Neural network dynamic surface position control of n‐joint robot driven by PMSM with unknown load observer

Abstract

To solve the problems of low accuracy and poor stability due to modeling error, external disturbance and unknown load, which exist in the position servo control of permanent magnet synchronous motor (PMSM) driven joint robot, this article is to propose the radial basis function (RBF) neural networks dynamic surface control strategy with the Sage-Husa adaptive Kalman filter load torque observer. For the unknown load torque of the robot, the PMSM load torque observer is established by using the Sage-Huga adaptive Kalman filter. The RBF neural network dynamic surface controller is designed using the online approximation capability of the neural network, which is used to approximate the modeling error, external interference and filtering error generated by the dynamic surface control of the joint robot online. Combining the above strategies, the n-joint robot position controller is designed. The stability of this control strategy is demonstrated by stability analysis. Simulations and experiments on the two-joint robot show that the control strategy ensures the accuracy and stability of the joint robot position control.

A refined anti‐disturbance control method for gimbal servo systems subject to multiple disturbances under constraints

“…In addition, different from the ideal dynamic models in refs. [19,[31][32][33], multi-channels disturbances are fully explored in the studied model. Specifically, the inner loop (current loop) considers the distortion voltages and the outer loop (speed loop) considers the imbalance disturbance and nonlinear friction.…”

Adaptive neural network control of a non‐linear two‐degree‐of‐freedom helicopter system with prescribed performance

“…By selecting a sufficient number of neuron nodes and reasonably distributed base function centre points, Ω z → R p can be obtained over a compact range. The smooth and bounded function h(Z ) can be accurately approximated by the RBFNN [41,42] via the following equation…”

“…Lemma By selecting a sufficient number of neuron nodes and reasonably distributed base function centre points, $$\Omega _z\rightarrow R^p$$ can be obtained over a compact range. The smooth and bounded function $$h(Z)$$ can be accurately approximated by the RBFNN [41, 42] via the following equation $$$$\begin{equation} h(Z)=W^{*T}S(Z)+\eta (Z), \forall \in \Omega _z, \end{equation}$$$$where $$W^*$$ is the ideal constant weight vector and $$\eta (Z)$$ is the function approximation error. There is an ideal $$W^*$$ that satisfies $$|\eta (Z)|\le \eta ^*$$, where $$\eta ^*>0$$ is the upper limit to a very small approximation error.…”

Wavelet neural network sliding mode control of two rigid joint robot manipulator

“…28,29 Recently, researchers have developed a considerable interest in using the Wavelet Neural Network (WNN) to approximate the bound value of uncertainties. 30,31 In summary, each drawback of the conventional (SMC) has been tackled by a corresponding suitable approach, however, there are no approaches in the literature that intended to design a single controller that considers all the drawbacks of (SMC) together and solve them simultaneously.…”