ارزیابی عملکرد برای کنترل یادگیری تکراری واحدهای دسته ای

A new method is developed to estimate the minimum variance bounds and the achievable variance bounds for the assessment of the batch control system when the iterative learning control is applied. Unlike continuous processes, the performance assessment of batch processes requires particular attention to both disturbance changes and setpoint changes. Because of the intrinsically dynamic operations and the non-linear behavior of batch processes, the conventional approach of controller assessment cannot be directly applied. In this paper, a linear time-variant system for batch processes is used to derive the performance bounds from the routine operating batch data. The bounds at each time point computed from the deterministic setpoint and the stochastic disturbance for the controlled output variance can help create simple monitoring charts. They are used to track the progress easily in each batch run, to monitor the occurrence of the observable upsets, and to accordingly improve the current performance. The applications are discussed through simulation cases to demonstrate the advantages of the proposed strategies.

In the past two decades, the chemical industry has undergone significant changes as the energy cost is rising and the global competition in price and quality is increasing. In agile manufacturing, product objective changes dynamically with customers’ demands. The manufacturing trend is shifted from conventional continuous manufacturing plants toward flexible batch plants with multiple products [1]. This is especially true for manufacturing of high value-added products, such as bioproducts, pharmaceuticals, polymers, specialty chemicals, and semiconductor materials. The control design in batch processes is quite different from that in continuous ones. When the process is operated continuously, there is a variety of methodologies in the feedback control loop system to ensure closed-loop stability and to achieve acceptable steady-state performance with respect to setpoint and disturbance inputs. Because of the intrinsically dynamic operations of batch processes, the conventional approach of the controller assessment cannot be directly applied. The conventional controller design for the continuous process may not be able to achieve a specified profile response and to process raw materials into products in finite duration. The operational challenges of dynamic batch control have been discussed [2] and [3]. Interest in research and development of batch control based on iterative learning control has increased steadily since the term, iterative learning control (ILC), was first presented [4]. ILC of batch operation allows the extraction of information from the past batches to refine the new batch run and to improve the performance of tracking control for product quality. ILC utilizes a feedback controller for stabilizing the closed-loop system. It also uses a feedforward controller for designing the transient response of the operating profile. Numerous ILC schemes have been developed in the past decades. They enhanced the control performance over a fixed time interval iteratively [5], [6] and [7]. The effect of ILC on a continuous controlled system to improve the performance has been proven [8] and [9]. Comprehensive review of this topic is shown in Refs. [5] and [10]. The webpage for iterative learning control research is linked (http://www.ece.usu.edu/csois/ilc/ILC/index.html). Lee et al. integrated iterative learning control into conventional model predictive control and applied it to batch processes [11]. Because of the large variation of batch processes in the operation condition during a batch run, model predictive control with artificial neural networks came into place for on-line optimization [12]. Huzmezan et al. applied an adaptive control to a PVC reactor and an ethoxylated fatty acid reactor [13]. However, these control research papers focused mainly on design strategies. They did not show how good the current controller performance of the batch operation was in comparison with benchmark control. In control engineering practice, the control method is designed based on a nominal model. However, plant models are often subject to certain kind of uncertainty. The robust issues in ILC, like parametric uncertainties, iteration-domain frequency uncertainty, and iteration-domain stochastic uncertainty, have been discussed [14], [15], [16] and [17]. The aim of the controller performance assessment is to determine and measure the capability of control systems in order to improve the degradation performance. If the deterioration of controller performance cannot be identified in time, the malfunction would cause inconsistent product quality and monetary loss or even a significant impact on personnel, environmental and equipment safety. The performance assessment of the control loop based on the minimum variance was first presented by Harris [18]. Several techniques using the minimum variance have been proven useful in prioritizing the activities of process engineers, including monitoring and assessing the controller performance [19] and [20]. In the research, the controller performance is evaluated based on the output variance of the stochastic performance coming from the unmeasured disturbance driven by white noise. However, the controller performance of the batch operation is influenced not only by the unmeasured disturbance but also by the deterministic regulation which is defined by the setpoint changes. Qin mentioned that the performance assessment techniques could be categorized into stochastic performance monitoring and deterministic performance monitoring [21]. Since the deterministic regulation is very different from the stochastic one, their achievable performance bounds should be separated [22]. Although there were many research papers on assessing continuous control systems, to our best knowledge, the assessment of the batch control system was never mentioned. In this paper, two issues are addressed to assess the controller performance of the batch operation. First, the minimum variance performance bound is developed for batch operation systems. The performance bound can subsequently be used for the performance assessment of the batch system control loop. It is considered a benchmark of performance. Interestingly, the performance bound at each operating time point can be achieved by the traditional minimum variance control law. However, because of the controller structure, researchers in the past believed that the designed controller could not meet the theoretical variance benchmark when processes contained non-stationary disturbance [21] and [23]. Secondly, an achievable minimum variance performance bound is estimated for the controllers used in the batch system. The multiple objective optimization technique is adopted. It allows tracking the setpoint and rejecting the disturbance simultaneously. Once the performance bounds at each time point are set up, like traditional performance assessment approaches in the continuous system, they can create simple monitoring charts to track the progress in each batch run, to monitor the occurrence of the observable upsets, and to accordingly improve the current performance. The remaining paper is structured as follows: The problem of the performance assessment in the batch operation system is defined in Section 2. Without specifying the control task, it is difficult to assess the performance of the control loop. In Section 3, the performance assessment bounds of the batch control system are derived. With only the past operating data, the achievable performance and the corresponding optimal parameters based on the closed identification schemes are developed in Section 4. The effectiveness of the proposed method and its potential applications are demonstrated through two computer simulation problems, including a simple linear system and a non-linear batch reactor in Section 5. Finally, concluding remarks are made.

As the market is highly competitive, the product life cycle is becoming shorter and shorter. Batch processes played an important role in the chemical industry over the last couple of decades. Different ILC control strategies have been widely used in this field; however, improvement should be made to ensure the control objectives are accomplished within the specified performance for each batch run. Many processes have been around for years and engineers have acquired lots of experience, but many operational problems still go undiagnosed for a prolong period of time. In this paper, the performance bounds of ILC benchmark for the LTV batch operation system are developed. For practical operation consideration, the process data can be accessed from any time period at the touch of a button because most modern chemical processes utilize computer systems in which large amounts of data is stored cheaply and efficiently. Thus, developing the data mining technique that assesses the performance of the ILC system is a reflection of this emerging need to maintain good performance of the operating batch unit. On the basis of the operating data from the repeated tracking task which is run at a finite time interval in the time domain and an infinite repetition along the iteration domain, the two-stage method for modeling the process and the disturbance in turn is proposed. Eventually, the performance bounds at each time point can be estimated by multiple objective optimizations with the pre-identified process and disturbance models. They can be used to assess the performance of the setpoint tracking and the disturbance rejection. From the routine operation data, the advantages of the proposed method are demonstrated through simulated examples that explain how to build the achievable performance bounds and accurately identify the control performance of the current batch operation at each time point. This prototype will be used for field testing in our future study to investigate if the proposed approach is practically good enough for industrial batch applications. Extended research on the assessment of the control performance in MIMO batch systems will be included. In the MIMO system, there is an interaction between the variables. The performance bound cannot be directly estimated because each variable is completely independent of each other. This work will be our potential research topic in the future.