Händelser: Studentarbete, Arkitektur, Bygg- och miljöteknik, Data- och informationsteknik, Energi och miljö, Kemi- och bioteknik, Matematiska vetenskaper, Material- och tillverkningsteknik, Mikroteknologi och nanovetenskap, Produkt- och produktionsutveckling, Rymd- och geovetenskap, Signaler och system, Sjöfart och marin teknik, Teknikens ekonomi och organisation, Fysik, Tillämpad IT, Tillämpad mekanik, Rymd-, geo- och miljövetenskap, Arkitektur och samhällsbyggnadsteknik, Bioteknik, Elektroteknik, Industri och materialvetenskap, Informations- och kommunikationsteknik, Mekanik och maritima vetenskaperhttp://www.chalmers.se/sv/om-chalmers/kalendariumAktuella händelser på Chalmers tekniska högskolaMon, 14 Jun 2021 17:52:33 +0200http://www.chalmers.se/sv/om-chalmers/kalendariumhttps://www.chalmers.se/sv/institutioner/tme/kalendarium/Sidor/Presentationer-av-examensarbeten-vid-Teknikens-ekonomi-och-organisation-vt-2021.aspxhttps://www.chalmers.se/sv/institutioner/tme/kalendarium/Sidor/Presentationer-av-examensarbeten-vid-Teknikens-ekonomi-och-organisation-vt-2021.aspxPresentationer av examensarbeten vid Teknikens ekonomi och organisation vt 2021<p>Online via Zoom</p><p>​ Presentationer av examensarbeten från masterprogram samt programmet Ekonomi och produktionsteknik.</p>​<span class="text-normal page-content"><a href="/sv/institutioner/tme/Utbildning/examensarbeten/Sidor/presentationer.aspx">​Läs mer och se presentationsschema</a></span>https://www.chalmers.se/sv/institutioner/fysik/kalendarium/Sidor/Masterpresentation-Morad-Mahmoudyan-och-Arianit-Zeqiri-210615.aspxhttps://www.chalmers.se/sv/institutioner/fysik/kalendarium/Sidor/Masterpresentation-Morad-Mahmoudyan-och-Arianit-Zeqiri-210615.aspxMorad Mahmoudyan och Arianit Zeqiri, MPCAS<p>Online via Zoom</p><p>​Titel på masterarbete: Retail Demand Forecasting With Neural Networks Följ presentationen online Lösenord: 552071</p><strong>​Sammanfattning:</strong><div><div> A third of the food produced for human consumption is wasted annually, amounting to 1.3 billion tons of food waste per year GlobalFoodLoss. Minimizing these enormous quantities of waste would not only be beneficial for the planet but also help feed the ever increasing human population. One way of minimizing this waste is by helping retail sellers better plan their logistical operations by accurately predicting the demand of goods using forecasting models. The usual procedure in time series forecasting has traditionally been to use statistical models such as autoregressive integrated moving average (ARIMA) and exponential smoothing methods. These methods have been shown to be limited in their predictive capabilities as the sizes of data sets and the number of variables increases. In the last decade new machine learning algorithms have been used extensively in various fields and has opened the door for utilization of neural network models in time series forecasting. A subset of these new machine learning algorithms, such as transformer based models and recurrent neural networks, have been proven to be especially suitable for temporal data. In this thesis we investigate two models, Temporal Fusion Transformers (TFTs) and Deep Temporal Convolutional Networks (DeepTCNs), showcasing their abilities to generate accurate forecasts of retail sales on a real-world data set. We demonstrate, using several metrics, their ability to outperform baseline models. </div> <div><br /></div></div>https://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Avanish-Raj.aspxhttps://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Avanish-Raj.aspxAvanish Raj, MPAUT<p>Webseminarium</p><p>​ Optimal torque split strategy for BEV powertrain considering thermal effects</p><div>​</div> <div><a href="https://chalmers.zoom.us/j/62450159105">Anslut till seminariet via Zoom.</a></div> <div>Lösenord: 059487</div> <div><br /></div> <div>Examinator: Sebastien Gros, Inst för elektroteknik</div> <div>Handledare: Anand Ganesan, Inst för elektroteknik</div> <div>Opponenter: Filip Tagesson och Oskar Edholm</div> <div><br /></div> <h2 class="chalmersElement-H2">Sammanfattning</h2> <div><br /></div> <div>A common architecture for electric vehicles is to have electric machines on both the front and rear axle. This work is an attempt to derive an optimal torque split strategy between the two machines to reduce the overall energy consumption. A clutch is implemented on the front axle and its engagement is dynamically controlled to reduce the magnetic drag losses. As electric machine and inverter losses are also dependent on temperature, a power loss map based on torque, machine RPM and temperature is used. An upper temperature limit for both electric machine and inverter is also imposed for component protection. Thermal models for electric machine, inverter and coolant circuit are simplified using system identification and model order reduction approach. Dynamic programming is used to investigate the benefits of including thermal losses and to generate a benchmark solution for optimal torque split strategy. Further, an online controller is developed based on non-linear model predictive control. The controllers are verified in a high-fidelity simulation environment.<br /><br />Keywords: Optimal torque-split control strategy, battery electric vehicles, thermal model of electric machines, system identification, model order reduction, co-simulation<br /><br /></div>https://www.chalmers.se/sv/institutioner/see/kalendarium/Sidor/Exjobbspresentation-Klaudia-Mur.aspxhttps://www.chalmers.se/sv/institutioner/see/kalendarium/Sidor/Exjobbspresentation-Klaudia-Mur.aspxExjobbspresentation, Klaudia Mur<p>Online</p><p>​A Network Analysis of a Company’s Internal Email Communications</p>​<span style="background-color:initial">Klaudia Mur presenterar sitt examensarbete ”​A Network Analysis of a Company’s Internal Email Communications”, utfört vid institutionen för rymd-, geo- och miljövetenskap. </span><div><br /></div> <div>Handledare och e<span style="background-color:initial">xaminator:  Claes Andersson</span></div> <div>Opponent: Georgios Triantafyllou</div> <div><br /></div> <div><a href="https://chalmers.zoom.us/j/62060584441">Ta del av presentationen via Zoom</a>. </div> <div>Lösenord: 884254</div> <div><br /></div>https://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Singh-Vestlund.aspxhttps://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Singh-Vestlund.aspxAditya Singh, MPCAS och Niclas Vestlund, MPSYS<p>Webseminarium</p><p>​ Prediction of energy consumption for battery electric vehicles</p><div>​A study based on a data-driven machine learning approach</div> <div><br /></div> <div><a href="https://chalmers.zoom.us/j/64988046382">Anslut till seminariet via Zoom.</a></div> <div><br /></div> <div>Examinator: Jonas Fredriksson, Inst för elektroteknik</div> <div><br /></div> <h2 class="chalmersElement-H2">Sammanfattning</h2> <div><br /></div> <div>The market for battery electric vehicles (BEVs) is growing steadily because of the environmental benefits, compared to conventional vehicles. The limited range and long charging times are however large issues for BEVs. To diminish the issue of range-anxiety among drivers, the energy consumption prediction needs to be accurate when setting a destination in the satellite navigation.<br /><br />This thesis aims to develop machine learning approaches that can predict the energy consumption for BEVs, split into propulsive and auxiliary consumption. The data used will consist of vehicle collected data from Volvo Cars and geographical data from a navigation supplier. Since a machine learning model’s performance relies on its inputs, a major part of the thesis will be spent on choosing the correct parameters and modifying the data to create better predictions.<br /><br />Linear regression, multi-layer perceptron, recurrent neural network, and gradient boosting were all machine learning models that were compared in the study. After the study it can be concluded that, for the prediction of the propulsive energy consumption, the choice of model did not matter significantly. The performance<br />of the auxiliary prediction was very similar for all models except the RNN, which showed worse results. It was also concluded that a more reasonable approach to show the predicted energy consumption for a trip, is to show a span of possible consumption instead of an absolute value. The main reason being that there is always a difference between the predicted and actual speed, acceleration, and total time for a trip.<br /><br /></div>https://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Presentation-Khalil-Almeamar.aspxhttps://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Presentation-Khalil-Almeamar.aspxMohammed Khalil och Abdulkrim Almeamar, TIELL<p>Webseminarium</p><p>​ Undersökning av el-konsumtion i en byggnad</p><div>​</div> <div><a href="https://chalmers.zoom.us/j/68594339149">Anslut till seminariet via Zoom.</a></div> <div>Lösenord: Thesis</div> <div><br /></div> <div>Examinator: Torbjörn Thiringer, Inst för elektroteknik</div> <div>Handledare: Thomas Hammarström och Patrik Ollas, Inst för elektroteknik</div> <div>Opponenter: Hadi Alzaher och Nour Aldeen Alassadi</div> <div><br /></div> <h2 class="chalmersElement-H2">Sammanfattning</h2> <div><br /></div> <div>The main goal of this report is how to calculate the electrical consumption of a house by calculating the electrical consumption of all electrical devices that could exist in the house. Based on the energy consumption of these devices, the electrical losses that occur in the home were calculated. In this report, the electrical losses were divided into two main parts, which are electrical losses in cables and electrical losses in switching units. There are many components that cause electrical losses, but in this report some component which cause the electrical losses in conversion units were studied, as well as the conversion losses, the cable conduction losses which occur with increasing of the temperature or overloads were also studied. In this job, the range of the electrical energy that produced by the solar panels has been studied through practical experiments. The amount of energy that produced by solar panels was recorded during one year for the mentioned house. The results of electrical consumption were also compared with solar energy and without solar energy. The amount of energy required from the electrical grid was calculated to compensate for the solar energy.<br /></div>https://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Arslan-Waltersson.aspxhttps://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Arslan-Waltersson.aspxGabriel Arslan Waltersson, MPSYS<p>Webseminarium</p><p>​ Planning and control for cable-routing with dual-arm robots</p><div>​</div> <div><a href="https://chalmers.zoom.us/j/67921125373">Anslut till seminariet via Zoom.</a></div> <div>Lösenord: 730647</div> <div><br /></div> <div>Examinator: Yiannis Karayiannidis, Inst för elektroteknik</div> <div>Handledare: Rita Laezza och Yiannis Karayiannidis, Inst för elektroteknik</div> <div>Opponent: Akos Vass</div> <div><br /></div> <h2 class="chalmersElement-H2">Sammanfattning</h2> <div><br /></div> <div>Deformable objects’ representation and manipulation is an active research area in Robotics. Though there are many types of deformable objects, all with different properties, this work considers deformable linear objects (DLO), such as ropes or cables. A framework for solving cable routing or wire-harness problems with a dual-arm robot is proposed, with the objective of clipping a rope into several fixtures. <br /><br />Dual-arm robots have two manipulators that are in close proximity and can collaborate on a single task. Building on inverse differential kinematics, two control methods are implemented. The manipulators can either be controlled individually as two separate entities or together as a single system. Hierarchical Quadratic Programming (HQP) is used to solve the inverse kinematics problem with secondary control objectives. A computer vision system is implemented for both tracking the DLO in real-time with structure preserved registration and to estimate the poses of the fixtures. A path planner is developed, that can generate trajectory parameters to solve the cable routing problem. The path planner builds a roadmap from predefined tasks and the trajectory parameters are evaluated for any issues. If any issues arise, a genetic optimization algorithm is used to find a solution.<br /><br />The system is tested with real-world experiments on an ABB YuMi robot. The results demonstrate successful cable routing through several fixtures and problem-solving capabilities. <br /><br />Keywords: Deformable linear objects, Dual-arm robotics, coordinated manipulation, structure preserved registration, Cable routing, Stochastic optimization, Genetic algorithms.<br /><br /></div>https://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Tell-Åstrand.aspxhttps://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Tell-%C3%85strand.aspxSara Tell och Alexander Åstrand, MPSYS<p>Webseminarium</p><p>​ Modelling measured and expected Doppler velocity of detections for inlier and outlier identification in radar odometry</p><div>​</div> <div><a href="https://chalmers.zoom.us/j/62229476264?pwd=bzJBbVMrWlFmK0s5c1VYYkFnampNUT09">Anslut till seminariet via Zoom.</a></div> <div>Lösenord: 414785</div> <div><br /></div> <div>Examinator: Lars Hammarstrand, Inst för elektroteknik<br /></div> <div><br /></div> <h2 class="chalmersElement-H2">Sammanfattning</h2> <div><br /></div> <div>Autonomous driving and driver assistance functions for autonomous vehicles need an accurate estimation of the rotational and translational ego-velocity to guarantee safe execution. Some sensors used for odometry are wheel encoders and IMUs, where the first are sensitive to tire slip and the second suffers from drift in the accelerometer. Other sensors which can be used for this purpose are radars, LiDARs and cameras, which perceive point clouds of the surrounding. However, the optical nature of the LiDAR and camera makes them sensitive to weather and lighting conditions, an issue the radar does not have. Another advantage of the radar is the ability to measure the relative radial range-rate, called the Doppler velocity, to the objects which the radar waves reflect upon. These characteristics make the radar a promising choice for odometry.<br /><br />The methods for estimating the ego-velocity with radars can be split into two types; scan matching methods and instantaneous methods. The first utilize the spatial measurements to estimate the rigid-transformation of the point clouds between two radar scans of stationary detections and the second utilize the relative range-rate, which is the inverse of the ego-vehicle sensor velocity if the object is stationary, often adopted in a RANSAC algorithm. There are also hybrid methods which combine scan matching and instantaneous algorithms.<br /><br />Dynamic and noisy measurements lead to errors in the estimations, therefore there is a need for an accurate outlier removal. This project models the behaviour of stationary detections and outliers based on the measured and expected Doppler velocity, where the models find the probability that the radar measurement is a stationary object. The probability is adopted in two radar odometry algorithms, one hybrid method and one instantaneous method, to refine the outlier removal. The methods are evaluated on the open dataset RadarScenes, with data from four automotive radars in urban environments and ground truth on the ego-vehicle longitudinal and rotational velocity. The affects of filtering with a Curbature Kalman Filter (CKF) is also assessed.<br /><br />The results show that the proposed model of the measurements can reduce the influence of outliers in the ego-motion estimations. Moreover, the hybrid methods are determined to be a less ideal radar odometry choice in many regards in comparison to the instantaneous methods, with (in worst case) more than 4 times larger standard deviation of the estimation error on the rotational velocity. The algorithm also suffers from drift, heavy computations and needs an accurate prior.<br />With only 5 iterations in a modified RANSAC algorithm, the drift-free instantaneous method achieves a standard deviation of 0.056 m/s and 1.6 deg/s of the longitudinal and rotational velocity estimation error respectively. <br /></div>https://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Do-Combrink.aspxhttps://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Do-Combrink.aspxStephie Do och Alvin Combrink, MPSYS<p>Webseminarium</p><p>​ Automatic shift scheduling for healthcare personnel using satisfiability modulo theory</p><div>​</div> <div><a href="https://chalmers.zoom.us/j/69902085413">Anslut till seminariet via Zoom.</a></div> <div>Lösenord: 1</div> <div><br /></div> <div>Examinator: Knut Åkesson, Inst för elektroteknik</div> <div>Handledare: Kristofer Bengtsson, Inst för elektroteknik</div> <div><br /></div> <h2 class="chalmersElement-H2">Sammanfattning</h2> <div><br /></div> <div>Shift scheduling involves allocating shifts to healthcare personnel in a way that complies with regulations and preferences. The task is laborious and time-consuming since it is often done manually. This is particularly true in the healthcare sector where many administrative tasks are allocated to qualified health care providers instead of dedicated administrators. With this in mind, a system for automating shift scheduling has been developed in this thesis to be used in healthcare services. To achieve this, a mathematical model is created that consists of 10 logically formulated constraints that are able to model a wide range of requirements. Using two independent optimization methods based on Satisfiability Modulo Theory (SMT) and Genetic Algorithms, it is demonstrated that SMT is better suited to handle a problem at this level of complexity under such restricting constraints. With the developed automatic scheduling system, a schedule that satisfies all hard constraints for a representative test case is generated in a considerably short time.<br /></div>https://www.chalmers.se/sv/institutioner/math/kalendarium/Sidor/Examensarbete210615.aspxhttps://www.chalmers.se/sv/institutioner/math/kalendarium/Sidor/Examensarbete210615.aspxPresentation av masterarbete<p>Online</p><p>​Noel Waters: Gaussian Process regression for modelling blood glucose dynamics -- A study based on clinical data from subjects with Type 2 diabetes mellitus</p><p><br />Abstract: Type 2 Diabetes is a disease characterized by poor control of blood glucose levels. Continuous Glucose Monitoring (CGM) is an increasingly popular technology for<br />studying glucose levels and evaluating treatment effects. Although CGM technology gives potential for granular insights into disease characteristics, more can be done in terms of exploiting this rich and dense data source to the fullest. This study aims to investigate the usefulness of Gaussian Process Regression as a framework for modelling blood glucose dynamics. The CGM data were collected from a previous clinical trial on a cohort of overweight and obese Type 2 Diabetes patients. Gaussian Process modelling tools were used to capture short-term and recurring trends while adjusting for long-term changes in glucose control. Results indicate that structure such as periodicity can be successfully modelled. Interpreting specific modelling results showed to be challenging due to a high degree of uncertainty in the model hyperparameters. Non-stationary models should be considered to better account for the irregular occurrence of meal-related glucose spikes and differences between day and night glycemic variability. Finally, the periodic properties of blood glucose dynamics should be further explored.</p> <p>Handledare: <span>Umberto Picchini (Chalmers/GU), Michail Doulis (AstraZeneca)<br />Examinator: Moritz Schauer</span></p>https://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Hultberg-Wu.aspxhttps://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Hultberg-Wu.aspxGustav Hultberg Hermansson och Simon Wu, MPCAS<p>Webseminarium</p><p>​ Machine learning threat assessment algorithms</p><div>​</div> <div><a href="https://chalmers.zoom.us/j/65296508064">Anslut till seminariet via Zoom.</a></div> <div><br /></div> <div>Examinator: Jonas Fredriksson, Inst för elektroteknik</div> <div><br /></div> <h2 class="chalmersElement-H2">Sammanfattning</h2> <div><br /></div> <div>Traffic fatalities remains a great problem for society today, being the leading causes of deaths for children and young adults. Following the fast development in technology, Advanced Driving Assistance Systems in vehicles take an ever increasing role in the struggle to reduce traffic related deaths and injuries. A key component of ADAS is to accurately and robustly predict potential future threats within its limits and capabilities. In this thesis, an Operational Design Domain (ODD) was determined to first define the capabilities and limitations of the model to be developed. Then, a Machine Learning (ML) based Threat Assessment (TA) model was developed for detecting unintended lane departures during driving. The model was constructed using an artificial neural network trained on real-world time series driving data, and a proposed decision-making logic. The threat assessment performance was evaluated and benchmarked against a kinematic model as well as a reference Lane Keeping Aid system.  The model's performance in the ODD was evaluated by investigating the underlying causes for the model's failure modes. A simulation tool was developed and used together with on-vehicle cameras for this purpose. The benchmarking results show that the ML based TA model has increased performance compared to the kinematic model and the reference system. Moreover, the results of the ODD evaluation indicate that the main underlying cause for the model's failures lies in its inability to correctly interpret driving behaviours. <br /></div>https://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Lingyun-Deng.aspxhttps://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Lingyun-Deng.aspxLingyun Deng, MPBME<p>Webseminarium</p><p>​ Development of a crash detection algorithm for motorcycle drivers using machine learning</p><div>​</div> <div><br /></div> <div><a href="https://chalmers.zoom.us/j/65236055654">Anslut till seminariet via Zoom.</a></div> <div>Lösenord: 1022</div> <div><br /></div> <div>Examinator: Stefan Candefjord, Inst för elektroteknik</div> <div><br /></div> <h2 class="chalmersElement-H2">Sammanfattning</h2> <div><br /></div> <div>Motorcycle is a popular vehicle that can be used for multiple purpose, such as sport and commuting. It is<br />estimated that only in Europe there are already 23 million motorcycles. However, statistics show that its risk<br />of severe or deadly crashes is over-represented in traffic fatalities, of which one important reason is the lack of support when the motorcyclists are driving alone so that the timely medical treatment is missed to improve the wounded outcome. To tackle this problem and save life, Detecht Technologies AB has developed a smartphone application for sending rescue information to emergency service provider when the app automatically detects a crash happening by measuring the built-in sensors in the smartphones. The aim of this thesis is to improve the motorcycle crash detection algorithm with machine learning, based on the driving data provided by Detecht.<br /><br />A literature review was conducted to study the existing methods of motorcycle crash detection. In addition,<br />several deep learning based temporal anomaly detection methods were researched and candidates methods<br />are compared. Furthermore, an exploratory data analysis was conducted to understand the composition and<br />limitation of the dataset. Then a crash detection algorithm pipeline was proposed based on the findings. The<br />method is autoencoder based to learn the generic normality features and then to define the anomalous patterns as possible crash events. Finally, the motorcycle crash detection results were compared with the given true crashes and normal driving, from the sense of sensitivity and specificity.<br /><br />Several models were tested their performance on the same dataset, and an optimal model was standing out<br />by comparing their ability of detecting true crashes while maintaining low false alarm rate. It shows that the<br />optimal model can detect the provided true crash events while raise a false alarm per 25 hours on average.<br /><br />The thesis analyzed these results and discussed the relationship between crash detection and false alarms<br />arousing. In addition, several suggestions were proposed and analyzed as a future development direction.<br /><br />This thesis developed a machine-learning-based algorithm classifying whether a motorcycle crash has happened and optimized it. The results of this thesis can form a foundation for future better crash detection algorithm pipeline and a better data engineering.<br /></div>https://www.chalmers.se/sv/institutioner/ims/kalendarium/Sidor/Environmental-impact-of-ilmenite.aspxhttps://www.chalmers.se/sv/institutioner/ims/kalendarium/Sidor/Environmental-impact-of-ilmenite.aspxEnvironmental impact of ilmenite from mines to product<p>Online</p><p>​James Andrén &amp; Hussein Abbas presenterar sitt examensarbete. Online presentation.</p>​<span style="background-color:initial">Program: Maskinteknik</span><div>Examinator: Martina Petranikova</div> <div>Handledare: Pavleta Knutsson</div> <div>Opponent: Johan Nguyen</div>https://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Liden-Holmström.aspxhttps://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Liden-Holmstr%C3%B6m.aspxMartin Lidén och Marcus Holmström, MPCAS<p>Webseminarium</p><p>​ Indoor radar localisation</p><div>​</div> <div><a href="https://chalmers.zoom.us/j/63471629802">Anslut till seminariet via Zoom.</a></div> <div><br /></div> <div>Examinator: Tommy Svensson, Inst för elektoteknik</div> <div>Handledare: Rickard Nilsson, Aptiv AB; Björn Langborn, Inst för elektroteknik</div> <div><br /></div> <h2 class="chalmersElement-H2">Sammanfattning</h2> <div><br /></div> <div>In this thesis indoor localisation using Simultaneous Localization And Mapping (SLAM) based on radar data has been investigated. As radar already is available in many vehicles in the automotive industry, improving the radar functionality to work better indoors would increase the confidence in the vehicle's overall positioning. The data used has been recorded at Aptiv's facilities and the data set is from a basement parking garage where the material in the area mainly are concrete and sheet metal, with the addition of a few corner reflectors. An analysis of the Radar Cross-Section (RCS) from these materials is presented as well as an algorithm for filtering, augmenting and processing the radar detections. This algorithm, called Detection Density Augmentation Filter (DDAF), was designed for compatibility with an open source SLAM module and is compared to a preexisting processing standard. A SLAM module has been implemented, into the Aptiv project City Mobility On Demand (CMOD), and tested. The SLAM results are subpar as the module was found to not fully account for higher velocities of the host vehicle. The faulty output could be reproduced and the issue traced down to the open source module not interpolating correctly between vehicle positions. Despite the subpar results the algorithm DDAF mentioned above shows promising results for indoor SLAM using radar.<br /></div>https://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Gardell-Månsson.aspxhttps://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Gardell-M%C3%A5nsson.aspxTherese Gardell och Olle Månsson, MPENM<p>Webseminarium</p><p>​ Lane detection with attention mechanisms</p><div>​</div> <div><a href="https://chalmers.zoom.us/j/69288181560">Anslut till seminariet via Zoom.</a></div> <div>Lösenord: 072474</div> <div><br /></div> <div>Examinator: Lennart Svensson, Inst för elektroteknik</div> <div><br /></div> <h2 class="chalmersElement-H2">Sammanfattning</h2> <div><br /></div> <div>For autonomous vehicles, perception of the surroundings is key. In order to avoid collisions and to keep vehicles on the road, the identification of lane markers is of utmost importance. Lane detection, the task of detecting lanes on a road from a camera, is traditionally performed with convolutional neural networks. These networks excel when you want to capture local relationships in an image, which is often key in computer vision tasks. However, lanes can stretch across the input image or be occluded by vehicles, debris or other alien objects on the road. This raises the question of whether global perception or long-range dependencies are also needed to effectively detect lanes under challenging conditions similar to those described above. Convolutional neural networks, by design, have a limited receptive field and a limited capacity to capture long-range dependencies. <br /> <br />In this thesis, we explore how attention mechanisms, from the field of natural language processing, can be used to augment convolutional neural networks and improve their ability to capture long-range dependencies. We combine a residual neural network encoder with a U-net style decoder and a self-attention mechanism in the form of a Transformer encoder. The model architecture is evaluated over a range of encoder-decoder configurations and the addition of an attention mechanism improves semantic segmentation performance across the board. The results are validated on both a synthetic dataset engineered to require long-range dependencies as well as on a lane detection dataset captured in real traffic.<br /><br /></div>https://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Zeqiri-Sanden.aspxhttps://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Zeqiri-Sanden.aspxArbnor Zeqiri och Ellen Sandén, MPCAS<p>Webseminarium</p><p>​ Machine learning for automotive cybersecurity</p>​Anomaly detection in CAN<br /><div><br /></div> <div><a href="https://chalmers.zoom.us/j/61515505739">Anslut till seminariet via Zoom.</a></div> <div>Lösenord: 975016</div> <div><br /></div> <div>Examinator: Martin Fabian, Inst för elektroteknik</div> <div><br /></div> <h2 class="chalmersElement-H2">Sammanfattning</h2> <div><br /></div> <div>As modern cars get increasingly computerized, they become more susceptible to hacker attacks. One of the main attack surfaces in a car is the CAN bus, which is a network that allows all electrical components in the vehicle to communicate with each other. The lack of encryption in CAN makes it vulnerable and in need of an external attack detection system. One way such a system can operate is by learning the normal behavior of the bus, and then analyse the incoming messages to search for anything that diverges from the norm. Any large enough deviation is deemed a potential attack.<br /><br />In this thesis, a two branched anomaly detection system is developed using several types of machine learning algorithms. The two main components of the CAN messages, the ID and the data load, are analysed in real time on both branches simultaneously to see whether the messages are normal. A detected deviation on either branch is deemed a potential attack, and will trigger a warning.<br /><br />To cover as many different types of potential attacks as possible, the two branches are designed to be able to detect different types of anomalies. The first branch consists of a combination of a one class support vector machine and a neural network auto encoder, where the former is able to detect anomalies in the frequency of IDs as well as time between messages and the latter detects anomalies in the content of the individual data loads. The second branch consists of a neural network auto encoder with convolutional and long short term memory layers, which detects anomalies in the correlations in time between messages, as well as correlations between ID and data load.<br /><br /></div>https://www.chalmers.se/sv/institutioner/fysik/kalendarium/Sidor/Masterpresentation-Juan-Diego-Arango-210615.aspxhttps://www.chalmers.se/sv/institutioner/fysik/kalendarium/Sidor/Masterpresentation-Juan-Diego-Arango-210615.aspxJuan Diego Arango, MPCAS<p>Online via Zoom</p><p>​Titel på masterarbete: Application of Curriculum Learning on de novo design Följ presentationen online Lösenord: 410785</p><strong>​Sammanfattning:</strong><div><div>The basis of the master thesis project is REINVENT which is an application for de novo design of small molecules. REINVENT is a generative model. The model is trained with a dataset derived from ChEMBL; it is a manually curated chemical database of bioactive molecules with drug-like properties. The molecules are represented in strings using SMILE (Simplified molecular-input line-entry system) line notation. For the generative model, it necessary to direct its output to a section of the parameter space where the generated compounds have the desired properties. How well a generated molecule adheres to the desired properties is measure by a scoring function. The scoring function is a composite of different user-defined components. The scoring is used as a reward in a Reinforcement Learning loop where the generative model will try to maximize the outcome of a scoring function. The objective of the master thesis is to investigate the possibility to enhance/speed up the generative process by exploring multi-agent scenarios as opposed to the current single agent implementation. Instead of having a single agent explore the parameter space trying to maximize a &quot;big&quot; scoring function. The idea is to maximizes each component of the overall scoring function, and aggregate all the components into a single scoring a function which is used in the update of the generative model.</div></div> <div><br /></div>https://www.chalmers.se/sv/institutioner/m2/kalendarium/Sidor/Blade-element-momentum-method-for-a-counter-rotating-pump-turbine.aspxhttps://www.chalmers.se/sv/institutioner/m2/kalendarium/Sidor/Blade-element-momentum-method-for-a-counter-rotating-pump-turbine.aspxBlade element momentum method for a counter-rotating pump-turbine<p>Online i Zoom</p><p>​Cristóbal Ibáñez presenterar sitt examensarbete med titeln &quot;Blade element momentum method for a counter-rotating pump-turbine&quot;.</p>​<div>Student: Cristóbal Ibáñez<br /></div> <div><br /></div> <div>Handledare: Håkan Nilsson, Hamidreza Abedi och Jonathan Fahlbeck</div> <div><br /></div> <div>Examinator: Håkan Nilsson</div> <div><br /></div> <div>Opponent: Sourav Nandakumar</div>https://www.chalmers.se/sv/institutioner/ims/kalendarium/Sidor/Flexibility-of-automotive-product.aspxhttps://www.chalmers.se/sv/institutioner/ims/kalendarium/Sidor/Flexibility-of-automotive-product.aspxFlexibility of automotive product architectures and its impact on platform lifecycles<p>Online</p><p>​Motasom Ibrahim och Hamza Hassan  presenterar sitt examensarbete. Online presentation.</p>​<span style="background-color:initial">Program: MSc Product Development</span><div>Examinator: Massimo Panarotto, IMS</div> <div>Handledare: Iñigo Alonso Fernandez, IMS</div> <div>Opponent: Oskar Strandh Tholin</div> <div><br /></div> <div><strong>Abstract</strong></div> <div>The market demand for differentiated product variants keeps increasing. A platform strategy is followed by companies to develop differentiated products to adapt to the forever changing customer needs while cutting the development lead times. In the automotive industry, this strategy has been used for decades to produce different variants of vehicles based around a common architecture. </div> <div>Platforms’ longevity is expected to decrease due to the faster changing market. To fulfil the customers’ needs with minimal changes to the production architecture, a modular platform is a good candidate. This eliminates the hefty cost of completely redesigning a new platform or heavily modifying an existing platform architecture and its associated production setup. There is currently a lack awareness in practice of how modularisation methods can enhance the flexibility of platforms and companies generally rely on subjective approaches.  </div> <div>This thesis work was carried out in collaboration with Volvo Cars, with the goal to investigate the flexibility of the current platforms, the trade-offs required to achieve a highly flexible platform, and the role of geometry and space allocation of components regarding flexibility level.  In-depth interviews were conducted with employees working at different departments at Volvo Cars. The interviews provided insights into the company's way of working with regards to modularization, the drawbacks with it and expected future challenges.</div> <div>The interview data was analysed in detail using reflective thematic analysis, comparing and categorizing the data according to the relevant themes associated with the research questions. Furthermore, a case study of a component was conducted to get an understanding of the design and interfaces across vehicles of different variants. </div> <span style="background-color:initial">The practices followed for modularization is at different levels of sophistication between departments at Volvo Cars, and generally trailing behind the state-of-the-art. The trade-offs identified were well understood and the qualitative analysis showed that the modularization of pure mechanical modules is better defined and documented in comparison to electrical and software modules.  The new developments like the electrification of cars pose challenges affecting the decision-making process and modularization strategy at Volvo Cars.</span>https://www.chalmers.se/sv/institutioner/fysik/kalendarium/Sidor/Masterpresentation-Thomas-Hoffmann-210615.aspxhttps://www.chalmers.se/sv/institutioner/fysik/kalendarium/Sidor/Masterpresentation-Thomas-Hoffmann-210615.aspxThomas Hoffmann, MPCAS<p>Online via Zoom</p><p>​Titel på masterarbete: Quantum models for Word-Sense Disambiguation Följ presentationen online Lösenord: 419949</p><strong>​Sammanfattning:</strong><div><div>In recent years, developments in machine learning and artificial intelligence had a tremendous impact on Natural Language Processing (NLP). Statistical vector space models based on the Transformers architecture have successively broken almost any benchmarks in a wide variety of NLP domains. Nevertheless, there has been criticism in the recent past. State-of-the-art language models (like BERT, GPT-3, ...) contain billions of parameters that require vast computational resources for optimization. Furthermore, purely statistical models capture syntactic rules only from data, which does not allow an extensive analysis of the underlying logic of language. To reduce the parameter space of NLP models and close the gap between logic-based language models and statistical vector space models, Coecke et al. introduce a compound framework called Compositional Distributional Model of Meaning, based on Lambeks Pregroup grammar and Quantum Theory.</div> <div>This thesis investigates applying the Compositional Distributional Model of Meaning on the word-sense disambiguation task by Kartsaklis, Sadrzadeh, and Pulman. Different quantum embeddings are evaluated in terms of disambiguation power, given a matching context. One focus lies on the description of ambiguous words as mixed states. Mixed states are probabilistic quantum states expressed as density matrices which entail a lack of knowledge about the underlying system. Empirical data was gathered from experiments using quantum circuits and classical computations. We evaluate the performance and discuss the challenges and limitations of the current quantum computing models. The results confirm the comprehensiveness of the compositional distributional model of meaning and show statistical evidence for a richer representation of words by density matrices.</div></div> <div><br /></div>https://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Hagström-Wenzel.aspxhttps://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Hagstr%C3%B6m-Wenzel.aspxFredrik Hagström och Arvid Wenzel Wartenberg, MPCAS<p>Webseminarium</p><p>​ Leveraging monocular depth estimation for 3D object detection</p><div>​</div> <div><a href="https://chalmers.zoom.us/j/61457744759">Anslut till seminariet via Zoom.</a></div> <div>Lösenord: 452231</div> <div><br /></div> <div>Examinator: Lennart Svensson, Inst för elektroteknik</div> <div><br /></div> <h2 class="chalmersElement-H2">Sammanfattning</h2> <div><br /></div> <div>3D object detection (3DOD) is a central task in the development of autonomous driving (AD). Camera-based 3DOD methods make use of camera data to produce detections of objects in 3D space. However, a central problem of camera-based 3DOD is that camera data inherently lack depth information, and as such, the performance of these methods have left much to be desired. Recently, deep learning-based algorithms have shown great progress in estimating pixel-wise depth information from monocular RGB-images. Deep learning-based 3DOD models which incorporate these depth estimations have recently been proposed, and, as a result, the performance of camera-based 3DOD has greatly increased. <br /> <br />The contribution of this work is two-fold. Firstly, we investigate how a state-of-the-art approach for 3DOD based on monocular depth-estimation performs in long-distance settings. We find that the performance improvements from incorporating depth estimations in camera based 3DOD models decrease drastically for very faraway objects. Secondly, we introduce a trainable point selection based on a multi-head attention mechanism. This allows our model to selectively attend to specific regions of the input. We show that the inclusion of our proposed mechanism yields significant performance improvements on Zenseacts data set, especially for problematic settings, such as higher distances. Additionally, we show qualitative results, where the attention mechanism helps the model focus on objects which are heavily occluded.<br /></div>https://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Master-presentation-Haitham-Babbili,-Olalekan-Peter-Adare.aspxhttps://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Master-presentation-Haitham-Babbili,-Olalekan-Peter-Adare.aspxHaitham Babbili och Olalekan Peter Adare<p>Online</p><p>​Power Control in Integrated Access and Backhaul Networks</p>​<div><span style="background-color:initial"><strong>Supervisors: Tommy Svensson (Chalmers); Behrooz Makki (Ericsson); Charitha Madapatha (Chalmers)</strong></span><div><strong>Examiner: Tommy Svensson (Chalmers)</strong></div> <div><b><br /></b></div> <div><b>Abstract</b><br />The integrated access and backhaul (IAB) network is a novel radio access network (RAN) solution, proposed by the 3rd Generation Partnership Project (3GPP). It is one of the interesting aspects of the fifth-generation (5G) RAN. The IAB network thrives on the advantage of using all ranges of spectrum defined for the 5G new radio (NR) to interconnect the mobile and fixed access users to the network, and still bulk transmit their data towards the 5G core network. IAB network may reduce the dependency on the optical fiber network and out-of-band frequencies for backhauling. </div> <div><br /></div> <div>However, there are inherent challenges that come with this approach. A possible constraint may be interference, which reduces the received signal quality. This, in turn, reduces the user data rate. Therefore, interference should be mitigated or canceled optimally. Using power control combined with adaptive beamforming, resource allocation and routing techniques may help to deploy efficient IAB networks, with higher spectral efficiency and better service coverage. </div> <div><br /></div> <div>This thesis focuses on a power control solution in uplink communication within IAB networks. A power control solution may effectively reduce the effects of interference among all the transceivers and keep the network operating at a network-defined minimum signal-to-interference-plus-noise-ratio (SINR). The solution is built on a genetic algorithm (GA), which offers reasonable solutions in multi-objective problem formulations. </div> <div><br /></div> <div>The performance of the solution is then evaluated using the service coverage probability. Service coverage probability is the probability of the event that the users are provided with a minimum data rate. First, we model the wireless channel resources using millimeter wave characteristics. Then, we build a finite coverage area with randomly distributed users and statically positioned base stations with transmit power constraints, as proposed by 3GPP. Afterwards, we model a wireless access channel that takes into account the predominant channel constraints, where the duplexing mode is time division duplex (TDD). </div> <div><br /></div> <div>Finally, we optimize the SINR and transmit power of all mobile terminals (MT) at every epoch using the GA. The GA offers a proper convergent solution for power control in IAB networks. Furthermore, based on our simulation, the IAB network with a well implemented power control scheme can achieve a better service coverage probability in uplink communication.</div> <div><br /></div> <div><p class="MsoNormal"><b><span lang="EN-US">Location:</span></b><span lang="EN-US"> Zoom online:</span><span lang="EN-US"> <a href="https://chalmers.zoom.us/j/64506447298">https://chalmers.zoom.us/j/64506447298​</a></span></p></div> </div>https://www.chalmers.se/sv/institutioner/ims/kalendarium/Sidor/Design-and-implementation-of-a-PLM-.aspxhttps://www.chalmers.se/sv/institutioner/ims/kalendarium/Sidor/Design-and-implementation-of-a-PLM-.aspxDesign and implementation of a PLM system for Chalmers Formula Student<p>Online</p><p>​Rikard Karlsson &amp; Abhishek Venkatesh presenterar sitt examensarbete. online presentation.</p><span style="background-color:initial">Program: Product Development</span><div><span style="background-color:initial">Examinator: </span><span style="background-color:initial">Dag Henrik Bergsjö, IMS</span></div> <div><span style="background-color:initial">Handledare: </span><span style="background-color:initial">Björn Pålsson</span></div> <div><span style="background-color:initial">Opponent: </span><span style="background-color:initial">Tim Ganzer</span></div> <div><span style="background-color:initial"><br /></span></div> <div><span style="background-color:initial"><div><strong>Abstract</strong></div> <div>Complex products typically consists of many different parts which needs to interact and function together. In the modern world, the development and verification of parts, sub-assemblies and entire products are done to a large extent with the help of computers. This in turn generates a lot of digital documentation and information which needs to be stored and controlled throughout the life cycle of a product. In order to keep track of the information, a Product Lifecycle Management, or PLM, system can be used. </div> <div><br /></div> <div>The study which has been carried out has investigated the current digital documentation structure of Chalmers Formula Student, CFS, with the aim of identifying possible areas of improvement and whether the implementation of a specific PLM system could be beneficial for their continued development efforts. Information about PLM and the aspects related to it has been gathered through a literature review and served as a basis for the study. Information about the current documentation structure was gathered through interviews with current and previous members of the CFS team. In an effort to extend the number of viewpoints of PLM, an additional round of interviews were held with professors and engineers with various experience with PLM system.</div> <div><br /></div> <div>Based on the theoretical framework and the held interviews, it became apparent that there were potential improvements to be made mainly regarding the aspects of knowledge transfer and finding needed documentation concerning changes made during the development efforts. The main underlying reasoning was mentioned as lack of time to document properly, last minute changes made without proper updating and a difficulty of navigating the folder structure. The investigation and evaluation of the benefits of a PLM system resulted in the suggestion of a gradual implementation, starting with the PLM system being used to manage Computer Aided Design, CAD files.</div> <div><br /></div></span></div>https://www.chalmers.se/sv/institutioner/math/kalendarium/Sidor/Examensarbete210615C.aspxhttps://www.chalmers.se/sv/institutioner/math/kalendarium/Sidor/Examensarbete210615C.aspxPresentation av masterarbete<p>Online</p><p>​Max Blom: Classifying Strictly Tessellating Polytopes</p><p><br />Abstract: This thesis consists of a paper and additional results. The paper shows a connection between the geometry of polytopal domains in Euclidean space and the eigenfunctions of the Dirichlet Laplacian. The necessary and sufficient geometric properties of a polytopal domain are shown for the first eigenfunction to extend to a real analytic function on the whole space. Furthermore, alcoves are essential for the proof of the main theorem. Additionally, the paper discusses how the results relate to crystallographic restrictions and lattices. Strictly tessellating polytopes are defined and used in connection to the main theorem. The paper concludes by formulating a conjecture akin to Fuglede’s, replacing tessellation by translation with strict<br />tessellation. In addition to the paper, results on the geometric properties of strictly tessellating polytopes are shown, and bounds on the number of strictly tessellating polytopes up to equivalence are given.</p> <p>Handledare: Magnus Goffeng<br />Examinator: Julie Rowlett</p>https://www.chalmers.se/sv/institutioner/math/kalendarium/Sidor/Examensarbete210615B.aspxhttps://www.chalmers.se/sv/institutioner/math/kalendarium/Sidor/Examensarbete210615B.aspxPresentation av masterarbete<p>Online</p><p>​Konrad Pohl: Machine Learning for NCC&#39;s Concrete Pile Production</p>​<br />Handledare: Holger Rootzénhttps://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Andreas-Tilo.aspxhttps://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Andreas-Tilo.aspxAndreas Tilo, MPEPO<p>Webseminarium</p><p>​ Evaluation of using coincidence factors for peak load estimation in a distribution grid</p><div>​A case study based on the Gothenburg power grid </div> <div><br /></div> <div><a href="https://chalmers.zoom.us/j/64114868273">Anslut till seminariet via Zoom.</a></div> <div>Lösenord: 123456</div> <div><br /></div> <div>Examinator: Peiyuan Chen, Inst för elektroteknik</div> <div>Handledare: Ferruccio Vuinovich, Göteborg Energi; Kristoffer Fürst, Inst för elektroteknik</div> <div>Opponent: Linnea Sundberg</div> <div><br /></div> <h2 class="chalmersElement-H2">Sammanfattning</h2> <div><br /></div> <div>In order for the power grid to have sufficient capacity for supplying demands in the future and for it to be used effectively, proper planning and forecasting of future demands is needed. One way to forecast increased demand when connecting new loads is by using coincidence factors which measures the ratio between the sum of the individual peak loads fed by a unit and the peak load of the aggregated load seen by the unit. The study is done in collaboration with the distribution system operator in Gothenburg, Göteborg Energi Nät AB, and focuses on a primary substation located in central Gothenburg. Real power and current measurement data taken during the year 2020 is used to calculate the coincidence factors of the radial 10 kV lines, the three 130/10 kV transformers as well as for the whole primary substation. Information on energy declarations was requested from the Swedish National Board of Housing, Building and Planning and was used to determine a customer distribution in all the individual lines and transformers based on the following customer categories; Industry, Commercial, Residential with district heating and Residential with electric heating. <br /> <br />To verify the validity of using coincidence factors to forecast increased demand, four type curves representing the yearly load profiles of the aforementioned customer categories were created based on the measurement data. These were then added with increasingly larger peak load up to a peak load of 3 MW to the yearly load profile of a chosen 10 kV line in the primary substation. The resulting peak load was then compared with the estimated peak load calculated using coincidence factors and the error was found to be within 2-3 % when adding loads with similar profiles to the initial load profile in the line. However when adding the Industrial and Commercial loads the error was greater due to a rapidly decreasing coincidence factor in the line. A similar analysis was done for the three main 130/10 kV transformers by readding secondary substations from the transformers as new loads and doing this yielded an error of up to ±4 % for 3 MW of added peak load as well as spans for the resulting coincidence factor in the transformers based on added peak load. The results containing the existing coincidence factors in part of the Gothenburg power grid as well as the knowledge of the accuracy the coincidence factors provide will help Göteborg Energi Nät AB prognose future demand more accurately and expand the grid more efficiently in the future. <br /></div>https://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Edvin-Alfredsson.aspxhttps://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Edvin-Alfredsson.aspxEdvin Alfredsson, MPSYS<p>Webseminarium</p><p>​ Floor quality inspection system</p><div>​</div> <div><a href="https://chalmers.zoom.us/j/69829234747">Anslut till seminariet via Zoom.</a></div> <div>Lösenord: 289271</div> <div><br /></div> <div>Examinator: Bengt Lennartson, Inst för elektroteknik</div> <div>Opponenter: Hampus Andersson och Divya Kara</div> <div><br /></div> <h2 class="chalmersElement-H2">Sammanfattning</h2> <div><br /></div> <div>The process of grinding a surface is today very reliant on the user. There does not exist many tools for measuring grinded surfaces, and those that are used are slow and only give profile measurements. As a consequence of this a lot of experience and intuition is required of the user to correctly decide when to use what grit. For this reason a better method for measuring the roughness of a grinded surface is wanted. This project evaluates an approach where 3D reconstruction is used to determine the roughness of a grinded surface. To start with several camera based approaches are considered in a theory study. Later on a Photometric stereo algorithm is developed for testing on concrete surfaces. The algorithm is divided into two parts where the first part estimates normals from light intensities in several images. The second part reconstructs the surface from the normals where two robust algorithms are tested which utilize diffusion tensors and continuous weights. Since Photometric stereo gives a relative 3D measurement it is not possible evaluate against old profile measurements. Because there is also very little is known about the shape of grinded surfaces Photometric stereo is instead tested on 3 different grinded concrete surfaces as to evaluate what grits are possible to measure. It was found that Photometric Stereo can see large scratches in the surface, but has problems with flatter surfaces caused by high grit. The algorithm is also very sensitive to the used light sources and is believed to cause the largest problems. Future work should therefore focus on utilizing more robust and consistent hardware as to improve the results. The algorithms can also be improved by either considering a more realistic light model or combining reconstruction algorithms.<br /></div>https://www.chalmers.se/sv/institutioner/ims/kalendarium/Sidor/Automatisad-montering-av-balansaxelhus.aspxhttps://www.chalmers.se/sv/institutioner/ims/kalendarium/Sidor/Automatisad-montering-av-balansaxelhus.aspxAutomatisad montering av balansaxelhus<p>Online</p><p>​Herman Wäpling, Maskinteknik  och Alexander Andersson, Mekatronik  presenterar sitt examensarbete. Online presentation.</p>​Examinator: Göran Gustafsson, IMS<div>Opponenter: Mattias Wallin och Linus Hermansson</div> <div><br /></div> <div><span style="background-color:initial"><strong>Abstract </strong></span><br /></div> <div>The goal with this bachelor thesis was to develop an automated assembly station to</div> <div>replace an manually operated assembly station at the internal combustion engine</div> <div>producer Powertrain Engineering Sweden AB in Skövde, Sweden. The thesis has been</div> <div>carried out at the Department of Industrial and Materials Science at Chalmers University</div> <div>of Technology, Gothenburg, Sweden.</div> <div>A complete robotic station for automated assembly of balance shaft housings was</div> <div>designed. After the requirements specification was compiled, a investigation was</div> <div>initiated where, among other things, sub-solutions were investigated. The sub-solutions</div> <div>were then used in morphological matrices to generate concepts followed by a elimination</div> <div>and pughs matrix to systematically reduce the number of concepts. The final elimination</div> <div>was done using a kesselring matrix.</div> <div>Function tests and simulations was conducted to verify the functionality. Drawings and</div> <div>CAD files were created. These will be used to build the station as well as a references</div> <div>when maintenance or further rebuilds of the production line is carried out.</div> <div>The finished robotic assembly station will improve ergonomics for the operators as well</div> <div>as offer a possibility to reduce the staff needed to run the production line.</div> <div><br /></div> <div><br /></div>https://www.chalmers.se/sv/institutioner/mc2/kalendarium/Sidor/s_farajzadeh.aspxhttps://www.chalmers.se/sv/institutioner/mc2/kalendarium/Sidor/s_farajzadeh.aspxModal Characterization of 940 nm VCSELs for 3D-Sensing and LIDAR Application<p>Online</p><p>​​Välkommen till Sahar Farajzadehs (MPWPS) exjobbspresentation.Handledare: Anders Larsson and Mehdi Jahed Examinator: Anders Larsson</p>​​Join ​from PC, Mac, Linux, iOS or Android: <span style="font-size:11pt;font-family:calibri, sans-serif"><a href="https://chalmers.zoom.us/j/69777829471">https://chalmers.zoom.us/j/69777829471</a></span><div>Password: 934879</div>https://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Vojic-Jansson.aspxhttps://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Vojic-Jansson.aspxWalter Vojic Frankmar och Anton Jansson, MPSYS<p>Webseminarium</p><p>​ Implementation of visual simultaneous localisation and mapping</p><div>​</div> <div><a href="https://chalmers.zoom.us/j/69218715725">Anslut till seminariet via Zoom.</a></div> <div><br /></div> <div>Examinator: Jonas Fredriksson, Inst för elektroteknik</div> <div><br /></div> <h2 class="chalmersElement-H2">Sammanfattning</h2> <div><br /></div> <div>Autonomous mobile units are increasingly popular in the endeavour of environmentally<br />friendly, safe and self-governing systems. In this pursuit it is often crucial to<br />keep track of the surroundings and the mobile units place in it in parallel, namely by<br />simultaneous localisation and mapping. This task can be done with different sensors,<br />such as LiDAR, IMU, RADAR or cameras, or fuses thereof. However since cameras<br />are relatively cheap and ofttimes indispensable for instance for reading signs, this<br />thesis investigated how this implementation can be done with monocular vision, i.e.<br />using only one camera.<br /><br />The focus of this thesis was mainly the implementation itself of a sparse SLAM<br />algorithm on a JetRacer AI Kit unit using a NVIDIA Jetson Nano. In the development<br />process, the suitability of the platform and the framework was evaluated.<br />The algorithm was based on feature detection and matching between image frames.<br />This matching was performed using an ORB feature detector, in the Python code<br />base pySLAM. For verification of the relevance of the 3D reconstructions, visualisation<br />tools were developed. Finally, the impact of certain design parameters were<br />analysed.<br /><br />The resulting 3D reconstruction of the surroundings, as well as the camera’s poses<br />in it, successfully resembled the real scene up to a scaling factor even despite the<br />occasional unusable blurry frame. By re-scaling the reconstruction using one ground<br />truth measurement, the resulting sparse mapping accurately depicted the distance to<br />feature rich objects in the environment. As for real-time application legitimacy, the<br />algorithm execution speed was dependent on the available computational resources.<br />This issue could most likely be resolved by removing the burdensome visualisation<br />and data analytic tools only important for the development process as well as substituting<br />Python for C, remastering the code base or utilising parallelism features of<br />the GPU instead of the CPU. The performance in relation to the design parameters<br />were typically found to plateau until a critical point from which the reconstruction<br />quality declined linearly rapidly.<br /><br />In conclusion, it was found that Python-based monocular vSLAM is a viable alternative<br />for sparse reconstruction as a base for navigation, but the lack of Python support<br />in combination with the limited JetRacer AI Kit platform architecture makes such<br />an implementation difficult and unnecessarily cumbersome. <br /></div>https://www.chalmers.se/sv/institutioner/see/kalendarium/Sidor/Exjobbspresentation-Cameron-Tinkler.aspxhttps://www.chalmers.se/sv/institutioner/see/kalendarium/Sidor/Exjobbspresentation-Cameron-Tinkler.aspxExjobbspresentation Cameron Tinkler<p>Online</p><p>​Techno-Economic Study of Solid Cycles for Thermochemical Energy Storage and Carbon Capture</p><div>Cameron Tinkler, student på masterprogrammet Nordic Masters in Innovative Sustainable Energy Engineering presenterar sitt examensarbete ”Techno-Economic Study of Solid Cycles for Thermochemical Energy Storage and Carbon Capture”, utfört vid institutionen för rymd-, geo- och miljövetenskap. </div> <div><br /></div> <div>Handledare: Diana Carolina Guio-Perez, Guillermo Martinez Castilla </div> <div>Examinator: David Pallarès </div> <div>Opponent: Gopi Subramanian, Jon Ahlgren, Pontus Wageborn</div> <br /><div><a href="https://chalmers.zoom.us/j/66595822791">Ta del av presentationen via Zoom</a>. </div> <div>Lösenord: 685109​</div> <div></div>https://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Johan-Olsson.aspxhttps://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Johan-Olsson.aspxJohan Olsson, MPEPO<p>Webseminarium</p><p>​ Playing music with an permanent magnet synchronous motor</p><div>​</div> <div><a href="https://chalmers.zoom.us/j/61068464024">Anslut till seminariet via Zoom.</a></div> <div>Lösenord: Thesis</div> <div><br /></div> <div>Examinator: Stefan Lundberg, Inst för elektroteknik</div> <div>Handledare: Joachim Härsjö, Volvo Cars</div> <div>Opponenter: Amer Mesic och Oscar Arvering Walldén</div> <div><br /></div> <h2 class="chalmersElement-H2">Sammanfattning</h2> <div><br /></div> <div>Electric vehicles become more common and since they have no internal combustion engine, they are much quieter which causes more accidents involving pedestrians and cyclists. Demands are made by authorities that electric vehicles should make some sound when traveling below some speed limit. In some electric vehicles the permanent magnet synchronous machine (PMSM) is used as traction motor. Due to a physical phenomenon called magnetostriction the PMSM can be temporarily deformed, and that causes the PMSM to make a sound. <br /><br />The main goal of the thesis was to play music with the PMSM without impacting the average torque production. This was done by testing two different methods that was believed to create sound. The first method superimposed a voltage on top on the voltage that are supplied to the machine. The second method disconnects all transistors in the inverter to create a voltage dip and thereby a resulting current dip. It is found that it is possible to play music on an PMSM with both methods without impacting the average torque within the boundaries of measurement error and that the frequency response of the PMSM depend on what method is used.<br /></div>https://www.chalmers.se/sv/institutioner/fysik/kalendarium/Sidor/Masterpresentation-Olle-Fager-210615.aspxhttps://www.chalmers.se/sv/institutioner/fysik/kalendarium/Sidor/Masterpresentation-Olle-Fager-210615.aspxOlle Fager, MPCAS<p>Online via Zoom</p><p>​Titel på masterarbete: Real-Time Multi-Object Tracking and Segmentation with Generated Data using 3D-modelling Följ presentationen online</p><strong>​Sammanfattning: </strong><div>Multi-Object Tracking and Segmentation (MOTS) is an important branch of computer vision that has applications in many different areas. In recent developments these methods have been able to reach favorable speed-accuracy trade-offs, making them interesting for real-time applications. In this work different deep learning based MOTS methods have been investigated with the purpose of extending the DeepTrack framework with real-time MOTS capabilities. Deep learning methods rely heavily on the data on which they are trained. The collection and annotation of the data can however be very time-consuming. Therefor, a pipeline is developed and investigated that automatically produces synthetic data by utilizing 3D-modelling. The most accurate tracker achieves a MOTSA score of 94 and the tracker with the best speed-accuracy trade-off achieves a MOTSA score of 88. It is also observed that satisfactory results can be achieved in most situations with a quite general data generation pipeline, indicating that the developed pipeline could be used in different scenarios.</div> <div><br /></div>https://www.chalmers.se/sv/institutioner/math/kalendarium/Sidor/Examensarbete210616.aspxhttps://www.chalmers.se/sv/institutioner/math/kalendarium/Sidor/Examensarbete210616.aspxPresentation av masterarbete<p>Online</p><p>​Johan Gustafsson: Entity Extraction From Images</p><p>​<br />Abstract: Websites provide valuable information to threat intelligence analysts operating within cybersecurity. Their work is often carried out manually, requiring searching through massive amounts of data based on uncertain leads. The high variability between analysts further motivates automation to reduce potential mistakes. Information gathering in this space is typically text-based, whereas this thesis aims to assist the effort through computer vision, focusing on entity extraction from screenshots of websites. This methodology applies to general entity extraction, but we concentrate on logotype detection because of its immediate business value. A model using the Faster R-CNN architecture is trained on the QMUL OpenLogo dataset, reaching 55.89 % mAP and exceeding the authors' results by 4.86 % mAP. Since this dataset consists primarily of natural images and does not accurately reflect the actual setting, a novel dataset comprising websites featuring logotypes is created. Given the small number of instances, the model is trained to recognize logotypes in general rather than specifically, reaching 61.7 % mAP.</p> <p>Handledare: Mats Kvarnström, Recorded Future<br />Examinator: Magnus Röding</p>https://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Bengtsson-Sy.aspxhttps://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Bengtsson-Sy.aspxTilde Bengtsson och Tommy Sy, MPSYS<p>Webseminarium</p><p>​ Investigasting irrigation management strategies with reinforcement learning and AquaCrop</p><a href="https://chalmers.zoom.us/j/67062616925"> <div>Anslut till seminariet via Zoom.</div> </a> <div>Lösenord: 170695</div> <div><br /></div> <div>Examinator: Torsten Wik, Inst för elektroteknik</div> <div>Opponenter: Joel Wall och Martin Håkansson</div> <div><br /></div> <h2 class="chalmersElement-H2">Sammanfattning</h2> <div><br /></div> <div>The world's population is expected to increase from 7.7 to between 9.4 to 10.2 billion people in the coming thirty years and a growth of water demand is thereby inevitable. Globally, the largest domain of water demand is agriculture consuming 70 % of the world's water usage. More effective usage of water within the agriculture sector is thereby required. AquaCrop is a crop model software program that allows its user to calculate properties as crop yield is given input parameters like climate, crop cultivar, and irrigation management system. Along with this software, two reinforcement learning algorithm were created which would find the optimal policy representing two irrigation strategies, i.e. net amount irrigation and soil moisture targets. The goal was to find a policy minimizing irrigation amount while maintaining a yield above certain percentage values. In addition, two grid searches were created for comparisons. The reinforcement algorithm was able to find certain optimal policies which fulfilled the lower constraint but failed the higher constraint. <br /></div>https://www.chalmers.se/sv/institutioner/ims/kalendarium/Sidor/Design-and-Development-of-Robust.aspxhttps://www.chalmers.se/sv/institutioner/ims/kalendarium/Sidor/Design-and-Development-of-Robust.aspxDesign and Development of Robust Removable Attachment System for Fastening Hand-held Objects on Tough Gear<p>Online</p><p>​Pranjal Jhaveri &amp; Aarush Bhardwaj presenterar sitt examensarbete. Online presentation.</p><span style="background-color:initial">Program: Product Development (MPPDE)</span><div><span style="background-color:initial">Handledare/examinator: Johan Malmqvist, IMS</span><br /></div> <div><span style="background-color:initial">Opponenter : Abhishek Venkatesh &amp; Rikard Karlsson</span></div> <div><span style="background-color:initial"><br /></span></div> <div><span style="background-color:initial"><div><span style="font-weight:700">Abstract </span></div> <div><span style="background-color:initial">Today multiple products exist in the market that helps fasten things together. The most common ones are buttons, clips, pins, Velcro, hooks, buckles, magnets, straps, and a combination of these. However, all these products need some defined installation (pre-process), e.g., stitching on the fabric to be usable for their respective applications. Lumeo technology AB's novel solution is a device that excludes this pre-process of attaching objects to the fabric. This device is a stand-alone system that can be quickly attached to and removed from the fabric while holding a hand-held object with it (mobile phone, wallet, IoT wearables, etc.). </span></div> <div><br /></div> <div>However,  the existing design is limited in its load-carrying capacity, reducing its utility for some potential applications where the flexibility and ease of removably attaching can be beneficial. This potential application is identified for the first responders (Police personnel, Security guards, Fire-fighters) of the company. This thesis aims to improve the existing design of the device and maximize its load-carrying capacity, also improving other significant functions of the device. This is done in three main phases; in the first phase, the technology of the device is investigated thoroughly and a  brief user understanding; in the second phase performance of the device is investigated and analysed; and in the third phase, concept development is done, resulting in a final concept and prototype. </div> <div><br /></div> <div>The design investigation and development process resulted in intriguing findings of the various parameters that affect the functioning and performance of the device. An improved concept design is tested and suggested based on these parameters, which increases the device's load-carrying capacity and improves the quality of attachment.</div> <div>​</div></span></div> https://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Tingberg-Lindström.aspxhttps://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Tingberg-Lindstr%C3%B6m.aspxAndreas Tingberg och Holger Lindström, MPENM<p>Webseminarium</p><p>​ The advanced ghost-vehicle algorithm</p><div>​</div> <div><a href="https://chalmers.zoom.us/j/61347273658">Anslut till seminariet via Zoom.</a></div> <div>Lösenord: 425655</div> <div><br /></div> <div>Examinator: Jonas Sjöberg, Inst för elektroteknik</div> <div>Handledare: Mikael Andersson, Volvo</div> <div>Opponent: Lorents Landgren</div> <div><br /></div> <h2 class="chalmersElement-H2">Sammanfattning</h2> <div><br /></div> <div>In the strive towards building safer cars, Volvo Cars are developing increasingly sophisticated active safety systems. In the process of developing active safety systems and driver assistance systems, validation is important. To complement real life validation, a virtual environment is used. A driver model, that mimics the behaviours of a human driver, is required if the validation in the virtual environment is to be veridical. In this thesis, the data that enables the driver model to simulate how a human anticipates and plans is generated. This is done using a Ghost-vehicle, that generates a trajectory for the main Ego-vehicle to follow. By allowing the Ghost-vehicle to regenerate a trajectory whenever the Ego-vehicle is obstructed or otherwise triggers on something, the algorithm becomes more fit to handle dynamical elements in traffic. The Advanced Ghost-vehicle Algorithm developed in this project has been successful in generating trajectories for most of the situations it set out to do.<br /></div>https://www.chalmers.se/sv/institutioner/bio/kalendarium/Sidor/Exjobb-FHadi-EAndreasson.aspxhttps://www.chalmers.se/sv/institutioner/bio/kalendarium/Sidor/Exjobb-FHadi-EAndreasson.aspxExjobbspresentation: Fahim Hadi & Emil Andreasson, MPBIO<p>Online</p><p>​&quot;Evaluation and improvement of filamentus fungi for casein production&quot; &amp; &quot;Bioelectroactivity of Clostridium ljungdahlii at different electric potentials</p><strong>​Examinator</strong>: Johan Larsbrink &amp; Lisbeth Olsson<br /><strong>Handledare</strong>: Elena Cámara &amp; Chaeho Im<br /><div><strong>Opponent</strong>: Fahim Hadi &amp; Emil Andreasson</div> <br />https://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Olof-Olivecrona.aspxhttps://www.chalmers.se/sv/institutioner/e2/kalendarium/Sidor/Masterpresentation-Olof-Olivecrona.aspxOlof Olivecrona, MPSYS<p>Webseminarium</p><p>​ Efficient algorithms for temporal logic verification</p><div>​</div> <div>Nytt datum: 2021-08-18<br /></div> <div><br /></div> <div><a href="https://chalmers.zoom.us/j/61793306876">Anslut till seminariet via Zoom.</a></div> <div>Lösenord: 123456</div> <div><br /></div> <div>Examinator: Bengt Lennartson, Inst för elektroteknik</div> <div>Opponenter: Sarah Torstensson och Joakim Lönn</div> <div><br /></div> <h2 class="chalmersElement-H2">Sammanfattning</h2> <div><br /></div> <div>Many real-world systems can be modelled as discrete event systems (DESs), which have a number of events and discrete states. When an event occurs, the system transitions from one state to another. Two ways of representing such systems are transition systems and Petri nets. Finding the reachable states in transition systems is a very central problem within DESs, and solving more complex problems often requires finding them. This makes it very interesting to create an efficient reachability algorithm, which this thesis contributes to by re-implementing a Matlab algorithm in C++ and Python. Interesting differences between the implementations in terms of efficiency are observed. Another important problem within DESs is to verify that systems have certain desirable properties. These properties can be described using temporal logic, where logical formulas can specify not only what should be true in the present, but also what should become true in the future. These specifications can then be verified using nuXmv, which is a symbolic model checker. While transition systems are easy to describe in nuXMV code, Petri nets are not. This thesis presents a parser that translates a Petri net description into nuXmv code, which is shown to greatly reduce the code that the user needs to write. This effectively extends nuXmv such that it may be used for formal verification of Petri nets as well.<br /></div>