Littérature scientifique sur le sujet « Q-system »

Thesis (Ph. D.)--Computing, Georgia Institute of Technology, 2004.<br>Pande, Santosh, Committee Member ; Pu, Calton, Committee Member ; Schwan, Karsten, Committee Chair ; Franke, Hubertus, Committee Member ; Eisenhauer, Greg, Committee Member. Vita. Includes bibliographical references (leaves 158-169).

Training autonomous agents that are capable of performing their assigned job without fail is the ultimate goal of deep reinforcement learning. This thesis introduces a dueling Quantile Regression Deep Q-network, where the network learns the state value quantile function and advantage quantile function separately. With this network architecture the agent is able to learn to control simulated robots in the Gazebo simulator. Carefully crafted reward functions and state spaces must be designed for the agent to learn in complex non-stationary environments. When trained for only 100,000 timesteps, the agent is able reach asymptotic performance in environments with moving and stationary obstacles using only the data from the inertial measurement unit, LIDAR, and positional information. Through the use of transfer learning, the agents are also capable of formation control and flocking patterns. The performance of agents with frozen networks is improved through advice giving in Deep Q-networks by use of normalized Q-values and majority voting.

Detta examensarbete genomfördes i samarbete med Trafikverket och analyserar möjliga orsaker till avvikelser mellan karaktärisering utförd på kärnborrhål vid förundersökningar, uppförande av ingenjörsgeologisk prognos och karaktärisering och klassificering utförd i tunnel vid byggnadsskedet för Citybanan. Två av projektets delentreprenader utvärderades – Norrströms- och Norrmalmtunneln. Totalt analyserades 4596 löpmeter tunnel (Norrströmstunneln) och 2557 meter (Norrmalmstunneln) i syfte att identifiera varför och vart i processen som avvikelser uppstod. Den utförda studien tyder på att inom de två undersökta delentreprenaderna uppstod en skillnad i resultat mellan karaktärisering utförd i kärnborrhål och karaktärisering och klassificering utförd i tunnel. Övergripande visar resultat från karaktäriseringen i samband med tunnelkartering generellt sämre bergförhållanden jämfört med det som prognostiserats. Skillnaderna  otsvarade sämre förhållande uttryckt i 5 (för Norrströmstunneln) respektive 7,5 enheter (Norrmalm) i RMR-systemet från borrhål till tunnelkaraktärisering. Analysen av det utvärderade resultatet indikerar att avvikelserna mellan karaktärisering på borrkärnor jämfört med karaktärisering och klassificering utförd i tunnel antagligen till viss del beror på den använda metodiken vid karaktärisering och klassificering av bergkvaliten med RMR- och Q-systemet. Detta på grund av att parametrarna i systemen inte bestäms på exakt samma sätt vid karaktärisering i förundersökningsskede och karaktärisering och klassificering i byggnadsskede. Dessutom indikeras att anvisningarna för karteringsrutiner vid karaktärisering och klassificering hanteras olika av karterande geologer vid tunneldrivning. Baserat på resultaten från den utförda analysen diskuteras även om det för vissa parametrar för karaktärisering och klassificering finns ett behov av uppdatering i den använda metodiken. Detta för att få förundersökningsprocessen att samspela med det arbete som sker i en tunnel. Resultaten tyder också på att geologerna vid tunnelkartering har en tendens till att kartera sämre bergförhållanden jämfört med det som är prognostiserat samt att vissa egenskaper hos berget kan vara lättare att beskriva på plats i tunneln än vid karaktärisering av berget från en borrkärna. Genom att harmonisera processen för beskrivning av ingående parametrar i de system som används vid karaktärisering mellan förundersökning och karaktärisering och klassificering i byggskede kan skillnader mellan stegen reduceras och risken för prognosavvikelser kan minska för tunnelprojekt i framtiden.<br>This thesis was carried out in cooperation with the Swedish Transport Administration and analyzes the possible causes of differing rock mass quality assessed in boreholes, engineering geological forecast and tunnel mapping. Two contracts from the City Line project where investigated - in total has 4596 meters (the Norrström tunnel) and 2557 meters (the Norrmalm tunnel) of tunnel mapped during construction been analyzed. The purpose was to identify the reason and where differences in the process of rock evaluation arose. The performed study indicates that in the two investigated tunnels it arose a difference in the results between the characterization performed in boreholes and the subsequent geological forecast compared with the characterization in the tunnel. The assessed rock mass quality became generally successively poorer as the process progresses. This was particulary the case when comparing the geological forecast with tunnel mapping during construction The differences corresponded to a difference of a downgrade of 5 (for the Norrström tunnel) and 7,5 units (the Norrmalm tunnel) in the RMR-system from bore holes to tunnel mapping. The analysis of the evaluated results indicates that the differences between the characterization of boreholes and geological forecasts compared to the characterization and classification performed in the tunnel is probably due to the methodology in the characterization and classification of rock quality with the RMR- and Q system. This is because the parameters of the  ystems is not determined exactly in the characterization of the preliminary investigation stage and the characterization and classification in the construction stage. Also the instructions for mapping routines for the characterization and classification are handled differently by the mapping made by geologists during tunneling. Based on the results of the analysis it is recommended that the way in which some parameters in the systems of characterization and classification of rock are determined should be updated. This is in order to standardize the process for characterizing the rock mass when mapping boreholes, creating geological forecast and conducting tunnel mapping. It is also indicated  that geologists tend to conservatively scale down the rock quality in the tunnel and that certain features of the rock is easier to determine in the tunnel than during characterization of bore holes. By harmonizing the description of parameters of the systems used in the characterization of bore holes and characterization and classification in the construction stage, differences between the steps can be reduced for projects in the future.

This dissertation presents work conducted at the Center for Power Electronics Systems (CPES) at Virginia Polytechnic Institute and State University. Chapter 1 introduces the concept of impedance measurement, and discusses previous work on this topic. This chapter also addresses issues associated with impedance measurement. Chapter 2 introduces the analyzer architecture and the proposed algorithm. The algorithm involves locking on to the voltage vector at the point of common coupling between the analyzer and the system via a PLL to establish a D-Q frame. A series of sweeps are performed, injecting at least two independent angles in the D-Q plane, acquiring D- and Q-axis voltages and currents for each axis of injection at the point of interest. Chapter 3 discusses the analyzer hardware and the criteria for selection. The hardware built ranges from large-scale power level hardware to communication hardware implementing a universal serial bus. An eight-layer PCB was constructed implementing analog signal conditioning and conversion to and from digital signals with high resolution. The PCB interfaces with the existing Universal Controller hardware. Chapter 4 discusses the analyzer software. Software was written in C++, VHDL, and Matlab to implement the measurement process. This chapter also provides a description of the software architecture and individual components. Chapter 5 discusses the application of the analyzer to various examples. A dynamic model of the analyzer is constructed, considering all components of the measurement system. Congruence with predicted results is demonstrated for three-phase balanced linear impedance networks, which can be directly derived based on stationary impedance measurements. Other impedances measured include a voltage source inverter, Vienna rectifier, six-pulse rectifier and an autotransformer-rectifier unit.<br>Ph. D.

About 400 million years ago sharks developed a separate co-processor in their brains that not only made them faster but also more precisely coordinated. This co-processor, which is nowadays called cerebellum, allowed sharks to outperform their peers and survive as one of the fittest. For the last 40 years or so, researchers have been attempting to provide robots and other machines with this type of capability. This thesis discusses currently used methods to create artificial cerebellums and points out two main shortcomings: 1) framework usability issues and 2) building blocks incompatibility issues. This research argues that the framework usability issues hinder the production of good quality artificial cerebellums for a large number of applications. Furthermore, this study argues that the building blocks incompatibility issues make artificial cerebellums less efficient that they could be, given our current technology. To tackle the framework usability issues, this thesis research proposes the use of a new framework, which formalizes the task of creating artificial cerebellums and offers a list of simple steps to accomplish this task. Furthermore, to tackle the building blocks incompatibility issues, this research proposes thinking of artificial cerebellums as a set of cooperating q-learning agents, which utilize a new technique called Moving Prototypes to make better use of the available memory and computational resources. Furthermore, this work describes a set of general guidelines that can be applied to accelerate the training of this type of system. Simulation is used to show examples of the performance improvements resulting from the use of these guidelines. To illustrate the theory developed in this dissertation, this paper implements a cerebellum for a real life application, namely, a cerebellum capable of controlling a type of mining equipment called front-end loader. Finally, this thesis proposes the creation of a development tool based on this formalization. This research argues that such a development tool would allow engineers, scientists and technicians to quickly build customized cerebellums for a wide range of applications without the need of becoming experts on the area of Artificial Intelligence, Neuroscience or Machine Learning.

CONSELHO NACIONAL DE DESENVOLVIMENTO CIENTÍFICO E TECNOLÓGICO<br>O nível de serviço fill rate é uma alternativa viável de quantificação do custo de falta em um modelo de estoque para o Sistema de Controle (Q, R). Aproximações deste modelo são importantes para tornar mais simples a busca por sua solução. Uma aproximação é proposta e outras duas são analisadas. Experimentos foram feitos com as três aproximações para verificar o erro da solução encontrada em relação à solução resultante do modelo exato. Foram analisados os casos em que as aproximações são aceitáveis.<br>The fill rate service level is a viable alternative to quantify the penalty cost in a inventory model for control system (Q, R). Approximations of this model are important to make a simplest solution. An approximation is proposed and two other are analyzed. Experiments were made with the three approximations to check the error of the solution in relation to the resulting solution of the exact model. So, we defined cases where the approximations are acceptable.

The objective of this study was two-fold: 1) to determine the accuracy of the Elements Apex Locator and 2) to compare the accuracy of the alginate and Endo Q models as in-vitro apex locator testing devices. Twenty teeth were decoronated at the CEJ and triplicate measurements were made using the Elements Apex Locator. All measurements were made to the apex reading of the apex locator. True length was established by visualizing the file tip at the apex with a dental operating microscope. Measurements were then taken with the teeth mounted in an alginate model and then in the Endo Q model. The two models were compared using a repeated-measure ANOVA. Statistically significant differences occurred between the alginate and Endo Q models. The results showed that in 95% (n=19) of the cases, an accurate location to within + 0.5 mm of the apical foramen was obtained with the Endo Q model and 55% (n=11) with the alginate model. With a + 1.0 mm tolerance level, an accuracy of 95% (n=19) was found with the use of the alginate model. In conclusion, it seems that the Endo Q system was more suitable for testing the electronic apex locator than the alginate model. The Elements Apex Locator with the use of the Endo Q model was highly accurate in locating to within + 0.5 mm of the apical foramen (Mean deviation = .17 mm).