This thesis presents the development of self adaptive flight controller for an unmanned helicopter system under hovering manoeuvre. The neural network (NN) based model predictive control (MPC) approach is utilised in this work. We use this controller due to its ability to handle system constraints and the time varying nature of the helicopter dynamics. The non-linear NN based MPC controller is known to produce slow solution convergence due to high computation demand in the optimisation process. To solve this problem, the automatic flight controller system is designed using the NN based approximate predictive control (NNAPC) approach that relies on extraction of linear models from the non-linear NN model at each time step. The sequence of control input is generated using the prediction from the linearised model and the optimisation routine of MPC subject to the imposed hard constraints. In this project, the optimisation of the MPC objective criterion is implemented using simple and fast computation of the Hildreth's Quadratic Programming (QP) procedure. The system identification of the helicopter dynamics is typically performed using the time regression network (NNARX) with the input variables. Their time lags are fed into a static feed-forward network such as the multi-layered perceptron (MLP) network. NN based modelling that uses the NNARX structure to represent a dynamical system usually requires a priori knowledge about the model order of the system. Low model order assumption generally leads to deterioration of model prediction accuracy. Furthermore, massive amount of weights in the standard NNARX model can result in an increased NN training time and limit the application of the NNARX model in a real-time application. In this thesis, three types of NN architectures are considered to represent the time regression network: the multi-layered perceptron (MLP), the hybrid multi-layered perceptron (HMLP) and the modified Elman network. The latter two architectures are introduced to improve the training time and the convergence rate of the NN model. The model structures for the proposed architecture are selected using the proposed Lipschitz coefficient and k-cross validation methods to determine the best network configuration that guarantees good generalisation performance for model prediction. Most NN based modelling techniques attempt to model the time varying dynamics of a helicopter system using the off-line modelling approach which are incapable of representing the entire operating points of the flight envelope very well. Past research works attempt to update the NN model during flight using the mini-batch Levenberg-Marquardt (LM) training. However, due to the limited processing power available in the real-time processor, such approaches can only be employed to relatively small networks and they are limited to model uncoupled helicopter dynamics. In order to accommodate the time-varying properties of helicopter dynamics which change frequently during flight, a recursive Gauss-Newton (rGN) algorithm is developed to properly track the dynamics of the system under consideration. It is found that the predicted response from the off-line trained neural network model is suitable for modelling the UAS helicopter dynamics correctly. The model structure of the MLP network can be identified correctly using the proposed validation methods. Further comparison with model structure selection from previous studies shows that the identified model structure using the proposed validation methods offers improvements in terms of generalisation error. Moreover, the minimum number of neurons to be included in the model can be easily determined using the proposed cross validation method. The HMLP and modified Elman networks are proposed in this work to reduce the total number of weights used in the standard MLP network. Reduction in the total number of weights in the network structure contributes significantly to the reduction in the computation time needed to train the NN model. Based on the validation test results, the model structure of the HMLP and modified Elman networks are found to be much smaller than the standard MLP network. Although the total number of weights for both of the HMLP and modified Elman networks are lower than the MLP network, the prediction performance of both of the NN models are on par with the prediction quality of the MLP network. The identification results further indicate that the rGN algorithm is more adaptive to the changes in dynamic properties, although the generalisation error of repeated rGN is slightly higher than the off-line LM method. The rGN method is found capable of producing satisfactory prediction accuracy even though the model structure is not accurately defined. The recursive method presented here in this work is suitable to model the UAS helicopter in real time within the control sampling time and computational resource constraints. Moreover, the implementation of proposed network architectures such as the HMLP and modified Elman networks is found to improve the learning rate of NN prediction. These positive findings inspire the implementation of the real time recursive learning of NN models for the proposed MPC controller. The proposed system identification and hovering control of the unmanned helicopter system are validated in a 6 degree of freedom (DOF) safety test rig. The experimental results confirm the effectiveness and the robustness of the proposed controller under disturbances and parameter changes of the dynamic system.

A novel system identification (SI)-based structural health assessment (SHA) procedure has been developed integrating several theoretical and implementation aspects. The procedure assesses health of structures using limited noise-contaminated dynamic responses and without using input excitation information. Since most practical structures are three dimensional (3D), the procedure has been developed for general 3D structures, represented by finite elements (FEs). The procedure identifies defects by tracking the changes in the stiffness of the elements in the FE representation. Once a defective element is identified, defect spot can be identified accurately within the defective element. The procedure is denoted as 3D Generalized Iterative Least-Squares Extended Kalman Filter with Unknown Input (3D GILS-EKF-UI) and implemented in two stages. In Stage 1, based on the available responses, substructure(s) are selected and the 3D GILS-UI procedure is used to generate the unknown input excitation, stiffness parameters of the elements in the substructure, and two Rayleigh damping coefficients. Using information from Stage 1, stiffness parameters for the whole structure are identified using EKF with Weighted Global Iteration (EKF-WGI) in Stage 2. The procedure accurately identified defect-free and defective states of various 3D structures using only analytically generated limited responses. To increase the robustness, 3D GILS-EKF-UI has been extended to develop an integrated structural health assessment strategy, denoted as Iterative Least-Squares Extended Kalman Filter with Unknown Input and Advanced Digital Integration Technique (ILS-EKF-UI-ADIT). The procedure has been implemented in three stages. In Stage 1, an advanced digital integration technique (ADIT) is implemented for post-processing of noise-contaminated acceleration time-histories, addressing all major challenges of digital integration. It also overcomes non-convergence issue in Stage 2 that arises due to phase-shift and amplitude errors. In Stage 2, substructure(s) are identified using the least-squares procedure. In Stage 3, stiffness parameters for the whole structure are identified using the EKF-WGI procedure. ILS-EKF-UI-ADIT has been verified in presence of relatively large noise in the acceleration time-histories, measured at small part(s) of defect-free and defective structures, without using excitation information. The SHA procedure is robust and has the potential to be applied for the health assessment, maintenance, retrofitting, and life extension of existing structural systems.

Parameterised analytical models that describe the trimmed inflight behaviour of classical aircraft have been studied and are widely accepted by the flight dynamics community. Therefore, the primary role of aircraft parameter estimation is to quantify the parameter values which make up the models and define the physical relationship of the air vehicle with respect to its local environment. Nevertheless, a priori empirical predictions dependent on aircraft design parameters also exist, and these provide a useful means of generating preliminary values predicting the aircraft behaviour at the design stage. However, at present the only feasible means that exist to actually prove and validate these parameter values remains to extract them through physical experimentation either in a wind-tunnel or from a flight test. With the advancement of UAVs, and in particular smaller UAVs (less than 1m span) the ability to fly the full scale vehicle and generate flight test data presents an exciting opportunity. Furthermore, UAV testing lends itself well to the ability to perform rapid prototyping with the use of COTS equipment. Real-time system identification was first used to monitor highly unstable aircraft behaviour in non-linear flight regimes, while expanding the operational flight envelope. Recent development has focused on creating self-healing control systems, such as adaptive re-configurable control laws to provide robustness against airframe damage, control surface failures or inflight icing. In the case of UAVs real-time identification, would facilitate rapid prototyping especially in low-cost projects with their constrained development time. In a small UAV scenario, flight trials could potentialy be focused towards dynamic model validation, with the prior verification step done using the simulation environment. Furthermore, the ability to check the estimated derivatives while the aircraft is flying would enable detection of poor data readings due to deficient excitation manoeuvres or atmospheric turbulence. Subsequently, appropriate action could then be taken while all the equipment and personnel are in place. This thesis describes the development of algorithms in order to perform online system identification for UAVs which require minimal analyst intervention. Issues pertinent to UAV applications were: the type of excitation manoeuvers needed and the necessary instrumentation required to record air-data. Throughout the research, algorithm development was undertaken using an in-house Simulink© model of the Aerosonde UAV which provided a rapid and flexible means of generating simulated data for analysis. In addition, the algorithms were further tested with real flight test data that was acquired from the Cranfield University Jestream-31 aircraft G-NFLA during its routine operation as a flying classroom. Two estimation methods were principally considered, the maximum likelihood and least squares estimators, with the aforementioned found to be best suited to the proposed requirements. In time-domain analysis reconstruction of the velocity state derivatives ˙W and ˙V needed for the SPPO and DR modes respectively, provided more statistically reliable parameter estimates without the need of a α- or β- vane. By formulating the least squares method in the frequency domain, data issues regarding the removal of bias and trim offsets could be more easily addressed while obtaining timely and reliable parameter estimates. Finally, the importance of using an appropriate input to excite the UAV dynamics allowing the vehicle to show its characteristics must be stressed.

Approved for public release; distribution is unlimited / The development of a dynamic computational model capable of predicting, with the requisite design certainty, the transient thermal response of jet vane thrust control systems has been undertaken. The modeling and simulation procedures utilized are based on the concept that the thermal processes associated with jet vane operation can be put into a transfer function form commonly found in the discipline of automatic controls. Well established system identification methods are employed to formulate and verify the relationships between the various gains and frequencies of the transfer function model and experimental data provided by Naval Weapons Center, China Lake. / http://archive.org/details/modelingofjetvan00hatz / Lieutenant, United States Navy

The plasma vertical position system on the COMPASS–D tokamak is studied in this thesis. An analogue P+D controller is used to regulate the plasma vertical position which is open loop unstable. Measurements from inside the vessel are used for the derivative component of the control signal and external measurements for the proportional component. Two main sources of disturbances are observed on COMPASS–D. One source is 600Hz noise from thyristor power supplies which cause large oscillations at the control amplifier output. Another source is impulse–like disturbances due to ELMs (Edge Localized Modes) and this can occasionally lead to loss of control when the control amplifier saturates. Models of the plasma open loop dynamics were obtained using the process of system identification. Experimental data is used to fit the coefficients of a mathematical model. The frequency response of the model is strongly dependent on the shape of the plasma. The effect of shielding by the vessel wall on external measurements when compared with internal measurements is also observed. The models were used to predict values of gain margins and phase crossover frequencies which were found to be in good agreement with measured values. The harsh reactor conditions on the proposed ITER tokamak preclude the use of internal measurements. On COMPASS–D the stability margins of the loop decrease when using only external flux loops. High order controllers were designed to stabilize the system using only external measurements and to reduce the effect of 600Hz noise on the control amplifier voltage. The controllers were tested on COMPASS–D and demonstrated the improved performance of high order controllers over the simple P+D controller. ELMs cause impulse–like disturbances on the plasma position. The optimal controller minimizing the peak of the impulse response can be calculated analytically for COMPASS–D. A multiobjective controller which combines a small peak impulse response with robust stability and noise attenuation can be obtained using a numerical search.

Heat and electricity production along with waste management are two modern day challenges for society. One of the possible solution to both of them is the incineration of household waste to produce heat and electricity. Incineration is a waste-to-energy treatment process, which can reduce the need for landfills and save the use of more valuable fuels, thereby conserving natural resources. This report/paper investigates the performance and emissions of a municipal solid waste (MSW) fueled industrial boiler by performing a system identification analysis using Principle Component Analysis (PCA) and Partial Least Squares Regression (PLS-R) modeling. The boiler is located in Västerås, Sweden and has a maximum capacity of 167MW. It produces heat and electricity for the city of Västerås and is operated by Mälarenergi AB. A dataset containing 148 different boilers variables, measured with a one hour interval over 2 years, was used for the system identification analysis. The dataset was visually inspected to remove obvious outliers before beginning the analysis using a multivariate data analysis software called The Unscrambler X (Version 10.3, CAMO Software, Norway). Correlations found using PCA was taken in account during the PLSR modelling where models were created for one response each. Some variables had an unexpected impact on the models while others were fully logical regarding combustion theory. Results found during the system analysis process are regarded as reliable. Any errors may be due to outlier data points and model inadequacies.

Unmanned Air Vehicles (UAVs) are becoming more widely used in both military and civilian applications. Some of the largest UAVs have power systems equivalent to that of a military strike jet making power management an important aspect of their design. As they have developed, the amount of power needed for loads has increased. This has placed increase strain on the on-board generators and a need for higher reliability. In normal operation these generators are sized to be able to power all on-board systems with out overheating. Under abnormal operating conditions these generators may start to overheat, causing the loss of the generator's power output. The research presented here aims to answer two main questions: 1) Is it possible to predict when an overheat fault will occur based on the expected power usage defined by mission profiles? 2) Can an overheat fault be prevented while still allowing power to be distributed to necessary loads to allow mission completion? This is achieved by a load management algorithm, which adjusts the load profile for a mission, by either displacing the load to spare generators, or resting the generator to cool it down. The result is that for non-catastrophic faults the faulty generator does not need to be fully shut down and missions can continue rather than having to be aborted. This thesis presents the development of the load management system including the algorithm, prediction method and the models used for prediction. Ultimately, the algorithms developed are tested on a generator test rig. The main contribution of this work is the design of a prognostic load management algorithm. Secondary contributions are the use of a lumped parameter thermal model within a condition monitoring application, and the creation of a system identification model to describe the thermal dynamics of a generator.

The need for mechanisms that assist human movements has been increasing due to the rising number of people that has some kind of movement disability. In this scenario, it is of great importance the development of control methods that assist the interface between a motor assistive device and its user. This work proposes a controller for an exoskeleton with one degree of freedom, using surface electromyography signals from the user as the input signal. An exoskeleton was adapted to serve as platform for the developed control method. To create an EMG-to-Angle model, a set of experiments were carried out with six subjects. The experiment consisted of a series of continuous and discrete elbow flexion and extension movements with different load levels. Using the experimental data, linear (ARIMAX) and non linear (Hammerstein-Wiener) system identification methods were evaluated to determine which is the best candidate for the estimation of the EMG-to-Angle model, based on its accuracy and ease of implementation. A new experiment wasconducted to develop a real-time controller, based on FIR model and tested in a real-timeapplication. Tests showed that the controller is capable of estimating the elbow joint angle with correlation above 70% and root-mean-square error below 25° when compared to the measured elbow joint angles. / A necessidade por mecanismos que auxiliam os movimentos do ser humano vem crescendo devido ao aumento do número de pessoas portadores de deficiências que afetam a capacidade motora. Nesse cenário, é de grande importância o desenvolvimento de métodos de controle que auxiliem a interface entre o dispositivo de assistência motora e o seu usuário. Esse trabalho propõe um controlador para um exoesqueleto com um grau de liberdade, usando sinais de eletromiografia de superfície do usuário como sinal de entrada. Um exoesqueleto foi adaptado para servir de plataforma para o método de controle desenvolvido. Para criar um modelo EMG-ângulo, um conjunto de experimentos foi conduzido com seis voluntários. O experimento consistiu em uma série de movimentos de flexo-extensão do cotovelo contínuos e discretos com diferentes níveis de carga. Utilizando os dados do experimento, métodos de identificação de sistemas linear (ARIMAX) e não linear (Hammerstein-Wiener) foram avaliados para determinar qual o melhor candidato para a estimação do modelo EMG-ângulo, baseado em sua acurácia e facilidade de implementação. Um novo experimento foi conduzido para desenvolver um controlador em tempo real, baseado no modelo FIR e testado em uma aplicação em tempo real. Testes indicaram que o controlador é capaz de estimar o ângulo da junta do cotovelo com valores de correlação acima de 70% e raiz do erro quadrático médio menor que 25º, quando comparados aos valores medidos de ângulo da junta do cotovelo.

Compared to conventional machine tools, industrial robots offer great advantages such as multitasking, larger workspace, and lower price. However, these advantages of robots are undermined by their high structural flexibility leading to excessive deflections, severe vibrations, and ultimately violating dimensional tolerances and poor surface finish. Modeling the dynamics of robots under machining (e.g. milling and drilling) forces is essential for reducing deflections and vibrations during the process. Although modeling the dynamics of traditional machining systems is a well-studied subject, the existing modeling approaches are not applicable to robotic manipulators because of the posture-dependent dynamics of industrial robots. Within this context, the presented thesis aims to predict the stability of vibrations during robotic machining operations through prediction of posture dependent dynamic behavior of robots. A rigid-body modeling approach is used to identify the dynamic parameters of the robotic manipulator based on least squares estimation method. Next, by adopting a rigid link flexible joint model and employing experimental modal analysis to identify the joint stiffness and damping parameters, posture dependent dynamic response prediction of the robot is achieved. Finally, the posture-dependent milling stability is presented as a function of the predicted tool center point transfer function, spindle speed, and axial depth of cut. A Staubli TX200 robot and a Kuka KR90 robot are used as experimental case studies. / Graduate

The development of Advanced Driving Assistance Systems (ADAS) produces comfort and safety through the application of several control theories. One of these systems is the Adaptive Cruise Control (ACC). In this work, a distribution of two control loops of such system is developed for an embedded application to a vehicle. The vehicle model was estimated using the system identification theory. An outer loop control manages the radar data to compute a suitable cruise speed, and an inner loop control aims for the vehicle to reach the cruise speed given a desired performance. For the inner loop, it is used two different approaches of model predictive control: a finite horizon prediction control, known as MPC, and an infinite horizon prediction control, known as IHMPC. Both controllers were embedded in a microcontroller able to communicate directly with the electronic unit of the vehicle. This work validates its controllers using simulations with varying systems and practical experiments with the aid of a dynamometer. Both predictive controllers had a satisfactory performance, providing safety to the passengers. / A inclusão de sistemas avançados para assistência de direção (ADAS) tem beneficiado o conforto e segurança através da aplicação de diversas teorias de controle. Um destes sistemas é o Sistema de Controle de Cruzeiro Adaptativo. Neste trabalho, é usado uma distribuição de duas malhas de controle para uma implementação embarcada em um carro de um Controle de Cruzeiro Adaptativo. O modelo do veículo foi estimado usando a teoria de identificação de sistemas. O controle da malha externa utiliza dados de um radar para calcular uma velocidade de cruzeiro apropriada, enquanto o controle da malha interna busca o acionamento do veículo para atingir a velocidade de cruzeiro com um desempenho desejado. Para a malha interna, é utilizado duas abordagens do controle preditivo baseado em modelo: um controle com horizonte de predição finito, e um controle com horizonte de predição infinito, conhecido como IHMPC. Ambos controladores foram embarcados em um microcontrolador capaz de comunicar diretamente com a unidade eletrônica do veículo. Este trabalho valida estes controladores através de simulações com sistemas variantes e experimentos práticos com o auxílio de um dinamômetro. Ambos controladores preditivos apresentaram desempenho satisfatório, fornecendo segurança para os passageiros.