TL;DR
This paper introduces a novel Adaptive Radial Basis Function Neural Network Fuzzy Sliding Mode Controller (ARFSMC) for precise trajectory tracking of hovering mining vehicles (HMVs) in deep-sea operations. By combining an adaptive neural network for uncertainty compensation with a fuzzy logic system to mitigate control signal chatter, the ARFSMC demonstrates superior tracking accuracy and significantly reduced oscillations compared to conventional methods.
What Problem Does This Paper Address?
Deep-sea mining is crucial for accessing valuable mineral resources, but traditional track-driven mining vehicles cause significant environmental disruption. Hovering mining vehicles (HMVs) offer a more eco-friendly alternative. However, HMVs present a significant control challenge due to their complex, nonlinear, and coupled dynamics, coupled with susceptibility to unpredictable external disturbances like ocean currents. Achieving precise trajectory tracking is essential for efficient and safe deep-sea mining operations. Existing control methods often struggle with robustness to these uncertainties and can lead to actuator wear due to control signal "chattering."
Methods
The proposed ARFSMC controller builds upon the principles of Sliding Mode Control (SMC) but introduces several key enhancements:
- Adaptive Neural Network for Dynamics Approximation: Instead of relying on a precise model of the HMV's dynamics, the ARFSMC uses an offline-trained Radial Basis Function Neural Network (RBFNN) to approximate the complex, unknown parts of the vehicle's dynamics. This allows the controller to adapt to system uncertainties.
- Adaptive Mechanism for Robustness: To further enhance disturbance rejection, an adaptive law is applied to the RBFNN's output weighting matrix. This allows the network to continuously adjust its parameters to better compensate for external disturbances and model uncertainties.
- Fuzzy Inference System for Smooth Control: The traditional, discontinuous "signum" function in SMC, which causes chattering, is replaced by a Fuzzy Inference System (FIS). This FIS acts as a smooth switching term, effectively reducing high-frequency oscillations in the control signal, thereby protecting actuators and improving overall system performance.
- Lyapunov Stability Analysis: The stability of the proposed ARFSMC controller is rigorously proven using Lyapunov stability theory, ensuring that the system will converge to the desired trajectory even in the presence of unknown disturbances.
Datasets
The paper does not use datasets for training. Instead, it relies on a simulated 4-DOF model of the "HaiMa" ROV, using its known parameters. Three distinct simulation scenarios were designed to mimic expected operational conditions:
- Vertical Deployment: Simulating the HMV descending into the mining zone.
- Linear Translation with Time-Varying Mass: Mimicking the HMV gathering minerals, which alters its mass.
- Spiral Motion: Testing the HMV's maneuverability in a complex environment.
These scenarios, along with defined external disturbances (sinusoidal functions with random noise), serve as the benchmark for evaluating the controller's performance.
Key Insights and Ideas
- Hybrid Control is Powerful: Combining the robustness of SMC with the adaptive learning capabilities of ANNs and the smooth switching of FIS offers a potent solution for complex control problems like underwater vehicle navigation.
- Chattering is a Major Hurdle: The paper highlights that chattering in control signals is a significant issue for actuators in underwater vehicles, and the proposed FIS-based switching term effectively addresses this.
- Adaptability is Key for Deep-Sea: The deep-sea environment is inherently unpredictable. An adaptive mechanism that can adjust to changing conditions (like varying mass or unknown disturbances) is crucial for reliable operation.
- RBFNNs as Effective Dynamic Approximators: The study demonstrates that RBFNNs are well-suited for approximating the complex, nonlinear dynamics of underwater vehicles, especially when combined with adaptive learning.
Key Results
The ARFSMC demonstrated superior performance across all three simulation cases when compared to Conventional Sliding Mode Control (CSMC) and Double-Loop Sliding Mode Control (DSMC):
- Reduced Tracking Error: The Root Mean Square Error (RMSE) for translational degrees of freedom was reduced by 15.0% compared to CSMC and 11.4% compared to DSMC.
- Significant Chattering Reduction: The ARFSMC achieved a remarkable 97.8% reduction in chattering compared to CSMC, as measured by the Total Variation (TV) metric. This indicates a much smoother control output.
- Faster Convergence: The ARFSMC exhibited faster convergence rates for tracking errors in all degrees of freedom.
- Robustness to Disturbances and Dynamics: The controller maintained high tracking accuracy even with time-varying system inertia and external disturbances.
Limitations and Future Work
- Offline RBFNN Training: The RBFNN in this study was trained offline. Future work could explore online training of the RBFNN for even greater adaptability.
- Simplified RBFNN Structure: The RBFNN used has a simplified structure. Incorporating features like input weightings and connection thresholds could offer additional flexibility.
- Prototype Development and Real-World Validation: The authors plan to develop a prototype vehicle based on these simulations and conduct real-world validations to further assess the ARFSMC's performance.
- Disturbance Observer Integration: Introducing a disturbance observer into the algorithm is suggested to further enhance tracking performance.
- Impact of Mass Variation: Further investigation into the impact of mass variation on the HMV model is recommended.
Summary updated