E-News May 2015 Edition

The Future is Now by Professor Mike J Grimble

It is difficult for companies to keep up with every technological development and this applies to all areas of engineering. In the general area of advanced control systems, modelling, simulation and condition monitoring, ISC and the ACTC provide some services to help that process including this newsletter. However, companies often need more than this to make sure they are up to date and have a good view on future developments that will affect their business. There is therefore increasing interest and activity in undertaking technology reviews which may involve a review of the main practices and procedures that are followed by the control and instrumentation groups within the company.

Such reviews are undertaken in some cases at the very practical level of considering equipment installed, procedures to handle faults and maintenance, and guidelines needed in the case of emergencies. In other circumstances it may be main design methods and strategies being used by a company’s development team that are considered. Such companies need design tools and procedures which are easily transferable between staff and which provide systematic approaches to the design of systems. There is great value in ensuring that standard design procedures are followed using methods which can accommodate expanding needs without radical changes. Modern day simulation, design and real-time code generation methods aid this process considerably. However, the value and importance of simulation for example may be well understood by the manufacturers producing for example aero engines, but it is not so well utilised by companies that manufacture for example chemical products. Technology reviews do often point the way to these obvious areas of improvement but can also highlight more subtle problems relating to product quality, energy utilisation and even waste management.

In years gone by the ACTC provided a service of regular meetings with companies which was intended to help in keeping abreast of new developments in technology. In the main, these general services of the ACTC have been replaced by more bespoke, tailored, training activities. The latter include much more targeted technology reviews requiring a much deeper understanding of the product being manufactured or the processes involved. In addition to reports written for the project there are usually presentations to both engineering staff and management to ensure the benefits are fully appreciated.

Further details can be obtained from Professor Mike Grimble

Mike Grimble


Sad News

We are sorry to report the sad passing of the former Managing Director and past Chairman of ISC Ltd., Andrew Buchanan. Andrew was the Managing Director of ISC when the ACTC service was at its height and many members of the ACTC will remember his conscientious management, professional organisation and enthusiastic interest in both technical developments and the members themselves. Since his sad passing we have received many messages from ACTC members who considered Andy a true friend. We thank them all for their kind remarks. Everyone within ISC had the greatest respect for Andy and we will sorely miss his kindness, patience and leadership.


Multivariable Control Training

The training course on multivariable control that is now offered to all was originally developed for Boeing in Seattle. The Multivariable Control training course provides an engineering approach to the design of multivariable control systems and aims to provide an intuitive understanding of behaviour as well as practical advice on the problems and implementation.

This has recently been revamped and presented very successfully to Cummins Diesel Engines in Columbus, Indiana. The interaction in multivariable control systems adds new difficulties to control design. Multivariable systems can add confusion in loop tuning or fault finding. For example, in a multivariable system, it is difficult to know whether disturbances in a loop are caused by interaction with other loops or by some external disturbances. Moreover, the relatively benign transfer-function models, for the different elements of a multivariable system, can become more difficult transfer function matrices when other paths are taken into account. The problems of uncertainty and robustness are also of course much exacerbated by the multivariable nature of the problem.

An important feature in all training courses are the hands-on sessions which enable engineers to gain an instinctive feel of the type of results generated by the design methods described. In most training courses a hands-on session follows every one or two lectures. The delegates receive copies of the Matlab/Simulink system simulation and design examples, and the PowerPoint presentations. These can be used later as an aid memoir to refresh the knowledge gained. Further details can be obtained from Dr Meghan McGookin


New Nonlinear Control Training Course

The topic of nonlinear control systems has been covered in existing courses and very short nonlinear courses have been provided on demand. However, a new more general 3-day training course has now been produced. In fact, this course can even be extended to 4 days by including more application examples. The course covers all aspects of nonlinear system modelling, control design and implementation. The main theme is the design of simple industrial controllers for nonlinear systems.

Another feature of the course is the introduction to predictive control for nonlinear systems. Model predictive control (MPC) is one of the most successful and widely applied advanced control methods used in industry. Many commercial industrial controllers solve the predictive control problem using a set of linearized models for a nonlinear system. However, it is believed that higher performance may be obtained by using true nonlinear predictive controls. A range of nonlinear control design methods are described in the MPC course and the benefits outlined. The course includes extensive hands-on material for a range of different industrial applications including automotive and wind energy. Further details can be obtained from Dr Meghan McGookin


The World of LMI’s

Ever since Stephen Boyd of Stanford University started to promote the use of linear matrix inequalities (LMI’s) they have gained in popularity and are now a staple tool for the control engineer. In fact, an LMI solves a particular type of optimisation problem for many areas including control engineering. There are many applications in control where LMI’s are valuable if not essential.

The most important application area for LMI’s is probably when dealing with uncertainty in systems. By their very nature these are not precisely defined. However, one can often describe the uncertainties by using sets of inequalities. The parameter ranges can be determined and parameter bounds may be specified. One of the most active areas for research is the solution of predictive control problems using LMI’s, where there are great uncertainties in disturbance or system models. The LMI approach has great applications potential in areas like machinery controls.

Since this topic is becoming so important it is being introduced into the ACTC training courses and there will be new hands-on to compliment the lecture material. Further details can be obtained from Dr Meghan McGookin


Energy Optimisation in Marine Systems

There is a lot of interest in different ways of saving energy in marine systems. It is of course quite an old topic to consider energy savings in autopilot controls. There are also well known application areas in ship positioning systems and support vessels, even limiting what might be termed the hotel loading costs in offshore accommodation facilities.

Staff of the Industrial Control Centre at the University of Strathclyde have been investigating ways to limit energy usage in areas such as fin roll stabiliser control systems. In fact, the NGMV controllers developed at Strathclyde University, and used within ISC and the ACTC have been applied to this problem. One of the most interesting areas researched is that of quantifying the energy savings that can be achieved. Further details on control methods to limit energy usage in marine system may be obtained from Professor Mike Grimble


Hybrid Control Systems

There have been many developments in hybrid systems control in the last 20 years. By hybrid systems in this case we are concerned with controllers that have a mixture of continuous control together with discrete decision making or switching. The continuous control can of course be implemented digitally and the hybrid nature refers to the link with systems like switching logic. There are many application areas where total integrated control, merging decision making with regulatory control would provide a better overall control solution. Automotive companies do for example have an interest in this area for powertrain control and management.

ISC the parent company of the ACTC has been one of the most active in the use of state-dependant and linear parameter varying models for systems. It is an extension of this concept which enables hybrid control to be implemented relatively simply. One of the problems with hybrid systems is the mathematical complexity of the theory and the rather unwieldy nature of the results for implementation. By the use of state dependant models much simpler algorithms can be defined for a certain class of Hybrid systems.

Hybrid methods have great potential for supervisory control systems. There is often a requirement to mix management decision making at the upper level with the set point optimisation for the lower level multivariable control loops. One of the useful application areas for the approach is in hybrid vehicle control, providing many opportunities in applications. Further details of the company’s work on hybrid systems may be obtained from Professor Mike Grimble


Bringing Wind and Waves Together

There is now a lot of interest in floating wind turbine structures. For deep water locations some form of floating wind turbine may be needed but this brings with it several problems. The cost of wind turbines is greatly influenced by the maintenance and breaks down costs over their period of life. One of the major problems that arises is the difficulties due to fatigue loading effects due to the wind acting on both blades and turbine structure. A floating wind turbine has the additional difficulty that there is wave motion introducing disturbances.

The control solutions are available partly from wind turbine control philosophies themselves but also from marine systems experience. The methods of modelling ship motions due to waves, current forces and even wind forces are reasonably well established. Experience is also available from systems like dynamic ship positioning of the effect on vessels of different forces and actuators. The control of floating wind turbines is therefore a difficult problem but one where the basic multivariable control tools are well established and available. Further information may be obtained fromfrom Professor Mike Grimble


Robustness in Predictive Control

Predictive control for industrial applications has become a very popular topic and has gone through several stages of development. In the early days of Model Predictive Control (MPC) there was little attention to achieving good theoretical stability properties and even less to achieving robust solutions. One of the main research areas today in predictive control systems is now in this area of robust control design.

People do of course mean different things by robustness and it is worth reviewing what a robust control should be. The most obvious robustness is the requirement that disturbances are rejected efficiently, even if there is some uncertainty regarding their nature. The second very critical requirement for robustness is that stability properties should not be compromised by gain or time-constant variations within the system. Finally, there is the robustness of the control law with respect to tuning sensitivity. That is, a well behaved and robust control systems will be one where the tuning of the controller parameters will not be too critical to stability properties.

This last measure of system robustness is not so well established. Clearly if the tuning parameters (like PID gains or in optimal control the cost weighting), are such that making small changes completely changes the transient behaviour of the system then it is not very robust in some sense. This is related to the so called controller fragility problem. Controllers can be fragile, non-fragile or resilient. A controller is fragile if very small perturbations in the coefficients of the controller destabilize the closed-loop control system. To assess controller fragility an index can be defined which relates the loss of robustness of the control loop when controller parameters change, to the nominal robustness of the control loop.

To address the deficiencies in current predictive control systems there are many approaches to building in robustness by allowing for uncertainty. The main difficulty is the increase in complexity that inevitably results, detracting from a major advantage of predictive control, which is its relative simplicity. A number of Linear Matrix Inequality (LMI) based robust predictive control algorithms have been proposed. Unfortunately an LMI optimization must often be solved online at each time step.

Modelling uncertainties are of course inherent in any control design model development and the first question that arises is how to configure a system model to adequately represent uncertainties. A robust MPC forecasts the system behaviour for all possible models in range of uncertainty or more precisely the uncertainty set.

Stability can be ensured in robust MPC with a finite prediction horizon by adding a zero-terminal equality at the end of the prediction horizon (adding the constraint x (t + N) = 0). For example, Professor David Mayne and co-workers from Imperial College have proposed a min-max control law that steered the state into an invariant region in which the state feedback law guaranteed convergence to the origin for all states. However, a disadvantage of the min-max control approaches is the computational complexity that is often associated.

Some more pragmatic methods of improving robustness may also be applied. It has been suggested that uncertainty first be neglected and an MPC design completed. That is, an MPC regulator can first be synthesised for the nominal plant model. Then, a so-called robustifying controller can be designed using an H8 approach. This is for the augmented system that includes the first-step MPC. This is appealing from the structural viewpoint of just adding another control function dedicated to the one purpose but achieving real robustness with standard H8 procedures can be elusive.

There is also a range of methods and results that link linear MPC designs to state-feedback LQ regulator results to try to inherit the good robustness properties of these controllers. In practice of course an observer of some type is needed so the results will not be as good as hoped for, and moreover the very good robustness margins were established for the continuous-time case and all controllers are normally implemented digitally.

There is therefore much further to go in the development of truly robust MPC solutions and particularly for fast processes.


Linear Matrix Inequalities

For some time the control world has been excited about the possibilities offered by Linear Matrix Inequalities (LMI’s). Standard algorithms have been produced for the solution of LMI problems, and a very large number of control problems can be posed in terms of LMI’s. However, the most important factor is that various forms of uncertainty can be included in the problem description and the LMI approach used to find the best solution.

Stephen Boyd of Stanford University in California was very active in promoting and developing the methods in control applications. Research effort in control culminated in the LMI book by Boyd, El Ghaoui, Feron and Balakrishnan in 1994. The development of a Matlab control design facility has also been very valuable in promoting the approach. The LMI toolbox was produced by Pascal Gahinet, Arkadi Nemirovski, Alan J. Laub and Mahmoud Chilali, and other tools have been produced using this numerical approach. It has been evaluated for applications in many industries and notably in the automotive industry. The ubiquitous predictive control methods have also been modified for use with LMI based solutions. Very traditional problems with additional requirements can be put in LMI based form, such as multi-criterion LQG/LQR problems.

For those working in model based advanced controls the LMI method provides a further design option. There is some overhead in learning the approach, and the solutions are generated by numerical algorithms with a certain element of black box mystery. However, the controllers can be provided in a traditional form and implementation should not be more difficult than with a standard classical controller.


Formation Flying

Problems in the control of multiple airborne vehicles or even ground vehicles or ships have led to considerable interest in problems of complexity. Tools like multi-agent systems have been introduced leading to an explosion of academic papers that can be used in the control of complex processes.

New analysis methods have also been introduced to consider the stability of complex systems using for example, methods based on chaos theory.

In parallel with new developments in systems theory there are new innovations in computer hardware and software introduced by companies such as National Instruments. All these developments are needed to cope with the increase in needs of modern systems where high performance is demanded in the presence of uncertainties such as varying transport delays, model dynamics and parametric variations. Some of the developments in control theory are not strictly necessary for practical applications engineers but it is always beneficial to have a better understanding of systems and to have tools available if needed.


Advanced Control of Hot and Cold Rolling Mills

Up to a few years ago the ACTC either ran Rolling Mill Academies at the University of Strathclyde or cooperated with their hosting at locations around the world. Since the recession there has not been the demand for such an event in Europe. However, recently an Advanced Control for Rolling Mills Course has been produced by the ACTC/ISC in a form, where it can be tailored for different company’s requirements. It is expected that the course will mostly be presented at other company premises and will include hands-on training material. The course deals with many of the common problems found in Hot and Cold Rolling Mills including:

  • both stand and integrated systems controls
  • gauge control and back up roll eccentricity
  • looper controls
  • flatness or shape control
  • interstand transport delays
  • supervisory control systems
  • fault tolerant control in presence of tension measurement failures
  • reheat furnace controls
  • The course can be tailored from 2 days to 5 days depending on requirements. Further details may be obtained from from Dr Meghan McGookin


    Courses on Signal Processing and System Identification

    The subject of signal processing is very wide indeed and covers topics in communication systems that are not so relevant to control engineers. However, a signal processing course for control engineers has been produced dealing with the usual areas of Kalman filtering, linear and nonlinear estimation, system identification, Wiener filtering, classical filtering techniques and many other topics. The aim of the course if to provide the control engineer with the digital signal processing tools required for applications like fault monitoring and detection, condition monitoring and use in control loops. The signal processing course is 5 days in duration but it can also be tailored to a particular company’s requirements. Further details may be obtained from Dr Meghan McGookin


    Book Review: Cooperative Control of Multi-Agent Systems: A Consensus Region Approach by Zhongkui Li and Zhisheng Duan, Published by CRC Press, £89

    This new book is published by CRC Press and it is in the automation and control engineering series. Multi-agent system is a rapidly growing area of control engineering mostly concerned with topics like cooperative control. With large computing systems that are interconnected a number of autonomous agents must work cooperatively whether for civilian or military applications. Such multi-agent systems can significantly improve operational effectiveness, cost and also provide a degree of redundancy in systems.

    The use of multiple autonomous agents that work together to achieve collective group behaviour is referred to as being cooperative control of multi-agent systems. This often involves multivehicle systems in many applications. Multi-agent systems are needed in satellite formation flying, distributed computing systems, electric power systems and intelligent transportation systems amongst others. The book aims to offer a systematic framework for designing distributed controllers for multi-agent systems having linear agent dynamics.

    The language of multi-agent systems is somewhat different to regular control problems, for example, consensus is a term used to mean that a team of agents reached an agreement for a common value through interactions with each other, and using a sensing or communicating network. There is also formation control where unlike the consensus problem the final states of all agents can be more diversified under a formation control scenario. Another term which is used is flocking. This is a typical collective behaviour in autonomous, multi-agent systems and the terms does of course stem from the behaviour of birds. In fact, by understanding flocking in animal groups the hope is that this will help in the development of artificial autonomous systems for the control of unmanned aerial vehicles, mobile robots or even automated highway systems.

    This monograph is certainly for a specialist in multi-agent systems. It will be useful to researchers and to advanced course control engineers where multi-agent systems are covered. It’s useful as a reference text and it has a good bibliography.

    Book Nerd.


    Book Review : Stability, Control, and Computation for Time-Delay Systems by Wim Michiels and Silviu-Iulian Niculescu, Published by: SIAM, $114 USD

    Transport delays are one of the main sources of control difficulties and possible instabilities in industrial systems. This text is published by SIAM and describes an eigenvalue approach to the control of such systems. It is a second edition and the cost is around $114 USD.

    As would be expected with a text published by the Society of Industrial and Applied Mathematics it mainly concerns the theory of such systems rather than covering engineering design matters. It is therefore most suitable for researchers in the subject but it does cover many areas of value to engineers in industry. The first two parts of the text cover the theory of systems but the third most interesting part from an industrial engineers perspective is concerned with everything from Smith Predictors to congestion control in networks. Delay models in bio sciences are even included.

    The Smith Predictor is important to engineers in industry, and particularly in the process industries. One of its main difficulties is the lack of robustness to various types of uncertainty. The mismatch between plant dynamics and the plant models used in the Smith predictor can be particularly troublesome. The Chapter 11 of the text provides a very nice introduction to such matters and even covers the multivariable problem. A Smith Predictor cannot in its basic form control open loop and stable systems so this type of problem is considered in Chapter 12 of the text.

    The book is therefore mostly of value as a reference text to researchers but there are particular chapters that will be of value to engineers facing real problems due to transport delays.

    Book Nerd