Examensarbeten för masterexamen // Master Theses
Länka till denna samling:
Browse
Browsar Examensarbeten för masterexamen // Master Theses efter Program "Engineering mathematics and computational science (MPENM), MSc"
Visar 1 - 20 av 34
Sökresultat per sida
Sortera efter
- PostAdjoint-based air duct topology optimisation(2018) Trangsrud, Ida; Chalmers tekniska högskola / Institutionen för mekanik och maritima vetenskaper; Chalmers University of Technology / Department of Mechanics and Maritime SciencesThis thesis was performed at the Climate Comfort department at Volvo Cars. The department works with the overall climate in the car, considering parts such as the ventilation system and defrosters. The aim of the thesis was to investigate the possibility and furthermore if it is bene cial to apply the software HELYX-Adjoint for creation of ducts for the climate system. HELYX-Adjoint is an adjoint CFD solver that, given an initial geometry, automatically produces an optimal shape. For this purpose, the software was applied on a manifold duct with two outlets. The objective of the optimisation process was to obtain as low pressure loss as possible between the inlet and both outlets. Di erent approaches and applications of the software were investigated, such as the dependence on speci c parameters, choice of objective function (the function that is to be optimised) as well as the in uence of the mesh size. The improvement of the resulting optimised design compared to the original geometry were 29 % and 27 % for the short and the long duct respectively. Compared to an already existing duct, the improvement were 34 % and 25 % for the two ducts. Hence, the performance of the optimised geometry in terms of loss in total pressure was better compared to both the original geometry and the already existing duct. The overall conclusion is that the use of HELYX-Adjoint for creation of ducts for the climate system has large potential.
- PostBehavior Classification based on Sensor Data - Classifying time series using low-dimensional manifold representations(2015) Rosén, John; Chalmers tekniska högskola / Institutionen för tillämpad mekanik; Chalmers University of Technology / Department of Applied MechanicsThis master´s thesis focuses on developing and testing methods that can automatically classify a given time series as having a certain behavior, chosen from a set of pre-specified, known behaviors. The first part of the thesis focused on finding statistical values where the empirical cumulative distribution of these values could be used for classification. The inverse of the cumulative distributions where then sampled at equally distanced sampling points and the resulting vector of sample values were treated as points in a high-dimensional Euclidean space. These points were then dimensionally reduced using projections onto a 2-dimensional manifold, where the manifold was warped in the high-dimensional Euclidean space using the elastic map and Kohonen Self-Organizing Map methodologies. The outputs from the manifold projections were then clustered using a 𝑘-nearest-neighbor algorithm. Both methodologies gave fairly good classification result for the two behaviors under consideration (86.5% / 80.3%, class 𝐶1 / 𝐶2 for elastic map, 83.6% / 78.3%, class 𝐶1 / 𝐶2 for Kohonen SOM). It was also shown that there truly were convergence in distribution for the behaviors under consideration.
- PostBenchmark and validation of Open Source CFD codes, with focus on compressible and rotating capabilities, for integration on the SimScale platform.(2014) Winter, Magnus; Chalmers tekniska högskola / Institutionen för tillämpad mekanik; Chalmers University of Technology / Department of Applied MechanicsThe use of CFD is widely spread in industry today. However, smaller business have difficulty to make use of this tool, due to high costs for commercial licenses and hardware. As SimScale provides the combination of cloud computing with open source CFD software, the prerequisites to use CFD are lowered. To provide more diverse solvers for the platform, different CFD software are integrated onto the platform. As OpenFOAM was the sole CFD software on the platform at the start of this thesis, two other open source software were researched and compared to OpenFOAM. OpenFOAM was compared to Gerris and SU2 in three different validation cases. OpenFOAM and Gerris rotating body capabilities are investigated in an analytical solvable case of laminar Taylor-Couette ow. Two different approaches were used in OpenFOAM, where AMI (Adaptive Mesh Interface) and MRF (Multiple Reference Frame) were used to solve the problem. OpenFOAM and SU2's turbulent capabilities are compared using a turbulent at plate case from Wieghardt [1]. Compressible transonic ow over a RAE2822 airfoil was set up for OpenFOAM and SU2. The results were compared to data from NASA [2] It was found that the rotating body capabilities (MRF and AMI) of OpenFOAM were suitable for integration onto the SimScale platform. Gerris as a software was deemed to be easy to work with, but would require a new work ow on the platform to be integrated. SU2 was integrated onto the platform during this thesis work, but was deemed to still su er from some early development issues as its first release was done in January 2013.
- PostCFD evaluation of a turbine rear structure with an integrated heat exchanger(2023) Skoglund, August; Chalhoub, Elias; Chalmers tekniska högskola / Institutionen för mekanik och maritima vetenskaper; Chalmers University of Technology / Department of Mechanics and Maritime Sciences; Chernoray, Valery; Andersson, SonnyThe expanding aviation industry is currently experiencing a substantial transition in order to become more sustainable. A more environmentally friendly alternative to traditional jet fuel is liquid hydrogen, an option that has been investigated by GKN Aerospace among others. Employing liquid hydrogen imposes many engineer ing challenges, one of which is the design on the turbine rear structure part (TRS) of the engine. This section of an aeroplane engine is of special interest to GKN as it is one of their main products. As liquid hydrogen has to be kept at low temperatures in order to remain in its liquid state, a state most suited for use in aircraft due to the low energy density of gases, and as the air exiting an engine is typically very hot this opens the possibility, if not necessity of heat exchange between the fluids. A model of a heat exchanger that can be used for this purpose was constructed by GKN and tested at Chalmers university of technology where the hydrogen was replaced by heated water. A patent for the heat exchanger model had been filed but not obtained during the thesis work which meant that the exact geometry of it was omitted from the thesis. In this thesis the tests at Chalmers were used in order to perform CFD simulations of the turbine rear structure with an integrated heat exchanger. The results obtained from the CFD were then validated by comparing turbulence models and boundary condition specification methods after which the results were compared to the test data. The validity of the used CFD model was confirmed using tests and reasoning although the CFD model was shown to over predict the magnitude of separation zones. The measured outlet temperature of the water was shown to be accurately predicted by the CFD simulations. Additional simulations of a TRS without the heat exchanger were also carried out to highlight the difference between using and not using a heat exchanger. Finally the performance of the heat exchanger was assessed by calculating its effectiveness for the different simulations which gave indications for the optimal air flow through it.
- PostCFD for Underhood Modeling. Development of an Efficient Method(2014) Ljungskog, Emil; Nilsson, Ulf; Chalmers tekniska högskola / Institutionen för tillämpad mekanik; Chalmers University of Technology / Department of Applied MechanicsToday's development of cooling systems for passenger cars relies on one-dimensional system simulations, which are heavily dependent on accurate input data obtained from testing or from CFD. Current three-dimensional CFD methods are deemed too computational and workload expensive to be used, which is why the present work was aimed at investigating the possibilities to develop a CFD model of the complete underhood compartment, with a high degree of automation. In the present work, two methods for CFD analysis of underhood flow in passenger cars were developed and compared. Both methods used ANSA for geometry cleanup and STAR-CCM+ as CFD solver, while the volume meshing differed. Method A used a polyhedral mesh created in STAR-CCM+, while a hexahedral mesh was used in method B. Surface wrapping was extensively used in both methods to reduce the manual workload in the geometry cleanup step. In order to validate the methods, they were both applied to a validation case. Both methods underestimated the heat rejection in the radiator with approximately 13% compared to wind tunnel measurements, while the heat rejection in the Charge Air Cooler was underestimated by 4:5% and 6:0% in method A and B, respectively. However, there was a large uncertainty in the experimental results, why further research will be needed to determine the validity of the developed methods. It was concluded that method A showed a larger potential to be used in production cases due to its lower time consumption and superior convergence behavior, even though more work is needed before it can be fully incorporated into the work flow. The time reduction compared to similar methods currently in use at Volvo cars was estimated to roughly 30 %.
- PostComplex Ventilation of Multiple Electrical Enclosure Systems Development and comparision of 1D pipe-flow models(2022) Andersson, Carl; Chalmers tekniska högskola / Institutionen för mekanik och maritima vetenskaper; Davidson, Lars; Sand, UlfThe nature of fluid flows in pipes is highly relevant to planning and construct ing cooling systems. Although general, conventional computational fluid dynamics solutions such as ANSYS FLUENT, OpenFOAM or similar software exists, these solutions are computationally expensive and require hours, if not days, to give re sults. This study aims to explore and compare implicit solutions for complex pipe networks that can be generated in a much quicker fashion. In this context, implicit solutions refer to 1D-implementations which solve the entire pipe network as a set of simultaneous equations, or in other words, a matrix. In order to evaluate the accuracy of any such implicit solver, the results that the solver produces are validated against a simulation run in ANSYS FLUENT using, to the extent that it is possible, identical settings. A set of different cases are run through this validation procedure in order to observe how the implicit flow rate solver compares across a number of cases. The results show that while the implicit flow rate solver manages to mimic the fluid flow of the ANSYS FLUENT simula tions with only small errors that arise primarily from observed asymmetries in the ANSYS FLUENT simulation. The implicit flow rate solver does however produce significant errors in temperature prediction when cooling systems that have heat sources approximating 2 · 105 W/m3 applied at smaller regions. Additionally, the results show an extreme benefit in terms of run-time, on the order of 103 compared to ANSYS FLUENT. These results suggest that the implicit solver fails to capture a variety of phenomena, particularly related to cases applying larger additional heat sources. These phenom ena would need to be captured if it is to act as a predictor of the behavior of such cases. For cases without the previously mentioned large heat source however, the results suggest that the implicit solver can serve as a semi-accurate run-time efficient predictor. Additionally, the results suggest that the usage of minor loss coefficients or length-equivalents for estimating the pressure drop over a pipe bend, T-junction or similar structures is insufficient to capture combined effects of said structures in more complex pipe networks. Finally, observation of the errors in temperature pre diction suggest an approximately linear behavior. On this basis, the implicit solver or a variant of it could very well serve as an early predictor when iterating through system design in order to quickly find designs that will fail to fulfill certain criteria.
- PostComplex Ventilation of Multiple Electrical Enclosure Systems. Development and comparision of 1D pipe-flow models.(2022) Andersson, Carl; Chalmers tekniska högskola / Institutionen för mekanik och maritima vetenskaper; Davidson, Lars; Sand, UlfThe nature of fluid flows in pipes is highly relevant to planning and constructing cooling systems. Although general, conventional computational fluid dynamics solutions such as ANSYS FLUENT, OpenFOAM or similar software exists, these solutions are computationally expensive and require hours, if not days, to give results. This study aims to explore and compare implicit solutions for complex pipe networks that can be generated in a much quicker fashion. In this context, implicit solutions refer to 1D-implementations which solve the entire pipe network as a set of simultaneous equations, or in other words, a matrix. In order to evaluate the accuracy of any such implicit solver, the results that the solver produces are validated against a simulation run in ANSYS FLUENT using, to the extent that it is possible, identical settings. A set of different cases are run through this validation procedure in order to observe how the implicit flow rate solver compares across a number of cases. The results show that while the implicit flow rate solver manages to mimic the fluid flow of the ANSYS FLUENT simulations with only small errors that arise primarily from observed asymmetries in the ANSYS FLUENT simulation. The implicit flow rate solver does however produce significant errors in temperature prediction when cooling systems that have heat sources approximating 2 · 105 W/m3 applied at smaller regions. Additionally, the results show an extreme benefit in terms of run-time, on the order of 103 compared to ANSYS FLUENT. These results suggest that the implicit solver fails to capture a variety of phenomena, particularly related to cases applying larger additional heat sources. These phenomena would need to be captured if it is to act as a predictor of the behavior of such cases. For cases without the previously mentioned large heat source however, the results suggest that the implicit solver can serve as a semi-accurate run-time efficient predictor. Additionally, the results suggest that the usage of minor loss coefficients or length-equivalents for estimating the pressure drop over a pipe bend, T-junction or similar structures is insufficient to capture combined effects of said structures in more complex pipe networks. Finally, observation of the errors in temperature prediction suggest an approximately linear behavior. On this basis, the implicit solver or a variant of it could very well serve as an early predictor when iterating through system design in order to quickly find designs that will fail to fulfill certain criteria.
- PostDesigning a survey to examine an anonymisation method for driver videos(2011) Jansson, Marcus; Chalmers tekniska högskola / Institutionen för tillämpad mekanik; Chalmers University of Technology / Department of Applied MechanicsData collected during naturalistic driving studies often includes video of the driver’s face. Due to the Data Privacy Act (“Personuppgiftslagen” (PuL) in Swedish), it is desirable to find a good method that can make the driver’s face anonymous, while keeping the original driver‘s facial expressions. In the unEye project, an attempt to create such method was made. Using this method, the original driver’s facial expressions are coded into action units which are then used to construct a corresponding video in which the driver face is replaced by an animated face model. This thesis project was conducted to validate whether the anonymisation method built in the unEye project could really achieve its objectives. In particular, this thesis contributes in validating whether a human viewer can identify who the original driver was from the animated version of the video and examining how well the facial expressions are translated when creating the animated driver videos. For this, a computer-based survey was designed and performed. The survey has two different parts corresponding to the two different validation tasks. The results from the survey showed some unexpectedly high number of correct classifications in the identification part. However, some explanations were found which suggest that these high numbers were most likely due to other reasons than the ability of the test participants to identify the driver. For the expressions part, the result showed that surprise was the best preserved expression and that anger was the most poorly preserved one. The average recognition proportion of the expressions was 24% and some patterns were found in the results which indicate that the details in the face are not translated well enough, especially in the regions around the mouth, eyes and eyebrows. In conclusion, while the method is able to make the driver face anonymous, it should be improved so that the facial expressions are better translated. The results were also examined with respect to the age, gender, area of occupation and education level of the test participants and no clear difference could be found within these groups. This indicates that the result could be generalised to a larger population.
- PostDriver Interaction with Pedestrians at Intersections - Quantifying the influence of environmental factors on driver comfort boundaries(2016) Jaber, Leila; Thalya, Prateek; Chalmers tekniska högskola / Institutionen för tillämpad mekanik; Chalmers University of Technology / Department of Applied MechanicsActive safety systems are constantly being enhanced, featuring more and more environmental factors and advanced algorithms. The system needs to be able to evaluate many different driving situations, and take appropriate action, at a safety critical moment. In this study, focus has been on quantifying factors’ influence on driver comfort boundaries when interacting with pedestrians at intersections. Seven factors, on two levels, were included in the study; Car speed, Pedestrian speed, Pedestrian size, Crossing angle, Crossing presence, Crossing entry and Lane width. The experiment was designed with a fractional factorial layout with the seven factors combined into 32 tasks. Additional 4 tasks were created, 3 to compare the study against two studies previously conducted, and another to analyze the learning/ adaptation effect. Participants were divided into two groups; Group 1, frequent drivers and Group 2, occasional drivers. Statistical analysis based on Linear mixed-effects models was carried out with a significance level of 0.01 for both group of drivers separately, and by combining the two groups. Factors Car speed and Crossing entry have a significant influence on Time To Collision (TTC) at brake onset for all group of drivers. The factors decreased TTC when changed from low to high level. Factors Pedestrian size and Lane width also had a significant influence on TTC at brake onset for all group of drivers. The factors increased TTC when changed from low to high level. The study was conducted with a low-fidelity simulator available at SAFER. Limitations of the simulator included participants’ lack of input from auditory and sensory sources and narrow field of view. Also, the expectancy of the pedestrian might have changed the comfort boundaries as there was only one pedestrian crossing the road in every scenario. However, even though the simulator was developed with open source software, still nearly 90% of the participants stated they experienced the environment to be natural or sort of natural and more than 95% said they behaved, or sort of behaved, as normally as they would have in a real traffic situation.
- PostEvaluation of Novel Wingsail Concepts in Terms of Aerodynamic Efficiency(2023) Ljungberg, Erik; Chalmers tekniska högskola / Institutionen för mekanik och maritima vetenskaper; Chalmers University of Technology / Department of Mechanics and Maritime Sciences; Yao, Hua-Dong; Zhu, Heng; Ramne, Bengt; Lee, Henry; Marimon Giovannetti, LauraIn this project the evaluation and comparison of five different designs for rigid wingsails were made by performing simulations in the CFD software STAR-CCM+ and two parameters, gap distance between foils and camber angle for three of the wingsails were varied to understand its effects. The wingsails in question were the crescent-shaped profile, two-element foil sails and three element foil sails. Three different designs were created for the two element foil sails with the following ratios of the total chord length between the front and aft foils; 70/30, 80/20, 90/10, and only one design for the three element foil sail with ratio 15/70/15. The parameters of gap distance and camber angle were only altered for the two element foil wingsails, while only the camber was altered for the three element foil wingsail. The comparisons that were made between the different designs were between the propulsive performances as well as stall angles. This study has found that, regarding the two element foil designs, the absence of a gap between the front and aft foils yield superior propulsive performance and that the difference in performance between different non-zero gap sizes is not very significant. The crescent-shaped profile performs significantly better than the non-cambered multi-element foil sails, but appeared to be a less favourable design when compared with the no-gap, cambered multi-element foil sails. However 3D simulation results revealed that the performance of the best multi-element foil design, the two element foil sail with ratio 70/30, no gap and 25° camber had been severely overestimated, and performed worse through all apparent wind angles when compared to the crescent-shaped profile. If we were to assume that the relation between the multi-element foil designs remained the same, in terms of performance over all apparent wind angles, when performing 3D simulations, then they would be ranked as follows. The best designs in terms of overall propulsion, in order of greatest to worst, are the crescent-shaped profile followed by the two element foil sails with ratio 70/30 and 80/20 with no gap and 25° camber and lastly ratio 90/10 (also no gap and 25° camber) and the three element foil profile with ratio 15/70/15 and 15° camber. It was discovered that the crescent-shaped profile began to experience stall, often much later than the multi-element foil designs, and reached its critical angle of attack earlier as well.
- PostFibre modelling in venturi flow and disc refiner. Implementation and development of models for turbulent fibre suspension flow.(2015) Ingelsten, Simon; Chalmers tekniska högskola / Institutionen för tillämpad mekanik; Chalmers University of Technology / Department of Applied MechanicsAs a part of a larger project in developing a new technology for increased energy efficiency in the pulp and paper industry a model for turbulent fibre suspension flow was needed. The model was also desired to be suitable for applications alongside additional multiphase flow models to study cavitation in fibre suspension flows. Two models with different levels of complexity were studied. A relatively simple model was the Bingham model where a fibre suspension is described as a non-Newtonian fluid. The model acts through modification of the viscous stresses, as the suspension viscosity is computed as a function of shear rate. A more complex model, denoted the ODF model, was implemented by modifying the viscous stress tensor with additional stresses computed from the orientation distribution of suspended fibres. A model for the orientation distribution function (ODF) was proposed and developed within the thesis work and used to construct explicit expressions for the additional stress tensor components as functions of the flow field with the use of Fourier series. Both fibre models were implemented to Fluent through user-defined functions (UDF) written in C programming language. The models were validated and the best performing parameter setting was identified by performing simulations of turbulent fibre suspension flow and comparing the results to experimental data from literature. Both fibre models resembled experimental data fairly well and had a reasonable computational cost. In comparison the performances of the models were roughly similar. Although both models showed potential the Bingham model gave slightly better resemblance of the experimental data was at the same time fairly simple to implement. In addition the ODF model was judged still needing further and more rigorous study. The Bingham model was therefore identified as the better and more reliable model at the current stage. Further validation of the chosen model, i.e. the Bingham model, was made by using it in an example application. A simplified model of a disc refiner was used and the fibre suspension flow was simulated with the Bingham model. The simulation yielded results to be expected and that resembled results from literature.
- PostFSI-analysis on vibrations of a slender rod exposed to axial flow. Calculations for nuclear power applications.(2016) Bengtsson, Kajsa; Chalmers tekniska högskola / Institutionen för tillämpad mekanik; Chalmers University of Technology / Department of Applied MechanicsVibration on components caused by flow, so called flow induced vibrations (FIV), is an important area in many industrial fields. Fluid-structure interaction problems can be solved by coupling a structure solver to a fluid solver and in each time step iterate to a solution. Different solvers can be used, but the coupling codes still need to be tested and evaluated to be used for industrial purposes. Experiments have been performed to create data for FSI-software validation and to see how an axial flow along a slender structure can cause vibrations of the structure. In order to prove the reliability of the FSI-simulation software the purpose of this master’s thesis is to see if the rod vibrations, induced by the axial flow, can be predicted with coupled FSI-simulations in ANSYS. An FSI-analysis of the same geometry as in the experiments has been carried out and the simulation data were compared to the experiment data. Different meshes, different turbulence models and structural damping were also investigated on how they affected the solution. The LES turbulence model could induce vibrations, while the URANS turbulence model could not. The vibration frequencies match the eigenfrequencies for the tube. The amplitudes increase with increased mass flow. The amplitudes were far too high and the frequencies were a bit higher in the simulations compared to the experiment. The differences could be because of discrepancies between the ANSYS model and the experiment, since there were some uncertainties in the documentation of the experiment. The simulations seemed not to be sensitive to time step or damping, but a coarse mesh resulted in lower amplitudes compared to a finer one.
- PostGeometrical Optimization of Plate-Fin Heat-Sink(2016) Irawan, Dani; Chalmers tekniska högskola / Institutionen för tillämpad mekanik; Chalmers University of Technology / Department of Applied MechanicsIn this thesis the geometrical configuration of a processor heat sink is optimized. Two objective functions, operational cost and maximum temperature, have been chosen and the design space has been limited to three degrees of freedom represented by number of fins, fin height and fin thickness. The optimization is simulation-based and carried out using multi-objective particle swarm optimization (MO-PSO). Two variants of MO-PSO have been selected, adapted and implemented. The first is a variant introduced by Coello and Lechuga in "MOPSO: A Proposal for Multiple Objective Particle Swarm Optimization" which is based on density of the pareto front. The second is a variant introduced by Fieldsend and Singh in "A Multi-Objective Algorithm Based Upon Particle Swarm Optimisation, an Efficient Data Structure and Turbulence" which is based on distance to the pareto front. Simulations have been performed using a state-of-the-art immersed boundary flow solver called IBOFlow developed by Fraunhofer-Chalmers research centre. Comparisons presented in this thesis show that FS-method requires less evaluations to find the Pareto front, whereas the CL-method explores the Pareto front more evenly. The number of particles does not seem to give an apparent effect on exploration it is rather the distribution of initial points that determines the exploration. Many evaluated points have been observed to be clustered in objective space, especially in FS-method. This observation triggered the idea of a filter. In the current work it has been found that the number of simulations can be significantly reduced by using filters. However, the threshold used in the filter must be chosen conservatively to prevent coarsening of the Pareto front. It has also been found that the Pareto front can be refined by projection of infeasible points onto the boundary of the feasibility region. The improvement from projection comes at the expense of additional evaluations. Optimal designs of the heat sink are proposed based on lexicographic method, Analytical Hierarchy Process, and Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS), respectively.
- PostImpact of Feature Representation for Imitation Learning in Autonomous Drive(2019) Dahl, Malin; Ramle, Elvira; Chalmers tekniska högskola / Institutionen för mekanik och maritima vetenskaper; Chalmers University of Technology / Department of Mechanics and Maritime SciencesAutonomous drive in complex traffic scenarios is a demanding task to solve. With high-dimensional input data available, problems related to redundancy and irrelevance are often implicated, hence determining what features bring the most useful information is of vital importance. The purpose of this thesis is to investigate how different dimensionality reduction methods affect the driving performance and how to determine what features are most relevant. Specifically, these questions were studied in a simulated environment where a car is manoeuvred using deep neural networks through a sequence of signalised intersections. Four different dimensionality reduction methods have been studied: choice of features based on reason, Principal Component Analysis, Auto-Encoders and Integrated Encoders. The results showed that the models which used a feature representation based on reason were shown to perform best. Also, the weight distributions of a model using all available features indicated that influential features may be partially identified by studying the spread of the weights. Therefore, an approach is proposed where the choice of features should be based on reason as well as a study of the features’ respective set of weights. In conclusion, establishing the most relevant feature representation is important since it may benefit the training of the models.
- PostImplementation of a Vision System for an Autonomous Railway Maintenance Vehicle: Track and Object Detection with YOLO, Neural Networks and Region Growing(2021) Warnicke, Albin; Jönsson, Jesper; Chalmers tekniska högskola / Institutionen för mekanik och maritima vetenskaper; Wolff, Krister; Wolff, KristerRailway infrastructure is often expensive to maintain. To improve efficiency and lower these costs, the use of autonomous railway vehicles for such maintenance has begun to be explored. A railway vehicle requires several components to achieve complete automation, including systems for navigation, decision-making, and sensors such as cameras. This project aims to develop the vision system used by an autonomous track trolley under development at Chalmers University of Technology. The proposed vision system can detect railway tracks and switches by a region-growing algorithm based on the image intensity gradient. Object detection is achieved by the use of a YOLOv4-tiny neural network and is developed to detect persons, vehicles, railway signs and signals, road crossings and catenary support poles. The signal and speed sign messages are further classified by additional convolutional neural networks. The vision system is implemented as a ROS node on a single-board computer, a NVIDIA Jetson Nano, and is running in real-time at up to 15 FPS. The vision system is accurate and robust enough to be used as a prototype in simple environments. The track that the vehicle is traveling on was detected in 98.4 % of the evaluated video frames, with the sidetracks correctly identified in 70-80 % of the time. Several of the considered objects were detected with 90-100 % accuracy, for example vehicles and road crossings. Other objects, particularly railway switches and incoming tracks, were however only correctly recognized in about 60 % of their occurrences. Signals and speed signs were detected with high accuracy. Some features can be improved or added before the vision system can be applied to a complete autonomous railway vehicle. The main limitation of the implemented object detection is the lack of large training datasets. With more available video data, datasets with an increased number of labeled objects and greater diversity could be created. Utilizing the full capabilities of larger datasets would eventually require the use of more complex neural networks. The currently used hardware however limits the possible methods to simpler algorithms. The track detection algorithm can serve as a base for further improvement, with the region growing based on the image intensity gradients not being robust enough to handle large variations in lighting and environment conditions. An approach with semantic segmentation neural networks is instead suggested to achieve robust track detection.
- PostImplementation of interpretable methods for paraphrasing and text disambiguation(2023) Carlström, Klara; Chalmers tekniska högskola / Institutionen för mekanik och maritima vetenskaper; Chalmers University of Technology / Department of Mechanics and Maritime Sciences; Wahde , Mattias; Wahde, MattiasIn this project, starting from an interpretable language model based on knowledge graphs, four essential methods for natural language processing (NLP) have been developed, namely (i) paraphrasing, (ii) part-of-speech tagging, (iii) semantic similarity analysis, and (iv) text simplification. The methods yield good results on a small dataset and thus offer promising prospects for continuing research on interpretable NLP. Applications of NLP are becoming increasingly embedded in our daily lives in applications such as voice assistants, automatic language translation, opinion mining and medical diagnostics. One of the reasons behind the exponentially growing interest in NLP is the development of deep neural network (DNN) models that have achieved outstanding performance on various NLP tasks. However, the domination of DNN models has been followed by deep concerns regarding the black-box nature of such systems. By contrast, the language model used here is fully interpretable, paving the way for safe and accountable NLP.
- PostImproving Accuracy and Efficiency of Aerodynamic Simulations for Heavy Vehicles(2020) Johansson, Simon; Chalmers tekniska högskola / Institutionen för mekanik och maritima vetenskaper; Davidson, Lars; Tenstam, Anders; Lundberg, AntonHeavy-duty vehicles are a necessity in the current society but they answer to 6% of the total CO2 emissions in Europe. Therefore new regulations are set by the European Union to limit those emissions. These limits make the losses of the vehicles an important factor and especially aerodynamic drag. Reducing the aerodynamic drag calls for good developing methods and accurate predictions. This means that validations and limitations of the methods are needed. This thesis will investigate a CFD method by creating a base case simulation and validate this against wind tunnel test. The method will use unsteady simulations to capture transient behaviour in the flow. A statistical confidence intervals for the averaged values as well as a starting point for the averaging will be presented. Further, the spatial mesh will be studied and the effects of the number of inner iterations will be presented. The validation of the base case will be stated as differences in the coefficients Cd , Cs , and Cp to understand advantages and disadvantages of the method. The frequency content from the time-resolved coefficients will also be studied. When a well-defined base setup has been validated the temporal resolution will be investigated to see how it affects the results. This will be done for increased time steps as well as a significantly reduced time step. The significantly reduced time step ensures the CFL number to be less than unity in the entire domain. The validation of the base case shows that there is an error in Cd of 45 ± 7 drag counts for the base case and that the error in pressure on the rear end of the truck is yaw angle dependent. The error in Cd is almost fully explained by the error in pressure on the rear end of the truck. The result from the simulations with different time steps show that for a fully resolved flow very fine time steps are needed to keep the CFL number less than unity in the majority of the domain, but if global force coefficients are of primary interest the time step can be increased moderately without significant changes. It is also found that the force coefficients alone are not a good estimators of how the method performs since errors can cancel each other. This means that a coarser mesh or a longer time step can generate a smaller error compared to wind tunnel tests but this is not reflecting the actual accuracy of the method.
- PostInterday news-based prediction of stock prices and trading volume(2015) Söyland, Christian; Chalmers tekniska högskola / Institutionen för tillämpad mekanik; Chalmers University of Technology / Department of Applied MechanicsThis thesis investigates the predictive power of online news on one-day stock price up or down changes and high or low trade volume of 19 major banks and nancial institutions within the MSCI World Index, during the period from January 1 2009 to April 16 2015. The news data correspond to news articles, press releases, and stock exchange information, and were obtained by a web-crawler, which scanned around 6000 online sources for news and saved them in a database. The news are partitioned and labeled into two classes according to which price change class, or trade volume class, it corresponds. A supervised automated document classi cation model is created and used for prediction. The model does not succeed in predicting the one-day stock price changes, but the percentage of correctly labeled documents in the one-day trade volume experiment was 78:3%, i.e. a classi cation accuracy of 78:3% was achieved, suggesting that online news does contain some valuable predictive information.
- PostInvestigation of under-reporting and the consistency of injury severity classifications in Swedish police crash data compared to hospital injury data based on the Swedish Traffic Accident Data Acquisition (STRADA)(2016) Held, Felix; Chalmers tekniska högskola / Institutionen för tillämpad mekanik; Chalmers University of Technology / Department of Applied MechanicsPolice reported road crash data is used for Sweden’s official road traffic statistics and is collected by the Swedish TRaffic Accident Data Acquisition (STRADA), which in turn provides a road crash database combining police and emergency hospital reported information. The general goal of this thesis was the investigation of police road crash data from STRADA for the purpose of gaining a deeper understanding of how certain parts of the police data might need to be weighted and how consistent police and hospital injury severity classifications are. Specifically it was investigated how different sub-populations of road crashes are represented in police data and to which extent they were under-reported. Furthermore, the consistency of police and hospital injury severity classifications was investigated. This was achieved by taking advantage of the availability of pre-linked hospital data in STRADA. Data from non-fatal crashes that occurred during the years 2003-2013 were used for the examination of under-reporting of road crashes while those that occurred during the years 2007-2013 were used for the investigation of consistency of injury severity classifications. A direct comparison of the available police and hospital data was used to establish a lower bound for the under-reporting. Capture-recapture methods were used to estimate the actual prevalent amount of under-reporting. Additionally, the longitudinal behaviour of under-reporting was investigated. The influence of different factors on the consistency of injury severity classifications was tested for statistical significance by chi-squared tests and the calculation of odds ratios. In total, under-reporting was found to be at least 48%. This lower bound decreased to 35% when single bicycle crashes were excluded. Two groups of months that consistently exhibited distinct rates of under-reporting were identified and crash severity was identified as a large influence on the amount of under-reporting. A dependence of under-reporting on the involved vehicles was found as well. For the investigation of the consistency of police and hospital reported classifications of injury severity it was found that male gender, age above 60, ambulance or helicopter transport and a rural traffic environment lead to higher odds for different classifications by the police and hospitals compared to their alternatives.
- PostJoint Optimization of Aerodynamics and Aeroacoustics of Side View Mirrors(2018) Stridh, Mattias; Chalmers tekniska högskola / Institutionen för mekanik och maritima vetenskaper; Chalmers University of Technology / Department of Mechanics and Maritime SciencesThis thesis is carried out at China Euro Vehicle Technology (CEVT) and aims at developing a method for performing multidisciplinary optimization for automotive applications. Such applications could be to optimize side mirrors, a-pillars, spoilers etc. for disciplines such as aerodynamics, aeroacoustics, contamination etc. For this thesis the optimization is performed in terms of aerodynamics and aeroacoustics on the side view mirrors of a passenger car. Another part of the thesis is to investigate the possibilities to post process and analyse the results, in order to find and understand the design parameters and how they affect the different disciplines. Two design parameters are used in the study. One parameter is the position of the mirror along the car and the other is the angle between the side of the car and the inside of the mirror. The optimization procedure follows four steps. The geometry is first morphed in the pre processing program ANSA into the design that will be simulated. A surface mesh is saved as a geometry representation. The second step is that the surface mesh is loaded into FLUENT meshing which creates the simulation domain and the volume mesh. Step two is performed in two separate session simultaneously, one for the aerodynamic and one for the aeroacoustic simulations. The next step is that the volume mesh is read by FLUENT which simulates the flow and calculates the optimization parameters, drag for the aerodynamic simulations and sound pressure level for the aeroacoustic simulations. The final step is that the optimization program HEEDS determines the new morphing parameters to send to ANSA based on the results from the previous simulations. A bash script was written which reads the design parameters from a separate file and runs ANSA and FLUENT in the correct order. HEEDS changes the design parameters in this separate file and then runs the bash script to obtain the output variables. After the output is obtained the process starts over. It was found that the angle of the inside of the mirror should be increased slightly to reduce drag and SPL. The optimal choice of the x-position seemed to be outside of the investigated interval, moving the mirror as far back on the car as possible was best in terms of drag. The effect the x-position had on the noise on the side window was very small.