The complexity of industrial systems and the mathematical models to describe them increases. In many cases point sensors are no longer sufficient to provide controllers and monitoring instruments with the information necessary for operation. The need for other types of information, such as audio and video, has grown. Suitable applications range in a broad spectrum from microelectromechanical systems and bio-medical engineering to papermaking and steel production. This thesis is divided into five parts. First a general introduction to the field of vision-based and sound-based monitoring and control is given. A description of the target application in the steel industry is included. In the second part, a recursive parameter estimation algorithm that does not diverge under lack of excitation is studied. The focus is on the stationary properties of the algorithm and the corresponding Riccati equation. The third part compares the parameter estimation algorithm to a number of well-known estimation techniques, such as the Normalized Least Mean Squares and the Kalman filter. The benchmark for the comparison is an acoustic echo cancellation application. When the input is insufficiently exciting, the studied method performs best of all considered schemes. The fourth part of the thesis concerns an experimental application of vision-based estimation. A water model is used to simulate the behaviour of the steel bath in a Linz–Donawitz steel converter. The water model is captured from the side by a video camera. The images together with a nonlinear model is used to estimate important process parameters, describing the heat and mass transport in the process. The estimation results are compared to those obtained by previous researchers and the suggested approach is shown to decrease the estimation error variance by 50%. The complexity of the parameter estimation procedure by means of optimization makes the computation time large. In the final part, the time consumption of the estimation is decreased by using a smaller number of data points. Three ways of choosing the sampling points are considered. An observer-based approach decreases the computation time significantly, with an acceptable loss of accuracy of the estimates.

The topic for this thesis is the analysis of human movement, or more specifically, markerless analysis of human movement from video material. By markerless analysis is meant that the full image material is used as input in contrast with traditional marker systems that only use the positions of marker centers. The basic idea is to use more of the information in the images to improve the analysis. Starting of with the aim of markerless analysis an application is designed that use, to the subject added texture to estimate the position of the knee joint center in real images. The approach show the plausibility of using subject texture for estimation purposes. Another issue that is addressed is how one can generate synthetic image data. Using basic tools of graphics programming a virtual environment used to synthesize data is created. This environment is also used to evaluate some different camera solutions. One method to make three dimensional reconstruction from multiple images of an object is tested using the synthetic data. The method is based on a "brute force" approach and does not show good performance in terms of computing speed. With appropriate representations of the three dimensional objects, mathematical methods might speed up the analysis.

Viscoelastic materials can today be found in a wide range of practical applications. In order to make efficient use of these materials in construction, it is of importance to know how they behave when subjected to dynamic load. Characterization of viscoelastic materials is therefore an important topic, that has received a lot of attention over the years. This thesis treats different nonparametric methods for identifying the complex modulus of a viscoelastic material. The complex modulus is a frequency dependent material function, that describes the deformation of the material when subjected to uniaxial stress. With knowledge about this and other material functions, it is possible to simulate and predict how the material behaves under different kinds of dynamic loads. The complex modulus is often identified through wave propagation testing. An important aspect of identification is the accuracy of the estimates. For the identification to be as accurate as possible, it is important that the experimental data contains as much valuable information as possible. Different experimental condition, such as sensor locations and choice of excitation, can influence the amount of valuable information in the data. The procedure of determining optimal values for such design parameters is known as optimal experiment design. The first two papers of the thesis treats optimal experiment design for nonparametric identification of the complex modulus, based on wave propagation tests on large homogenous specimens. Optimal sensor locations is treated in the first paper, and optimal excitation in the second. In the third paper, a technique for estimating the complex modulus for a small pellet-sized specimen is presented. Three different procedures are considered, and an analysis of the accuracy of the estimates is carried out.

No description available.

The report describes how a control system has been developed to dampen known resonance frequency and optimize excitation at other frequencies of a seismic vibrator source. This vibrator source is used to excite energy at the surface of reflection seismic surveys.The resonance frequency dominates the excited signal, which results in limitations of the systems performance. The vibrator source is excited with a power amplifier that receives a control signal from a chirp signal generator. During the project, a chirp generator, as well as a control system, was programmed in Labview, and tested boththrough simulations and in the laboratory. Initially, it was investigated whether it was possible to regulate the system, partly for resonance attenuation and also for optimization of excitation at other frequencies, using a PID controller. However, this proved to be overly advanced and time consuming to perform in a ten-week project. Therefore, the system was regulated by envelope compensation, which both dampened resonance and optimized excitation at other frequencies. / vibratorkälla, reglersystem, reglerteknik

Detection of faults, such as clogged valves, broken bearings or biased sensors, has been brought more and more into focus during the last few decades. There are two main reasons why faults are important to detect at an early stage. Firstly, faults in safety critical applications, such as aircraft, nuclear reactors, cars and trains, may create risks of personal injuries. Secondly, faults in the manufacturing or process industry, e.g. flotation processes and steel plants, may cause decrease in quality or interruptions of production. A fault detection algorithm consists of two parts, the residual generator, which generates a residual, and the residual evaluator, which compares the residual, or a function of it, with a threshold to determine if a fault is present. The residual generation contains a process model and the residual can be described as a filtered difference between the measured and estimated process outputs. When no fault is present, the residual will be nonzero due to residual disturbances, i.e. measurement disturbances, process disturbances and model uncertainties. Therefore, the residual evaluation must be robust against these disturbances to avoid false alarms. Due to the model uncertainties, the residual is affected by the known input signals, which are, in general, time varying. To achieve a threshold that is as tight to the residual as possible, the threshold should also depend on the known input signals. To make this possible, parametric uncertainty in the process model is considered in this thesis. The dynamic threshold generator is introduced, a dynamic system whose output is the threshold and the inputs are the known process inputs. A dynamic threshold generator is developed for full-state measurement systems, assuming that the residual disturbances are constant and unknown but bounded. This dynamic threshold generator is then generalized to non-full state measurement systems with time-varying but bounded residual disturbances. Both generators depend on the unknown upper bounds of the residual disturbances. These upper bounds are replaced by design parameters, which are determined by minimizing the threshold for a set of fault free data. A nonlinear optimization solution is discussed. It is also shown that the residual generator state vector can always be parameterized such that the designing of the parameters can be done by linear optimization. A part of the generalized dynamic threshold generator is a system whose impulse response is an upper bound to another impulse response. Automatic methods to find realizable upper bounds are derived. To validate the methods in this thesis, two applications have been considered, detection of clogging in the valves of a flotation process and detection of faults in the compressor inlet temperature sensor of a jet engine. / Godkänd; 2005; 20061004 (ysko)

Access to measurements is a necessity in most technical applications, in order to detect faults, monitor performance, or exercise control. In some cases, however, installing measurement equipment is very expensive or even impossible. In such a case, estimates can be produced instead. In an observer, this is done by combining process knowledge, in the form of an analytical process model, with information, in the form of indirect measurements. If the process model is in the form of a system of linear differential equations, then the problem of constructing an observer is essentially solved by the Kalman filter and the Luenberger observer. For a system of nonlinear differential equations, however, there is no generic solution, which is the reason for extensive research in this area for the past decades. This thesis treats the development and analysis of nonlinear observers for three applications in the steel industry. The first application is the detection of gas leakages in a pulverized coal injection plant. An observer whose residual is sensitive to the gas leakage flow, has been designed for a nonlinear process model. A Generalized Likelihood Ratio test was applied to the residual to distinguish between different types of leakages. The method has been implemented in the plant and tested successfully with actual leakages. Furthermore, a Laguerre spectrum representation of the residual was utilized, to reduce disturbances and computational effort. The second application is the detection of clogging in pulverized coal injection lines. An observer, with a state variable that represents clogging, has been designed for a time-varying process model. An adaptive threshold for the estimated clogging variable was calculated. In experiments with data from the plant, the method was shown to detect clogging successfully, without producing false alarms. The third application is the estimation of metal analysis in the steel converter process. A nonlinear, physical process model was utilized and an observer was proposed, whose feedback is weighted by the sensitivity of the output with respect to the state. Experiments with data from a converter plant show that this strategy provides accurate estimates of the carbon content in the converter. Furthermore, a generalization of the proposed observer structure has been analyzed in terms of asymptotic stability and region of attraction. / Godkänd; 2001; 20061113 (haneit)

Fault detection and isolation is a potentially powerful tool for achieving security and effective maintenance in various types of processes. The motivation for performing leakage detection in the coal injection plant is mainly the inflammability of pulverized coal. A leakage of air into an injection vessel could have catastrophic consequences. Nonlinear physical gray-box models of the plant are developed. Values of the unknown parameters are estimated by identification. Observers are constructed for these models and the residual is shown to be an estimate of the leakage flow.The Generalized Likelihood Ratio is employed to compare the residual to predefined typical leakage functions. When evaluating the residual, it is desirable to represent the essential dynamics concisely while removing irrelevant behaviour and noise. In order to ease the computational burden while preserving the essential dynamic behaviour of a leakage, a truncated Laguerre series representation of the signals is used. The developed algorithms are implemented in the commercially available product SafePCI and installed at SSAB Tunnplåt, Luleå. / Godkänd; 1999; 20070403 (ysko)

In distributed applications knowing the topological properties of the underlying communication network may lead to better performing algorithms. For instance, in distributed regression frameworks, knowing the number of active sensors allows to correctly weight prior information against evidence in the data. Moreover, continuously estimating the number of active nodes or communication links corresponds to monitoring the network connectivity and thus to being able to trigger network reconfiguration strategies. It is then meaningful to seek for estimators of the properties of the communication graphs that sense these properties with the smallest possible computational/communications overheads. Here we consider the problem of distributedly counting the number of agents in a network. This is at the same time a prototypical summation problem and an essential task instrumental to evaluating more complex algebraic expressions such as products and averages which are in turn useful in many distributed control, optimization and estimation problems such as least squares, sensor calibration, vehicle coordination and Kalman filtering. Being interested in generality, we consider computations in anonymous networks, i.e., in frameworks where agents are not ensured to have unique IDs and the network lacks a centralized authority. This setting implies that the set of distributedly computable functions is limited, that there is no size estimation algorithm with uniformly bounded computational complexity that can provide correct estimates with probability one, and thus that scalable size estimators are non-deterministic functions of the true network size. Natural questions are then: which one is the scheme that leads to topology estimators that are optimal in Mean Squared Error (MSE) terms? And what are the fundamental limitations of information aggregation for topology estimation purposes, i.e., what can be estimated and what not? Our focus is then to understand how to distributedly estimate cardinalities given devices with bounded resources (e.g., battery/energy constraints, communication bandwidth, etc.) and how considering different assumptions and trade-offs leads to different optimal strategies. We specifically consider the case of peer-to-peer networks where all the participants are required to: i) share the same final result (and thus the same view of the network) and ii) keep the communication and computational complexity at each node uniformly bounded in time. To this aim, we study four different estimation strategies that consider different tradeoffs between accuracy and convergence speed and characterize their statistical performance in terms of bias and MSE.

<p>Godkänd; 1989; 20080410 (ysko)</p>