Bringing Energy-Efficient AI to Embedded Devices
How can we bring intelligent machine learning directly to small, resource-constrained devices?
Description
AI is increasingly moving beyond cloud environments and into sensors, machines, and embedded devices. This creates exciting opportunities for industrial applications, but also introduces challenges involving energy consumption, memory limitations, and computational performance.
At Knowit Connectivity, we are offering a thesis opportunity to develop a reusable toolchain for Edge AI and TinyML applications in industrial sensor systems.
About the Project
The objective is to develop a modular engineering process that makes it easier to create, optimize, deploy, and evaluate machine learning models on embedded hardware.
Rather than building a single AI application, you will investigate how to create a reusable approach that works across different sensors, hardware platforms, and industrial use cases.
Potential applications include anomaly detection, machine fault identification, equipment-state classification, and predictive maintenance.
An important part of the project is understanding how to balance model accuracy against computational requirements, memory usage, inference speed, and energy consumption.
What You Will Explore
Your work may include:
Developing reusable workflows for sensor data preparation and feature extraction.
Training and comparing machine learning models for industrial sensor data.
Exploring TinyML techniques such as quantization, pruning, and model optimization.
Building tools for model conversion, packaging, and deployment to embedded devices.
Measuring inference latency, memory consumption, and energy efficiency.
Investigating event-triggered and energy-aware inference strategies.
Evaluating the toolchain across at least two representative industrial applications or datasets.
Developing documentation and processes that allow engineers to reuse the solution.
Who We're Looking For
This thesis could suit students in Computer Science, Artificial Intelligence, Machine Learning, Embedded Systems, Engineering Physics, or related technical disciplines.
Relevant interests include:
Machine learning and neural networks.
Python and AI development frameworks.
Embedded systems and microcontrollers.
Signal processing and time-series data.
Model optimization and performance benchmarking.
Industrial applications of AI.
You should enjoy experimenting with different technical approaches and evaluating how they perform under real hardware constraints.
Expected Outcomes
The thesis aims to produce a reusable Edge AI toolchain, a documented development process, and working demonstrations using representative industrial sensor applications.
You will evaluate how different models and optimization strategies affect performance, resource requirements, and energy consumption.
The project should provide practical guidance for future engineers developing AI applications for embedded and industrial environments.
The final scope and research questions will be defined together with your supervisors.
Interested in This Thesis?
Want to explore the intersection of artificial intelligence, embedded systems, and energy-efficient computing?
Apply through our thesis application page and indicate your interest in "Energy-Efficient Edge AI and TinyML".