Machine learning models for predicting non-linear features in reinforced concrete beams - Master’s thesis

Aug 5, 2026

Upload by Asher

This master’s thesis develops machine learning models in Python with TensorFlow to predict specific non-linear response features in reinforced concrete beams. Three distinct models are trained on data generated by non-linear analyses in the finite element program DIANA FEA. Beam geometric and reinforcement parameters are used as inputs, while load/displacement results from NLFEA in DIANA serve as outputs. Case study 1 focuses on classifying failure modes, while case studies 2 and 3 address regression tasks including peak load and load/displacement curve prediction.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages119

Reading Time5 h 0 min

AccessFree to download

Summary

Model development

Data generation and inputs/outputs

Case study 1: classification

Case studies 2–3: regression

Evaluation and hyperparameter tuning

Preface

Study background and motivation

Thesis execution and acknowledgements

Abstract

Master’s thesis Mohammed Isac Saaliti Yasin Salah Hassan Machine learning models for predicting non-linear features in reinforced concrete beams Master’s thesis in Civil and Environmental Engineering Supervisor: Daniel Cantero June 2024 NT NU Norwegian University of Science and Technology Faculty of Engineering Department of Structural Engineering Mohammed Isac Saaliti Yasin Salah Hassan Machine learning models for predicting non-linear features in reinforced concrete beams Master’s thesis in Civil and Environmental Engineering Supervisor: Daniel Cantero June 2024 Norwegian University of Science and Technology Faculty of Engineering Department of Structural Engineering Department of Structural Engineering Faculty of Engineering NTNU- Norwegian University of Science and Technology MASTER THESIS 2024 ACCESSIBILITY Open | SUBJECT AREA: Machine
learning and concrete structures | DATE: 10.06.2024 | NO. OF PAGES: 112 | | --- | --- | --- | | TITLE:
Machine learning models for predicting non-linear features in reinforced concrete beams
Maskinlæringsmodeller for å predikere ikke-lineære trekk i armerte betongbjelker | | | | --- | --- | --- | | BY:
Mohammed Isac Saaliti Yasin Salah Hassan | | | SUMMARY: This master's thesis aims to develop machine learning (ML) models in Python TensorFlow to predict specific non-linear features in reinforced concrete beams (RC beams) . Three distinct ML models are developed through case studies, each trained with data generated from non-linear analyses performed in the finite element program DIANA FEA. Each case study seeks to optimise the developed models through various evaluation metrics and hyperparameter tuning. The objective is to explore the application of machine learning to the non-linear analysis of concrete structures and, simultaneously, gain a better understanding of these complex fields. All ML models in each case study are trained using beam parameters (beam length, height, width, and transverse and longitudinal reinforcement area) as input data and with measured load/displacement from NLFEA in DIANA as output data. Case study 1 attempted to develop a model based on a binary classification algorithm that could predict bending or shear failure modes. Additionally, this case study determined that Sobol sampling is the most efficient sampling method for ML models in our case studies. Case studies 2 and 3 developed models based on regression algorithms. The model in study 2 could predict the peak load for an RC beam, while in study 3, it could predict the expected load/displacement graph for an RC beam. The developed ML models were evaluated using standard various evaluation metrics from machine learning. Some models from the case studies showed good results, especially after implementing hyperparameter tuning, while others needed further improvement. The models with good results exhibited low error measurements and predictions that were closely aligned with the true data values. However, after a final assessment of the models, it was clear that they had room for improvement. Enhancing the training dataset's quality and quantity would provide the much-needed information for the models. Exploring alternative hyperparameter tuning methods could also be beneficial in finding an optimal model fit. Implementing machine learning into the already complex field of non-linear analysis is challenging, but the models have shown promising results, which means they are worth further exploring. RESPONSIBLE TEACHER: Daniel Cantero SUPERVISOR(S): Daniel Cantero CARRIED OUT AT: Department of Structural Engineering PREFACE This master’s thesis marks the conclusion of our five-year studies in Civil and Environmental Engineering at the Norwegian University of Science and Technology (NTNU) . It was undertaken over a period of 20 weeks, starting in January and concluding in June. The work in this thesis, which accounts for 30 credits, was carr

FAQs

What is the main goal of the thesis?

To develop machine learning models that can predict specific non-linear features of reinforced concrete beams based on simulation-generated data.

How are the training data and outputs obtained?

Training inputs use beam geometry and reinforcement parameters, while outputs use measured load/displacement data produced from non-linear finite element analyses in DIANA FEA (NLFEA).

What differences exist between the three case studies?

Case study 1 uses a binary classification approach for predicting bending versus shear failure modes, while case studies 2 and 3 use regression methods for peak load prediction and for estimating the load/displacement graph.

Aug 5, 2026

Upload by Asher