Bachelor and Master Theses

To apply for conducting this thesis, please contact the thesis supervisor(s).
Title: Prediction of CPU usage in embedded device using ML and regression techniques
Subject: Embedded systems, Applied Artificial Intelligence, Industrial Systems
Level: Advanced
Description:

Intelligent Electronic Devices (IEDs) are CPU-based embedded devices essential for power grids. These devices collect data from the grid, e.g., current and voltage, to integrate a multitude of fault-detection mechanisms to maximize grid availability. For example, modern IEDs implement functions such as differential current or time overcurrent protection that operators can configure to locate and isolate faulty sections of the grid.

Given the vastly varying needs of power grid operators, several IEDs at different price points and CPU capabilities are presented to the customer to help them reduce operating costs. How[CB1] ever, an operator might create a heavy configuration that exceeds the CPU capabilities of some less potent[CB2]  IEDs. Currently, some cases of overload are not detected before testing preparing for commissioning.

The purpose of this thesis is to find a way to identify if the IED has enough CPU power to execute a given configuration. Already several techniques using machine learning has been used which shows promising accuracy when access to proper training data. However, the data is complicated to collect, creates relatively large models and not complete due to missing conditions in test setup.

As the number of configurations are too large to categorize, a proposed solution is to predict, before execution, if the operator configuration can be executed or not in its IED.

Your tasks will be to:

  • Identify which are the crucial factors and variables in a configuration to predict the CPU usage.
  • Analyze the problem to understand characteristics needed for test and validation of data sets.
  • Compare proposed machine learning techniques between them, and vs. linear evaluation.
  • Evaluate complexity of solution vs. accuracy of results in order to provide a performant model to user.
  • Research different methods to accelerate predictions.
Start date: 2026-01-10
End date: 2026-06-08
Prerequisites:

Prerequisites

Academic Background

1. A solid understanding of digital systems engineering principles.

2. Basic knowledge of machine learning - prior coursework on embedded systems, machine learning / AI.

3. General good skills in programming

4. Proficiency in English, both spoken and written.

 

Knowledge of power system engineering is not required, but it would be a significant advantage.

 
IDT supervisors: Tiberiu Seceleanu Edin Jelacic
Examiner: Ning Xiong
Comments:

The work is to be done in cooperation with Hitachi Energy, largely at the company's premises in Västerås.

According to Hitachi Energy a standard compensation may apply.

 

 
Company contact:

The supervisor at Hitachi Energy is Sachin Srivastava, Ph.D.