This newsletter summarises the activities carried out and the results achieved in MONA LISA between July and December 2025. The period was dedicated to establishing the common foundations of the project: shared requirements, a common data format and the architecture for analysis and visualisation.
Project management
The first project deliverables were released: D1.1 Project Management Plan, D1.2 Data Management Plan and D1.3 Project Consortium Agreement. Technical coordination advanced to the definition of the MONA LISA Technical Items, which serve as the central technical milestones of the project and allow progress to be tracked consistently across all partners. The consortium also prepared the first annual project review, which took place in January 2026.
Industrial use cases
Dedicated working groups were set up for each use case provider to ensure close alignment with the industrial needs and the early identification of technical challenges. Deliverable D2.1 Use Case Requirements was submitted, describing the abstract and specific use cases of the project. The refinement of the use cases into detailed scenarios, to be reported in D2.2, was started.
Technical progress
Common trace data format
The partners agreed to use a subset of the Google Trace Event Format (GTEF) as the common format for run-time data. HCL defined the subset required for traces captured by Model RealTime and Code RealTime and published an open-source script on GitHub that converts these traces into GTEF. The converted traces were successfully tested in the Chrome Trace Viewer and in Perfetto. Work was also started on trace filtering, as very large trace files can exceed the limits of trace viewers and downstream analysis tools.
Model re-engineering, analysis and visual analytics
The following deliverables define the technical basis for the coming project phases:
- D4.1 defines the models of computation and the tools used to process them. The model re-engineering framework will support several modelling formalisms and will use large language models to reconstruct system models from requirements, design and test documentation.
- D5.1 maps the analysis needs of the use cases, root cause analysis of defects and predictive analysis, to the solutions available in the consortium.
- D6.1 specifies the architecture and requirements for visual analytics. The architecture connects the systems developed in the technical work packages as data sources and uses Eclipse Trace Compass for processing and visualisation.
Language models for testing PLC software
A controlled experiment investigated whether large language models can generate and amplify test suites for PLC programs written to the IEC 61131-3 standard. Using a locally hosted model and a data set of 53 industrial function block diagram programs, test generation and two forms of test suite amplification were evaluated with zero-shot and few-shot prompting and different input formats. The results indicate that AI-assisted methods can accelerate test development and increase test coverage.
Data collection at the BDÇ steel plant
For the Bilecik Demir Çelik use case, the data to be collected from the induction furnaces and the communication protocols were identified. The hardware to be integrated and the standardisation of data formats were planned, and a detailed schedule for the data collection activities was prepared.
Publications
The paper Benchmarking Large Language Models for Root Cause Analysis in Train Control Software Testing by researchers at Mälardalen University was published at MoDeVVa 2025. It evaluates seven large language models on execution logs from Alstom's digital twin-based testing environment. The paper is available on the publications page.
Next steps
- Collection of example data from every use case provider and conversion into the common format.
- Completion of the conversion tools, including validation of the completeness and consistency of the converted traces.
- Implementation of the data processing workflows defined in D4.1.
- Continued experimentation with large language models and AI-based analysis for root cause analysis and predictive analytics.