Skip to main content Skip to main navigation


Predicting semantic building information (BIM) with Recurrent Neural Networks

Burak Mete; Jessica Bielski; Viktor Eisenstadt; Christoph Langenhan; Frank Petzold; Klaus-Dieter Althoff
In: European Conference on Product and Process Modeling. European Conference on Product and Process Modeling (ECPPM-2022), ECPPM, No. 14th, European Association of Product and Process Modelling, 9/2022.


Recent advances in technology established artificial intelligence (AI) as a crucial domain of computer science for both industry and research, but also for everyday life. However, while computer-aided architectural design (CAAD) and digital semantic building models (BIM) became essential aspects of the contemporary architectural design process, AI cannot be seen as a leading support-ive computational method due to its absence in the established design software and the challenging acquisition of proper data. An option to acquire rich design data, for example in the form of time slices and relations of atomic design steps, is the reproduction of design protocol studies (Lawson, 2004). However, this data is still unstructured and requires a framework for pre-processing and training artificial neural networks (ANN). In this paper, we present our research on BIM and AI, dedicated to autocompletion of design steps for architectural design, based on the methods of the [omitted] projects. Autocompletion is achieved through the suggestion of further design steps to improve the quality and speed of the design process of the early design stages. It is inspired by other autocompletion methods that have been applied for data-driven decision-making. Assuming the position of Lawson (Ibid.), we propose an approach for a recurrent neural network (RNN) model to predict future de-sign steps through sequential learning. Thus, we propose a model based on cognitive sequences of the architectural design process as relational sequences (Lawson, 2004), using sketch data quantified through custom labelling via an open-source tool assigning the respective design phase (Lawson, 2004; Laseau, 2000). We adapt to the idiosyncrasies of the user by identifying the current cognitive processes to predict further mental activities and thus, future design steps.