master thesis or bachelor thesis

Are you a student and keen to gain practical experience in an innovative environment?

As a job student at Vintecc, you will work on real projects and discover how software, AI and engineering come together to make companies smarter and more efficient. A perfect opportunity to gain knowledge, expand your network and prepare yourself for your professional career.

What to expect:

✔ Practical experience in a high-tech environment

✔ A cool and instructive work experience

✔ A team open to your ideas

Do you have a topic in mind or are you curious about our proposals?

Get in touch and we will be happy to discuss!

You can find some approved topics below:

COMPUTER VISION

  1. Low-Latency Logging
    • Goals
      • Enable fast communication between processes. This includes buffering for optimal packet size, buffering during communication interruptions, fall-back paths, error handling, and so on.
  2. Sensor Fusion
    • Goals
      • To map information from different sensors (visible light, ultraviolet, infrared, radar, lidar) onto each other in 3 dimensions.
      • Compare early vs late sensor fusion for deep learning on calibrated inputs.
  3. Learning Visual Embeddings
    • Goals
      • Re-identify objects (re-ID).
      • Classification of new classes without re-training the entire network.
  4. Keypoint Detection
    • Goals
      • The goal of this stage is to create a Python package that can effectively train and infer on COCO datasets with key point annotations.
  5. GANs for synthetic data
    • Goals
      • The goal of this internship is to create a workflow for data augmentation that seamlessly integrates with an existing Python package. With this package, we can easily create realistic-looking datasets starting from synthetic data.
  6. Copy-Paste Data Augmentation
    • Goals
      • The goal of this internship is to create a data augmentation workflow that seamlessly integrates with an existing Python package. This package would allow us to easily create new datasets with real or synthetic backgrounds or objects with their masks. It would also allow us to perform augmentation on-the-fly, during the training phase of deep learning.
  7. 3D Deep learning at Point Clouds
    • Goals
      • The goal is to create a Python package that contains tools to work with (load & save / transform / sample / visualise) 3D point clouds. These points can contain coordinates, intensity, colour, normal vectors, ... contain. They can be structured or unstructured.
      • Another important goal is to directly perform semantic/instance segmentation on these 3D point clouds.

Profile/required skills

  • Education: computer science student industrial engineer or computer science student.
  • Be proficient in the Python programming language (and have experience with C/C++).
  • Background knowledge about computer vision and practical experience with deep learning. Apply below!

SIMULATION/DIGITAL TWINS

  1. Tuning of physical parameters
    • Goals
      • The aim of this thesis is to investigate and develop best practices for tuning a virtual physical simulation to an existing ‘counterpart’. Ideally, this will result in a methodology for addressing tuning and various methods for (potentially) automating parts of this task (auto-tuning).
    • Profile / skills required
      • Electromechanics / Mechanics / Control & Automation/...System identification / Control engineering / Modelling / Mechatronics/...Basic skills in programming and software development

MECHATRONICS & CONTROLS

  1. Scalable Test Framework for Matlab/Simulink Models
    • In the development of autonomous systems using Matlab/Simulink, reusable libraries are often employed to enhance efficiency. These libraries contain essential functions such as motion control and sensor processing but quickly grow in complexity. To ensure stability and seamless integration into projects, a robust and scalable test framework is essential.
    • This thesis focuses on developing a flexible test framework that enables efficient verification of library functions.
    • Key objectives include:
      • Establishing a scalable test environment in Matlab/Simulink.
      • Automating tests using Functional Mock-up Units (FMUs) to accelerate simulations and enable execution independent of Matlab.
      • Facilitating efficient test expansion for future developments and evolving requirements.
    • This project lays the foundation for a sustainable and reusable testing process, ensuring the long-term reliability and stability of Matlab/Simulink libraries.
  2. Acceleration of Simulations with Reduced Order Modelling
    • Context
      • In model-based software development, physical systems are simulated to test control software. However, high-fidelity simulations require significant computational power and time. This thesis explores how Reduced Order Modelling (ROM) and Physics-Informed Neural Networks (PINN) can increase simulation speed without significant loss of accuracy.
    • Objectives
      • Accelerate automated testing in CI/CD by replacing computationally intensive models with efficient surrogate models.
      • Identify bottlenecks in simulation time.
      • Explore ROM and data-driven techniques for physical models.
      • Evaluate the trade-off between computational speed and accuracy.
    • Methodology
      • A literature study will analyse existing ROM and PINN techniques. Subsequently, several Simulink models will be selected to test suitable ROM methods. The accuracy and speed of the surrogate models will be evaluated in different test scenarios.
    • Result
      • A demonstration of ROM and PINN, with a comparative analysis of accuracy and computational efficiency for physical simulations.
  3. Trajectory Optimization for Yarn Tension Control in Composite Material Weaving
    • Context
      • When weaving high-tech composite fabrics, weft yarns such as carbon or glass fibers must remain free of twists or kinks. This is achieved by actively regulating yarn tension using a buffer arm positioned between the bobbin and the loom. The bobbin unwinds the yarn at a constant speed, while the buffer arm adjusts the required yarn length to maintain stable tension.
      • After each cycle, the buffer curve is updated, determining how much yarn needs to be buffered. A feedforward curve for the buffer arm is then generated. Issues arise when the buffer arm needs to buffer more yarn than physically possible or too little. In these cases, the bobbin speed must be adjusted, leading to high torque peaks that negatively impact both the weaving process and the motors.
    • Objective
      • This thesis aims to develop a real-time algorithm that optimally combines the feedforward curves of both the buffer arm and the bobbin based on the buffer curve. The goal is to minimize acceleration and torque peaks while ensuring all weaving constraints are met.
    • Methodology
      • Analysis Phase: Study of the current algorithm and literature review of state-of-the-art optimization techniques that address similar challenges.
      • Implementation Phase: Development, testing, and validation of the algorithm using real data. If successful and time permits, integration into the machine software and testing on the actual machine will follow.

apply now

at Vintecc, you get the chance to work on your master thesis or bachelor thesis.