Master Thesis Proposal: TinyML for Predictive Maintenance in Embedded Systems

AFRY is a European leader in engineering, design, and advisory services, with a global reach. We accelerate the transition towards a sustainable society. With 19,000 experts in infrastructure, industry, energy, and digitalization, we create sustainable solutions for generations to come.

Many of our employees started their journey at AFRY through their thesis work. It’s a great opportunity to build valuable relationships and explore future career paths.

Location: Solna, Stockholm, Sweden

Start: Spring 2026

Workload: Full-time (30 ECTS)

Language: English

Number of students: 2

Thesis Topic

We are looking for two master's students to explore the use of TinyML for predictive maintenance and fault diagnosis in resource-constrained embedded systems.

The project will start by designing and building a small controllable electromechanical system that can operate normally as well as reproduce different types and levels of faulty behavior. The students will then investigate how sensor data from the system can be used by machine-learning models running directly on a microcontroller to detect and classify faults. The thesis therefore combines two main areas:

- Design and implementation of a controllable embedded test system
- Development and optimization of TinyML models for fault diagnosis

The final demonstrator should be able to monitor the system, identify abnormal behavior, determine the likely type and severity of the fault, and provide an appropriate maintenance recommendation.

We are looking for two students in their final year of a relevant master’s program. You can apply individually (to be paired with someone) or together with a partner.

Responsibilities

Develop the embedded software for real-time system control, fault injection, and synchronized data acquisition.; Implement configurable and reproducible fault conditions with different severity levels.; Build a labeled dataset covering normal operation and the selected fault conditions.; Investigate suitable machine-learning approaches for fault detection and classification using embedded sensor data.; Develop and train models using Python and relevant ML frameworks.; Investigate different sensor combinations and their impact on fault-diagnosis performance.; Deploy suitable models on a resource-constrained microcontroller.; Evaluate the trade-offs between accuracy, memory consumption, computational requirements, latency and energy consumption.; Investigate model optimization techniques such as quantization, pruning and, where appropriate, knowledge distillation.; Evaluate the resulting models on the physical system under different operating conditions and fault severities.; Investigate how detected faults can be mapped to appropriate maintenance actions.

Requirements

Suggested master programs: Engineering Physics, Embedded Systems, Computer Science with focus on Graphics or High-Performance Computing, Electrical Engineering, Data Science with specialization in Signal Processing or Electronics, Systems, Control and Robotics.; Good programming skills in C/C++ and Python.; Experience working with microcontrollers and embedded systems.; Familiarity with software development using Git.; Interest in combining software, hardware and machine learning.; Ability to work independently and systematically.; Good analytical and problem-solving skills.; Good communication skills in English, both spoken and written.

Nice to have

Experience with RTOS or real-time embedded systems.; Familiarity with TinyML or embedded machine-learning frameworks.; Experience with sensors, signal processing or data acquisition.; Basic knowledge of machine learning and model training.; Experience with electronics or motor control.