertainty and unknown disturbance is proposed.Using neural network to identify time-varying parameters,unknown disturbance and other model information.By designing appropriate intermediate virtual control variables and Lyapunov function,stable backstepping controller is proposed without building model.In the simulation process,it is assumed that the workpiece mass and disturbances are changing over time.In the experiment process,using three kinds of workpiece mass and reference position.Simulation and experiment results show that even with unknown disturbance,time-varying parameter and different reference position,compared with paralle neural network PID controller and adaptive sliding mode controller based on improved neural network,adaptive neural network backstepping controller based on flexible body model still has higher tracking precision.