The AI factory is driving the digital transformation of building services engineering
The AI Champion project, funded by Business Finland, is developing 100 AI agents designed to improve automation and the flow of information within building services supply chains. Researchers estimate that around 30 AI agents have already been created during the first part of this year.
The building services engineering and civil engineering field have entered a phase where artificial intelligence no longer merely analyses data or answers questions, but is capable of participating in entire work processes. At the heart of this development are AI agents and multi-agent systems, which are being used to automate design, production control, quality control and project management.
“An AI agent is a piece of software that utilises a language model and is capable of drawing on various data sources, using software and progressing independently through multi-stage tasks. In the AI Champion project, the development of the agent begins with identifying the problem, after which the consortium proceeds in accordance with the agreed use case process,” says Jussi Rasku, the work package leader, the deputy director of the GPT-Lab research group at the Tampere University, and a postdoctoral researcher.
The project will produce so-called customer service agents with whom a person can converse, as well as agents that streamline workflows either independently or in collaboration with other agents as part of a multi-agent system. In a multi-agent system, each agent has its own role, but they exchange information with one another and form a shared understanding of the situation.
“One new type of multi-agent system is the AI factory. Through this system, the automatic collection, processing and analysis of data by agents leads to the creation of an AI model within the factory. The model can be developed and tested using a virtual twin before it is rolled out for industrial use,” Rasku explains.
The first AI factory built to optimise HVAC control
The first researcher exchanges with companies as part of the AI Champion project took place in the summer. Project researcher Rajratan Wankhade, in collaboration with Koja-Chiller, built a system to control the ventilation units at the AI factory at Koja’s Jalasjärvi plant. Koja provided the AI researchers with access to data from three air-handling units within the production facility.
“The AI factory, which was set up over the summer, operates with around seven different agents, each with their own specific tasks. Machine learning models also play a key role, as all predictions and control decisions happen by automatically choosing the best ML models based on actual sensor data from the building,” Wankhade explains.
As an example, he cites the adjustment of ventilation according to population size, which helps save energy.
“One agent draws the topography of a building, which helps in understanding the size of a room, for example. Another agent collects data from such topography, another assesses whether the data can be used or is accurate, a third directs the other agents, and the final agents plan the agentic workflow and select which machine learning model to use. The chain goes on and on, illustrating how the agents and machine learning models work together and overlap,” Wankhade says.
The aim of this work is to develop a fully autonomous multi-agent system – in other words, an ‘AI factory’ – which generates an intelligent machine learning model for regulating ventilation in real time.
Over the summer, the performance of the machine learning models produced by the AI factory was measured using a data-driven virtual twin.
“The next logical step in the Koja-Chiller collaboration is to test the control models produced by the AI factory developed by Wankhade on actual air-handling units,” Rasku continues.
“The importance of AI agents is growing, particularly in construction projects, where vast amounts of data are generated by various parties and decisions must be made quickly throughout the project lifecycle. It is extremely valuable that, as part of this project, we are able to develop and test solutions in a real-world project environment in collaboration with companies in the sector. In this way, we can accelerate the digitalisation of the construction sector and establish practices that improve information management, collaboration and productivity in construction projects,” Associate Professor and project coordinator at Tampere University Piia Sormunen says.
The AI Champion consortium meets three times a year to share information on what has been developed, tested and learnt within the framework of the project. The final meeting of the year is AI Champion Day, to which we extend a warm welcome to anyone interested in the field. We recommend following the project’s
LinkedIn account
At the AI Champion project’s Show and Tell day, a panel discussion took place on the role of humans in the development of artificial intelligence. You can find out more about the discussion by watching the recording of the event (subtitles in English)
