Analysis And Guarantee Of ILC Systems With Measurement And Control Signals Transmitted Over Wireless Networks

Compared to systems using point-to-point connection methods, networked controlsystems have advantages such as easy setup and maintenance, reduced weight andwiring. Especially for the system over a wireless network, signals travel through thewireless network from the sensor to the controller and from the controller to the actuator.Therefore, the controller can be separated from the system platform, which yields moredesign flexibility. For periodic systems, it is a useful method to adopt iterative learningcontrol (ILC) method for controller design in the wireless networked control systems.This method uses the output error obtained from the last trial to improve the input forthe next one, the input is adjusted and the output error converges to zero as the iterationgoes on through learning the output error.However, the introduction of the wireless network brings greater challenge to theILC system. Three main issues, including channel noise, communication delay and datadropouts, occur when measurement and control signals are exchanged among deviceslinked by the wireless network. As a result, the learning process of the controller isinterfered by these issues, and the ILC system performance is degraded or evendestabilized if it is not properly considered with these issues taken into account.In this thesis, the effect of these issues on the convergence performance of the ILCsystem with measurement and control signals transmitted over the wireless network areanalyzed, and some signal processing methods are proposed, respectively. Specifically,our contributions lie in the following aspects:1. Based on the ILC system model in presence of channel noise, therelationship between output error vector and channel noise vector is obtainedby super-vector formulation firstly, and then the norm of output errorcovariance matrix is employed to analyze the convergence performance of theILC system. The result shows that the output error covariance matrix normconverges to a limited error rather than zero with appearance of channel noises.Upper bound of the limited error is a function of noise variances, which meansthe variance of noise variances affects the convergence performance of the ILC system. Particularly, the effect of controller-to-actuator (CA) noise is largerthan that of sensor-to-controller (SC) noise.The convergence of output error is determined by the convergence ofcontrol error, and the relation between control error, channel noise and learninggain reveals the fact that the contribution of the SC noise and the CA noise tothe control error is influenced by the learning gain. Based on this discovery, amethod adaptively selects the learning gain is proposed. Specifically, thelearning gain is selected adaptively through minimizing the trace of controlerror covariance matrix. With the adaptively selected learning gain, theconvergence performance of the control error is guaranteed, and the then theconvergence performance of the output error is improved significantly.2. Based on the ILC system model in presence of one-step random delay orfixed delay, the effect of which on the convergence performance of the ILCsystem are studied through analyzing the variation of eigenvalues and otherelements in the lower triangular of the transition matrix, respectively. For thecase of one-step random delay, the results show that the convergence speed ofthe ILC system is reduced, and the robust convergence of the ILC system isalso affected. Especially, the effect of random control signal delay on robustconvergence of the system is larger than that of random measurement signaldelay. For the case of fixed delay, the analysis results show that theconvergence performance of the ILC system is diverged, and the accuratetracking of desired trajectory is impossible.Based on the assumption that the fixed delay suffered by measurementsignals and control signals can be measured, a compensation method isproposed to guarantee the convergence performance of the ILC system inpresence of fixed time delay. This method compensates the delay at thecontroller, and then the disordered sequence of measurement signals andcontrol signals are adjusted. With the fixed time delay compensation method,the convergence performance of the ILC system is guaranteed in presence offixed delay in measurement and control signals, which is proved throughanalyzing the element values of the system transition matrix. 3. Based on the ILC system model in presence of data dropouts, the effect ofwhich on the convergence performance of the ILC system is also studiedthrough analyzing the variation of eigenvalues and other elements in the lowertriangular of the transition matrix,. The analysis results show that the randommeasurement signal dropouts only reduce the convergence speed of the ILCsystem, but the robust convergence of the ILC system is affected by randomcontrol signal dropouts significantly.Based on the attribute that the control signal converges in iteration domain,a compensation method is proposed to guarantee the convergence performanceof the ILC system in presence of control signal dropouts. This method uses thesignal at the same time with the lost one but in last iteration to compensate thedata dropout at the actuator. As a result, the effect of data dropouts on theconvergence performance of the ILC system is compensated in the case ofexpensing some convergence speed of the ILC system.