Kernel methods are able to exploit high-dimensional spaces for representational advantage, while only operating implicitly in such spaces, thus incurring none of the computational cost of doing so. They appear to have the potential to advance the state of the art in control and signal processing applications and are increasingly seeing adoption across these domains. Applications of kernel methods to fault detection and isolation (FDI) have been reported, but few in aerospace research, though they offer a promising way to perform or enhance fault detection. It is mostly in process monitoring, in the chemical processing industry for example, that these techniques have found broader application. This research work explores the use of kernel-based solutions in model-based fault diagnosis for aerospace systems. Specifically, it investigates the application of these techniques to the detection and isolation of IMU/INS sensor faults – a canonical open problem in the aerospace field. Kernel PCA, a kernelised non-linear extension of the well-known principal component analysis (PCA) algorithm, is implemented to tackle IMU fault monitoring. An isolation scheme is extrapolated based on the strong duality known to exist between probably the most widely practiced method of FDI in the aerospace domain – the parity space technique – and linear principal component analysis. The algorithm, termed partial kernel PCA, benefits from the isolation properties of the parity space method as well as the non-linear approximation ability of kernel PCA. Further, a number of unscented non-linear filters for FDI are implemented, equipped with data-driven transition models based on Gaussian processes - a non-parametric Bayesian kernel method. A distributed estimation architecture is proposed, which besides fault diagnosis can contemporaneously perform sensor fusion. It also allows for decoupling faulty sensors from the navigation solution.

Current OBD-II vehicle systems detect misfires by monitoring slight variances of crankshaft acceleration throughout power-strokes of each of the engine’s cylinders. If the PCM determines that the acceleration of the engine’s crankshaft is inappropriate, it concludes a misfire is detected. However, after this misfire is detected, the technician still needs to diagnose (isolate) the root-cause. Diagnosis is no easy task, especially with several potential subsystems that could be at fault: fuel injection, air-intake, sparkignition, and engine-mechanical. With this being said, it is difficult for many technicians to isolate the fault causing a misfire because of the wide range of root-cause possibilities within each of the subsystems. The proposed On-Board Diagnostics III contributes to the computer-aided detection and diagnosis of future-production vehicle faults. Several data-mining algorithms were investigated and applied to data parameters collected from misfire and misfire-free fault instances. Rules were then used to accurately classify future engine misfire fault instances.

An application of a recently introduced framework for isolating unprecedented faults for an automotive engine EGR valve system is presented. Using normal behavior data generated by a high fidelity engine simulation, the Growing Structure Multiple Model System (GSMMS) is used to construct models of normal behavior for EGR valve system and its various subsystems. Using the GSMMS models as a foundation, anomalous behavior of the entire system is then detected as statistically significant departures of the most recent modeling residuals from the modeling residuals during normal behavior. By reconnecting anomaly detectors to the constituent subsystems, the anomaly can be isolated without the need for prior training using faulty data. Furthermore, faults that were previously encountered (and modeled) are recognized using the same approach as the anomaly detectors. / text

This report introduces the concept of three-dimensional (3D) radial plots for the visualization of multivariate large scale datasets in plant operations. A key concept of this representation of data is the introduction of time as the third dimension in a two dimensional radial plot, which allows for the display of time series data in any number of process variables. This report shows the ability of 3D radial plots to conduct systemic fault detection and classification in chemical processes through the use of confidence ellipses, which capture the desired operating region of process variables during a defined period of steady-state operation. Principal component analysis (PCA) is incorporated into the method to reduce multivariate interactions and the dimensionality of the data. The method is applied to two case studies with systemic faults present (compressor surge and column flooding) as well as data obtained from the Tennessee Eastman simulator, which contained localized faults. Fault classification using the interior angles of the radial plots is also demonstrated in the paper. / text

<p>The main topic of this licentiate thesis is fast faultdetection. The thesis summaries the work performed in theproject“Fast fault detection for distributionsystems”.</p><p>In the first chapters of the thesis the term“fast”is used in a general manner. The term is laterdefined based upon considerations and conclusions made in thefirst chapters and then related to a specific time.</p><p>To be able to understand and appreciate why fast faultdetection is necessary, power system faults and theirconsequences are briefly discussed. The consequences of a faultare dependent of a number of different factors, one of thefactors being the duration of the fault.</p><p>The importance of the speed of the fault detection dependson the type of equipment used to clear the fault. A circuitbreaker which interrupt currents only when they pass through anatural zero crossing might be less dependent on the speed ofthe fault detection than a fault current limiter which limitsthe fault current before it has reached its first prospectivecurrent peak.</p><p>In order to be able to detect a fault in a power system, thepower system must be observed, i.e., measurements of relevantquantities must be performed so that the fault detectionequipment can obtain information of the state of the system.The fault detection equipment and some general methods of faultdetection are briefly described.</p><p>Some algorithms and their possible adaptation to fast faultdetection are described. A common principle of many algorithmsare that they assume that either a signal or the power systemobject can be described by a model. Sampled data values arethen fitted to the model so that an estimate of relevantparameters needed for fault detection is obtained. An algorithmwhich do not fit samples to a model but use instantaneouscurrent values for fault detection is also described andevaluated.</p><p>Since the exact state of a power system never is known dueto variations in power production and load, a model of thepower system or of the signal can never be perfect, i.e., theestimated parameter can never be truly correct. Furthermore,errors from the data acquisition system contribute to the totalerror of the estimated parameter.</p><p>Two case studies are used to study the performance of the(modified) algorithms. For those studies it has been shown thatthe algorithms can detect a fault within approximately 1msafter fault inception and that one of the algorithms candiscriminate between a fault and two types of common powersystem transients (capacitor and transformer energization).</p><p>The second case study introduced a system with two sourceswhich required a directional algorithm to discriminate betweenfaults inside or outside the protection zone.</p><p>It is concluded that under certain assumptions it ispossible to detect power system faults within approximately 1msand that it is possible to discriminate a power system faultfrom power system transient that regularly occurs within powersystems but which not are faults.</p>

No description available.

No description available.

No description available.

No description available.

No description available.