Control Engineering Practice最新文献

The safety control of precision motion equipment in modern industrial fields is a key focus of industrial research, directly affecting the accuracy and lifespan of motion equipment. This paper presents a fault-tolerant gradient descent B-spline wavelet neural network (FTGDBNN) based controller of precision motion equipment for a dual-drive gantry system (DDGS), ensuring the safety and effectiveness of DDGS precision system control. The proposed controller contains the loss-of-effectiveness fault estimator and the gradient descent B-spline wavelet neural network (GDBNN) based compensator that can observe and compensate for loss-of-effectiveness and additive actuator faults in real time. In addition to the actuator additive faults, GDBNN-based compensators can suppress the impact of nonlinear disturbances such as system parameter uncertainties and fault estimator errors on precision equipment. Moreover, The boundedness of the fault estimator and the stability of the entire closed-loop system are theoretically proven. Finally, the safety and effectiveness of the proposed control strategy are validated through a series of fault experiments on the DDGS platform. The experimental results indicate that FTGDBNN has better safety and control performance compared to other control strategies applied to precision systems, especially in high curvature and extreme motion conditions.

In System of Systems Engineering (SoSE), robustness and reliability are essential for operations under various conditions. Distributed control approaches demonstrate superior performance for handling changes in highly scalable systems, making them suitable for this context. The study explores the efficacy of the Alternating Direction Method of Multipliers (ADMM) applied in a simulated case study and on a physical Quadruple Tank Plant system. In this work, a multi-agent coordination problem using the Model Predictive Control (MPC) paradigm, Linear Quadratic Tracking (LQT) techniques, and ADMM for the consensus has been formalized and a solution has been proposed as a control algorithm. During the experiments, there was no assumption of feasibility when the references were chosen. To improve the reliability of the controller, a Kalman filter to estimate the state has been added.

The fusion of industrial artificial intelligence with the Industrial Internet of Things (IIoT) can attain a heightened level of process monitoring in modern manufacturing processes. In general, the state variables of industrial processes collected through the IIoT encompass not only continuous variables but also numerous discrete variables. Owing to potential coupling factors, these variables frequently exhibit strong correlations. However, most existing methods deal only with continuous variables, which results in breaking the integrity of the state information and being incompetent to extract the useful information carried by discrete variables. To effectively address the joint monitoring challenges of continuous and discrete variables under the IIoT framework, hybrid variable dictionary learning (HVDL) is proposed in this paper. Specifically, considering that the values of discrete variables are finite sets, a specific discrete dictionary is built for data reconstruction. Besides, in order to consider the correlation between continuous and discrete variables, the alignment of them in the time dimension is achieved by sharing labels. The HVDL method can judiciously learn data dictionaries to extract multifaceted valid features across diverse data types, free from prior assumptions on data distributions. Finally, extensive experiments are conducted to demonstrate the superiority of the proposed method, including a numerical simulation case, a closed-loop continuous stirred tank reactor benchmark, and a real zinc smelting roaster. Experimental results indicate that the proposed method can fully consider the correlation between continuous and discrete variables, thus it is conducive to identifying early anomalies and mismatch anomalies.

This paper introduces a data-enabled predictive control methodology designed for Linear Parameter Varying (LPV) systems. By leveraging a polytopic representation of the LPV system, we formulate Willem’s lemma to accommodate parameter variations. The system trajectory is predicted over a finite horizon using specific trajectories generated offline. A notable advantage of this approach is its independence from a priori knowledge of parametric variations. Simulation studies conducted on both a numerical example and a simulator of a calendering process governed by partial differential equations substantiate the effectiveness of this approach.

To avoid human–robot confrontation, it is necessary to create a safe, accurate and stable rehabilitation environment for stroke patients when using upper limb exoskeleton robots for rehabilitation training. Considering these requirements, this work presents a novel compliant actuator based on torsion spring devices and linear spring devices. The establishment of dynamics model and control algorithm are challenging tasks for the novel mechanical structure. Firstly, the mechanical structure of the compliant actuator is simplified, and the interaction forces between gear trains are considered to establish the dynamics model. Secondly, the optimal control scheme is designed based on the dynamics model, and the stability of the actuator system is proved theoretically. Thirdly, it has been demonstrated that the optimal control scheme ensures precise and stable trajectory tracking for the compliant actuator through comparative simulations. The experimental results verify that the proposed optimal control scheme can enable the compliant actuator to complete trajectory tracking under different working conditions, as well as the buffering and self-protection performance of the torsion spring devices. In addition, the tracking accuracy of the optimal control scheme is further verified by conducting trajectory tracking experiments on the compliant actuator-based upper limb exoskeleton robot.

The accuracy of the underlying model predictions is crucial for the success of model predictive control (MPC) applications. If the model is unable to accurately analyze the dynamics of the controlled system, the performance and stability guarantees provided by MPC may not be achieved. Learning-based MPC can learn models from data, improving the applicability and reliability of MPC. This study develops a nonlinear sparse variational Bayesian learning based MPC (NSVB-MPC) for nonlinear systems, where the model is learned by the developed NSVB method. Variational inference is used by NSVB-MPC to assess the predictive accuracy and make the necessary corrections to quantify system uncertainty. The suggested approach ensures input-to-state stability (ISS) and the feasibility of recursive constraints in accordance with the concept of an invariant terminal region. Finally, a PEMFC temperature control model experiment confirms the effectiveness of the NSVB-MPC method.

Direct Torque Control (DTC) is widely used in motion control of motors. Classical DTC utilizes the Park transformation(dq-model) or observer to estimate the stator fluxlinkage and electromagnetic torque. However, the dq-model and observer lack descriptions of the cogging torque and magnetic saturation in permanent magnet synchronous motors (PMSM). This paper proposes a novel method based on the surrogate model to estimate stator fluxlinkage and torque for PMSM DTC. First, a finite element model (FEM) was established based on the parameters of PMSM. Using the FEM and Latin hypercube sampling (LHS), a surrogate model for estimating the stator fluxlinkage and torque of PMSM was constructed. The stator fluxlinkage and torque surrogate model, active flux observer, and dq current model under different control methods are constructed, respectively. The servo control parameters were tuned using the extended symmetric optimum algorithm, and simulation and experiments were conducted. Finally, electromechanical coupling models with a ball screw feed system were built based on various control and feedback methods. Simulation and experiment results indicate that the surrogate model reduces torque ripple, stator fluxlinkage offset, and feed system tracking error.

In industrial processes, quality variable prediction is important for process control and monitoring. Deep learning (DL) methods offer excellent prediction performance and potential paradigm shifts in quality variable modeling. However, in real-world production, the lack of offline labeled data and time-varying data distributions commonly exist, which seriously prohibits practical applications of DL-based predictive models. This paper introduces an enhanced quality variable prediction framework, Transfer-Incremental-Learning Parallel Stacked Autoencoders (TIL-PSAE), to address this challenge. TIL-PSAE integrates three key components: a parallel model structure, a transfer-learning (TL)-based offline training strategy that accumulates knowledge from multiple similar but different processes, and an incremental-learning (IL)-based online adaptation strategy. The model structure comprises two parallel SAEs for extracting process-invariant and target-process-specific features. Offline training involves sequential training using data from different processes, facilitating knowledge accumulation into different parts of model. During online adaptation, the accumulated knowledge remains unchanged while a new combination of knowledge is learned, thus improving online prediction accuracy and avoiding knowledge forgetting. The proposed model is applied to a sulfur recovery unit with four parallel sub-units. Experimental results demonstrate the effectiveness of the proposed model in both offline and online prediction performance.

Anthropomorphism of artificial systems is a key enabling factor to ensure effective and compelling human–machine interactions in different domains, including immersive extended reality environments and cobotics applications. Among the different aspects that anthropomorphism refers to, the generation of human-like motions plays a crucial role. To this aim, optimization-based techniques, whose functional cost is devised from neuroscientific findings, or learning-based approaches have been proposed in literature. However, these methods come with limitations, e.g., limited motion variability or the need for high dimensional datasets. In previous works of our group, we proposed to exploit functional Principal Component Analysis (fPCA) of human upper limb movements, to extract principal motion modes in the joint domain and use them to directly embed the human-like behaviour in the planning algorithm. However, this approach faces with translational issues related to the computational burden and to the application to kinematic structures different from the one used to describe human movements. To overcome this problem, we propose a general framework to generate human-like motion directly in the Cartesian domain by exploiting fPCA. This solution permits to perform obstacle avoidance with low computational time and it can be applied to any kinematic chain. To prove the effectiveness of our approach, we tested it against a state-of-the-art human-like planning algorithm both in terms of the accuracy of target reaching and human-likeness features of the generated movement.

Established model-based methods often use a combination of state feedback and observer to control complex systems. They rely on detailed mathematical models that are often hard to derive. Nonetheless, such methods may achieve a high level of accuracy, which justifies the cumbersome modelling. An alternative approach is model-free control, in a form introduced by Fliess and Join, where the system is approximated in a short time interval by a low-order differential equation with unknown parts, a so-called ultra-local model. This control method is a powerful tool, but the parametrisation and the concrete implementation may require time, effort, and experience. The present paper investigates the systematic tuning of a model-free controller for a magnetically supported plate that is modelled as an unstable multiple-input multiple-output system. Furthermore, the incorporation of model information into the model-free controller is investigated. These adaptations ultimately improve results by simplifying parameter tuning and interpretation of estimates. Several experiments are carried out on a test bed to show the capabilities of the proposed algorithms for set point stabilisation and trajectory tracking. The effects of the different parameters in the model-free controllers are addressed, and excellent robustness with respect to actuator faults is demonstrated. Filters for estimating derivatives and unknown quantities are designed using an open-source toolbox.