Developing a neural network-based run-to-run process controller for chemical-mechanical planarization

A new neural network-based run-to-run process control system (NNRtRC) is proposed in this article. The key characteristic of this NNRtRC is that the linear and stationary process estimator and controller in the exponentially weighted moving average (EWMA) run-to-run control scheme are replaced by two multilayer feed-forward neural networks. An efficient learning algorithm inspired by the sliding mode control law is suggested for the neural network-based run-to-run controller. Computer simulations illustrate that the proposed NNRtRC performs better than the EWMA approach in terms of draft suppression and adaptation to environmental change. Experimental results show that the NNRtRC can precisely trace the desired target of material removal rate (MRR) and keep the within wafer non-uniformity (WIWNU) in an acceptable range.     

Developing a neural network-based run-to-run process controller for chemical-mechanical planarization

A new neural network-based run-to-run process control system (NNRtRC) is proposed in this article. The key characteristic of this NNRtRC is that the linear and stationary process estimator and controller in the exponentially weighted moving average (EWMA) run-to-run control scheme are replaced by two multilayer feed-forward neural networks. An efficient learning algorithm inspired by the sliding mode control law is suggested for the neural network-based run-to-run controller. Computer simulations illustrate that the proposed NNRtRC performs better than the EWMA approach in terms of draft suppression and adaptation to environmental change. Experimental results show that the NNRtRC can precisely trace the desired target of material removal rate (MRR) and keep the within wafer nonuniformity (WIWNU) in an acceptable range.     

Adaptive Control of Neutralization Process using Neural Networks

Abstract

A neural network (NN) based adaptive interaction technique is proposed for controlling highly nonlinear neutralization processes. In this approach, the controller is decomposed into interconnected subsystems and adaptation occurs during the interactions. This approach is adaptive in structure and doesn't use an explicit model of the process in the design. The NN is used to establish the adaptive interaction technique for the development of a nonlinear pH controller, which calculates the necessary change in a manipulated variable to drive the system to the desired value. By applying this adaptive algorithm, the same adaptation as the back‐propagation algorithm is achieved without the need of backward propagating the error throughout a feedback network. This important property makes it possible to adapt the NN controller directly without a process model. This advantage reduces the computational complexity drastically in comparison to the well known back‐propagation algorithm based adaptive NN system and a model based system. The designed model‐free online adaptive controller was implemented to a laboratory scaled pH process in real time by use of a dSPACE 1104 interfacing card. The responses of pH and acid flow rate show good tracking for both the set point and load changes over the entire nonlinear region.     

A neural network-based adaptive algorithm on the single EWMA controller

The single EWMA controller has been proven to have excellent performance for small disturbances in the run-to-run process. However, incorrect selection of the EWMA parameter can have the opposite effect on the controlled process output. An adaptive system is necessary to automatically adjust the controller parameters on-line in order to have better performance. In this study, a simple and efficient algorithm based on neural networks (NN) is proposed to minimise the inflation of the output variance on line. The authors have shown that the sequence of EWMA gains, generated by a NN-based adaptive approach, converges close to the optimal controller value under IMA (1, 1), step and trend disturbance models. The paper also shows that the NN-based adaptive EWMA controller has a superior performance than its predecessors.     

NC polishing of aspheric surfaces under control of constant pressure using a magnetorheological torque servo

In this paper, a method of maintaining a constant polishing pressure is proposed for a NC polishing system by controlling the polishing force during the polishing process. First, the NC polishing system is developed to resolve the force–position coupling problem encountered in common polishing processes. It mainly consists of a force control subsystem based on a magnetorheological torque servo to provide a controllable torque to polishing tool to generate the polishing force and a position control subsystem based on a general CNC lathe to control the position of the polishing tool. Second, a constant polishing pressure model is established by controlling the polishing force according to the variation of the curvature of the aspheric surfaces, and the polishing parameters for model are planned. Then, the control model of the polishing system is proposed, and a PID controller is designed for torque tracking with the actual torque feedback from a torque sensor. Finally, polishing experiments are conducted with constant force and constant pressure, respectively. Experimental results show that the surface roughness is greatly improved, the aspheric surfaces can be polished more uniformly with constant pressure than with constant force, and the PID controller can meet the requirements for the polishing force control.     

Adaptive Control of Neutralization Process Using Recurrent Neural Networks

Abstract

This paper presents Recurrent neural Network (RNN) based adaptive control scheme for a pH neutralization process which is difficult to control due to its nonlinear dynamics with uncertainties. The proposed design comprises of both RNN estimator which adapts online and a RNN controller. Desired performance of the system is ensured by the parallel operation of both. The estimator weights are updated recursively by back propagation algorithm and controller weights are modified by steepest descent approach. Stability and convergence of proposed controller is guaranteed by Lyapunov stability analysis. Servo and regulatory performance of the system thus obtained by simulation is compared with a model based IMC controller. The RNN based controller is exhibits better performance as shown by the control simulation of a nonlinear pH neutralization process.     

Cascade control of superheated steam temperature with neuro-PID controller

In this paper, an improved cascade control methodology for superheated processes is developed, in which the primary PID controller is implemented by neural networks trained by minimizing error entropy criterion. The entropy of the tracking error can be estimated recursively by utilizing receding horizon window technique. The measurable disturbances in superheated processes are input to the neuro-PID controller besides the sequences of tracking error in outer loop control system, hence, feedback control is combined with feedforward control in the proposed neuro-PID controller. The convergent condition of the neural networks is analyzed. The implementation procedures of the proposed cascade control approach are summarized. Compared with the neuro-PID controller using minimizing squared error criterion, the proposed neuro-PID controller using minimizing error entropy criterion may decrease fluctuations of the superheated steam temperature. A simulation example shows the advantages of the proposed method.     

<Emphasis Type="Bold">Application of Neural Networks in Injection Moulding Process Control</Emphasis>

In order to produce precise injection moulding products, a closed-loop controller is employed instead of the open-loop control of a traditional injection moulding machine for monitoring the filling and post-filling phases of the injection processes. Since the injection moulding system has complicated and variable dynamics, the classical control theory is difficult to implement for the precise injection moulding processes. Here, two intelligent neural network control strategies are employed to adjust the injection speed of the filling phase and control the nozzle pressure of the post-filling phase. Since the neural controller has learning ability to track the variation of the injection processes, this control strategy has the advantages of adaptivity and robustness for general purpose application to an injection moulding machine. The experimental results show that this controller has good performance in the actual injection moulding processes.     

用于气动伺服系统的自适应神经模糊控制器

研究了一种基于压力比例阀的气动伺服系统自适应神经模糊控制器。其中的神经网络辨识器（NNI）通过高线训练可以充分逼近非线性动态系统的模型,并能够在线调整模糊控制器的控制规则。系统的位置控制精度和伺服特性有了很大改善。试验结果表明,所提出的控制器对该气动伺服系统具有很好的控制特性以及很强的自适应能力。     

Adaptive Control of Multivariable Process Using Recurrent Neural Networks

Abstract

Industrial processes are naturally multivariable in nature, which also exhibit non-linear behavior and complex dynamic properties. The multivariable four-tank system has attracted recent attention, as it illustrates many concepts in multivariable control, particularly interaction, transmission zero, and non-minimum phase characteristics that emerge from a simple cascade of tanks. So, the multivariable laboratory process of four interconnected water tanks is considered for modeling and control. For processes which show nonlinear and multivariable characteristics, classical control strategies like PIDs have performance limitations. Hence, intelligent approaches like Neural Networks (NN) is an important term in this juncture. The use of Recurrent Neural Network (RNN) is apt for modeling and control of nonlinear dynamic processes as it contains the past information about the process. The objective of the current study is to design and implement an adaptive control system using RNN for a nonlinear multivariable process.

The proposed adaptive design comprises an estimator based on RNN, which adapts online and predicts one step ahead output. A Recursive Least Square (RLS) based back propagation algorithm is used for training the network. The controller used is also a RNN, which minimizes the difference between the predicted output and reference trajectory. The objective function is minimized using a steepest descent algorithm which gives the optimum control input. Desired performance of the system is ensured by the parallel operation of both. The proposed control strategy is implemented in a laboratory scale four tank system. The trajectory tracking and disturbance rejection response obtained are compared with the response obtained by using a well designed decoupled, decentralized IMC controller.     

超精密车床溜板的模糊神经网络主动振动控制研究

提出了溜板的振动主动控制。作为器采用自行研制的主动空气轴承,实现无摩擦接触。控制方法彩和模糊神经网络控制,模糊控制为主控制器,用以逼近实际的复杂时变系统;神经网络控制补偿耦合效应。实验结果表明,基于主动空气轴承地模糊神经网络振动主动控制可以有铲减小溜板振动。     

非对称泵缸系统模型跟踪控制研究

用神经网络逼近非对称泵缸系统的非线性逆模型，通过神经网络模型参考控制实现电液斜盘位置和流量的控制，使系统的不确定性和非线性等得到补偿。仿真和实验表明，所开发的神经网络控制器能够较好地实现模型跟踪控制，有较好的自适应性和鲁棒性，跟踪性能有较大改善。     

基于神经网络的通用模型自适应控制

应用复合正交神经网络来实现过程的自适应逆控制方法,和通用模型控制器策略相结合,提出了一种基于神经网络的通用模型自适应控制方法,将非线性过程模型应用逆系统的方法可以在控制算法中直接嵌入过程模型,从而保证通用模型控制策略的可实现性.另一方面,在自适应逆控制中采用复合正交神经网络具有算法简单、学习收敛速度快等优点,可以克服常用的BP和RBF神经网络一些缺点.基于神经网络的通用模型自适应控制方法中的参考轨迹是一条典型的二阶曲线,该控制器参数具有明显的物理意义,参数整定方便.仿真验证了该控制策略的有效性.     

基于比例阀的气动伺服系统神经网络控制方法的研究

研究了用神经网络PID控制器对基于比例阀的气动伺服系统进行控制的方法。用神经网络辨识器来逼近非线性动力学系统,并在线修改控制参数。实验及分析表明,适当的选择网络参数,经过充分的离线训练,该控制器可以进行在线的自适应控制,系统的控制精度和动态特性有明显提高,且在环境参数变化时,控制器具有在线自学习和自整定参数的能力。     

High-accuracy servo press system for the clinching joint process

This paper focuses on the quality improvements on clinching joints using a servo press with a Radial basis function neural network and a sliding mode (RBFS) control strategy. Bottom thickness, which is affected by the press punch position, is usually used to monitor clinching joint quality. Traditional clinching presses are driven by pneumatic pistons or motors that provide feedback on punch force or motor position. However, this feedback is indirectly related to the joint bottom thickness. Clinching workers who set the control parameters on these presses depend on tests and statistics. Thus, this paper presents a servo press system that utilizes punch position feedback to directly control the joint bottom thickness. Transmission errors are considered for the movement accuracy of the servo press. A mathematical model of the servo press is established for analyzing. An algorithm, which combines RBF neural network and sliding mode, is proposed and applied for press position tracking. This algorithm adopts an RBF neural network to approximate the nominal model of the press system. The update law of the algorithm is based on the Lyapunov function used to prove the stability of a closed-loop system. The sliding mode controller compensates for the neural network error and disturbance. Finally, experiments are executed on the servo press with an RBFS controller. To evaluate the performance of the proposed method, a fuzzy PID controller is also applied to the press for comparison. The results indicate that the servo clinching press system with RBFS efficiently and accurately control the clinching jointing process.

     

焦化鼓风机系统智能控制策略研究及应用

针对焦化鼓风机系统具有非线性时变、多变量、强耦合及存在随机干扰的特点,通过采用基于最近邻聚类方法的RBF神经网络快速学习算法,实时在线辨识,建立被控对象的精确逆模型并用于控制,实现了将具有强耦合特性的多输入多输出(MIMO)系统解耦成单个独立的伪线性对象,并提出一种基于RBF神经网络逆控制与非线性比例积分微分(PID)控制相结合的智能控制策略,保证了系统稳定的同时改善了控制系统性能.仿真和应用结果证实了该控制策略具有快速适应对象和过程变化的能力及较强的鲁棒性.     

Adaptive neural network controller for the molten steel level control of strip casting processes

The twin-roll strip casting process is a steel-strip production method which combines continuous casting and hot rolling processes. The production line from molten liquid steel to the final steel-strip is shortened and the production cost is reduced significantly as compared to conventional continuous casting. The quality of strip casting process depends on many process parameters, such as molten steel level in the pool, solidification position, and roll gap. Their relationships are complex and the strip casting process has the properties of nonlinear uncertainty and time-varying characteristics. It is difficult to establish an accurate process model for designing a model-based controller to monitor the strip quality. In this paper, a model-free adaptive neural network controller is developed to overcome this problem. The proposed control strategy is based on a neural network structure combined with a sliding-mode control scheme. An adaptive rule is employed to on-line adjust the weights of radial basis functions by using the reaching condition of a specified sliding surface. This surface has the on-line learning ability to respond to the system’s nonlinear and time-varying behaviors. Since this model-free controller has a simple control structure and small number of control parameters, it is easy to implement. Simulation results, based on a semiexperimental system dynamic model and parameters, are executed to show the control performance of the proposed intelligent controller. In addition, the control performance is compared with that of a traditional PID controller.     

Research on intelligent launching control of dual clutch transmissions based on adaptive neural fuzzy inference system

Vehicle launching has an important influence on driving performance of the vehicle. For vehicles with dual clutch transmissions (DCT), the clutch torque control is the key to the launching control. Therefore, a data-driven control method for DCT launching process based on adaptive neural fuzzy inference system (ANFIS) is proposed. Firstly, the vehicle test data during launching process is collected and the optimal clutch torque is obtained based on multi-objective particle swarm optimization (MOPSO). Afterward, to learn the launching control rules from optimization results, the combination of neural network and fuzzy logic algorithm, referred to as an ANFIS, is established. The dataset of the optimized launching clutch torque is utilized to train the ANFIS controller. Finally, the simulation and test results show that the data-driven control can accurately learn the launching control rules from the optimality, thereby achieving the optimal control for different launching intentions.

     

基于小波神经网络的控制方法及其应用研究

提出一种基于小波神经网络的控制方法,该方法利用两个小波神经网络作为控制系统的辨识器和控制器来构成控制系统。小波神经网络辨识器能更准确逼近非线性对象,小波神经网络控制器能产生复杂的最佳控制规律。仿真结果表明系统具有逼近精度高、控制效果好、抗干扰能力强等优点。     

基于神经网络的缠绕过程张力积分鲁棒控制

纤维缠绕系统是典型的非线性系统,缠绕过程张力控制精度决定了缠绕制品成型品质,然而系统非线性特性、摩擦及外干扰等严重制约了缠绕过程张力控制性能的提升。目前现有方法主要以收/放卷两轴同步控制为基础,通过解耦等复杂操作展开张力控制研究,计算量大且不利于控制算法的应用。为了避免上述问题并准确描述缠绕系统张力产生机理和实际的摩擦特性,建立简化的缠绕系统非线性数学模型。同时以自适应作为神经网络权值训练方法,基于自适应神经网络设计出干扰量的逼近函数,在基于连续积分鲁棒（RISE）控制器设计的控制律中补偿扰动的影响,并基于Lyapunov稳定性理论证明了系统的渐近稳定性。最后,给出所提出控制器与现有方法的对比验证实例,结果表明所提出基于神经网络扰动补偿的积分鲁棒控制显著增强了系统抑制外干扰的能力,提升了张力控制精度。