MSc. Graduation project advanced real-time parameter estimation algorithms
DEMCON is a company that develops advanced mechatronic systems for its customers. One of the current developments is a mechanical system that performs real-time measurements of its mass, damping and stiffness properties. Such a system can be used in a variety of applications in the process control industry. The working principle is based on fitting mechanical mass-spring-damper models, see Figure 1, on dynamic system response measurements. To this end, actuators and sensors are placed to generate real-time measurement data on which the parametric model could be fitted.
Figure 1: Example of a mechanical system with unknown parameters that could be estimated online.
Current parameter estimation algorithms are often based on linear least squares (LLS) methods because their calculations are relatively fast and easy and it ensures convergence to a global minimum with a unique closed-form solution. However, LLS methods can only deal with relatively simple system models and can only minimize a limited set of cost criteria which might not fit the actual estimation problem. As such, the LLS solution is generally not the solution that provides the best estimations of the physical parameters. Other challenges in the application DEMCON is aiming for are the presence of time-varying dynamics, environmental vibrations and disturbance forces (which might be unknown or unobservable), sensor noise, and internal feedback control loops. All of these four aspects might significantly contribute to bias and variance in the estimated parameter values when compared to the actual physical values. Unfortunately, these aspects are not always included in standard solutions for parameter estimation found in literature, but they should be included in the algorithm as well to improve the estimation accuracy.
As a first step towards overcoming the shortcomings of LLS, DEMCON has been working on alternative estimation methods such as iterative weighted LLS, non-linear least squares, etc., including support for both periodic and non-periodic input forces and disturbances. Preliminary results show great potential on improving the estimation accuracy, but the estimation methods still do not include yet all the aspects mentioned before.
In this graduation project (7-9 months), the goal is to further develop the new real-time parameter estimation algorithms such that bias and variance in the estimated parameters is minimized under the presence of time-varying dynamics, disturbances, noise and control loops. An experimental validation of the estimator algorithm should be performed on a prototype system which is available in the laboratory in Enschede. A possible outline of the project could be:
- Study literature about system identification and parameter estimation;
- Develop and formulate the new parameter estimation algorithm(s) that is optimal under the given conditions;
- Compare the new and previous algorithms using simulations and experimental data;
- Write the report.
- MSc program in Mechanical Engineering, Applied Mathematics, Electrical Engineering or Systems and Control, including courses related to time series analysis, system identification, parameter estimation, dynamics and control.
- Must be currently living in the Netherlands;
- Studying and/or working experience in the Netherlands is a plus;
Contact and more information:
- ir. M.A. (Michiel) Beijen, firstname.lastname@example.org
Solliciteer direct Stel je vraag aan de desbetreffende leidinggevende