MPC MOTION PLANNING ON ROUGH TERRAINS (in bosnian)

STUDENT: Nadir Kapetanović

THESIS ABSTRACT:

Path planning and control for a mobile vehicle which has to accomplish some mission, some of the crucial problems in the field of mobile robotics, especially if the vehicle moves on a large scale rough terrain with obstacles. In order to guarantee that the mission will be accomplished, path planner has to find the most traversable path from some given starting location to the goal location in the map, while ensuring the physical integrity of the vehicle, but also avoiding obstacles, and preventing roll-over and slip-off phenomena. In this thesis one such planner is presented, which is based on merging model predictive control with Wavefront RbNF algorithm into one generic path planning framework, which takes into account path traversability, distance from the goal location, system model, and constraints on states and control. Wavefront RbNF algorithm is developed as an extension of the existing RbNF algorithm. It represents a fast approximation of the optimal cost-to-go map in terms of traversability on known large scale rough terrains, since computation of the optimal cost-to-go map for large scale terrains is very expensive in terms of computer resources, both in terms of execution time and RAM memory. Thorough simulation results show that the majority of cost-to-go map values, calculated for every location on the map with respect to the goal location, are near-optimal. As opposed to Dijkstra an A* algorithms, the proposed algorithm has an inherently parallel structure, which allows its execution time to be significantly faster, depending on the number of CPU cores used. The main advantages of using model predictive control for a path planner are twofold. First, such planner would generate trajectories which a mobile vehicle could follow, which is often not the case in the state-of-the-art path planning algorithms. Second, these trajectories would be optimal in terms of system model, constraints on
states and control, and the given optimization criterion. Simulation results show that the use of this planner is justified, since it gives very good results for the examples of path planning on large scale rough terrains with obstacles. This way, the most traversable path is generated for the vehicle to follow, i.e. the least rough path towards the goal location. Additionally, the vehicle avoids obstacles with an extra security requirement: that it must not come closer to the obstacles than some predefined distance,
and it also makes softer turns compared to a much simpler realisation gradient method generated paths.

TERRAIN CLASSIFICATION FOR MOBILE VEHICLE PLANNING ON ROUGH TERRAINS (in bosnian)

STUDENT: Amel Selimovic

THESIS ABSTRACT:

The aim of this work is to analyze how different terrain features influence autonomous vehicle mobility index. Some terrain features can significantly influence vehicles mobilitiy index, which mostly occurs when the vehicles face uneven terrains. An inadequate analysis of these features and an inadequate terrain classification may cause similar problems such as the one that NASA's Mars exploration mission faced, when its autonomous vehicle had been stuck for some weeks. Detection or estimation of physical terrain features is important for ensuring vehicle safety. Further knowledge of terain features can improve vehicles motion prediction and help the vehicles to avoid terrain regions that would prevent the vehicle to complete its mission. Furthermore, avoiding regions with low mobility indexes would also reduce necessary power consumption. This work presents several classification techniques used for detection of different terrain characteristics (e.g., terrain color, terrain texture) which are necesary to successfully determine type of terrain and its mobility index. In addition, a moethod of using these terrain classification techniques for autonomous vehicle path planning has been implemented and presented.

CHANCE CONSTRAINED MPC - HVAC (in bosnian) n bosnian)

STUDENT: Šalaka Edin

THESIS ABSTRACT:

The focus of this work is given to the robustness of model predictive control (MPC) with respect to uncertainties that might influence the outputs of the system and thereby the constraints imposed on those outputs. Unlike the robust approach that guarantee constraints are satisfied even for worst case scenarios, which can be rather conservative in some applications, chance constrained MPC optimization framework guarantees the imposed constraints will not be violated with some a priori chosen probability. Chance constraint MPC utilizes the probability density function of the uncertainty present in the system and solve repeatedly an appropriate convex optimization problem.
The proposed approach is presented and validated through the application of energy management in intelligent buildings to solve an HVAC (heating, ventilation and air conditioning) problem.

COOPERATIVE CONTROL OF MULTIPLE VEHICLES (in bosnian)

STUDENT: Bostan Aldin

THESIS ABSTRACT:

The paper presents a new solution to the multi-vehicle coverage problem. The proposed algorithm guarantees complete coverage and provides collaborative behaviors of vehicles, despite the fact that it does not explicitly exploit any computationally intensive optimization technique. The algorithm can deal with any mission domain, including regions with irregular shapes, multi-connected and disjoint regions. It gives reasonably good solutions for partially connected multi-vehicle systems.

CONSTRAINED COVERAGE PROBLEM WITH MULTIPLE VEHICLES (in bosnian)

THESIS ABSTRACT:

A cooperative way of solving problems has been popular for a long time in scientific
community, especially in control theory. Great attention is dedicated to cooperative control of multi - vehicle systems because of their benefits in terms of completing a task more efficiently and the extent of work a group of vehicles can do in comparison to a single one. Cooperative system control of vehicles or sensors has found its use in many areas among which applications arose in military field are the most popular ones. Some of the applications have been recognized as rather useful such as those related to military resource organization as in case of unmanned terrain and aerial vehicles. In this paper one aspect of cooperative control has been studied, which is known as the coverage problem. This problem is actually inherently included in most applications of the cooperative control, including the autonomous sensor arrangements, routing, planning, following, as well as rescuing, mowing, cleaning, extinguishing fires, and so on. Using GPOPS, which is a toolbox made for MATLAB users for solving optimal control problems by the means of pseudospectral methods, an algorithm for covering space with a heterogeneous multi - vehicle system, where each vehicle has a defined task duration time, has been developed. The vehicles have also a defined starting and ending point. In addition, simulation results have been shown in the paper and some appropriate conclusions have been derived with respect to some further possible research directions within this field.

MPC IN THERMAL POWER PLANTS (in bosnian))

THESIS ABSTRACT:

A thermal power plant is a complex system intended for multistep conversion of
energy. A very important factor for a safe and economical process in thermal power
plants is a quality realization of the control process. Nowadays, advanced control
methods introduced in modern power plants, independent or in the combination with
classical methods, cause enhancement of the power plants performances. The subject of this master thesis is the presentation of one advanced control tehnique – model predictive control (MPC) in the control process of a fossil fueled power plant. A thermal power plant control system based on the combination of classical and model predictive control is analized. Then, the superheated steam temperature control system based on the classical and model predictive control is presented, and a comparation between both control methods is conducted. Finally, based on the input – output data collected at the thermal power plant in Tuzla, a system identification of a two stage power plant superheater is carried out and an appropriate model predictive control is designed.

AUTOMATED INSERTION OF GEOPHONES USING LIGHT-WEIGHT ROBOTS (in bosnian and english)

THESIS ABSTRACT:

Manual installation of seismic networks, i.e., geophones in extraterrestrial applications or in extreme environments on Earth is risky, expensive and error-prone. A more reliable alternative of inserting sensors into soil is the automatic deposition with a leight-weight robot manipulator. However, inserting a sensor into soil is a challenging task for robotic control since the soil parameters are variable and di cult to estimate. Therefore, this thesis investigates an approach to accurate insertion and positioning of geophones using a Cartesian impedance controller with
a feed-forward force term. The feed-forward force component of the controller is estimated using the earth-moving equation and the Discrete Element Method. For  the rst time, both the geological aspects of the problem as well as the aspects of
robotic control are considered. Based on this consideration, the control approach is enhanced by predicting the resistance force of the soil. Experiments with the humanoid robot Rollin' Justin inserting a geophone into three di erent soil samples validate the chosen approach.

All experiments have been conducted at DLR, Germany.

MODEL PREDICTIVE CONTROL IN SOLAR POWER PLANTS (in bosnian)

THESIS ABSTRACT:

Solar radiation is an environmentally friendly, the largest, most affordable energy source on Earth. Adequate utilization of a small portion of this energy can meet Earths electricity needs. The research work has found that there are two basic technologies of producing electricity using solar radiation. In addition to economically dominant technology photo-cell, in the last 15th years arises the renewed interest for concentrating parabolic trough technology (since 1980) which can accumulate a greater amount of energy, and thus improving the conditions of availability, integration into the power grid. Price competitiveness of this technology, in addition to reducing the cost of individual elements of the system, can be also achieved by using an adequate control strategies that can increase efficiency, and the number of operational hours. This works describes the aforementioned technology, with emphasis on basic control needs, on which it is possible to accomplish these objectives. The result of these studies has shown that monitoring the movement od the Sun and an improved ability to reject disturbances, increase efficiency and the number of operational hours. For the purpose of detailed analysis, an adequate modeling of Acurex test plant Almeria (Spain) has been performed along with the identification of incident angle modifier, which defines parabolic troughs efficiency, using the CMA (Covariance Matrix Adaptation) evolution strategy. By using simulation analysis of this model dominant disturbances which may impair the efficient operation of this type of power plants, have been identified. These disturbances are: changes in solar radiation (fast disturbance), and changes in inlet temperature collector array (slow disturbance), and they might cause significant deviation from the optimum output temperature. Given the basic characteristics of this system in the form of non-stationarity, time delay, nonlinearity, disturbances, model predictive control has been turned out to be a good choice of control strategy. This control strategy uses a linear model in state space, which is a result of linearization at each sampling point. For the purposes of state estimation, nonlinear form of Kalman filter UKF (Unscented Kalman Filter) is used. This approach has resulted in a very good elimination of slow disturbances (increasing the number of operational hours at the beginning of the day), with appropriate behavior due to the effects of fast disturbances. A possible improvment of the control strategy, given in the form of model predictive control, regards the use of an appropriate prediction model of solar radiation within the optimization framework.

FLATNESS-BASED CONTROL (in bosnian and english)

THESIS ABSTRACT:

In this work, flatness-based control basics are presented. Two examples are presented: a mobile robot and a magnetically supported spindle. Mobile robot is described theoretically, while the part with the spindle was mostly experimental. A magnetically supported spindle can be driven at very high speeds, and as it levitates, there is neither friction nor mechanical wear. This type of spindles is increasingly popular in industry, but the control of the active bearings is challenging. There is a nonlinear relation between the coil current, shaft position and the force  which accelerates the shaft, and also a lot of nonlinearities in the system that need to be considered. To control one axial bearing, an exact relationship between these quantities is required. With a good model, an efficient real-time control can be implemented and the sensor features of the bearing can be used to measure the force or the position. Flatness-based control of the spindle is described, a mathematical model of an axial bearing is derived, and the parameters of the model are experimentally identified. The sensors used for the experiments are analyzed, as well as the deformation of the disc in the stator. Results of the identification are discussed and a suggestion for the further research is given.

MPC IN BIOMEDICAL SYSTEMS (in bosnian)

THESIS ABSTRACT:

Applying knowledge from engineering sciences to formulate solutions for medical problems has given fruitful results; many solutions have been successfully implemented. One such example of a modern application of a closed-loop control is an artificial pancreas system. The aim of the artificial pancreas is to treat diabetes by continuously maintaining glucose levels in the permitted boundaries, in a manner which is neither aggresive nor difficult for the patient. Controlling glucose levels is done by continuously delivering insulin by an insulin pump. The insulin pump is controlled by an algorithm, which uses glucose concentration levels measured in the blood. This thesis explores the implementation of advanced control methods in biological systems. Furthermore, the paper explains the different modelling methods of biological systems, a review of model forms and examples of specific models used for control algorithms. A detailed review is given for the artificial pancreas system, the accent on the needs and advantages of such a system. Furthermore, the characteristics of the system’s elements are given, the focus being on the mathematical models used for the glucose-insulin dynamics, and also the control algorithms which have already been implemented, or which will be implemented if the required research is conducted. For the implementation of the NMPC control, an augumented minimal model of the glucose-insulin system was used. Also, a model of glucose disturbance was given, which was used as a basis to implement the control. The NMPC control was implemented in the Matlab toolbox Gpops, which solves optimal control problem using pseudospectral methods. Correspondingly, an adaption of the optimal problem of open loop control was used to design the NMPC control. Finally, several different simulations were undertaken, each with a different approach to solve the control problem. The acquired results are quite satisfactory. Given results are analyzed, and references for further researches are given.

NONLINEAR MPC (in bosnian)

THESIS ABSTRACT:

Model predictive control (MPC) is a control strategy that is suitable for optimizing the performance of constrained systems. Constraints are present in all systems due to physical or environmental limits on plant operation. MPC is able to handle these constraints in a systematic way. Along with constraints, tighter performance specifications can only be satisfied by an explicit inclusion of process nonlinearities in the controller. In this thesis a framework for simulating nonlinear MPC is realized, based on the pseudospectral optimal control solver GPOPS. Different systems with different specifications and different constraints have been analyzed, and their sensitivity on prediction horizon and sampling time parameters has been tested. At last, stability issues of the model predictive controller were analyzed in regard of control Lyapunov functions.

MPC IN HYDROPOWER PLANTS (in bosnian)

THESIS ABSTRACT:

The importance of the usage of renewable power sources, such as hydropower, rises due to the necessity of the CO2 pollution reduction. That is why huge efforts are being made in the field of the hydropower efficiency enhancement. It is possible, using optimal distribution of power sources, to maximize efficiency of the hydropower valley. In this assignment, for the existing model of hydropower valley, we develop a Model Predictive Control (MPC) controller providing that the electrical production successfully follows demand side of the electrical power consumption while considering constraints regarding changes of water level and water flow rate in lakes and reaches, which should be within given range and as far as possible from the limits.