An artificial neural network-based model for material defects classification in oil and gas projects in Saudi Arabia

Maintaining the quality of materials supplied in construction projects is challenging, prompting government and businesses to dedicate substantial efforts and resources to ensure the highest quality standards. Any material defects can be a substantial threat to a project’s safety, schedule, cost, or scope. In addition, the literature review reveals a significant gap in the integration of data-driven models, such as Artificial Neural Networks (ANN), with construction material control processes. Furthermore, there is a lack of studies implementing these models specifically in the context of Saudi Arabia, which is currently experiencing an increase in construction projects. Consequently, this paper proposes a data-driven model based on a Feedforward Neural Network (FNN) to classify and predict construction material defects. The proposed model is designed to classify the material defects and propose a robust model that helps the stakeholders to avoid and minimize material defects in construction projects. The proposed model utilizes fully connected architecture with multiple hidden layers to learn complex relationships between input features that influence the quality of the supplied construction material. In addition, the proposed model is compared with support vector machine (SVM) model. These factors are identified based on previous studies and experts. Moreover, the relationship between the input factors and between the input and output is investigated. The proposed model is developed and trained using 494 real-world data points from Saudi construction projects based on sixteen input variables. The model classifies material defects, with 66.19% identified as minor and 33.81% as major, supporting quality control through data-driven insights. Furthermore, the proposed model is evaluated using 10-fold cross-validation. To the best of our knowledge, this is the first application of an ANN model specifically developed to classify construction material defects into major and minor categories using real-world industrial data. This also investigates the relative importance of each input factor in the classification process. The findings indicate that the proposed model achieves up to 90% F-score in classifying the case defects and outperforms the SVM. The developed model is extremely useful to decision-makers in predicting and classifying material defects in construction.