E-News August 2017 Edition
The Joys of PID by Professor Mike Grimble
The proportional, integral, derivative controller remains the most popular feedback controller available. In fact, the derivative "D term" is often absent and the resulting PI controller provides reasonable performance for a huge range of processes. Consider first the good news:
- No mathematics required and even trial and error tuning is satisfactory.
- There are many industrial process applications where commercial industrial PID controllers are left set at default settings and still work.
- They seem to have some natural inbuilt robustness.
- Even staff untrained in control engineering, like process operators, can retune the controllers.
There are of course some downsides to PID control, which might be listed as:
- Not very suitable for multivariable systems, where interaction is significant.
- Cannot address multiple performance requirements formally, like optimising performance and disturbance rejection.
- Only limited tuning variables to tweak the system design.
From a theoretical viewpoint, a very good case can be made that PID controllers are of such restricted capability they cannot provide good performance, compared with advanced control solutions. However, practical experience reveals PID controllers perform very well on real systems. A problem not mentioned is that of uncertainties and nonlinearities, which can upset many advanced linear control solutions. It is perhaps surprising that PID controllers will often shrug off such difficulties and perform much better than would be predicted theoretically.
The question that arises is what lessons can be learned from such comparisons, and what should be done with PID designs. Since they are so successful, all of the techniques, which improve basic PID, should be considered. Some of these may be listed as:
- Anti-wind up protection is normally essential.
- If derivative action is to be included a high frequency filter is necessary for measurement noise attenuation.
- When feedforward is possible this normally provides faster and better responses.
- If the system is nonlinear some form of partial nonlinear compensation can be used that will often improve responses.
- It is sometimes desirable to add a fourth PID control term so that the PID controller is extended by a time constant term. This might be chosen to be effective in the mid frequency region.
- On occasions setting up the PID controller using a plant simulation can often reduce commissioning time and provide improved performance.
- For multivariable systems, the use of pre and post compensation to provide a system, which is almost diagonal, can allow a set of diagonal PID controllers to be used. The design is then for a set of scalar feedback control loop designs.
A PID controller provides the most reliable solution for an extensive range of control problems. Where it has deficiencies, solutions have often been found that are of a bespoke nature. The Smith Predictor, for compensating systems with transport delays, provides a good example. It will probably be many years before our distinguished Royal Leader "PID" is dethroned.
Linear and Nonlinear Predictive Control Training Course, Glasgow
ISC is pleased to announce that registration is now open for the Linear and Nonlinear Predictive Control Training Course being held on 21st & 22nd Nov 2017 in Glasgow City Centre.
This course provides a brief overview of various popular techniques for nonlinear control systems design. It will also introduce a new family of nonlinear industrial controllers referred to as Nonlinear Generalised Minimum Variance (NGMV). The new NGMV controllers are simple to drive theoretically and easy to implement. The wider family of controllers contains versions that in the limiting case of linear systems revert to well-known generalised minimum variance, Linear Quadratic Gaussian, and Generalised Predictive Controllers.
Assessing Companies Advanced Control Needs
There have been rapid advances over the last decade in the application of advanced control methods in a range of sophisticated control applications. In fact, there has been a notable shift in conference papers on control engineering away from pure theoretical developments towards more design and application papers. This is particularly noticeable in the areas of nonlinear control. Nonlinearities are often the cause of poor performance, partly because of the difficulties in controller tuning.
There was an era, roughly 20 years ago, when advanced multivariable linear control methods were starting to be used more frequently in applications. Most companies managed to shift from classical to true multivariable controls very easily. Unfortunately, many of the advanced methods for nonlinear systems require a deeper understanding and possibly experience, to be able to be used effectively. Many more companies are therefore seeking advice on which advanced control tools to use for their particular application problems.
Fortunately, one can learn from experience on different nonlinear control applications. It is still the case that although applications look very different physically, the actual system models and problems are very similar. The ACTC organisation and its parent company ISC Ltd. have a long experience in assessing the value of advanced controls in different applications. It is now often called upon to provide such guidance either in quite substantial design projects or in very brief advisory studies.
One of the problems that does not receive much attention but is important is how to assess what is good control, particularly for systems with severe nonlinearities. This does of course overlap the subjects of controller benchmarking and performance assessment. This is also an area where rapid progress has been made in recent years.
If you would like further details please contact Dr Meghan McGookin.
Potential for Adaptive Systems
There is considerable potential for adaptive control and signal processing methods. Advances in software techniques and in computing power provide some of the necessary tools to enable a breakthrough in technology to be achieved. Unfortunately, theoretical advances in adaptive control or in adaptive estimation have not really progressed at the same speed. What engineers in industry often think of as adaptive control systems really date back a couple of decades, at least as regards the basic concepts?
This might also be said of auto-tuning methods, which provide an automatic controller tuning capability. However, at least in the case of auto-tuning, there has been wide acceptance in industry, and the various methods have been improved in the light of industrial experience. This is not true for adaptive systems, where progress has been painfully slow in industrial applications and in other application areas. The good news is that this provides a great opportunity for both new theoretical results and practical advances.
It is of course true that many papers on adaptive control and signal processing have been published in recent times, but for whatever reason the techniques described have not caught the imagination of engineers in industry. Most of the adaptive methods developed are of course universal solutions to general control problems and it may be that this is too loftier a goal. Simpler and more practical adaptive systems might be those that only treat special cases by exploiting the structure of the application involved. Since such solutions are not generally applicable they are not so interesting from an academic research perspective, but they may be more useful in applications.
An adaptive system of the future may remove much of the stress in designing new control systems and be able to account for changing conditions and uncertainties in a robust and reliable manner. The promise of such systems is therefore great. If a breakthrough can be made, which is soundly based and practical to implement, then it will revolutionise both the design and the commissioning of industrial controllers.
Forthcoming ISC Training Courses
EDF, Barnwood, September 2017
A bespoke 5-day training course in control engineering practices for the nuclear industry will be redeliver to EDF CCGT Barnwood this September. The training course is delivered by ISC staffs (Dr Andy Clegg and Dr Petros Savvidis) and also EDF control expert on particular EDF case study. Interactive computer based hands-on (in LabVIEW) are used throughout the training course to reinforce the lecture material.
EDF, West Burton, September 2017
ISC will be delivering a 3 days Control Fundamentals training course to EDF West Burton CCGT power plant. This particular training course focuses on applying control engineering for power generation power such as combustion control and also discussing common control issues.
Ford, Detroit Training course, November 2017
Prof Mike Grimble and Dr Pawel Majecki will be delivering two training courses to Ford, Detroit in November 2017. The first two days of the training course will focus on applying optimisation, estimation and predictive control to linear systems. And the subsequent 2 days training course will focus on applying similar techniques but to nonlinear systems.
If you would like further details on any of these courses please contact Dr Meghan McGookin.
Scheduling Controllers for Nonlinear Applications
Over the last two decades there have been many advances in nonlinear control so it is reasonable to ask whether the use of scheduling in nonlinear systems is still relevant, or it is old fashioned? In fact, the vast majority of industrial applications use scheduling type techniques. The main reason is of course they are simple to understand and relatively east to implement.
The basic idea is of course to compute a number of linearized models and complete designs and tuning for each local operating point. These controllers can then be stored away and used in an online control scheme, where the appropriate controller is switched in depending upon the region of operation. There are possible problems of switching between controllers, requiring bumpless transfer solutions, but if the controllers are realised in state space form these problems are mitigated.
There are also problems concerning the choice of representative operating points and the type of models computed at each point. For example, the order of the models may be different at different operating points. This would give rise to different numbers of states in any state-equation based controller. It is clearly desirable to avoid problems by having the same number of states for different operating points, to simplify the linear state-estimators or observers used.
The old ideas of scheduling can be placed in more of a modern context since they can be considered a form of multiple model control. In its most general form a multiple model control system can involve banks of estimators which provide a type of adaptive control solution. So called piecewise affine state space representations provide a formal framework for theoretical analysis of this type of problem. In fact the PWA model provides the "simplest" extension of linear system models that can model non-linear and non-smooth processes with arbitrary accuracy. The PWA model is reasonably general, covering discrete-time linear systems with static piecewise-linear characteristics, discrete-time linear systems with logic states and inputs, and switching/hybrid systems.
Many different types of control law can of course be used in such an approach, but optimal control methods, based upon linear quadratic or linear quadratic Gaussian approaches are common. The Hinfinity robust controls can also be applied to this type of problem, and in some cases, an effort has been made to use the uncertainty models in the Hinfinity design to cope with the modelling errors that inevitably arise.
Thus, whether the old ideas of simply switching in different controllers in different operating regions is used, or whether this is placed in a more formal modern framework, the use of scheduling, or multiple model control, is very likely to continue for many years.
Importance of Kalman Filtering
The Kalman filter is one of the most useful tools in control engineering. Surprisingly, it is not as well-known as might be expected. The name Kalman filtering is a little misleading, since the main purpose is normally to provide estimates of the states of the system. It is true that the measurements that feed the filter are assumed contaminated by white measurement noise. In this respect, the device does act as a filter since it can provide an estimate of the noise free output. However, it performs more like a state observer, in that it provides estimates of states or signals that cannot be measured directly. In most applications, there may be many more states than actual output measurements.
The difference between a state observer and a Kalman filter is that the estimates are optimal. That is, the Kalman filter gain is chosen to ensure the minimum variance of the state estimation error is achieved. The gain in an observer may be chosen based upon some other requirement such as the observer pole locations.
Note that a Kalman filter is not always part of a control loop. That is, the filter maybe applied to the output of a process to reconstruct states for condition monitoring purposes. In fact, many fault monitoring and detection schemes have utilised the Kalman filter or an extended Kalman filter in some way. The extended Kalman filter is of course needed for nonlinear systems and it provides an estimate, which is sub- optimal, relative to a true nonlinear estimator.
The way the Kalman filter works is that it includes a model of the process and any known plant inputs are input to the Kalman filter. Since the model of the process is known, the output of the model should follow that of the real plant. In practice, the real system will have disturbance inputs, which the model may not contain. There is therefore likely to be a divergence between the states of the real plant and those estimated. The Kalman filter includes a computation of the error between the plant output (measured) and that estimated. A proportion of this error is then fed through a gain, referred to as the Kalman-filter gain, to provide a correction to the state estimates.
If for example, the estimate of the plant output is lower than the actual plant output a positive signal is fed into the model for the Kalman filter until the estimated output reaches the actual output. If the output of the real system is higher than the Kalman filter output a negative signal is fed into the state estimate model to lower the estimated output signal. This action can be recognised as a classic negative feedback action and it involves a feedback loop around the filter itself.
The Kalman filter is one of the most effective devices and most flexible available to the control engineer. It may be used as part of a Linear Quadratic Gaussian feedback controller, or used in a fault detection system. It can be used to estimate unmeasurable signals, and be modified to provide linear or nonlinear parameter estimation capabilities. It is highly recommended.
Introduction to Pharmaceutical Process Control, ADDoPT Program, University of Leeds
ISC Ltd has successfully delivered a 2-days training course in University of Leeds as part of the ADDoPT Program with Perceptive Engineering. This training course focuses on illustrating the importance of process control within pharmaceutical industry. The training course is targeted at engineer or technician who as very little and no control engineering background. The next training course is scheduled in October 2017 in Daresbury.
Book Review: Control and Dynamics in Power Systems and Micro Grids, by: Lingling Fan, Published By: CRC Press,Cost: £92
The author of this book produced the main part of this text in the spring of 2016 when providing a course at the University of South Florida in Tampa. This is a challenging area and many examples were developed for the courses which are included in the text. The dynamics of synchronous generators is of course an important topic and these are treated using space vector concepts. Another element of the problem is the use of power electronics and important topics such as droop control used for power system power sharing is included.
Traditional texts on power system dynamics and control have not linked the renewable energy, power electronics and micro grids, which arise in modern international power systems. The material on renewable energy systems was particularly of interest and the text includes MATLAB code and numerous problems and examples. The text is aimed at students rather than engineers in industry, but it will be valuable to both. The text is nicely presented with a logical layout of the text and its presentation.
This is certainly a book that can be recommended for the bookshelves of engineers working on power systems, power electronics and renewable energies. The hard back is £92 but there is an e-book available at £64.40. It includes more than 200 pages of useful material.
Book Review: MM Optimization Algorithms, by: Kenneth Lange, Published By: SIAM,Cost: $74
This new text is on MM optimisation algorithms, a particular class of optimization algorithm, which satisfy the ascent or descent property. They have numerical advantages like avoiding large matrix inversions and can turn a non-differentiable problem into a smooth problem. Moreover, they can deal with equality and inequality constraints.
This book is really aimed at those working in high dimensional optimisation problems. The author is a well-known researcher in the subject and the text has been written for student use. Useful references, examples and problems are included. At just over 220 pages it is not too long and provides a very clear exposition of the subject. It should be useful to all those working in the areas of system optimisation, whether researchers or engineers in industry. SIAM members will obtain a reduction on the price of this book.
Book Review: Engine Modeling and Control, Modeling and Electronic Management of Internal Combustion Engines by: Rolf Isermann , Published By: Springer,Cost: £153 (Hardcover) £122 (Ebook)
This new text is very useful for engineers working on the control of both gasoline and diesel engines. It provides a very comprehensive treatment and goes into the details of engine control in much greater depth than many texts which are more aimed at control design methods.
One of the most important parts of the control problem is to generate suitable engine models from either physical equations, or using system identification methods. This is not usually the most interesting aspects from a control viewpoint, but it is an essential step and this text reflects its importance through the model information provided. Model based modelling, neural network and systems identification techniques are all covered.
There are separate chapters covering the different aspects of gasoline and of diesel engines. The very important and topical subject of emissions control in diesel engines is covered. There is a generous use of diagrams and figures to make the text easy to read.
Professor Rolf Isermann has been considered a leading researcher in advanced control methods over many years and this is a very authoritative and interesting contribution to the literature reflecting his long experience and attention to detail. This text is highly recommended both for engineers in industry and for University researchers that need to understand the practical as well as the theoretical challenges in engine control.
- Linear and Nonlinear Predictive Control Training Course, Glasgow
- Assessing Companies Advanced Control Needs
- Potential for Adaptive Systems
- Forthcoming ISC Training Training Courses
- Scheduling Controllers for Nonlinear Applications
- Importance of Kalman Filtering
- Introduction to Pharmaceutical Process Control, ADDoPT Program, University of Leeds
- Book Review: Control and Dynamics in Power Systems and Micro Grids
- Book Review: MM Optimization Algorithms
- Book Review: Engine Modeling and Control, Modeling and Electronic Management of Internal Combustion Engines
Forthcoming ISC/ACTC Events
Early Bird Discounts Avaliable
We are offering an "Early Bird Discount" if registration is received 2 weeks before the course start date (see online registration form). Employees of ACTC member companies are entitled to two places free of charge. Note that all courses offered can also be provided at your company premises through special arrangements, please contact us for more information.