<p>The main objective of this thesis is to use available knowledge about a process and combine this with measurement data from the same process to extract more information about the process. The combination of knowledge and measurement data is referred to as Multi Sensor Data Fusion, MSDF. This added information is then used to control the process towards a specified goal.</p><p>The process studied in this thesis is the process of drilling wells in a petroleum reservoir, while the oil is flowing from the reservoir. In the petroleum industry, this is defined as underbalanced drilling (UBD), where the bottom hole pressure (BHP) in the well is below the pore pressure in the reservoir.</p><p>Detailed knowledge of the process is of paramount importance when using multi sensor data fusion. Due to this, various process modelling efforts are examined and evaluated, from simple relations between parameters to a finite-element approach of modelling the fluid flow in the well during drilling. Several sensors are used in the various cases, and existing sensors such as pressure sensors and flow sensors are the main data source in the analysis. Future scenario with sensors such as pressure arrays and non-intrusive multiphase flow meters are evaluated. In addition, new positions of existing sensor systems are discussed.</p><p>The methods available for fusing the knowledge of the process represented as models together with the available data is ranging from artificial intelligent methods such as neural networks, to methods incorporating statistical analysis such as various Kalman filters. History matching techniques using gradient techniques are also examined.</p><p>The migration of reservoir fluids into the well during UBD influences the BHP of the well. The results in the thesis show that this reservoir influx can be calculated by estimating some of the important reservoir parameters such as reservoir pore pressure or reservoir permeability. These reservoir parameters can be estimated most efficiently by performing an MSDF using a detailed nonlinear model of the well and reservoir dynamic behaviour together with real-time measurements of the fluid flow parameters such as fluid temperature, fluid pressure and fluid flow rates. The unscented Kalman filter shows the best performance when evaluating both estimation accuracy and computational requirements.</p><p>Regarding available instrumentation for use during UBD, the analysis shows that there is a major potential in introducing new sensors. As new data transmission methods are emerging and making data from sensors distributed along the drillstring available, this can generate a shift in paradigm regarding real-time analysis of reservoir properties during drilling.</p><p>Controlling the process is an important usage of the information gained from the MSDF analysis. Various control methods for controlling the most important process variables are examined and evaluated. The results show that acceptable pressure control can be obtained when using the choke valve opening as the primary control parameter. However, the choke valve operation has to be closely coordinated with drilling fluid flow rate adjustments. The choke valve opening control is able to compensate for pressure variations during the whole drilling operation.</p><p>A suggested nonlinear model predictive control algorithm gives best results when looking at the control accuracy, and can easily be expanded to handle multiple control inputs and system constraints. This control algorithm uses a detailed model of the well and reservoir dynamics. The Levenberg-Marquardt algorithm is used to calculate the optimal future control variables. The main drawback of the control algorithm is computational burden. A linear control algorithm, which also is evaluated, uses less computational resources, but has less control accuracy and is more difficult to expand into a multivariable control system.</p><p>Recommendations for further work are to expand the suggested model predictive control algorithm to handle more control inputs, while reducing the computational burden by incorporating low-order models for describing the future behaviour of the well.</p>

The main objective of this thesis is to use available knowledge about a process and combine this with measurement data from the same process to extract more information about the process. The combination of knowledge and measurement data is referred to as Multi Sensor Data Fusion, MSDF. This added information is then used to control the process towards a specified goal. The process studied in this thesis is the process of drilling wells in a petroleum reservoir, while the oil is flowing from the reservoir. In the petroleum industry, this is defined as underbalanced drilling (UBD), where the bottom hole pressure (BHP) in the well is below the pore pressure in the reservoir. Detailed knowledge of the process is of paramount importance when using multi sensor data fusion. Due to this, various process modelling efforts are examined and evaluated, from simple relations between parameters to a finite-element approach of modelling the fluid flow in the well during drilling. Several sensors are used in the various cases, and existing sensors such as pressure sensors and flow sensors are the main data source in the analysis. Future scenario with sensors such as pressure arrays and non-intrusive multiphase flow meters are evaluated. In addition, new positions of existing sensor systems are discussed. The methods available for fusing the knowledge of the process represented as models together with the available data is ranging from artificial intelligent methods such as neural networks, to methods incorporating statistical analysis such as various Kalman filters. History matching techniques using gradient techniques are also examined. The migration of reservoir fluids into the well during UBD influences the BHP of the well. The results in the thesis show that this reservoir influx can be calculated by estimating some of the important reservoir parameters such as reservoir pore pressure or reservoir permeability. These reservoir parameters can be estimated most efficiently by performing an MSDF using a detailed nonlinear model of the well and reservoir dynamic behaviour together with real-time measurements of the fluid flow parameters such as fluid temperature, fluid pressure and fluid flow rates. The unscented Kalman filter shows the best performance when evaluating both estimation accuracy and computational requirements. Regarding available instrumentation for use during UBD, the analysis shows that there is a major potential in introducing new sensors. As new data transmission methods are emerging and making data from sensors distributed along the drillstring available, this can generate a shift in paradigm regarding real-time analysis of reservoir properties during drilling. Controlling the process is an important usage of the information gained from the MSDF analysis. Various control methods for controlling the most important process variables are examined and evaluated. The results show that acceptable pressure control can be obtained when using the choke valve opening as the primary control parameter. However, the choke valve operation has to be closely coordinated with drilling fluid flow rate adjustments. The choke valve opening control is able to compensate for pressure variations during the whole drilling operation. A suggested nonlinear model predictive control algorithm gives best results when looking at the control accuracy, and can easily be expanded to handle multiple control inputs and system constraints. This control algorithm uses a detailed model of the well and reservoir dynamics. The Levenberg-Marquardt algorithm is used to calculate the optimal future control variables. The main drawback of the control algorithm is computational burden. A linear control algorithm, which also is evaluated, uses less computational resources, but has less control accuracy and is more difficult to expand into a multivariable control system. Recommendations for further work are to expand the suggested model predictive control algorithm to handle more control inputs, while reducing the computational burden by incorporating low-order models for describing the future behaviour of the well.

A satellite in orbit about a planet needs some means of attitude control in order to, for instance, get as much sun into its solar-panels as possible. It is easy to understand that, for example, a spy satellite has to point at a certain direction without the slightest trembling to get a photo of a certain point on the earth. This type of mission must not exceed an error in attitude of more then about 1/3600 degrees. But, since high accuracy equals high cost, it is also easy to understand why a research satellite measuring solar particles (or radiation) in space does not need high accuracy at all. A research vessel of this sort can probably do with less accuracy then 1 degree. The first part of this report tries to explain some major aspects of satellite space-flight. It continues to focus on the market for small satellites, i.e. satellites weighing less than 500 kg. The second part of this final thesis work deals with the development of a program that simulates the movement of a satellite about a large celestial body. The program, called AOSP, consists of user-definable packages. Sensors and estimation filters are used to predict the satellites current position, velocity, attitude and angular velocity. The purpose of the program, which is written in MATLAB, is to easily determine the pointing accuracy of a satellite when using different sensors and actuators.

<p>A satellite in orbit about a planet needs some means of attitude control in order to, for instance, get as much sun into its solar-panels as possible. It is easy to understand that, for example, a spy satellite has to point at a certain direction without the slightest trembling to get a photo of a certain point on the earth. This type of mission must not exceed an error in attitude of more then about 1/3600 degrees. But, since high accuracy equals high cost, it is also easy to understand why a research satellite measuring solar particles (or radiation) in space does not need high accuracy at all. A research vessel of this sort can probably do with less accuracy then 1 degree. </p><p>The first part of this report tries to explain some major aspects of satellite space-flight. It continues to focus on the market for small satellites, i.e. satellites weighing less than 500 kg. The second part of this final thesis work deals with the development of a program that simulates the movement of a satellite about a large celestial body. The program, called AOSP, consists of user-definable packages. Sensors and estimation filters are used to predict the satellites current position, velocity, attitude and angular velocity. The purpose of the program, which is written in MATLAB, is to easily determine the pointing accuracy of a satellite when using different sensors and actuators.</p>

<p>A container crane of STS-type, Ship To Shore, consists of a spreader hanging underneath a railrunning trolly. As the container is under the influence of wind, it is likely that it starts to turn in a torsional pendulum. This report handles how the torsional pendulum of a container crane can be damped. </p><p>A number of different models have been developed to analyze how different placement of the actuators affects the system. Two differens types of controllers, LQG and MPC, have been developed and applied to these models. The different models and controlers were evaluated and compared by studying simulation results in timedomain. Moreover in order to make the simulations more realistic, a wind model has been developed and applied. </p><p>The models and controllers have been analyzed with bodediagrams and sensitivity functions. </p><p>The analyses shows clearly that the best placement of the actuators for control of the torsional pendulum on an STS-crane is in the trolly, pulling and relaxing the wires. This control is best handled by a state feedback control (LQG). Furthermore, the control should in this way, with addition of in the horizontalplane movable suspensions in the trolly, work acceptably in the whole operational area of a STS-crane.</p>

A container crane of STS-type, Ship To Shore, consists of a spreader hanging underneath a railrunning trolly. As the container is under the influence of wind, it is likely that it starts to turn in a torsional pendulum. This report handles how the torsional pendulum of a container crane can be damped. A number of different models have been developed to analyze how different placement of the actuators affects the system. Two differens types of controllers, LQG and MPC, have been developed and applied to these models. The different models and controlers were evaluated and compared by studying simulation results in timedomain. Moreover in order to make the simulations more realistic, a wind model has been developed and applied. The models and controllers have been analyzed with bodediagrams and sensitivity functions. The analyses shows clearly that the best placement of the actuators for control of the torsional pendulum on an STS-crane is in the trolly, pulling and relaxing the wires. This control is best handled by a state feedback control (LQG). Furthermore, the control should in this way, with addition of in the horizontalplane movable suspensions in the trolly, work acceptably in the whole operational area of a STS-crane.

One of the fundamental limitations of mobile radio communications is their time-varying fading channel. This thesis addresses the efficient use of channel state information to improve the communication systems, with a particular emphasis on practical issues such as compatibility with the existing wireless systems and low complexity implementation. The closed-loop transmit diversity technique is used to improve the performance of the downlink channel in MIMO communication systems. For example, the WCDMA standard endorsed by 3GPP adopts a mode of downlink closed-loop scheme based on partial channel state information known as mode 1 of 3GPP. Channel state information is fed back from the mobile unit to the base station through a low-rate uncoded feedback bit stream. In these closed-loop systems, feedback error and feedback delay, as well as the sub-optimum reconstruction of the quantized feedback data, are the usual sources of deficiency. In this thesis, we address the efficient reconstruction of the beamforming weights in the presence of the feedback imperfections, by exploiting the residual redundancies in the feedback stream. We propose a number of algorithms for reconstruction of beamforming weights at the base-station, with the constraint of a constant transmit power. The issue of the decoding at the receiver is also addressed. In one of the proposed algorithms, channel fading prediction is utilized to combat the feedback delay. We introduce the concept of Blind Antenna Verification which can substitute the conventional Antenna Weight Verification process without the need for any training data. The closed-loop mode 1 of 3GPP is used as a benchmark, and the performance is examined within a WCDMA simulation framework. It is demonstrated that the proposed algorithms have substantial gain over the conventional method at all mobile speeds, and are suitable for the implementation in practice. The proposed approach is applicable to other closed-loop schemes as well. The problem of (long-range) prediction of the fading channel is also considered, which is a key element for many fading-compensation techniques. A linear approach, usually used to model the time evolution of the fading process, does not perform well for long-range prediction applications. We propose an adaptive algorithm using a state-space approach for the fading process based on the sum-sinusoidal model. Also to enhance the widely-used linear approach, we propose a tracking method for a multi-step linear predictor. Comparing the two methods in our simulations shows that the proposed algorithm significantly outperforms the linear method, for both stationary and non-stationary fading processes, especially for long-range predictions. The robust structure, as well as the reasonable computational complexity, makes the proposed algorithm appealing for practical applications.

One of the fundamental limitations of mobile radio communications is their time-varying fading channel. This thesis addresses the efficient use of channel state information to improve the communication systems, with a particular emphasis on practical issues such as compatibility with the existing wireless systems and low complexity implementation. The closed-loop transmit diversity technique is used to improve the performance of the downlink channel in MIMO communication systems. For example, the WCDMA standard endorsed by 3GPP adopts a mode of downlink closed-loop scheme based on partial channel state information known as mode 1 of 3GPP. Channel state information is fed back from the mobile unit to the base station through a low-rate uncoded feedback bit stream. In these closed-loop systems, feedback error and feedback delay, as well as the sub-optimum reconstruction of the quantized feedback data, are the usual sources of deficiency. In this thesis, we address the efficient reconstruction of the beamforming weights in the presence of the feedback imperfections, by exploiting the residual redundancies in the feedback stream. We propose a number of algorithms for reconstruction of beamforming weights at the base-station, with the constraint of a constant transmit power. The issue of the decoding at the receiver is also addressed. In one of the proposed algorithms, channel fading prediction is utilized to combat the feedback delay. We introduce the concept of Blind Antenna Verification which can substitute the conventional Antenna Weight Verification process without the need for any training data. The closed-loop mode 1 of 3GPP is used as a benchmark, and the performance is examined within a WCDMA simulation framework. It is demonstrated that the proposed algorithms have substantial gain over the conventional method at all mobile speeds, and are suitable for the implementation in practice. The proposed approach is applicable to other closed-loop schemes as well. The problem of (long-range) prediction of the fading channel is also considered, which is a key element for many fading-compensation techniques. A linear approach, usually used to model the time evolution of the fading process, does not perform well for long-range prediction applications. We propose an adaptive algorithm using a state-space approach for the fading process based on the sum-sinusoidal model. Also to enhance the widely-used linear approach, we propose a tracking method for a multi-step linear predictor. Comparing the two methods in our simulations shows that the proposed algorithm significantly outperforms the linear method, for both stationary and non-stationary fading processes, especially for long-range predictions. The robust structure, as well as the reasonable computational complexity, makes the proposed algorithm appealing for practical applications.