Banca de DEFESA: Raí Luz Barbosa

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : Raí Luz Barbosa
DATE: 30/09/2022
TIME: 09:00
LOCAL: Plataforma Teams
TITLE:

Semantic Enrichment of BIM Models for Room Classification Using Machine Learning Techniques

KEY WORDS:

BIM, semantic enrichment, machine-learning, multiclass classification, ensemble.

PAGES: 97
BIG AREA: Engenharias
AREA: Engenharia Civil
SUBÁREA: Estruturas
SUMMARY:

The BIM process emerged in the AEC industry with the proposal to develop a digital model of a building to assist the project conception, the building process and the management and maintenance of the building. These models have a large amount of information that constantly needs to be exchanged between the various disciplines involved in the project. As these disciplines use different tools for reading the model’s information, it is important to guarantee a complete exchange of information to avoid correction and/or addition of data. The nonproprietary IFC file was created so that there would be interoperability between the different software that are based on the BIM process. However, due to the complexity of the digital models and the wide variety of software that may be involved in the project, it is still common for information to be lost. Because of that, there was a need to use methodologies that would help in the interoperability process and promote the semantic enrichment of IFC files. Among these methodologies, machine learning has been gaining ground, with its various techniques that use the learning from experience process to predict new information. This paper proposes and validates machine learning (ML) models and strategiesto promote the semantic enrichment of Building Information Modeling (BIM) to improve interoperability among architectural and engineering applications. Various ML techniques (k-Nearest Neighbor, Bagged Tree, SVMGaussian, SVM-Quadratic e SVM-Cubic) were implemented to classify room types of residential building models. In addition to traditional binary classifiers, it was proposed two voting techniques, namely multiclass-binary and ensemble, to improve the accuracy of the classifications and decrease overfitting during the training process. Real architectural design data of residential buildings with 8 different classes of rooms were used to train the classifiers in scenarios with different feature variables selection. The results demonstrate that the voting techniques can be successfully used as semantic enrichment strategies to accurately predict the residential room classes using few and easy-to-obtain features from IFC files. The proposed strategies and techniques improved the interoperability of BIM models in efficiently classifying residential rooms without the need for human intervention.

BANKING MEMBERS:
Presidente - 2047809 - FRANCISCO EVANGELISTA JUNIOR
Interno - 1752368 - ANTONIO CARLOS DE OLIVEIRA MIRANDA
Externo à Instituição - KLEOS M LENZ CESAR JUNIOR - UFV