Finite mixture models have had a profound impact on the history of statistics, contributing to modelling heterogeneous populations, generalizing distributional assumptions, and lately, presenting a convenient framework for classification and clustering. A novel approach, via Gaussian mixture distribution, is introduced for modelling receiver operating characteristic curves. The absence of a closed-form for a functional form leads to employing the Monte Carlo method. This approach performs excellently compared to the existing methods when applied to real data. In practice, the data are often non-normal, atypical, or skewed. It is apparent that non-Gaussian distributions be introduced in order to better fit these data. Two non-Gaussian mixtures, i.e., t distribution and skew t distribution, are proposed and applied to real data. A novel mixture is presented to cluster spatial and temporal data. The proposed model defines each mixture component as a mixture of autoregressive polynomial with logistic links. The new model performs significantly better compared to the most well known model-based clustering techniques when applied to real data. / Thesis / Doctor of Philosophy (PhD)

No / In the automotive industry, modern cyber-physical systems feature cooperation and autonomy. Such systems share information to enable collaborative functions, allowing dynamic component integration and architecture reconfiguration. Given the safety-critical nature of the applications involved, an approach for addressing safety in the context of reconfiguration impacting functional and non-functional properties at runtime is needed. In this paper, we introduce a concept for runtime safety analysis and decision input for open adaptive systems. We combine static safety analysis and evidence collected during operation to analyse, reason and provide online recommendations to minimize deviation from a system’s safe states. We illustrate our concept via an abstract vehicle platooning system use case. / This conference paper is available to view at http://hdl.handle.net/10454/17415.

No

No / Reliability evaluation for ensuring the uninterrupted system operation is an integral part of dependable system development. Model-based safety analysis (MBSA) techniques such as Hierarchically Performed Hazard Origin and Propagation Studies (HiP-HOPS) have made the reliability analysis process less expensive in terms of effort and time required. HiP-HOPS uses an analytical modelling approach for Fault tree analysis to automate the reliability analysis process, where each system component is associated with its failure rate or failure probability. However, such non-state-space analysis models are not capable of modelling more complex failure behaviour of component like failure/repair dependencies, e.g., spares, shared repair, imperfect coverage, etc. State-space based paradigms like Markov chain can model complex failure behaviour, but their use can lead to state-space explosion, thus undermining the overall analysis capacity. Therefore, to maintain the benefits of MBSA while not compromising on modelling capability, in this paper, we propose a conceptual framework to incorporate complex basic events in HiP-HOPS. The idea is demonstrated via an illustrative example. / This conference paper is available to view at http://hdl.handle.net/10454/17423.

With ever-increasing numbers of engine actuators to calibrate within increasingly stringent emissions legislation, the engine mapping and calibration task of identifying optimal actuator settings is much more difficult. The aim of this research is to evaluate the feasibility and effectiveness of the Multidisciplinary Design Optimisation (MDO) frameworks to optimise the multi-attribute steady state engine calibration optimisation problems. Accordingly, this research is concentrated on two aspects of the steady state engine calibration optimisation: 1) development of a sequential Design of Experiment (DoE) strategy to enhance the steady state engine mapping process, and 2) application of different MDO architectures to optimally calibrate the complex engine applications. The validation of this research is based on two case studies, the mapping and calibration optimisation of a JLR AJ133 Jaguar GDI engine; and calibration optimisation of an EU6 Jaguar passenger car diesel engine. These case studies illustrated that: -The proposed sequential DoE strategy offers a coherent framework for the engine mapping process including Screening, Model Building, and Model Validation sequences. Applying the DoE strategy for the GDI engine case study, the number of required engine test points was reduced by 30 – 50 %. - The MDO optimisation frameworks offer an effective approach for the steady state engine calibration, delivering a considerable fuel economy benefits. For instance, the MDO/ATC calibration solution reduced the fuel consumption over NEDC drive cycle for the GDI engine case study (i.e. with single injection strategy) by 7.11%, and for the diesel engine case study by 2.5%, compared to the benchmark solutions. / UK Technology Strategy Board (TSB)

The advent of Software Defined Networking (SDN) and programmable data planes has revolutionized the networking domain, enabling the programming of networking functions down to the silicon level responsible for data packet switching. Unfortunately, while this programmability offers greater flexibility and control, it also increases the likelihood of introducing software bugs. To counter this risk, rigorous testing methodologies and strategies are essential to ensure the reliability, security, and stability of SDN deployments. A comprehensive approach should combine various techniques, including formal verification, fuzz, and performance testing. Model-Based Testing (MBT) is a technique that can significantly enhance the effectiveness of SDN testing. By leveraging formal models of the system under test, MBT automatically generates test cases that can help identify potential issues in network configuration, data plane programming, and network protocols. Utilizing MBT allows network administrators to systematically explore SDN components’ possible states and transitions, resulting in a higher level of coverage and confidence in the system’s overall stability and security. However, a lack of information on applying MBT in an SDN environment challenges its full implementation and utilization in this field. This master thesis aims to investigate and demonstrate the application of MBT to programmable data plane functions. This work uses VLAN tagging as the target data plane function, and AltWalker is employed as the MBT tool for generating and executing tests on an SDN switch. The results present an initial testing methodology that, when applied to the VLAN tagging function, can provide insights into the potential benefits and challenges of using MBT for SDN testing. This thesis lays the groundwork for further exploration and refinement of MBT methodologies in the context of SDN and programmable data plane functions.

This study aims to discover and create an overview of how model-based systems engineering (MBSE) is utilized in Scaled Agile Framework (SAFe) settings. The overall goal is to identify the challenges and opportunities, the best practices and to find different publication trends that exist for this combination. The combination is interesting to investigate given that MBSE is considered to traditionally be a waterfall-based way of working, whereas SAFe is an Agile way of working. The study was executed by conducting a systematic mapping study. The results we found that were linked to the publication trends where that the number of primary studies investigating the combination of MBSE and SAFe were relatively low, and that the amount of grey literature that were found were higher than peer-reviewed primary studies. The best practices of the combination of MBSE and SAFe were identified by analyzing different methods, tools, and processes. Identified challenges and opportunities of this combination were that it provides: enhanced collaboration and communication capabilities, centralized information about the system under development, reduction of complexity, and decreased development time.

Fault Detection and Diagnosis (FDD) constitutes an essential aspect of modern life, with far-reaching implications spanning various domains such as healthcare, maintenance of industrial machinery, and cybersecurity. A comprehensive approach to FDD entails addressing facets related to detection, invariance, isolation, identification, and supervision. In FDD, there are two main perspectives: model-based and data-driven approaches. This thesis centers on model-based methodologies, particularly within the context of control and industrial applications. It introduces novel estimation strategies aimed at enhancing computational efficiency, addressing fault discretization, and considering robustness in fault detection strategies. In cases where the system's behavior can vary over time, particularly in contexts like fault detection, presenting multiple scenarios is essential for accurately describing the system. This forms the underlying principle in Multiple Model Adaptive Estimation (MMAE) like well-established Interacting Multiple Model (IMM) strategy. In this research, an exploration of an efficient version of the IMM framework, named Updated IMM (UIMM), is conducted. UIMM is applied for the identification of irreversible faults, such as leakage and friction faults, within an Electro-Hydraulic Actuator (EHA). It reduces computational complexity and enhances fault detection and isolation, which is very important in real-time applications such as Fault-Tolerant Control Systems (FTCS). Employing robust estimation strategies such as the Smooth Variable Structure Filter (SVSF) in the filter bank of this algorithm will significantly enhance its performance, particularly in the presence of system uncertainties. To relax the irreversible assumption used in the UIMM algorithm and thereby expanding its application to a broader range of problems, the thesis introduces the Moving Window Interacting Multiple Model (MWIMM) algorithm. MWIMM enhances efficiency by focusing on a subset of possible models, making it particularly valuable for fault intensity and Remaining Useful Life (RUL) estimation. Additionally, this thesis delves into exploring chattering signals generated by the SVSF filter as potential indicators of system faults. Chattering, arising from model mismatch or faults, is analyzed for spectral content, enabling the identification of anomalies. The efficacy of this framework is verified through case studies, including the detection and measurement of leakage and friction faults in an Electro-Hydraulic Actuator (EHA). / Thesis / Candidate in Philosophy / In everyday life, from doctors diagnosing illnesses to mechanics inspecting cars, we encounter the need for fault detection and diagnosis (FDD). Advances in technology, like powerful computers and sensors, are making it possible to automate fault diagnosis processes and take corrective actions in real-time when something goes wrong. The first step in fault detection and diagnosis is to precisely identify system faults, ensuring they can be properly separated from normal variations caused by uncertainties, disruptions, and measurement errors. This thesis explores model-based approaches, which utilize prior knowledge about how a normal system behaves, to detect abnormalities or faults in the system. New algorithms are introduced to enhance the efficiency and flexibility of this process. Additionally, a new strategy is proposed for extracting information from a robust filter, when used for identifying faults in the system.

The purpose of this project was to perform model-based diagnosis on Li-ion batteries using real-world data and sensor fusion algorithms. The data used in this project was collected and distributed by NASA and mainly consists of voltage and current measurements collected on numerous batteries that were repeatedly charged and discharged from their beginning of life, and until surpassing their end of life. The health of the batteries was decided by estimating their state of health through the increase in internal resistance. To validate the results, the increase in resistance was later compared with the decrease in charge capacity. Firstly an attempt was made to parameterize equivalent circuit models by fitting the model parameters to the data with a forgetting-factor recursive least squares filter, and a batch-wise recursive least squares filter. The forgetting-factor recursive least squares filter proved unreliable and unable to consistently be able to parameterize the models. The batch-wise recursive least squares filter was able to consistently parameterize the models providing a root mean squared error of around 0.35V in simulated voltage response tests. In total 20 equivalent circuit models were parameterized for four batteries. The models were then used in conjunction with a standard Kalman filter and an unscented Kalman filter to fur-ther estimate the increase in internal resistance of the batteries throughout their lifetime. In addition, the state of charge of the batteries was tracked through Coulomb counting and later attempted to refine via the Kalman filters. The results were partly successful as both the standard and unscented Kalman filters were able to track the state of health of the batteries, albeit to a varying degree. However, the Kalman filters were unable to improve the state of charge estimation, thus highlighting a limitation in their application. While the Kalman filters showcased limitations in tracking the state of charge, their effectiveness intracking the state of health thus proved that model-based approaches and sensorfusion algorithms can be utilized with both the standard Kalman filter and the unscented Kalman filter for meaningful health tracking.

Maintenance and diagnostics play a vital role in the aviation sector. This is especially true for the engines, being one of the most vital components. Lack of maintenance, or poor knowledge of the current health status of the engines, may lead to unforeseen disruptions and possibly catastrophic effects. To keep track of the health status, and thereby supporting maintenance planning, model based diagnostics is a key factor.  In the work going into this thesis, various aspects of model based gas turbine diagnostics, focused on aero engines, are covered. First, the importance of knowing what health parameters may be derived from a set of measurements is addressed. The selected combination is herein denoted as a matching scheme. A framework is proposed where the most suitable matching scheme is selected for a numerically robust diagnostic system. If a sensor malfunction is detected, the system automatically adapts. The second subject is a system for detecting a burn-through of an afterburner inner liner. This kind of burn-through event has a very small impact on available on-board measurements, making it difficult to detect numerically. A method is proposed performing back-to-back testing after each engine start. The method has shown potential to detect major burn-through events under the preconditions, regarding data collection time and frequency. Increasing these will allow for more accurate estimations. The third subject covers the importance of knowing the airplane installation effects. These are generally the intake pressure recovery, bleed and shaft power extraction. Just like inaccurate measurements affect diagnostic results, so does erroneous installation effects. A method for estimating said effects in the presence of gradual degradation has been proposed by using neural networks. By retraining the networks throughout the degradation process, the estimation errors is reduced, ensuring relevant estimations even at severe degradations. Finally, an issue related to the general lack of on-board measurements for diagnostics is addressed. Due to lack of measurements, the diagnostic model tend to be underdetermined. A least square solver working without a priori information has been implemented and evaluated. Results from the solver is very much dependent on available instrumentation. In well instrumented components, such as the compressors, good diagnostic accuracy was achieved while the turbine health estimations suffer from smeared out results due to poor instrumentation.