E-News February 2015 Edition

Increase in Tailored Training Courses by Professor Mike J Grimble

Over the last decade there has been a shift from general courses on control engineering to more tailored courses run through the ACTC training programme. One of the most successful of these courses has been organised with BP on offshore oil production platform control. Although the control techniques used on production platforms are in some ways less advanced than those encountered downstream in refineries, there are nevertheless important functions to be performed from a regulation, safety and a reliability view point.

Another area, which has seen a steady growth, is in specialist automotive training courses at introductory, intermediate and advanced levels. The ACTC now offers tailored training courses for petrol or diesel engine control ranging from simple introductory material to advanced state of the art design.

The growth in wind energy around the world has also led to the development of specialist training courses, some concerned with individual wind turbine control and others with wind farm control; onshore and offshore. Again there are simple introductory courses and some courses which consider more sophisticated concepts, like reduced wear and tear and fatigue loading in offshore wind turbines.

Over the last decade the ACTC has worked closely with Boeing to ensure courses meet their very high professional standards. One of the innovations introduced was to structure the courses more clearly into the introductory, intermediate and advanced levels. Another development Boeing strongly encouraged was the use of very detailed design studies within training courses, in addition to the hands-on examples. Thus, in subjects like wind turbine control or automotive engine control half-day modelling, simulation and design exercises are provided for delegates so that they can use and really understand the techniques covered.

The ACTC develops courses according to the needs of customers and we are always pleased to received proposals for new material. Please contact us at iscmail@isc-ltd.com or call 0141 847 0515 with any ideas.

Mike Grimble


ISC News - NI Alliance Partner and How Simulations can help Control System Deployments

National Instruments Alliance Partner

Over the past four years ISC has undertaken some really nice control software development projects using National Instruments hardware and their LabVIEW software. We have even won the prestigious Worldwide Application of the Year for the Turbine Access System Control System we developed for Houlder Ltd.

It is therefore very pleasing to announce that ISC is now an NI Alliance Partner - meaning we have staff certified in using LabVIEW, access to NI's full suite of tools and a proven track record in developing applications with them. Our capability in using LabVIEW and its Control Design and Simulation tools is very well established and we will be writing many more case studies in the future, as well as having a training course on Control Fundamentals using LabVIEW.

In some of the control design and deployment projects we have undertaken, we have made very effective use of simulations to allow testing to be done as early possible, which can be to refine the mechanical design before it is finalised or to minimise the impact of software bugs when there are very tight deadlines for commissioning. These approaches can be referred to as Model-in-the-Loop (MIL) testing, Software in the Loop (SIL) or Hardware in the Loop (HIL - Eric Bradley from NI has a great summary of these various terms). There are many benefits in incorporating simulations into the design, deployment and through life support for control systems, and Andy Clegg at ISC has summarised our experience in a nice White Paper called "Using Simulations for Virtual Commissioning during Control Software Development". It is freely downloadable and we hope you find it useful.


Chrysler & Ford Training Courses

Chrysler Training

Two training courses have been presented to Chrysler at their headquarters near Detroit. The first course was on control fundamentals, and was open to engineering staff that support control engineers, such as calibration engineers and software engineers. It attracted the largest attendance ever for an in-house ACTC training course, with 60 delegates. The second course was held later in the year and was on advanced control, dealing with subjects such as nonlinear systems control and model predictive control.

Ford Training

The Ford motor company in Dearborn near Detroit organised two recent courses through the ACTC including a basic course in introducing control topics to non-specialist and a more advanced course on observers, Kalman filtering and estimation methods. This course also included estimation and filtering methods for nonlinear systems such as extended Kalman filtering.


Hot and Cold Strip Rolling Mill Course

In the past the ACTC helped organise the rolling mill academies that were very successful. Some of these were run at the University of Strathclyde in Glasgow and others were run around the world in locations such as Pittsburgh and Amsterdam. ISC has been working on a new course aimed at covering some of the material of the academies which can be held at company premises and may have three, four or five day formats according to the requirements of the company. Further details may be obtained from Professor Mike J Grimble.


ISC Training Course Question and Answer Sessions

The ACTC and ISC Ltd., continually strive to improve the training course material and the presentation methods. It listens to customers and innovates. One of the recent developments suggested by the Norwegian company Kongsberg is that we should include a question and answer session at the end of the course. Although this by itself was not novel the idea was to provide delegates with a record sheet they could complete at any time during the course, so they could raise problems and questions. There was then time to consider these and collect information, before they were addressed at the final course presentation. The trial revealed this was a very successful innovation that enabled the presenters to be ready with answers and back up material. It was also cited as the favourite part of the course by many delegates.

It will therefore become standard practice to ask companies that are having courses at their company premises whether they would like us to include such a session and to set aside time for such open discussions. It is even possible that companies will bring along particularly difficult problems they have experienced to air in such sessions.


New Developments and Results in Control Theory

Problems in the control of multiple airborne vehicles, or even multiple 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 and stability 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 increasing needs of modern systems. For example, 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 more concerned with the control of networks and very large systems. They are not so applicable to many industrial control problems. However, it is useful to have some understanding of the developments in large scale systems and to have the tools available if needed.

If you are interested in hearing more about this please contact Meghan McGookin or Andy Clegg.


Link between Fault Monitoring and Benchmarking

There have been many developments in model based fault monitoring which enable the early onset of possible fault conditions to be predicted. This is particularly important in systems where failure would cause expensive downtime and equipment replacement. At the same time there have been new developments in benchmarking controllers which rely on similar techniques. To benchmark a controller system models are often needed. These are used in controller benchmarking to predict some best possible control action and to then suggest controller tuning variables. The range of methods used to detect faults, such as parameter estimation, are also introduced in controller benchmarking algorithms.

Many SCADA manufacturers have built in controller benchmarking software as standard features and the implication of the above is that various alternative model based fault detection methods will also be introduced in future systems as a standard feature.

The bespoke control solutions offered by ISC are normally based upon good model information and simulations. It is not therefore surprising that fault detection, or condition monitoring capabilities, can also often be included with a little extra effort.


New Nonlinear Condition Monitoring Methods

ISC working with the Industrial Control Centre at the University of Strathclyde has developed new fault monitoring and detection methods for systems with significant nonlinearities. Most model based fault monitoring, detection and isolation methods are based on linear system assumptions but the new estimators can be used on nonlinear applications.

In for example automotive power train control the performance of the catalytic convertor is important and air fuel ratio is controlled. So called lambda sensors are employed and the engine should regulate the lambda to a normalised value of unity. However, these devices are highly nonlinear and new estimation methods which can accurately estimate lambda provide a useful condition monitoring tool.

These same estimation methods enable channel dynamics to be taken into account which can be nonlinear. There are many fault detection applications where residuals must be calculated for systems where the channel represents the path between the faulty components and the measurements. In addition to the estimation process there is also of course a need for thresholds to be used. These provide a way to detect when the fault is significant relative to other stochastic effects such as noise and disturbances. To further complicate matters there are uncertainties in modelling such systems. Nevertheless, the new nonlinear fault detection strategies seem to have great potential in real applications.

This work was completed in coperation with the University of Strathclyde and was led by Dr. Alkan Alkaya of Mersin University in Turkey.


Nonlinear Predictive Control

The ACTC is now providing training courses on nonlinear predictive control and ISC Ltd. (parent company of the ACTC) is undertaking research and development projects with companies on the topic. Linear predictive control is now accepted as being the most successful advanced control method for industrial applications. However, some applications like automotive engine control are so nonlinear a suitable nonlinear predictive control algorithm is desirable. Even linear predictive control algorithms have constrained versions so that hard actuator limits and output limits can be handled. This is of course a type of nonlinearity. If in addition a system has significant linearity in its dynamics then this ideally requires a nonlinear predictive control solution. Some commercial systems do of course use multiple models and scheduling to handle nonlinear dynamics. This is an attractive solution since at least conceptually it is simple. However, there are difficulties in deciding the mesh of points at which to linearize the system and there is then the need to switch between controllers.

A nonlinear predictive controller can provide what might be termed a global solution so that the controller is applicable across the range of operating points, even as nonlinear behaviour changes. There are many applications such as wind turbines, hot and cold strip mills, high speed servo systems and robotics where nonlinearities can dominate and severely limit performance. Traditionally controllers have to be detuned to provide stabilising solutions. A nonlinear control law can avoid detuning and substantially improve performance. Moreover, predictive control methods offer some of the most reliable techniques.

ISC in cooperation with the industrial control centre at the University of Strathclyde has continued to research nonlinear predictive controls, particularly for high speed applications. Predictive controls for the process industries are in one sense not so demanding since the computational time is insignificant compared with the process time constants. However, in applications such as automotive engine controls the computational times start to be significant compared with the plant time constants involved. Moreover, the computing devices available are of much more restricted capabilities. There is therefore a much greater challenge in developing suitable nonlinear predictive controls.

One of the areas particularly developed by ISC and the ACTC is that of the class of nonlinear predictive controls which can be accommodate state dependant or linear parameter varying models. In fact, the algorithms available can also include black box nonlinear terms that can even include a neural network model. The predictive control cost functions used can include nonlinear cost-function terms, dynamic frequency dependant weights and also time-varying weightings. Algorithms have also been produced for hierarchical structures with predictive control being applied at both supervisory and regulatory levels.

Some years ago the Strathclyde University group developed a nonlinear predictive control facility within National Instruments LabVIEW tools. ISC has now also developed a Matlab/Simulink design facility for many application sectors. It is interesting that the modelling methods, cost function descriptions and underlying problems for applications like automotive engine controls and wind turbine power controls are very similar. In fact, some of the most difficult problems such as reducing wear and tear and reducing fatigue require the same type of algorithm, even if the system models and performance objectives are very different.

ISCs experience in using the Matlab/Simulink design facilities in different applications can considerably reduce engineering effort needed on new application areas. The companies experience in implementing real control laws using National Instruments hardware such as Compact Rio and its software LabVIEW can also reduce implementation time.

University research and company development effort is currently aimed at improving the robustness of predictive controls, developing novel restricted structure solutions and considering so called hybrid system problems. That is, systems which have requirements for switching and logical decision making in addition to continuous control (even when implemented digitally). Further details may be obtained from Professor Mike J Grimble.


Multi-Agent Systems and their Role in Control

A multi-agent system is one that uses recent techniques in computer science arising from the needs of systems to work together and reach agreements or cooperate, or even compete, with other systems. An agent in a computer system is by definition able to take independent action to try to achieve its objectives. It may employ a number of methods to do this such as using artificial intelligence techniques. A multi-agent system is of course the rather obvious combination or a number of agents that interact with one another. The agents themselves may collaborate to complete tasks or they may have different objectives. A multi-agent system is one where the agents cooperate, coordinate and negotiate together to achieve the goals set.

The role of multi-agent systems in control engineering was brought into prominence by the rapidly increasing use of autonomous systems. A remotely operated vehicle on the sea bed may be difficult to instruct what to do continuously and will need its own autonomous control for periods. The same applies to spacecraft. The multi-agent system can replace a team of human operators in deciding what trade-offs to make, compromises to be made and actions to be taken.

On occasions multi-agent systems are referred to as self-organising systems. In fact, this is a very old term used to describe some early control system ideas and the purpose is fairly obvious from the name. Agent-oriented approaches are very suitable for the development of complex software systems and very relevant to large control systems design. They can be applied in applications as diverse as electrical power distribution management systems and vehicle manufacturing line control systems.

Multi-agent control systems containing distributed control and applications will collaborate dynamically to achieve the control objectives. The agents involved will undertake knowledge processing and feedback control actions using real-time distributed control environment. The integration of data acquisition, feedback and feed-forward control, and condition monitoring is possible in the multi-agent environment. Sophisticated methods such as dynamic reconfiguration of the control system can be undertaken.

One of the most perplexing problems facing the control engineer is the development of safe and reliable systems where the range of fault conditions can arise through poor control, not taking into account the nonlinearities or interactions, or maybe due to some component or system failures. The traditional tools of the control engineer to model and understand the system and subsequently to undertake a good and reliable design, based upon these modelling studies, are undermined when the system may change in unpredictable ways. The consequences in flight controls of aircraft, ship manoeuvring systems or even autonomous vehicles are of course life threatening. Multi-agent systems provide a possible way to tackle this problem. They themselves do of course introduce a degree of uncertainly, since things do not remain fixed. Nevertheless they offer an automated way of responding based upon well thought out rules and decisions.

Polakow and Metzger of the Silesiam University of Technology have described an agent based approach using LabVIEW distributed control systems. LabVIEW is of course a popular tool for use in automation and measurement systems and it is very suitable for the assessment of the benefits of this type of approach.

The agents can be based on artificial intelligence methods and use tools such as fuzzy systems and neural networks. They seem to have natural roles in the supervisory level of large industrial processes but most of the interest has been in manufacturing lines rather than large process systems. The subject of multi-agent controller design is very likely to move away from the mathematical control scientists towards the practical application stages and probably become an essential part of the control of large processes. Genetic algorithm based fuzzy control has been applied in multi-agent power plant control systems and this industry is also a natural user of such techniques.

There are of course dangers in relying upon distributed computing systems but our faith in computer systems is increasing. For example, fly by wire aircraft are soon to be followed by fly by wire car systems. This subject of multi-agent systems reminds us that the old idea of the control engineer tuning PID controllers, is far removed from the total requirements of the current profession, where machine learning, cybernetics and all the modern tools of computer science are also needed.


Book Review: Optimization Concepts and Applications in Engineering, By: Ashok D. Belegundu and Tirupathi R. Chandrupatla, 36.99 (paperback)

This is the second edition of this book that concerns optimisation concepts and applications in engineering. This is a softcover text and is very suitable for both students and engineers using optimisation methods. It begins with unconstrained minimisation problems and then considers constrained techniques. It covers the usual areas of linear programming and gradient optimization methods, and it also goes into areas of nonlinear system optimisation. It also covers multi-objective optimisation and dynamic programming, which is often used in optimal control problem solutions. There are some references to real applications but in the main this is a theoretical text which describes methods and algorithms.

The book is very nicely presented and organised and it includes numerous features to illustrate the optimisation process. It is particularly valuable from a student point of view. There are numerous examples and solutions provided and very appropriate references at the end of each chapter. This text is highly recommended for anyone interested in optimisation methods and for those that have to use them in applications for industrial systems.

Book Nerd.


Book Review : Modelling and Identification with Rational Orthogonal Basis Functions by Peter S.C. Heuberger, Paul M.J. Van den Hof, Bo Wahlberg, Published by: Springer, 108

This text on modelling and system identification is edited by the very well-known international researchers Peter S.C. Heuberger, Paul M.J. Van den Hof and Bo Wahlberg. It is published by Springer and is a useful addition to the bookshelves for all involved in modelling physical systems. The computation of the model is often the most difficult part if engineering projects and the quality of models can determine the success or otherwise of control designs.

This text makes a useful contribution to many areas of system identification and involves contributions from a number of international experts in this edited text. It covers system identification methods for use in both the time-domain and in the frequency-domain.

One of the key ideas in the text is the use of orthogonal basis functions. These are used for generating the approximations to the response of systems for the computation of what are simplified models. A basis function is an element of a particular basis for a function space. Every continuous function in the function space can be represented as a linear combination of basis functions. Models of physical systems are of course approximations to very complex behaviour but luckily control systems design only requires models that are adequate and not exact.

The theory of system identification considered builds on theory from the 1930s and this text provides an excellent overview of such developments. Even though it is an edited book the material ties together very well and the layout is consistent and very straight forward. There is a strong underlying mathematical theme but the book is written more from an engineering perspective. It is not therefore difficult to understand, and it is unusual in being very accessible to engineering practitioners as well as to researchers. In addition to playing a useful role in control engineering the text is also valuable for engineers working in circuit theory, signal processing, networks and telecommunication systems.

Book Nerd