Bachelor and Master Theses

Modern industrial systems are the result of joint efforts coming from a multitude of disciplines, like mechanics, electronics, software, and more. These domains were once clearly separate and managed in "silos", as their interactions could be clearly predicted. With the progressive ubiquity of software such a clear separation has becomes more and more blurrier: nowadays a large part of efforts are put in decision making, where it is critical to understand how domains interact with each others and what consequences a certain design choice might have for the resulting system. It is in these phases that even the most powerful LLM will never be able to substitute humans' contribution in a fully reliable way. Nonetheless, LLMs can be useful tools to reason about available information, elicit potential issues, etc., but need to be provided with contextual information about the settings in which they operate.

In this thesis we want to start exploring the encoding of contextual information for aiding industrial systems engineering. In particular, we want to investigate what information would be useful to provide in order to support decision making in integration phases. The overall plan could be drafted as follows:

- take as input documents from different sources (requirements specifications, Excel spreadsheets, standards, etc.);

- encode the inputs into an appropriate knowledge graph;

- use the information stored in the knowledge graph to reason about different design alternatives (i.e. decisions).

This process might be augmented by using LLMs, that given the potential sensitiveness of handled data, would be open-source/locally executable alternatives. The LLM might be used to both to extract the inputs to be encoded into knowledge graphs and to reason about the available information. Moreover, since humans are required to be kept in the loop, the different design alternatives are represented as SysML v2 models.

The thesis will be developed in collaboration with Volvo CE, which will provide realistic scenarios in terms of document to take into account, types of decisions, limitations about usable tools/solutions. The student(s) will collaborate with the industrial contact in looking for suitable validation scenarios.

A demonstrated knowledge of software engineering principles is required.

Demonstrated previous experience in applications development is required.

Previous experience in software and system modelling (UML, SysML, MatLab, etc.) is considered a plus.

Previous experience with knowledge graphs, ontologies, and/or LLMs-augmented development is considered a plus.

This thesis is also suitable for two students.

Volvo CE contact (co-supervisor): Johan Cederbaldh, johan.cederbladh@volvo.com