Research On Operational Feedback Control For Double-layer Networked Industrial Processes

Along with the development of global economy,especially the introduction of ”Industrial 4.0”,the scale of the industrial production process is much larger and is more complex than before.In addition,due to the fierce competition in the international market,more and more industrial companies focus on improving their production energy efficiency,product quality level,and price-performance ratio to improve the production competitiveness.Based on the aforementioned market analysis and the double-layer industrial production processes,and by considering the networked transition channels between device layer and operation layer,the network-based double-layer industrial production processes is introduced in this dissertation.The subsystems at the device layer of the double-layer architecture are considered as local nonlinear ones,while by using the sampled signals from device layer,the index prediction function is constructed by using neural network technique.The system performance at the operation layer is optimized by introducing nonlinear model predictive control strategy,which can also compensate the network-induced phenomena.In this dissertation,the nonlinear control methods are regarded as the basic technologies,combined with additive technologies like: Lyapunov function design,fault tolerant control scheme,stochastic system stability theory,the control design problem for network-based double-layer industrial production processes is preliminarily studied.The main contributions are summarized as follows:(1).The sampled controller is build by using sampled signals and associated control gains in the traditional sampled control design problem,and zero-order-holder is introduced to guarantee that the signals transmitted to the actuator are continuous.According to the behavior of zero-order-holder,the sampled signals are constant during one sampling period,which results in that the overall control law for the operating period is piecewise function.The system considered in the beginning of this dissertation is a class of sampled nonlinear systems,and the fuzzy state estimator is first designed to guarantee that the designed control law is continuous which not only used the sampled signals from original physical plants but also used the signals from fuzzy state estimator.In this case,the designed control law is more robust and more suitable for systems with large sampling period.Based on the control design results for nominal nonlinear system,the nonlinear H? control problem is further considered.By combining the designed fuzzy state estimator,the prescribed H? performance index,fuzzy logic systems,and Backstepping design procedure,the adaptive fuzzy controller is constructed to guarantee the system stability and its H? performance.(2).Then for a class of network-based double-layer industrial processes,the control problem for the local device layer nonlinear subsystems is first considered.In order to solve this problem,T-S fuzzy modelling technique,neural network function or fuzzy logic systems are proposed to model the nonlinear plant,the unknown nonlinear terms in the system,or the associated state observer.The index prediction function is constructed via radial basis function and the sampled signals from device layer with smaller sampling period.For the transmitted channels between device layer and operation layer,the stochastic packet dropouts and network-induced delays are considered for both channels.Depending on the above results,the prescribed performance function is then introduced,by combining the aforementioned intelligent control strategy and model predictive control,a hybrid intelligent optimal control scheme is designed to guarantee the overall stability.(3).For the case that there exist stochastic actuator faults and stochastic networkinduced delays at device layer,two independent Markov processes are given to describe these two stochastic processes.The device layer controllers are designed to guarantee the stochastic stability of the discrete-time stochastic subsystems.The index prediction function is constructed via radial basis function and the sampled signals from device layer with smaller sampling period.The setpoints dynamic equation is given as general nonlinear discrete-time system,where the setpoint and index prediction error are viewed as the state variable and disturbance variable.The two channels are modelled as one based on the constraints of associated variables,and the compensator is designed to optimize the overall system performance.(4).Depending on the first three obtained results,and considering the fact that some complex nonlinear industrial process model is difficult to obtain.The data-driven based controller and compensator are design by optimizing two objective cost functions,i.e.,the tracking error cost function and index prediction error cost function for the operation layer.The tracking performance and the overall prescribed performance for the operation layer are guaranteed by the designed controller and compensator.Finally,the conclusion is given,which summarize the main contributions of this dissertation,and then the future research work are given.