Advancing sustainable repairs: Development of an AI-driven fault detection model As part of the EU-funded ‘CIRC-UITS’ project, we are working together with BOSCH on an exciting sub-project: We are developing a computer vision model for the diagnosis and fault detection of control units used in cars - for example for ABS or ESP. The aim? To increase sustainability in the electronics industry! An interview with Alexandr Bystrov, student at the University of Oldenburg and junior researcher in the Production department at OFFIS.
Aleksandr, you are part of the CIRC-UITS project. Can you explain what this project is about and what role you play in it?
The CIRC-UITS project, which stands for ‘Circular Integration of Independent Reverse Supply Chains for the Smart Reuse of Industrial Relevant Semiconductors’, aims to make products with printed circuit boards (PCBs) more sustainable. In collaboration with BOSCH, we are working on a diagnostic model for electronic control units (ECUs). The project evaluates both the theoretical and practical reparability of products, and one of our pilot projects concerns Bosch ECUs used in systems such as ABS and ESP in cars.
What is the specific aim of this sub-project?
The aim is to develop a computer vision model that recognises faults in seven critical components of ECUs. Essentially, this model will classify these components as either functional or faulty. By using machine learning, we want to eliminate non-repairable products as early as possible and thus increase the sustainability of the repair process. We not only optimise the repair itself, but also reduce the unnecessary consumption of resources for transport, cleaning and sorting. Ultimately, the key question is: ‘Is this product worth repairing?’
You mentioned that this project will develop a model of computer vision. Where are you in this process?
We are in the early stages - building a solid framework for image capture and labelling. This may sound simple, but we have already faced challenges. For example, some ECU components are reflective while others are not, which complicates the exposure settings for image capture. We have developed several approaches to solve this problem, and each will result in the creation of multiple data sets. These datasets will later be compared against each other based on model accuracy. Despite the challenges, this phase makes perhaps the most boring part of an image processing project - labelling - even more exciting.
You started working on this project at OFFIS in 2024. What experiences have you had so far?
I joined OFFIS in mid-July 2024 and was immediately involved in the project. My first tasks focused on a literature review and researching potential challenges such as class imbalance, data augmentation and model architecture. Now I'm working on both the theoretical and practical aspects of the project, which is helping us to find a good balance as we move towards our goals. One of the exciting aspects is the number of challenges we face: the complexity of the calculations, the imbalance of classes and the small dataset we currently have for training. These are both motivating and challenging, which keeps me focused.
Is your degree programme helping you with this project?
I am currently studying for a Master's degree in Applied Economics and Data Science at the University of Oldenburg, which I started in October 2023. This project allows me to put my theoretical knowledge into practice, which is a fantastic learning experience. The team at OFFIS has been incredibly supportive and I am lucky to have such a positive working environment. It's also great to be able to build on the skills I've gained on my Master's programme.
It sounds like you've experienced a lot of changes in the last year. Would you like to talk about it?
Almost a year ago, at the age of 32, I started a new chapter in my life: I changed jobs, countries (I'm from St Petersburg, Russia), languages and people. I started my studies at the University of Oldenburg in the Master's programme Applied Economics and Data Science. During this time, I gained a lot of valuable theoretical and practical knowledge. To cut a long story short: I enjoy studying at UOL from a multidimensional perspective - studies, people, city and facilities. As a working student, I now appreciate this opportunity even more. OFFIS is a well-known potential employer among students, and I first heard about it when I was advised about subject-related job opportunities during my studies. I was lucky enough to get a job there in my second semester.
And how do you spend your time outside of work and studying?
In addition to my studies and work, I prioritise the following in my free time: Meditation, playing the guitar, yoga and road cycling - although learning German is of course the top priority. All of these activities are great, but meditation is my life's passion. Maybe one day I'll tell you more about it.
