Research on motion control algorithms for the blanket remote maintenance robot

Abstract

The China Fusion Engineering Test Reactor (CFETR) is a tokamak-type fusion reactor designed to bridge the gap between the International Thermonuclear Experimental Reactor and the demonstration fusion reactor. After the shutdown of the fusion reactor, the blankets in the vacuum vessel (VV) need to be transported to the hot cell for further maintenance work through the remote handling system. In the CFETR, the blanket module is a high-centroid elongated and heavy-load component, with each blanket module weighing 60 tons and being 7.6 meters high. The blanket remote maintenance robot (BRMR) is a crucial component of the blanket remote maintenance system in the CFETR. During the maintenance process, because of the special structure of the blanket, it requires collaborative maintenance by dual BRMRs. The dual BRMRs consist of the upper BRMR (BRMR-U) and lower BRMR (BRMR-L) that can lift the blanket and transport it from its working position to the maintenance port in the VV.

During the blanket transportation process, even a millimeter-level synchronization error of the dual BRMRs can lead to unpredictable hazards. The primary purpose of this study is to determine a valid method to achieve precise motion control and synchronous control of the BRMR, thereby enabling the accurate and efficient transportation of the blanket. However, the BRMR consists of a highly nonlinear and asymmetric valve-controlled hydraulic system, making it difficult to build a high-precision model. Additionally, due to the elongated structure of the blanket and the large load of the BRMR (60 tons), significant load discrepancies exist between BRMR-U and BRMR-L, which can cause synchronization errors in the dual BRMRs. Any minor synchronization errors can lead to changes in the loads of the BRMR-U and BRMR-L, which in turn affect the synchronization accuracy of the dual BRMRs, creating a vicious cycle. These factors all contribute to the challenges in motion and synchronization control accuracy of the BRMR. To achieve stable and high-precision transportation of the blanket, it is necessary to propose effective control algorithms and motion controllers that precisely synchronize the asymmetric multi-hydraulic system.

This study first utilized a physics-based analytical modelling approach to establish mathematical models of the valve-controlled asymmetric hydraulic system of the BRMR. To improve system stability and simplify the mathematical models, the system was linearized near its operating point. Then, this study employed a model predictive control (MPC) algorithm to establish the MPC controller for the BRMR. Additionally, innovatively combining cross-coupling control (CCC) theory, this dissertation proposes an MPC-CCC synchronization controller tailored for the dual BRMRs.

Furthermore, this dissertation optimized the aforementioned work and established nonlinear state-space mathematical models for the BRMR. To retain model nonlinearity while enhancing controller stability, a method of nonlinear feedback linearization was employed to process the model. To obtain a more accurate nonlinear mathematical model, this dissertation used a variance-based sensitivity analysis method to identify critical system parameters affecting model accuracy, followed by employing an improved adaptive particle swarm optimization-simulated annealing (APSO-SA) algorithm to identify these parameters. Since the weight matrix significantly affects the receding horizon optimization process of MPC-type controllers, a fuzzy regulator (FR) combining error feedback and fuzzy theory was proposed to adaptively adjust the weight matrix in real-time for optimal control output. To mitigate the effects of disturbances, a deep neural network (DNN) inverse model was innovatively come up with and established for the system. The system’s DNN inverse model serves as a control feedforward to provide disturbance rejection control and enhance controller effectiveness. Subsequently, based on the aforementioned nonlinear models and control algorithms, the nonlinear model predictive control with fuzzy regulator and deep neural network feedforward motion controller and nonlinear model predictive control and cross-coupling control with fuzzy regulator and deep neural network feedforward synchronization controller of the BRMR were designed.

Finally, simulations and experiments were conducted to validate the effectiveness of the proposed parameter identification algorithm, control algorithms, and controllers. The results demonstrate that the APSO-SA algorithm enhances model accuracy. Compared with the original parameters, the root mean square error (RMSE) of the system output with the optimal parameters can be reduced by up to 90.5%. Additionally, the proposed control algorithms and controllers significantly reduce control errors for the BRMR. The RMSE of the MPC-type control algorithms decreased by up to 68.9% compared with proportional integral derivative type control algorithms. The FR control algorithm reduced the system RMSE by 33.1%. The DNNF algorithm reduced the system RMSE by 48.9%. And the CCC algorithm reduced the system synchronization RMSE by up to 75.8%. These results all demonstrate the effectiveness of the proposed algorithms and controllers.

This study establishes a theoretical and application foundation for ensuring the future safe, stable, and efficient maintenance and transportation of the blanket. Moreover, the findings of this research can be extended to motion control and synchronous control applications in various fields such as robotics, hydraulic servo systems, process control systems, etc.