Laws map advanced control onto chips
Tom Shelley reports on a multivariable control strategy expected to bring dramatic improvements to a wide range of systems including car engines
By performing complex optimisations offline, instead of online, it is possible to greatly improve car engine efficiency, as well as significantly reduce energy consumption in chemical processing and other applications.
The trick is to map out regions where different control laws offer best solutions and apply those laws on chip that can typically be addressed in less than 1ms loop time.
The theoretical breakthroughs behind the technology were originally established by groups led by Professor Stratos Pistikopoulos at Imperial College and Professor Manfred Morari at ETH Zurich. It has since been spunout as Parametric Optimisation Solutions or ParOS, with Dr Roderick Ross as acting CEO and Dr George Boustras as sales and marketing consultant.
Dr Ross describes the technology as, "Model based multivariable predictive control, computed offline", so that results of the computations of best solutions in all possible circumstances can be loaded onto a chip or PLC as lookup functions. When encountered by Eureka, Dr Ross likened it to an examination in which the user already has a sheet of possible correct answers, and is given rules with which to select the right one in any given situation.
The method is called, 'Parametric' because the sensor measurements that would provide the inputs for a conventional predictive control system become parameters in the offline computation process. The different possible inputs become optimisation variables.
The first successful application is with a "Large chemical company", which has, apparently, successfully used the technology to reduce energy consumption by around 5%. The control problem involves two inputs, two disturbances and two outputs.
The second application is for control of a camless active valve train in a research prototype engine that will hopefully come to market in about 2010. Although the control problem only involves one input and one output, Dr Ross explains that it is, "Highly non linear", so that the controller has to switch between the results of 24 different models according to circumstance.
Dr Ross and Dr Boustras considered that the technique is currently limited to situations with no more than about five inputs and five outputs. "It would not work on an oil refinery" as Dr Ross put it. It is nonetheless suitable for a very wide range of applications. Those currently being considered include: automotive catalytic converters, traction control, artificial organs for humans, drug delivery devices, missile guidance, control of unmanned vehicles, robotics and precise motion control.
Parametric Optimization Solutions
* Technique allows the advanced multivariable, model based predictive control to be put on chip and run orders of magnitude faster than is possible with a conventional system
* This is achieved by solving the problem for all possible combinations of sensor input values and inserting the answers in lookup functions
* The technique is currently limited to systems with up to about five inputs and five outputs
This material is protected by MA Business copyright
see Terms and Conditions.
One-off usage is permitted but bulk copying is not.
For multiple copies contact the