On-line adaptive control

Adaptive control has been an active area of research for many years. The fullblown ideal adaptive controller continuously identifies (on-line) the parameters of the process as they change, and retunes the controller appropriately. Unfortunately, this on-line adaptation is fairly complex and has some pitfalls that can lead to poor performance. Also, it takes considerable time for the on-line identification to be achieved, which means that the plant may have already changed to a different condition. These are some of the reasons why on-line adaptive controllers are mot widely used in the chemical industry. 

However, the main reason for the lack of wide application of on-line adaptive control is the lack of economic incentive. On-line identification is rarely required because it is usually possible to predict with off-line tests how the controller must be retuned as conditions vary. The dynamics of the process are determined at different operating conditions, and appropriate controller settings are determined for all the different conditions. Then, when the process moves from one operating region to another, the controller settings are automatically changed. This is called openloop-adaptive control or atn scheduling. 

The one notable case where on-line adaptive control has been widely used is in pH control. The wide variations in titration curves as changes in buffering occur makes pH control ideal for on-line adaptive control methods. 

Several instrument vendors have developed commercial on-line adaptive controllers. Difficulties have been reported in two situations. First, when they are applied in a multivariable environment, the interaction among control loops can cause the adaptation to fail. Second, when few disturbances are occurring, the adaptive controller has little to work with and its performance may degrade drastically. 

The design of digital control systems will be the subject of Chapter 30, and Chapter 31 will treat the question of experimentally modeling a process. Finally, the on-line coordination of an experimental modeling procedure with a control algorithm will be examined in Chapter 31, in an attempt to develop on-line adaptive control systems. 

In this chapter we present an experimental approach, known as process identification, which can be used to construct a reliable model, either before or after the process has been placed in operation. In addition, we study how process identification can be coupled with various control systems to yield on-line adaptive control strategies. 

It should be emphasized that process identification and on-line adaptive control require extensive computations, which can be per-... 

In this chapter we have presented a rather simplistic view of the on-line adaptive control systems. There are a number of very important questions which have not been addressed, such as whether the parameter estimates are biased, the interplay between estimation and control, and the stability characteristics of the adaptive controller. A thorough examination of these questions is beyond the scope of this text. The interested reader can consult the relevant references at the end of Part VII. 

In the following papers the reader can find some typical examples of on-line adaptive control for various processes of interest to chemical engineers ... 

The one notable case where on-line adaptive control has been widely used is in pH control. The wide variations in titration curves as changes in buffering occur make pH control ideal for on-line adaptive control methods. Several instrument vendors have developed commercial on-line adaptive controllers. Seborg, Edgar, and Shah (AIChE Journal 32 88 1, 1986) give a survey of adaptive control strategies in process control. 

Image analysis is an important aspect of many areas of science and engineering, and imaging will play an important role in characterizing self-assembled structures as well as in on-line process control. Development of effective noise identification and suppression, contrast enhancements, visualization, pattern recognition, and correlation algorithms should be co-opted where possible and adapted to the analysis of self-assembled structures. 

Figure 31.3 dramatizes the effect of the on-line adaptation on the quality of the closed-loop response versus the case of controlling without controller adaptation. 

On-line adaptation is not limited to feedback systems. On-line process identification can be coupled easily with feedforward, inferential, and other control systems, thus expanding the range of their applicability. Adaptation is particularly valuable for feedforward and inferential systems because they rely heavily on good process models for their successful implementation. 

Why do you need on-line adaptation of control systems Can you improve the quality of a process model using input-output data when the process output remains close to the desired set point Explain your answer. 

Describe an on-line adaptive procedure for a typical feedforward control system (see Chapter 21). Do the same for the inferential control of a distillation column (see Example 22.S). 

Neural Network-Based Optimizing Controller With On-Line Adaptation... 

This work briefly shows some findings about the influence of RTE parameters on the process economical performance. As a result, a methodology for an adequate tuning of such parameters is proposed. It is shown how the parameters related to control variables can be tuned just by using the steady state model. On the other hand, the time parameter needs both, the characterisation of the disturbance in terms of amplitude and frequency and a further testing over a dynamic simulation of the process. In addition, a periodical characterisation of the disturbances allows an on-line adaptation of the parameters. 

Ryhiner, G., Dunn, I.J., Heinzle, E., Rohani, S., 1992. Adaptive on-line optimal control of bioreactors apphcation to anaerobic degradation. Journal of Biotechnology 22, 89—106. 

Pig. 4. Photo dissociation of ArHCl. Left hand side the number of force field evaluations per unit time. Right hand side the number of Fast-Fourier-transforms per unit time. Dotted line adaptive Verlet with the Chebyshev approximation for the quantum propagation. Dash-dotted line with the Lanczos iteration. Solid line stepsize controlling scheme based on PICKABACK. If the FFTs are the most expensive operations, PiCKABACK-like schemes are competitive, and the Lanczos iteration is significantly cheaper than the Chebyshev approximation. 

A wide variety of particle size measurement methods have evolved to meet the almost endless variabiUty of iadustrial needs. For iastance, distinct technologies are requited if in situ analysis is requited, as opposed to sampling and performing the measurement at a later time and/or in a different location. In certain cases, it is necessary to perform the measurement in real time, such as in an on-line appHcation when size information is used for process control (qv), and in other cases, analysis following the completion of the finished product is satisfactory. Some methods rapidly count and measure particles individually other methods measure numerous particles simultaneously. Some methods have been developed or adapted to measure the size distribution of dry or airborne particles, or particles dispersed inhquids. 

Feedback provided by on-line monitoring of self-assembling processes will play an increasingly important role in controlling the microscopic and macroscopic architecture of molecular assemblies. Successful adaptation of char-... 

G. Bastin and D. Dochain. On-Line Estimation and Adaptive Control of Bioreactors. Elsevier, Amsterdam, 1990. 

Of the analytical techniques available for process analytical measmements, IR is one of the most versatile, where all physical forms of a sample may be considered - gases, liquids, solids and even mixed phase materials. A wide range of sample interfaces (sampling accessories) have been developed for infrared spectroscopy over the past 20 to 30 years and many of these can be adapted for either near-lme/at-lme production control or on-line process monitoring applications. For continuous on-line measurements applications may be limited to liquids and gases. However, for applications that have human interaction, such as near-line measurements, then all material types can be considered. For continuous measurements sample condition, as it exists within the process, may be an issue and factors such as temperature, pressure, chemical interfer-ants (such as solvents), and particulate matter may need to be addressed. In off-line applications this may be addressed by the way that the sample is handled, but for continuous on-line process applications this has to be accommodated by a sampling system. 

---