A federated learning and blockchain-based data sharing framework for developing intelligent building design models

Abstract

Intelligent building design can reduce manual work and streamline the design process by automatically generating design content using artificial neural networks (ANNs). However, it is challenging to collect sufficient drawings to develop a high-performance ANN. Data owners may not be willing to share their drawings with untrusted parties due to privacy considerations. To address these challenges, this paper proposes a novel data sharing framework of confidential building design information to facilitate the development of intelligent auxiliary building design models. The data sharing framework utilises the federated learning technique and blockchain technology to encourage data sharing through fair benefits allocation based on the Shapley value. A case study was conducted to evaluate the effectiveness and feasibility of the proposed framework. The results show that the intersection over union is improved by more than 10%. More benefits are allocated to data owners who provide datasets with higher quality and quantity. Methodologically, the paper should facilitate the effective integration of the fragmented and confidential project data to train building design models and add much value by addressing the data sharing complexity and dynamics in modern construction. Practically, the paper demonstrates a novel way to train auxiliary design models for building designers.