La Tesis Doctoral estudia la observaci'on en tiempo real de la concentraci'on en el colector de escape de 'oxidos de nitr'ogeno (NOx) y del dosado en motores diesel sobrealimentados (¿ '1 ). Para ello se combinan dos fuentes de informaci'on diferentes: ¿ Sensores capaces de proporcionar una media de dichas variables, ¿ y modelos orientados a control que estiman estas variables a partir de otras medidas del motor. El trabajo parte de la evaluaci'on de la precisi'on de los sensores, realizada mediante la comparaci'on de su medida con la proporcionada por equipos anal'¿ticos de alta precisi'on, que son usados como est'andares de calibraci'on est'atica. Tambi'en se desarrollan en la Tesis m'etodos para la calibraci'on de la din'amica del sensor; dichos m'etodos permiten identi¿car un modelo de comportamiento del sensor y revelar su velocidad de respuesta. En general, estos sensores demuestran ser precisos pero relativamente lentos. Por otra parte, se proponen modelos r'apidos para la estimaci'on de NOx y ¿ '1 . Estos m'etodos, basados en relaciones f'¿sicas, tablas de par'ametros y una serie de correcciones, emplean las medidas proporcionadas por otros sensores con el ¿n de proporcionar una estimaci'on de las variables de inter'es. Los modelos permiten una estimaci'on muy r'apida, pero resultan afectados por efectos de deriva que comprometen su precisi'on. Con el ¿n de aprovechar las caracter'¿sticas din'amicas del modelo y mantener la precisi'on en estado estacionario del sensor, se proponen t'ecnicas de fusi'on de la informaci'on basadas en la aplicaci'on de ¿ltros de Kalman (KF). En primer lugar, se dise¿na un KF capaz de combinar ambas fuentes de informaci'on y corregir en tiempo real el sesgo entre las dos se¿nales. Posteriormente, se estudia la adaptaci'on en tiempo real de los par'ametros del modelo con el ¿n de corregir de forma autom'atica los problemas de deriva asociados al uso de modelos. Todos los m'etodos y procedimientos desarrollados a lo largo de la presente Tesis Doctoral se han aplicado de forma experimental a la estimaci'on de NOx y ¿ '1 . De forma adicional, la Tesis Doctoral desarrolla aspectos relativos a la transferencia de estos m'etodos a los motores de serie. / The dissertation covers the problem of the online estimation of diesel engine exhaust concentrations of NOx and '1. Two information sources are utilised: ¿ on-board sensors for measuring NOx and '1, and ¿ control oriented models (COM) in order to predict NOx and '1. The evaluation of the static accuracy of these sensors is made by comparing the outputs with a gas analyser, while the dynamics are identified on-board by perform- ing step-like transitions on NOx and '1 after modifying ECU actuation variables. Different methods for identifying the dynamic output of the sensors are developed in this work; these methods allow to identify the time response and delay of the sensors if a sufficient data set is available. In general, these sensors are accurate but present slow responses. Afterwards, control oriented models for estimating NOx and '1 are proposed. Regarding '1 prediction, the computation is based on the relative fuel-to-air ratio, where fuel comes from an ECU model and air mass flow is measured by a sensor. For the case of NOx, a set-point relative model based on look-up tables is fitted for representing nominal engine emissions with an exponential correction based on the intake oxygen variation. Different corrections factor for modeling other effects such as the thermal loading of the engine are also proposed. The model is able to predict NOx fast with a low error and a simple structure. Despite of using models or sensors, model drift and sensor dynamic deficiencies affect the final estimation. In order to solve these problems, data fusion strategies are proposed by combining the steady-state accuracy of the sensor and the fast estimation of the models by means of applying Kalman filters (KF). In a first approach, a drift correction model tracks the bias between the model and the sensor but keeping the fast response of the model. In a second approach, the updating of look-up tables by using observers is coped with different versions based on the extended Kalman filter (EKF). Particularly, a simplified KF allows to observe the parameters with a low computational effort. Finally, the methods and algorithms developed in this work are combined and applied to the estimation of NOx and '1. Additionally, the dissertation covers aspects relative to the implementation of the methods in series engines. / Blanco Rodríguez, D. (2013). Modelling and observation of exhaust gas concentrations for diesel engine control [Tesis doctoral]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/32666 / TESIS / Premios Extraordinarios de tesis doctorales

Dans l'apprentissage machine, la classification est le processus d’assigner une nouvelle observation à une certaine catégorie. Les classifieurs qui mettent en œuvre des algorithmes de classification ont été largement étudié au cours des dernières décennies. Les classifieurs traditionnels sont basés sur des algorithmes tels que le SVM et les réseaux de neurones, et sont généralement exécutés par des logiciels sur CPUs qui fait que le système souffre d’un manque de performance et d’une forte consommation d'énergie. Bien que les GPUs puissent être utilisés pour accélérer le calcul de certains classifieurs, leur grande consommation de puissance empêche la technologie d'être mise en œuvre sur des appareils portables tels que les systèmes embarqués. Pour rendre le système de classification plus léger, les classifieurs devraient être capable de fonctionner sur un système matériel plus compact au lieu d'un groupe de CPUs ou GPUs, et les classifieurs eux-mêmes devraient être optimisés pour ce matériel. Dans ce mémoire, nous explorons la mise en œuvre d'un classifieur novateur sur une plate-forme matérielle à base de FPGA. Le classifieur, conçu par Alain Tapp (Université de Montréal), est basé sur une grande quantité de tables de recherche qui forment des circuits arborescents qui effectuent les tâches de classification. Le FPGA semble être un élément fait sur mesure pour mettre en œuvre ce classifieur avec ses riches ressources de tables de recherche et l'architecture à parallélisme élevé. Notre travail montre que les FPGAs peuvent implémenter plusieurs classifieurs et faire les classification sur des images haute définition à une vitesse très élevée. / In the machine learning area, classification is a process of mapping a new observation to a certain category. Classifiers which implement classification algorithms have been studied widely over the past decades. Traditional classifiers are based on algorithms such as SVM and neural nets, and are usually run by software on CPUs which cause the system to suffer low performance and high power consumption. Although GPUs can be used to accelerate the computation of some classifiers, its high power consumption prevents the technology from being implemented on portable devices such as embedded systems or wearable hardware. To make a lightweight classification system, classifiers should be able to run on a more compact hardware system instead of a group of CPUs/GPUs, and classifiers themselves should be optimized to fit that hardware. In this thesis, we explore the implementation of a novel classifier on a FPGA-based hardware platform. The classifier, devised by Alain Tapp (Université de Montréal), is based on a large amount of look-up tables that form tree-structured circuits to do classification tasks. The FPGA appears to be a tailor-made component to implement this classifier with its rich resources of look-up tables and the highly parallel architecture. Our work shows that a single FPGA can implement multiple classifiers to do classification on high definition images at a very high speed.

A conceptual design of an innovative two-axle lightweight railway vehicle for commuter services is carried out at KTH Railway Group. An active wheelset steering is introduced to improve the curving performance of the vehicle, which is one of the critical performance requirements. This thesis aims to improve the steering performance of the active wheelset steering. Look-up tables for estimating time-varying wheel-rail contact parameters are introduced to supervise a simple PID controller of the active steering system in order to improve steering performance. The look-up table (LUT) estimation is focused on time-varying wheel-rail contact parameters, including creep coefficients and contact patch variables due to their direct influence on curving performance and lateral stability of the wheelset. As a result, the estimated longitudinal unit creep forces (UCF) have the potential to supervise the gains determination of PID controller because it can appropriately distinguish running conditions. The estimation of longitudinal UCF is achieved by the combination of the results from the LUT of creep coefficients and the LUT of contact patch variables. The result from longitudinal unit creep force estimation is shifted to the first quadrant to use as critical gain in the Ziegler-Nichols tuning method for the PID controller. The critical oscillation period for PID tuning can be expressed as a function of vehicle speed. Consequently, the PID controller for the active steering system uses time-varying gains with real-time tuning. The proposed control system for active wheelset steering is validated with nine running conditions using SIMPACK and MATLAB/Simulink co-simulation. The proposed control system provides a stable wheelset lateral displacement control regardless of the running condition. The active steering system significantly reduces wheel-rail wear, which demonstrates the effectiveness of the proposed active steering system. / KTH:s Järnvägsgruppen utvecklar en konceptuell design av ett innovativt, två-axligt, lättvikts järnvägsfordon för tunnelbana eller pendeltåg. En aktiv hjuparsstyrning introduceras för att förbättra kurvtagningsförmågan hos fordonet, vilket är ett av de kritiska prestandakraven hos dessa fordon. Det här examensarbetet har som målsättning att förbättra styrningsprestandan av den aktiva hjulsatsstyrningen. För att uppskatta tidsvarierande hjul-rälskontaktparametrar introduceras pre-definierade tabeller (LUT) som en övervakning av en enkel PID-kontroll för det aktiva styrningssystemet, för att förbättra styrprestandan. Uppskattningen som baseras på tabellen fokuserar på tidsberoende hjul-rälsparametrar, inklusive krypkoefficienter och kontaktytans storlek och form. Dessa variabler är i fokus på grund av deras direkta effekt på kurvtagningsförmågan och den laterala stabiliteten hos hjulparet. Den uppskattade longitudinala enhets krypkraften (UCF) har potential att bestämma förstärkningen hos PID-kontrollen på grund av att den, på ett lämpligt sätt, kan skilja mellan olika körtillstånd. Uppskattningen av longitudinell UCF uppnås genom en kombination av resultat för krypkoefficienter och kontaktytavariabler i LUT. Resultaten från den longitudinella UCF-uppskattningen skiftas till den första kvadranten för att användas som kritisk förstärkning i Ziegler-Nichols justeringsmetod för PID-kontroller. Den kritiska oscillationsperioden för PID-justering kan utryckas som en funktion av fordonets hastighet. Utgående från detta använder PID-kontrollen tidsvarierande förstärkning med realtidsjustering för den aktiva styrningen. Det föreslagna kontrollsystemet valideras mot nio körtillstånd med hjälp av SIMPACK och MATLAB/Simulink-simuleringar. Det föreslagna kontrollsystemet tillhandahåller en stabil lateral förflyttning av hjulparet oberoende av körtillstånd. Det aktiva styrsystemet reducerar hjul-räls slitaget signifikant, vilket demonstrerar effektiviteten hos det framtagna aktiva styrsystemet.

No description available.