PRR Project

Summary

The role involves exploring advanced computational models alongside point clouds, to enhance productivity and competitiveness in the Architectural, Engineering and Construction (AEC) sector. The focus is on extracting information from point clouds to augment existing workflows, leveraging BIM and AI. The prospective employee will perform tasks in three core areas: 1) create point cloud datasets to train, test and validate AI models; 2) develop AI models; and 3) apply the developed models to distinct AEC workflows.Area 1 addresses the shortage of open-source point cloud datasets in the literature by creating a dataset. This dataset will serve various purposes and requires proper labelling to answer them. For example, building elements will be individually labelled with class, instance (e.g., column, beam), surface material, geometric and structural information. Given the impracticality of manually creating the dataset, the employee will collaborate with other researchers to crowdsource its development. This involves issuing examples and guidelines to support contributions, grounded in research.Orientation and co-orientation of MSc and PhD thesis focusing on point clouds will also be a key task supporting the dataset, with BUILT leveraging its link to the academia to foster these relations. The employee will also develop automatic mechanisms for dataset augmentation, such as BIM-to-point cloud workflows, to create realistic synthetic data. Researching the integration of existing datasets (e.g., S3DIS, Semantic3D) will also take place, to enhance collaboration with other researchers, enable their work, and easily incorporate their innovations. Lastly, integrating different sources of data in the dataset (e.g., thermography, depth and infrared images) will also be focused, given the multimodal models promising performances in recent literature, despite their lacking research.Area 2 aims to advance various current point cloud-based AI problems, such as noise removal, registration, segmentation, classification, generation, and completion. Despite the focus on the AEC sector (e.g., segmenting building elements, recognizing construction equipment), this area will emphasize creating neural networks to address these challenges across multiple domains. Initially, the research will focus on 3D Convolutional Neural Networks, with subsequent exploration of different architectures (e.g., point-, voxel-, graph-, projection-based) and their unique answers to irregular and sparse data. Special attention will be given to Transformers and hybrid approaches.Since accelerating these networks is crucial for some workflows in Area (3) that demand real-time processing, techniques for diminishing computational demands (e.g., pruning, quantization, low-rank approximation, tensor decomposition, distillation) will be expanded. Dataset limitations (tackled in Area 1) will also be considered, with one-shot, transfer and ensemble learning techniques being researched to further the literature.Area 3 applies and fine-tunes the developed models to distinct problems facing the AEC sector, including: automated i) scan-to-BIM and ii) construction monitoring. BUILT has established itself in the field of scan-to-BIM, and the research aims to expand on this expertise. Specific areas of focus include Historic BIM modelling, virtually unexplored, and the identification of materials for integration in BIM. Regarding construction monitoring, the real-time analysis of the construction quality and progress by comparing as-built point clouds with as-planned BIM models should be further expanded, given recent advances that enable this technology. Particularly, the uptake of 5G allows large amounts of data to be transferred from site and the integration of 4D data in Industry Foundation Classes files fosters the usage of OpenBIM at reduced costs.Additionally, the employee will also: draft technical and scientific texts, promoting the dissemination of the achieved results in the industry and academia; participate in national/international conferences and events; research new technologies and methodologies that enhance the conducted work and keep it at the forefront of both industry and science.To fulfil these tasks, the prospective employee must showcase excellent domain over the Portuguese and English languages. Additionally, the employee must have several critical soft skills expected of an established researcher for both individual and team-oriented work, namely, excellent communication, organisation, autonomy, proactivity, and teamwork. Motivation and dedication are also key to have. Knowledge and hands-on experience with BIM, AI and point cloud data are also a must. Programming with IFC and .NET frameworks are expected.Altogether, hiring for the indicated area will bolster BUILT capabilities to support the academia and industry in tackling current state-of-the-art problems, which solutions can solve long standing issues in the AEC sector.