Data-driven Monitoring System Design And Online Optimation Methods

With the continuous expansion of the scale of modern industrial system,its complexity is also increasing.It also follows with problems such as external disturbance,non-linearity,system uncertainty and so on.How to effectively monitor the possible faults and anomalies of complex industrial system has also attracted more and more attention.As a result of the development of data network technology,the application of data acquisition and monitoring system(SCADA)is becoming more and more extensive.Therefore,through the reasonable use of the data collected by the SCADA system,bypassing the identification part of the model,designing the process monitoring system,monitoring the possible failure of the system is also a very important research direction.In this paper,a residual generator is constructed by establishing a diagnostic observer based on data-driven method,and an adaptive approach is designed for the possible disturbances in the system,so as to effectively monitor the industrial system online.Firstly,this paper introduces the design method of data-driven open-loop control system.In in order to diagnose industrial systems effectively and provide reliable diagnostic results for fault-tolerant control afterwards,the parameterized form of fault diagnosis observer is studied by means of coprime decomposition technology.On the basis of existing data-driven design methods,the one-to-one relationship between Luenberger equation and stable kernel representation function is established,so that the one-dimensional observer-based residual generator is realized.To expand to multidimensional,the design of process monitoring system based on multidimensional residual generator is studied.Secondly,due to the wide application of closed-loop system in industrial process,this paper extends the open-loop control method based on stable kernel representation to closed-loop control systems.Using the data generated by the closed-loop system,the kernel of the system is reconstructed,and the residual generator of the system can also be constructed using kernel,so as to monitor the the closed-loop system reasonablely.At the same time,considering the coupling between the input and output signal,the evaluation function is constructed by reconstructing the residual covariance matrix,and the real-time monitoring of the system performance is realized by comparing with the threshold.Furthermore,based on the adaptive observation technology,a reliable and efficient adaptive updating strategy is designed for low frequency disturbances which are commonly appeared in industrial systems.On the basis of not identificating the whole system,the adaptive strategy can only update the key parameters online to achieve the goal of tracking changes.The adaptive method is used to identify the possible disturbance signals such as sine wave in the actual system.At the same time,the adaptive method is improved by sliding window,and the estimation effect is improved when the system has random disturbance signals,so that the industrial system can be monitored reasonably.Last but not least,rolling systems are widely used in morden industries.Rolling systems use the hot rolling or cold rolling process in steel rolling.Strips can be classified by rolling extrusion into different specifications catagories.The application in the rolling mill model is presented to verify the effectiveness of the methods proposed above,and to deal problems that may occur in the process of steel rolling roll eccentricity,so as to realize real-time monitoring for rolling system.