Overview

Model Predictive Control (MPC) is the most successful advanced multivariable control technique, which is used in applications across industrial sectors, particularly where high performance is required. It is valuable for complex systems that are difficult to control because of difficult dynamics, interactions and the multivariable nature, noise and disturbances, transport delays or because of constraints and nonlinearities.

The training course is aimed at engineers in industry that are interested in using MPC and need to know the design procedures or enhancements that can make practical implementation faster and more effective. The event therefore focuses on the basic ideas, design stages and the steps in implementation that can avoid difficulties in commissioning or normal operation.

The basic principles and concepts of MPC will be introduced but the full mathematical background is not needed to use an MPC algorithm. Nevertheless, it is important to have an instinctive understanding of the various stages in the MPC design stages and computations. An intuitive introduction to the solution of MPC problems will therefore be provided. It will detail the various aspects of model choice, cost-function definition and the type of optimization method and form of optimal control solution that is obtained.

The MPC algorithm can handle multiple input/output variables and constraint conditions reliably. It introduces “prediction capabilities” to improve the accuracy and performance of industrial processes or electromechanical control systems over a given future horizon.

Course Structure

Early presentations will motivate the use of MPC and introduce the physical system modelling methods or system identification techniques that may be needed. The great majority of predictive control laws are obtained by optimizing a criterion that is representative of the form of control needed. This may be motivated by the physical system problem that could involve energy minimization, or it could be a general criterion where the cost-function weightings act rather like the tuning parameters in PID controls.

Optimization methods will therefore be introduced that are applicable to both constrained and unconstrained MPC system problems. Most physical processes do of course have hard constraints of some type and the way in which these are described and implemented in the optimization algorithms will be introduced.

After the basic modelling, optimization and optimal control algorithms have been presented attention will then turn to the design of such systems to satisfy industrial requirements and specifications. Care is needed when specifying the type of system model and the noise and disturbance models. The specification of the form and parameterization of the cost function that should embody the requirements of the physical problem is also important.

The design stage will therefore consider how to choose the cost-function, and the weighting terms involved. After an MPC controller is designed, it can of course be assessed using simulation. However, the implementation of MPC in real systems may involve many additional problems since the system model will have inaccuracies, uncertainties, and robustness issues that may arise. Fortunately, there are various engineering enhancements to basic MPC that can mitigate problems with uncertainties. There are also changes to the standard MPC problem that allow practical improvements to be made such as introducing integral action, allowing for different transport delays in input/output channels and ways to deal with unknown disturbances that may be present.

It is also recognised that artificial intelligence and machine learning will have a significant impact on the future use and implementation of MPC controllers. There are already significant advances in the use of adaptive AI based predictive controls and there are also simple additions to existing MPC designs that take advantage of data driven methods. The AI predictors can for example provide longer range prediction capabilities than is used in the MPC’s own cost-function prediction horizon. The way in which the basic MPC algorithms can be enhanced to provide improved performance by using data driven methods, such as neural networks or support vector machines will be covered. The final presentation is a look to the future of AI Enhanced MPC to indicate the developments that companies should be aware of to maintain a competitive edge.

To ensure the ideas on modelling and on optimization are fully understood hands-on Simulink exercises will be provided throughout the course, so that delegates can assess the methods and solution methods. A servosystem will be used for all the hands-on examples from initial modelling to final implementation, although different application options can be offered through tailored versions of the course.

Instructors

Professor Michael Grimble

Professor Mike Grimble has a good understanding of the needs of industry having first worked for Ciba Geigy and then for Associated Electrical Industries which became GEC at Rugby. During his time on the Industrial Automation Group at Imperial College he worked on the modelling for control of cold rolling mills. He later set up industrial groups first at what is now Sheffield Hallam University and secondly at the University of Strathclyde where he still works as a Research Professor. His interests are on the design of high performance and robust control systems for applications across industrial sectors. His industrial background is valuable to provide motivation and insights on the industrial training courses.

Dr Pawel Majecki

Dr Pawel Majecki undertook his research studies in the Industrial Control Centre at Strathclyde University and later joined Industrial Systems and Control Ltd. ISC was established to encourage technology transfer and Pawel has worked for major international companies on the application of advanced control methods, often involving predictive or optimal controls. He has also been involved in control training courses run in the UK, Norway, Italy, Spain and USA. His extensive experience in the use of MATLAB/Simulink is valuable on the training courses hands-on where delegates often gain greater insight into these tools in addition to their value in explaining how the design methods are applied.

Delegates will find that both instructors are very happy to answer questions or discuss industrial problems they may have in coffee breaks, lunch breaks and at the end of the day.

Additional Information

Event Venue

Scottish Engineering, 105 W George St, Glasgow G2 1QL

Accommodation

Glasgow City Centre offers a wide range of accommodation, you can find our recommendations here.

Course Summary and Agenda (Download PDF)

Timings may change slightly.

Registration

Please complete the Online Registration Form.

Can't Find the Course You Need?

Try our bespoke course service.

Tell us your training needs.