In this study, we propose a real-time nonlinear predictive control of a thermal process. The nonlinear dynamics of the process are mathematically modeled using a fractional Hammerstein structure. This estimated model is then employed as a prediction model in a nonlinear model predictive control scheme over a finite prediction horizon. The proposed algorithm generates the control sequence by optimizing a cost function with iterative nonlinear optimization methods, including the Nelder-Mead and gradient-based approaches. To evaluate its effectiveness, the nonlinear predictive control algorithm is implemented on an electric oven using the STM32F407VG micro-controller and compared in the tracking performances to a conventional proportional integral controller.

In this paper, an output feedback implicit Lyapunov control (OFILC) method is proposed to address the positioning challenges of rigid chain lifters under load variations and unmeasurable velocity conditions. OFILC represents a novel integration of a linear extended state observer (LESO) and implicit Lyapunov control (IL), a combination that has not been previously explored. Firstly, OFILC introduces friction and gravity moment compensation to reduce the uncertainty of the dynamics model. Secondly, a LESO-based velocity observer is developed to overcome the measurement limitation, providing clean state estimation without the noise amplification issues inherent in signal differentiation. Finally, to address the disturbance problem caused by parameter uncertainty and load differences, LESO is introduced to estimate and compensate for model disturbances online, thereby enhancing robustness and control accuracy compared to the IL approach. Theoretical analyses demonstrate that this controller is uniformly ultimately bounded convergence. Experimental results demonstrate that OFILC offers advantages in terms of high precision, robustness, and low power consumption with different load conditions.

For the direct-driven wave energy converters (WEC) with time-varying disturbance and uncertain parameters of model, a maximum wave energy capture strategy based on input saturation L 1 adaptive control is proposed. By analyzing the hydrodynamic model of a direct-drive WEC and the mathematical model of permanent magnet linear synchronous generator (PMLSG), a state-space model of the direct-drive WEC is established; the maximum power capture condition is obtained by using the equivalent circuit method; the input saturation L 1 adaptive controller with µ-modification is designed considering the controller limited case, which can effectively suppress the effects of time-varying disturbances and uncertain parameters of model to make the system track the desired current, and then achieve the maximum power tracking control. The convergence proof of the observation and reference errors is given. Simulations verify the effectiveness of the proposed control strategy, and the system has good steady state tracking performance and transient performance.

Physical systems are often affected by nonlinearities, such as saturation, which leads to the degradation of performance and stability. However, it is lacking of compensation design methods to improve transient performance for finite-time control under input saturation. To address the issue, we propose a new dynamic adjustment compensation (DAC) method for finite-time sliding mode control of linear systems with input saturation. It constructs a dynamic function where a time-varying parameter can be adjusted to enhance transient performance. The method that is applied to modify fast terminal sliding mode control designs yields a new controller that guarantees finite-time stability. An estimate of region of attraction for linear systems with input saturation is obtained by using the induced matrix norm. The advantages of the proposed method are evaluated through some simulation results of the closed loop control system of a Bernoulli air suspension gripper such that the closed-loop system is finite-time stable within the estimated region of attraction with the improved transient performance.

This study proposes an innovative attitude and guidance control system for unmanned ships, emphasizing an adaptive approach to solve ship navigation and attitude control challenges. By constructing a dynamic state space model of the ship, this research integrates two advanced adaptive algorithms, namely adaptive gradient moment estimation (AMSGrad) and adaptive gradient algorithm (AdaGrad), within the linear quadratic regulator (LQR) framework. These algorithms can be used to enhance on-the-fly convergence of control parameters, ensure robust system performance under dynamic conditions, and reduce modeling errors or installation errors. Unlike traditional control methods that rely heavily on exhaustive modeling and lack on-the-fly adaptability, elastic and flexible systems for autonomous ship operations can be achieved. Research results show that this technology has made a significant contribution to unmanned navigation technology, achieving reliable control, precise navigation, and expanding the scope of applications such as automatic transportation and environmental monitoring. The proposed system highlights key advances in linking theoretical control strategies with practical, real-world implementation in autonomous vessel systems.

Waist assistive exoskeleton is a wearable robot used to alleviate muscle fatigue and prevent lower back injury for workers during repetitive load lifting. Multi-joint human-robot coupling, the effect of human control and the model uncertainties like load variations make the robust controller design of waist assistive exoskeleton become challenged. Most existed control methods are based on a simplified dynamic model and neglect model uncertainties, which leads to a limited control performance. This paper focuses on the dynamic modeling and high-performance force control of waist assistive exoskeleton. In order to obtain a dynamic model which is accurate as well as suitable for controller design, a 5-DOF human-robot rigid body dynamics is established first. Then holonomic constraints are proposed to describe the control effect of the wearer, which helps convert the 5-DOF dynamics into a 1-DOF dynamics. Based on the established 1-DOF dynamics, adaptive robust force control strategy is proposed to effectively address various model uncertainties and disturbances. Comparative simulations and experiments indicate that the proposed control method can realize accurate and robust force control performance under different loads.

This paper investigates the state estimation problem for a class of two time-scale cyber-physical systems. A binary symmetric channel (BSC) associated with the binary encoding transmission (BCT) scheme is highlighted to enhance the reliability of data transmission. While binary bit strings are being transmitted through memoryless BSC, random error codes due to channel noise may occur. In light of this, a novel BCT-based state estimator is put forth, guaranteeing that the estimation error exponentially ultimately bounded in the mean square sense. To alleviate the numerical stiffness resulting from the two-time-scale property, a novel parameter-based Lyapunov function is developed, along with corresponding parameter-based criteria. Finally, a simulation example is provided to confirm that the theoretical results are valid.

In this work, a novel broad learning neural network-based adaptive control (BLNNAC) scheme is designed for a class of lateral thrust/aerodynamic force composited high-speed unmanned aerial vehicles with external perturbations, and unknown uncertainty under the time-varying output constraints. The proposed control strategy incorporates several key innovations. Firstly, an innovative tan-type barrier Lyapunov function is introduced to successfully avoid violations of time-varying output constraints. Secondly, the fast response capability and robustness to external disturbances inherent in the integral sliding mode control (ISMC) scheme are integrated into the strategy, enhancing its overall performance. Finally, a novel broad learning neural network (BLNN) is designed to effectively suppress the detrimental effects of unknown uncertainties, thereby significantly improving the system’s approximation performance. The results indicate that all signals are well-constrained, and the transient states of the output signals satisfy the constraint conditions constantly. Finally, the effectiveness and advantages of the proposed scheme are demonstrated through simulation results.

Considering the occurrence of actuator faults in the wind turbine generator main drive system, a fault model of the wind turbine generator main drive system is established. First, an adaptive sliding mode observer is designed to estimate the actuator faults, which is able to identify and reconstruct the faults quickly and provides the necessary information for the implementation of fault-tolerant control. In order to improve the design process of the observer, the gain matrix solution problem is transformed into an optimization problem subject to linear matrix inequality constraints. Secondly, considering the existence of unmodeled dynamics and external noise disturbances in the main drive system, an adaptive feedback fault-tolerant controller is designed using the fault reconstruction information after an actuator failure occurs in the system to compensate for the system errors caused by actuator failures, unmodeled dynamics, and external noise disturbances, which ensures a stable operation of the system. Finally, a simulation study of a 5 MW wind turbine system shows that at an average wind speed of 7.5 m/s and a turbulence intensity of class A, the system is restored to a normal state after the addition of the fault-tolerant controller, and the generator electromagnetic torque error is only 1.89%, which verifies the validity of the proposed method.

Fault detection in wind turbine systems remains a significant challenge due to variable operational conditions, the complexity of Supervisory Control and Data Acquisition (SCADA) signals, and the high dimensionality of real-time data. Traditional machine learning and reinforcement learning approaches often encounter limitations such as manual hyperparameter tuning, slow convergence, and susceptibility to local minima. These issues contribute to high false alarm rates and hinder the effectiveness of predictive maintenance strategies. To overcome these challenges, we propose a novel Hybrid Quantum-Inspired Proximal Policy Optimization (QGA-PPO) framework. This method combines the exploratory power of Quantum Genetic Algorithms (QGA) with the adaptive learning capabilities of Proximal Policy Optimization (PPO). The QGA component autonomously optimizes hyperparameters and refines the feature space, thereby enhancing the stability and robustness of PPO policies in complex SCADA environments. We evaluated the proposed framework using real-world SCADA data from 2.5 MW wind turbines. The QGA-PPO model achieved a 97.5% fault detection precision, reduced false alarms by 20%, and exhibited a 30% improvement in convergence speed compared to baseline PPO models. These results confirm the model’s effectiveness for advanced, real-time fault monitoring. Moreover, the framework demonstrates strong scalability, making it suitable for both individual wind turbines and large-scale wind farm systems. This research highlights the potential of quantum-inspired reinforcement learning for enabling autonomous fault tolerance and predictive maintenance in next-generation wind energy infrastructures.