Designing optimal control algorithms for complex real systems development is a difficult task, at best.  Plant models based on mathematical formulas and control system design techniques undergo significant time-consuming refinement to approach the behavior of the real system.  Perhaps there is a more efficient way to develop these complex plant models?

Our client used conventional modeling techniques, including finite element modelling and multibody dynamics, to develop mathematical plant models of complex dynamic systems. While this process gave a first principle understanding of the behavior of the system dynamics, it was time consuming and created a bottleneck in the control algorithm development activity. To help our client accelerate their development process, we introduced them to System Identification Methods.

Using our detailed knowledge of MathWorks toolboxes, Pilot Systems engineers developed a standardized system identification process for building our client’s linear plant models using measurements of the system’s physical input and output test bench data. Our process generated accurate high-order plant models which were then used to develop control algorithms. The performance of the controllers was evaluated in the laboratory (HIL testing) using dSPACE hardware and software. Results of closed-loop HIL testing revealed that our system identification process was a highly effective method for building mathematical plant models. Our client is now capable of building plant models that earlier required months of efforts in a matter of days or weeks. With this, we successfully helped our client significantly accelerate their control algorithm development activity.

System Identification Methods allow for the development of mathematical models of dynamic systems utilizing measured input-output responses from the subject system. Parameters are adjusted within a given model until model output coincides with measured output data. We use either time-domain or frequency-domain input/output data to estimate continuous-time and discrete-time transfer function process and state-space models.  This method is especially helpful in creating models of physical systems which are not easily described from first principles, for example chemical process control, industrial processes such as a rolling mill or in the automotive area internal combustion engine emissions control or climate modeling of vehicle cabins. The method is also helpful for systems which the parameters are not well known such as large mechanical structures.  

If this problem sounds familiar and you would like to reduce the development time of your control system algorithms, please give us a call for consultation on applying System Identification Methods to your design.

Areas of expertise

MATLAB, Simulink, MATLAB scripts, System Identification Toolbox, dSPACE hardware, dSPACE software, Signal Processing

For more information, please contact us.