Through a combination of machine learning and a generic building performance model, it possible to quickly gain insights into the indoor climate conditions of the majority of rooms in a building. With support from ELFORSK, Artelia and Aalborg University are working to develop a new method for indoor climate calculations that will be particularly valuable in the early design phases.<\/p>\n<\/div>
The design team must meet increasing requirements for indoor climate while simultaneously ensuring a low energy demand. Finding the perfect balance between these is a complex task. The design team has several tools they can use when ensuring thermic comfort, adequate daylight, good air quality, etc. The indoor climate may be influenced by a long list of design parameters related to window size and properties, solar shading, ventilation system, constructions, and user behaviour.<\/p>\n
To ensure optimal indoor climate solutions, the preferred tool is currently building simulations. However, due to the complexity of indoor climate calculations, this is often very time-consuming. As a consequence, building simulations are often not utilised until late in the project by which decisive decisions have already been made. Quite contradictory, this results in a lot of iterative adjustments that often turn out to be equally resource demanding. In consideration of the time available, a limited number of simulations are usually made for a few, critical rooms, assessed to largely represent the building. Put simply, the majority of rooms are designed to comply with the worst-case scenarios and this leaves unresolved potential for the rooms that appear less problematic.<\/p>\n
The industry needs new methods for automation of this process that will allow for the best solutions to be implemented at the right time, early in the process. A method that will simultaneously contribute to a more in-depth assessment of the indoor climate conditions as a whole.<\/p>\n
A holistic approach The project group has first set up a generic space for an open-floor office which makes up a large quantity of new commercial developments. Open floor offices are known for high occupant density, heat-emitting IT-equipment and for having high requirements for daylight utilisation. The machine learning model for the generic office space is trained on 70,000 simulations which reflect industry guidelines, standards, and a statistical analysis of 11,000 indoor climate simulations of office buildings. From this data, 37 influential design parameters have been identified.<\/p>\n Using the so-called Monte Carlo approach, it is possible to change and adapt all the different parameters simultaneously and see how they influence one another, the energy demand, and the indoor climate. The ability to rapidly change the many design parameters, means that the generic room can estimate the performance of the majority of office spaces. As the design combinations are endless, the simulations are subsequently used to create fast and precise meta-models using machine learning.<\/p>\n Promising results More user-friendly \u00a0<\/strong><\/p>\n<\/div>
\n<\/strong>Artelia and Aalborg University are working to develop exactly that. A new method, with the main objective to set up an indoor climate model in the form of a generic room that can be adjusted and changed quickly via machine learning, allowing for an examination of a large number of rooms in the building.<\/p>\n![\"\"](\"https:\/\/www.arteliagroup.dk\/wp-content\/uploads\/2021\/06\/Elforsk-nyhed-Figur-1-scaled.jpg\")
<\/p>\n
\n<\/strong>To test the concept, the team compared a machine learning model of the generic room to a completed model of an office room with precise geometry, placement of windows, shade conditions, etc. The objective was to determine how easy it was to see the effect of different design variations and measure their influence on the total energy demand and the indoor climate. The results were positive and strongly indicate that the machine learning models allow for direct feedback on all potential variations for a great number of rooms in a building. This makes the concept very useful in the early project stages and the assessment of different design options while, at the same time, allowing for assessment of a lot more rooms. The fact that there are minimal deviations from the simulation results from the detailed rooms means that the potential deviations are easily managed in the later design stages.<\/p>\n
\n<\/strong>The machine learning model of the generic office room can be used in several ways without ever using a simulation software, which is usually required. Halfway through the research project, the concept was made accessible for Artelia\u2019s consultants via the Google Colab platform. Thus, the concept is already being tested on actual cases. This provides valuable feedback that supports the continued development of the method. In the future, the models will be made available via an API, to be used for parametric studies in BIM tools such as Dynamo or Grasshopper. This will make it possible for both architects and engineers to gain access to fast feedback in the early design-shaping phases and structural analyses. This is a considerable advantage as BIM models, because of their visual capabilities, are the preferred platform for collaborations in the building industry.<\/p>\n