Modeling and control of a grinding machine
One of the railroad maintenance operations is to improve the profile of the rail with a grinding machine. This operation reduces noise and vibration transmitted to train and residences next to the railway. Rail grinding is carried out by a special train equipped with several grinding stones which are set at different angles and rotate whilst the train is moved along.
A grinding stone rotates by an asynchronous motor and should apply a constant force on the surface of rail. The applied force is controlled by a pneumatic jack using a PI controller. The grinding stone should be moved up and down when there is an obstacle on the railway (e.g. a level crossing).
The objective of this project is to develop a mathematical model for the system and to propose alternatives for controller and actuator to improve the performance of the system especially by reducing the fall time of the grinding stone after a level crossing.
This project is proposed in collaboration with Scheuchzer Ltd, 1030 Bussigny, Switzerland, one of the leaders in rail grinding.
Modeling of the glucose-insulin relationship for diabetic patients
In type 1 diabetes, the pancreas does not produce insulin any longer. This deficiency should be compensated by an exogenous delivery of insulin. Ideally, this is performed using an insulin pump. Such a pump makes it possible to adjust continuously the infusion rate. The insulin profile should be optimized in order to maintain the blood glucose concentration in a reasonable range. From a control point of view, the insulin infusion rate represents the manipulated variable and the blood glucose concentration the controlled variable. However, the relationship between these two variables is not fully understood. It has been shown that numerous factors have a significant influence. A precise quantification of these influences is tedious because of a lack of understanding of the underlying phenomena on one side and the lack of measurements which would make it possible to decouple these effects on the other side.
The goal of this project is the development of a new dynamical model without any supposed knowledge of the physiology. This model would be developed using experimental clinical data. Different identification approaches will be considered.
Estimation of Kinetic Model Parameters
Knowledge about reaction pathways and kinetics is important in process development and scale-up of chemical reaction systems. The chemist typically proposes various reaction pathways and kinetic expressions with unknown parameters. The aim of this project is to estimate the kinetic model parameters using experimental data. It is proposed to study hydrogenation of nitrobenzene to aniline which is an important dyestuff reaction in the agrochemical industries. Parameters will be estimated, and various reaction pathways proposed in the literature will be discriminated based on calorimetric and infrared measurements. Several experimental batch runs are available from ETH Zürich. It will be important to compare two data fit methods: Factor Analysis [1-3] using Matlab PCM_Toolbox and Pareto optimization using a Matlab toolbox from ETH Zürich.
 Amrhein, Michael, “Reaction and Flow Variants/Invariants for the Analysis of Chemical Reaction Data”, PhD thesis No. 1861, EPF Lausanne, Switzerland (1998).
 Bijlsma, Sabina et al., “Rapid estimation of rate constants of batch processes using on-line SW-NIR”, AIChE Journal 44 (1999), 2713-2723.
 Zogg, A. et al, “A new approach for a combined evaluation of calorimetric and online infrared data to identify kinetic and thermodynamics parameters of a chemical reaction”, Chemometrics and Intelligent Laboratory Systems 71 (2004), 165-176.
Atmospheric disturbance compensation in the VLTI Telescope
For improving the resolution of the Very Large Telescope Interferometer (VLTI) by rejecting the atmosphere disturbance on the spatial observations, a mechanical prototype is being developed at the EPFL.
The system is made of a double-stage structure. The first stage is a blade guiding structure driven by a high-accuracy nonlinear microstepper motor with a theoretical precision of 100nm. The second stage mounted on the first one is a fine positioning piezo actuator with a theoretical nanometer accuracy. The overall structure is supposed to be able to follow a reference trajectory at very high accuracy (nanometer scale) over a course of 70 mm for frequencies up to 200 Hz.
In this context, it is proposed to develop and implement a LQG multivariable optimal controller, the performance of which will be compared to a decoupled dual-stage SISO design. As the prototype is available, it will be possible to measure, implement and test the algorithms in real conditions.
Robust PID control of a flexible arm
In this project, the robust control of a flexible arm is considered. A DC motor rotates a flexible arm from one end in the horizontal plane. The motor end of the arm is instrumented with a strain gage that can detect the deflection of the tip. The system parameters can be changed by adding an extra mass in different positions.
Robust controllers are designed to stabilize and satisfy the performances for a set of models. This type of controllers can be applied to nonlinear systems linearized in different operating points, time-varying systems as well as systems with parametric uncertainty.
The objective of this project is to design a robust controller which stabilizes the flexible arm with different mass positions and ensures a certain gain and phase margin and crossover frequency for all models. The main idea is to compare the performances of a design method developed in LA with the classical PID or RST controllers. The new method is based on open-loop shaping in the frequency domain using a set of non-parametric models. Linear programming or quadratic programming is employed to optimize the controller parameters for all models.
First of all, the model identification will be carried out for different mass positions. Then, different methods will be used to design several controllers which will be compared in simulation and on the experimental setup.
Data driven controller design for a magnetic suspension system
The goal of this project is to design a high performance controller for a magnetic suspension system using a direct model matching technique. In model matching approaches a reference model is used to define the control objectives. The design method then minizes the difference between the real controlled system and the reference model. Such design procedures usually exist of two steps, model identification and ‘optimal’ controller design. Another approach is to combine the identification and controller design steps, resulting in a direct data-driven technique. Such direct approaches include only one approximation step, which should improve the achievable performance.
The magnetic suspension system considered in this project is unstable in open loop. Starting from an available first principle model of the system a stabilizing controller will be designed using a technique as seen in the courses of the LA. This controller will be implemented on the real system in order to be able to perform identification experiments. This closed loop data will then be used for the data-driven controller tuning. Several controllers will have to be implemented in order to compare the performance achieved.
Hysteresis compensation for a piezoelectric stack actuator
The EPFL is involved in the development of an optical differential delay line (DDL), whose purpose is to compensate for the atmospheric perturbation on the VLTI (Very Large Telescope Interferometer) located in Chili. A (PZT) piezoelectric stack actuator is used to actuate the fine stage of the DDL. The hysteresis is a well-known limitation for all nanometer-accurate positioning devices based on piezoelectric materials.
It is proposed to develop and identify a non-linear model of the piezoelectric including the hysteresis. This model will then be used for compensation purpose. The performances of this compensation scheme will be compared to the classical first-harmonic methodology.
As the prototype is available, it will be possible to measure, implement and test the algorithms in real conditions.
Application of an adaptive Iterative Learning Control Algorithm to a linear motor
Iterative Learning Control (ILC) is a relatively new control algorithm used to enhance the tracking performance of systems carrying out repetitive tasks. The technique uses information from the previous repetitions to adjust the input signal to the system at the next iteration so as to reduce the tracking error.
Iterative Learning Control is very successful at improving tracking when disturbances affecting the system repeat from one repetition to the next. However, when they are not repetitive, like measurement noise, the achieved precision is reduced.
In  an adaptive ILC algorithm is proposed which takes the measurement noise into account. The project proposes to test the algorithm, first in simulation, and then experimentally on a linear motor system.
The project is proposed in collaboration with the Swiss company, ETEL.
 M. Norrlof, “An Adaptive Iterative Learning Control Algorithm with Experiments on an Industrial Robot”, IEEE Transactions on Robotics and Automation, 18(2), pp.245-251, 2002.
Robust pole placement of a flexible arm
Robust controllers are designed to stabilize and satisfy the performances for a set of models. This type of controllers can be applied to nonlinear systems linearized in different operating points, to time-varying systems as well as to systems with parametric uncertainty. An approach to robust control design is to formulate the problem as a convex optimization problem, which can be solved efficiently using numerical methods.
In this project, the robust control of a flexible arm is considered. The system contains a flexible arm attached to a servomotor. The parameters of the system can be changed by adding an extra load or changing the load position on the arm. The objective is to design a robust controller via convex optimization, which stabilizes the system for different loads and assigns the closed-loop poles around the desired places.
The project covers all basic controller design steps of a general control problem, which starts from system identification and analysis and terminates with controller design and validation both in simulation (MATLAB) and in real system. Moreover, the student will learn many basic control concepts and the basics of controller design by convex optimization.
Control of a unicycle robot
The Automatic control laboratory would like to develop an autonomous robot demonstrator based on the unicycle. The robot will be standing on a single wheel as funambulists do.
The mechatronic setup corresponding to the problematic will be entirely defined in the context of the project. In particular, the kind of actuation necessary to make the system controllable, will be determined based on the feasibility analysis.
Hence, the project has three objectives:
(i) Developing a dynamical model of a unicycle robot.
(ii) Analysis of the controllability.
(iii)Designing a stabilizing controller.
SwissCube attitude determination
The SwissCube is a small satellite (1kg) entirely designed by students from different Universities in Switzerland.
The SwissCube attitude is controlled through magneto-torquers acting throug the local Earth magnetic field and possibly an additional inertia wheel. The attitude is measured with magnetometers and gyroscopes.
The aim of this project is to design the attitude determination algorithm. The starting point will be the dynamical model and the control strategy already developed in previous projects. The goal is to determine the current position of the satellite with respect to the orbital frame based on the noisy and possibly incomplete available measurements.
Since the measurements and the actuation are partially based on the magnetic field, a model of the Earth’s field has to be embedded. Although numerous standard models are available, a suitable tradeoff has to be found between accuracy and computation complexity.