Abstract
In this paper, a scheme of designing neural network-based sliding mode controller is proposed for dealing with the chattering phenomenon existing in conventional sliding mode controller because of its discrete control law. Firstly, the sliding mode control law is designed using equivalent control technology. Then the neural network and adaptive control are used to delete the chattering of sliding mode controller. The stability of control system is analyzed by Lyapunov stability theory finally. Simulations and experiments demonstrate that the proposed neural network-based sliding mode controller not only achieves better control performance than the conventional sliding mode control system, but also is robust with regard to system parameter variation and external disturbance.
Introduction
Permanent Magnet Synchronous Motors (PMSMs) are critical components in modern industrial applications, ranging from robotics to electric vehicles. However, achieving high-precision control is challenging due to time-varying internal parameters and external load disturbances.
Sliding Mode Control (SMC) is often favored for its inherent robustness against such uncertainties. By forcing the system state onto a predefined sliding manifold, SMC ensures the system follows a desired trajectory regardless of disturbances. However, the standard implementation of SMC relies on high-frequency switching, which leads to "chattering"—violent oscillations that can degrade accuracy and cause mechanical wear. This paper presents a hybrid approach that integrates Radial Basis Function (RBF) Neural Networks with SMC to eliminate chattering while maintaining robustness.
Proposed Methodology
The proposed Neural Network-based Sliding Mode Controller (NNSMC) is developed through a multi-stage design process. First, a mathematical model of the PMSM is established in the rotating $d-q$ coordinate system. An "equivalent control" law is then designed to manage the known dynamics of the motor.
To address unknown nonlinearities, such as parameter variations and external noise, an RBF Neural Network is integrated into the control loop. Unlike traditional fixed-gain switching, the neural network's weights are updated in real-time using an adaptive law derived from Lyapunov stability theory. This ensures that the network effectively "learns" to cancel out disturbances and chattering without compromising the mathematical stability of the system.
Furthermore, the harsh "signum" switching function typically used in SMC is replaced with a smoother saturation function. This modification further dampens output spikes, ensuring a cleaner control signal to the inverter.
Experimental Results and Discussion
The performance of the NNSMC was validated through both Matlab/Simulink simulations and hardware experiments using a TMS320F2812 DSP processor. The system was tested against traditional integer-order sliding mode control methods.
Response Speed and Tracking
In experimental step-response tests, the proposed controller achieved a rise time of 0.25625 seconds, a significant improvement over the 0.4625 seconds required by the standard controller. This represents a roughly 45% increase in response speed. As shown in the tracking plots, the "Responsed Sine Signal" aligns closely with the "Given Sine Signal" after a brief initial transient period.
Chattering Reduction
The effectiveness of the neural network in mitigating chattering is most evident in the sliding mode surface plots. While initial oscillations occur as the system reaches the sliding manifold, the adaptive neural network quickly compensates for these errors, reducing the chattering to negligible levels. This allows for smooth motor operation even under parameter perturbations, such as changes in resistance or inductance.
Conclusion
This research demonstrates that combining Neural Networks with Sliding Mode Control effectively solves the chattering problem inherent in traditional SMC. By treating chattering as an approximation error that can be learned and cancelled in real-time, the proposed NNSMC achieves superior response speeds and high robustness. Future work will focus on optimizing the computational overhead of the adaptive updates to ensure compatibility with lower-cost microcontrollers.
Summary updated