<p>This thesis deals with three methods for integrating measurements from different sensors for an underwater vehicle. The sensors that were used are inertial measurement unit (IMU), Doppler velocity log (DVL), Hydro- acoustic position reference system (HPR) and tilt and heading measurements. The external measurements (DVL, HPR and attitude) are used to aid the inertial navigation system (INS) which uses the measurements from the IMU to calculate position, velocity and attitude. The different methods presented are extended Kalman filter (EKF), unscented Kalman filter (UKF) and a nonlinear observer. The two Kalman filters were implemented as indirect filters, while the nonlinear observer was implemented as a direct filter. The main difference between the EKF and UKF is that UKF does not make any linarizations such that it captures the covariance of the system more accurate than EKF. To compare the different approaches a navigation system was implemented using Matlab and simulations were carried out to test accuracy and robustness. The nonlinear observer has the most accurate position estimate when all measurements are available. It performed slightly better than UKF which again was more accurate than EKF. A greater difference was seen between UKF and EKF when the noise characteristics in the filters were wrong. For velocity and attitude all estimates were unbiased, but the nonlinear observer produced estimates with far more noise than what the Kalman filters did. All filters handled losing the HPR well. The nonlinear observer did not manage to limit the error in the case of DVL loss as opposed to both Kalman filter which have limited error. They performed with the same grade of degradation of the estimates during the loss. When the measurement returned both Kalman filters immediately regained accuracy but the nonlinear observer did not manage to recover. When losing the IMU measurements both Kalman filters had problems estimating changes in the attitude which again led to error in the position estimate. The position error is however much larger in EKF than UKF. The nonlinear observer has a structure with a separate attitude observer and therefore had a much better attitude estimate during the loss. From these results UKF is considered the best choice for implementation in a real system. It performs accurate estimates during noisy conditions, and suffers only from limited degradation when measurements are lost, both external and inertial.</p>

<p>The need of simulation models for physical systems is essential in all phases of its lifetime. During construction better decisions can be made and difficulties discovered. When running observers one can use the model to estimate states. Knowledge of the system is also needed to develop good control strategies. A compression system without surge avoidance recycling exists, and its mathematical development and simulation model was studied and simulated. Kalman filters were developed in theory, four filters were designed and lastly implemented. A study of industrial solutions for surge avoidance was performed, with attention to practical building, measurements, mathematical modeling and control. The study resulted in three simulation models for recycling that were fitted to the compressor at hand. Lastly a surge avoidance system was implemented on the model most likely to be built. A good simulation model of the existing compressor is available, and using this three well working Kalman filters were implemented. A fourth filter yielded non-unique and sometimes unstable estimates, and emphasized the importance of retrieving enough information from the measurements to estimate the correct state. All recycle system models used the characteristics of the existing compressor. The solution most likely to be implemented and form basis for a simulation model recycled gas directly downstream of the compressor. The other two models retrieved the recycled gas from the plenum, however this will change the dynamics of the system. These were not considered practical solutions for the specific system. The surge avoidance scheme worked well for production close to the surge line and small set-point changes. It also worked well for large changes in speed, but will surge if the speed becomes too low. The controller scheme did not work well when a large change in plenum throttle occurred, the recycle valve was then too slow. Future work consists of building a recycle loop, develop simulation model and implement surge avoidance scheme for academic purposes.</p>

<p>In this thesis a Sliding Mode Controller (SMC) is designed for an electro-pneumatic clutch actuator controlled by two on/off valves with PWM. The areas of application of the clutch actuator is in Automated Manual Transmission (AMT) and Clutch-by-wire (CBW) systems in heavy-duty trucks. As with most automated systems in the automation industry safety is the main priority, and in a cybernetic point of view this means that robustness of the control systems is very important. Sliding Mode Controllers are known for their excellent robustness properties and the focus of this thesis is to validate these properties for this particular application. The robustness properties of the SMC also indicates that a simple design model is adequate and since the existing mathematical models for this system are quite complex some simplifications are introduced. The controller design is performed in two phases. First an ideal SMC is designed, but since this introduces discontinuities in the control law any practical implementation would give heavy chattering at the output from the controller. Therefore a continuous linear approximation to the discontinuity is introduced. This controller is known as a boundary layer controller and it will reduce the control chattering to an acceptable level. The ideal controller is proven to be asymptotically stable, while for the boundary layer controller ultimate boundedness is achieved and a linearisation is performed for the case of a constant reference and this analysis shows that the origin of the linearised system is a stable focus. This indicates that the boundary layer controller might also be asymptotically stable. Since only a positionmeasurement is available to the control system the rest of the system states must be estimated. Velocity and acceleration are simply estimated as the first and second order filtered derivatives of the position measurement. The pressure is estimated based on the equation of motion for the clutch actuator. Through computer simulations and experimental testing the SMC has shown satisfying tracking performance and very good robustness with respect to parameter variations. Comparisons with a PD and a Backstepping controller shows that the performance of the SMC is superior to the PD controller and absolutely comparable to the Backstepping controller, though they have different strengths and weaknesses and therefore yield rather different results.</p>

<p>This thesis is a part of the SSETI (Student Space Exploration Technology Initiative) project, where students from several universities around Europe work together with the European Space Agency (ESA) with designing, building, testing and launching an Earth-Moon satellite orbiter (European Student Moon Orbiter (ESMO). A satellite model with reaction wheels placed in tetrahedron was deduced in a preliminary study together with an extended Kalman filter to estimate the attitude from star measurements. The stability and convergence properties of this system are studied in this thesis. Previous studies on the convergence of extended Kalman filter are presented and a proof of exponentially convergence of a system with extended Kalman filter is given and used to prove that ESMO with the extended Kalman filter converges exponentially. The most recent work and different methods to apply a nonlinear separation principle is presented. Three feedback controllers with proof of global asymptotic stability (GAS) is then introduced and implemented on ESMO. Based upon the global asymptotic stability of the feedback controllers, and the proof that the extended Kalman filter works as an exponentially observer, a nonlinear separation principle is deduced. The closed loop system can then be stated globally asymptotically stable based upon the deduced separation principle. The closed loop with the three different controllers is then simulated in Simulink for varying gains and different reference steps. The three controllers show stable characteristic as the theory implies. The robust controller shows best tracking and estimation properties, it is very accurate, simple, robust and adaptable to environmentally changes, and is therefore proposed as the most suitable controller for ESMO.</p>

<p>This report describes the modelling and performance of an attitude determination and control system (ADCS) for a small satellite in lunar orbit. The focus is on stability analyses of each of the components in the system, and of the system as a whole. In connection to this, the separation principle for nonlinear systems is investigated. Central background information is presented, covering necessary rigid body dynamics and stability properties. Three different controller types are analysed and compared herein, namely a model-dependent linearizing controller, a robust controller and a standard PD-controller. An observer is chosen based on earlier work, but some detail modifications are made to its structure. A state-space model of the satellite and environment is derived and implemented in Matlab, along with the observer and controllers. The observer and all three controllers are shown to be stable with Lyapunov analysis. The total ADCS including the observer is shown to have a cascaded structure, on which theory of nonlinear separation principles is used to establish stability properties of the total system. Finally, the ADCS is put to simulation tests imitating real-life scenarios and the performance of the different controllers are compared. The PD-controller shows the best performance, both in speed of convergence and robustness to model errors. While not completely satisfactory, the results give a basis on which to perform further work.</p>

<p>In this project a robust adaptive controller has been developed for Minerva, NTNU's research ROV. The controller was tested in simulation using Matlab/Simulink with a mathematical model of the vessel. It was also tested in a practical experiment at sea, with the ROV Minerva. The simulations, the control system performs very well. The results from the practical experiment are promising, but several improvements are necessary before the system works satisfactory. The single factor which is believed to degrade the perfomance the most is an error in the mapping between thrust force and rotational speed.</p>

<p>An adaptive Luenberger-type observer with the purpose of locating and quantifying leakages is presented. The observer only needs measurements of velocity and temperature at the inlet and pressure at the outlet to function. The beneficial effect of output injection in form of boundary conditions is utilized to ensure fast convergence of the observer error. This approach is different from the usual practice where output injection might appear as a part of the PDE’s. This makes it possible to employ OLGA, which is a state of the art computational fluid dynamics simulator, to govern the one-phase fluid flow of the observer. Using OLGA as a base for the simulations introduces the possibility to incorporate temperature dynamics in the simulations which in previous work was impossible. The observer is tested with both a straight, horizontal pipeline and an actual, long pipeline with difference in altitude. Both simulations with oil and gas are carried out and verification of the robustness of the observer is emphasized. In order to cope with modelling errors and biased measurements, estimation of roughness in the monitored pipeline is introduced.</p>

<p>In reservoir management it is important with reservoir models that have good predictive abilities. Since the models initially are based on measurements with high uncertainties it is important to utilize new available data. Ensemble Kalman Filter (EnKF) is a new method for history matching that has received a lot of attention the last couple of years. This method is sequential and continuously update the reservoir model states (saturations, pressures etc.) and parameters (permeabilities, porosities etc) as data become available. The EnKF algorithm is derived and presented with a different notation, similar to that of the Kalman Filter (KF) used in control engineering. This algorithm is also verified on a simple linear example to illustrate that the covariance of the EnKF approaches that of the linear KF in case of an infinite ensemble size. In control theory this method falls under the category of parameter and state estimation of nonlinear large scale systems. Interesting aspects as observability and constraint handling arises, and these are linked to the EnKF and the reservoir case. To determine if the total problem is observable is a nearly impossible task, but one can learn a lot from introducing this concept. The EnKF algorithm was implemented on a simple “shoe box” reservoir model and four different problem initializations were tested. Although decent results were achieved from some of the simulations other failed completely. Some strange development in the ensemble when little information is available in the measurements was experienced and discussed. An outline was presented for a reservoir management scheme where EnKF is combined with Model Predictive Control (MPC). Some challenges was pointed out and these involve computation time, predictive ability, closed-loop behavior etc.</p>

<p>This thesis investigates different methods of backstepping controller design for an electro-pneumatic clutch actuator used in heavy duty trucks. The first part of the thesis is a literature study, where the subject is control of nonlinear-sampled data systems in general. Sampled-data systems contain a continuous-time plant and a digitally implemented controller, which in general make them harder to analyze and control than systems that operate purely in the continuous-time or discrete-time domain. The available theory of nonlinear sampled-data control systems is scarce, but three different methods are described in this thesis; emulation design, direct discrete-time design, and sampled-data design. The electro-pneumatic clutch actuator is controlled using a continuous-time backstepping controller implemented digitally. This is essentially the procedure of emulation design and is the common, if not only, method used in practical engineering tasks so far. However, redesign of the continuous-time controller using the direct discrete-time method shows great potential of improving performance and robustness of sampled-data systems. Direct discrete-time design is based on an approximate discrete-time model of the plant, giving the controller a structure that accounts for the sampling of the hybrid system. Potentially, one can utilize slower sampling in the system by implementing a discrete-time controller into the digial computer instead of a continuous-time one. Examples and case studies that prove the improvement one can achieve by chosing the direct discrete-time design is included in the first part of the thesis. Both a third- and fifth-order model of the electro-pneumatic clutch actuator are presented, and used as a basis for continuous- and discrete-time state-feedback backstepping controllers. These controllers are simulated with different sampling intervals to show their performance under different circumstances. The continuous-time controllers prove good reference trajectory tracking of the pure continuous-time system, while the performance of the sampled-data systems descends as higher sampling intervals are used. And, as opposed to the mentioned examples and case studies, the controller designed when taking the sampling into account shows no sign to outperform the controller that was designed without considering the sampling, at least not for the relative fast sampling the clutch actuator operates with.</p>

<p>This report covers the development of a new joint mechanism and a robotic simulator. These are intended to be used in the further development of a climbing robot that has been previously worked on. The climbing robot is briefly described in this report. It is a robot with four arms that each has 6 joints with 1 degree of freedom each. At the end of each arm is a gripper that will serve as the robots main tool of interaction with its immediate environment, for instance a ladder the robot is climbing in. The weight of this robot is estimated and used to find torque requirements on the new joint mechanism. A lightweight joint mechanism is of interest not only for a climbing robot, but for many if not all other kinds of robots as well. Such a joint will not need to be as strong as its heavier counterparts, as less power is required to support the joint itself. The new joint mechanism is an attempt of creating a joint that is lighter than traditional joints. The main idea is to have a single powerful actuator driving several joints. The joints are mechanically powered by a rotating shaft that runs through all of them. Each joint uses a clutching mechanism to connect to the shaft and thus transfer torque. the clutching mechanism has been explored in detail, and a suitable clutch has been selected. This clutch is then used in four design proposals. The last of these proposals is assembled into a prototype, and the prototype is qualitatively tested. The tests demonstrate the concept, but it is not yet shown that this joint mechanism will be lighter that a traditional joint mechanism. The simulator is developed to be a design tool for further development of the climbing robot. A list of specifications for the simulator is presented, however the simulator is not able to meet all these requirements at the current stage of development. The remaining work on the simulator is discussed, as well as an evaluation of the software as it is at the time of writing. The report describes the kinematic modeling that is used to represent the translational and rotational relationship between the different elements of the robot. The kinematic relationships is then put into a mathematical representation of the robot which represents the robots as a chain of elements. Each element has several properties, among which are its relationship to other elements and a graphical representation of the element. The elements, and thus the robot, is represented graphically in a virtual reality environment. The simulator allows the user to specify several parameters of the robot, such as number and dimensions of the joints and the dimensions of the body. The user is also able to control the velocity of each joint. Both the simulator and the joint has been developed to be general enough for use on other kinds of robots, with small modifications in the case of the simulator.</p>