Gespeichert in:
Bibliographische Detailangaben
1. Verfasser: Bayazit, Ali Mert
Format: Recurso digital
Sprache:Englisch
Veröffentlicht: Zenodo 2026
Schlagworte:
Online-Zugang:https://doi.org/10.5281/zenodo.19795706
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866901576741814272
author Bayazit, Ali Mert
author_facet Bayazit, Ali Mert
contents <p class="font-claude-response-body break-words whitespace-normal leading-[1.7]">This Zenodo record contains the source code and dataset accompanying the M.Sc. thesis "Multi-Target Machine Learning for Stiffened Panel Stress Prediction", completed at Tallinn University of Technology in 2026.</p> <p class="font-claude-response-body break-words whitespace-normal leading-[1.7]">The framework includes an automated FEMAP–NX Nastran pipeline for generating a parametric finite element dataset of 1000 stiffened panel configurations using Latin Hypercube Sampling. It also includes a machine learning benchmarking suite evaluating nine regression architectures: Linear Regression, Decision Tree, Random Forest, Extra Trees, AdaBoost, Gradient Boosting, Hist-Gradient Boosting, XGBoost and LightGBM. The models are trained across five dataset sizes and twenty independent random seeds.</p> <p class="font-claude-response-body break-words whitespace-normal leading-[1.7]">The study uses a multi-target regression formulation that predicts plate and stiffener stresses as separate outputs, preserving the distinct mechanical behaviour of the two structural components. Tree-based ensemble models achieved high agreement with the finite element results, with Extra Trees reaching an average R² of 0.9991. The trained surrogate enables orders-of-magnitude faster stress prediction compared with repeated direct finite element analysis, with single-configuration inference in approximately 28.6 ms and batch inference of 10,000 designs in approximately 92 ms.</p> <p class="font-claude-response-body break-words whitespace-normal leading-[1.7]">The record also includes the Pearson correlation analysis code used to examine the relationship between input parameters, analytical scaling terms and FEM-derived stress outputs, and the inference benchmark script used to quantify the computational speedup against direct finite element analysis.</p> <p>Contents:</p> <ul> <li>FEMAP_data_generation.py</li> <li>FEMAPautomationv7_FULL_LOCATION</li> <li>ml_benchmark.py</li> <li>pearson_correlation_analysis.py</li> <li>inference_benchmark.py</li> <li>merge_datasets.PY</li> <li>STIFFENER_MULTI_TARGET_DATA.csv</li> <li>ML_BENCHMARK_DETAILED.csv</li> <li>ML_BENCHMARK_SUMMARY.csv</li> <li>STIFFENER_MASTER_DATASET.csv</li> </ul> <p>Supervisor: Mihkel Kõrgesaar, Tallinn University of Technology  <br>Co-supervisor: Muhammed Adil Yatkın, Tallinn University of Technology</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_19795706
institution Zenodo
language eng
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle Multi-Target Machine Learning for Stiffened Panel Stress Prediction: Source Code and Dataset
Bayazit, Ali Mert
Machine Learning
surrogate modelling
stiffened panel
finite element analysis
multi-target regression
Latin Hypercube Sampling
marine structures
Extra Trees
tabular data
FEMAP
ship structural analysis
<p class="font-claude-response-body break-words whitespace-normal leading-[1.7]">This Zenodo record contains the source code and dataset accompanying the M.Sc. thesis "Multi-Target Machine Learning for Stiffened Panel Stress Prediction", completed at Tallinn University of Technology in 2026.</p> <p class="font-claude-response-body break-words whitespace-normal leading-[1.7]">The framework includes an automated FEMAP–NX Nastran pipeline for generating a parametric finite element dataset of 1000 stiffened panel configurations using Latin Hypercube Sampling. It also includes a machine learning benchmarking suite evaluating nine regression architectures: Linear Regression, Decision Tree, Random Forest, Extra Trees, AdaBoost, Gradient Boosting, Hist-Gradient Boosting, XGBoost and LightGBM. The models are trained across five dataset sizes and twenty independent random seeds.</p> <p class="font-claude-response-body break-words whitespace-normal leading-[1.7]">The study uses a multi-target regression formulation that predicts plate and stiffener stresses as separate outputs, preserving the distinct mechanical behaviour of the two structural components. Tree-based ensemble models achieved high agreement with the finite element results, with Extra Trees reaching an average R² of 0.9991. The trained surrogate enables orders-of-magnitude faster stress prediction compared with repeated direct finite element analysis, with single-configuration inference in approximately 28.6 ms and batch inference of 10,000 designs in approximately 92 ms.</p> <p class="font-claude-response-body break-words whitespace-normal leading-[1.7]">The record also includes the Pearson correlation analysis code used to examine the relationship between input parameters, analytical scaling terms and FEM-derived stress outputs, and the inference benchmark script used to quantify the computational speedup against direct finite element analysis.</p> <p>Contents:</p> <ul> <li>FEMAP_data_generation.py</li> <li>FEMAPautomationv7_FULL_LOCATION</li> <li>ml_benchmark.py</li> <li>pearson_correlation_analysis.py</li> <li>inference_benchmark.py</li> <li>merge_datasets.PY</li> <li>STIFFENER_MULTI_TARGET_DATA.csv</li> <li>ML_BENCHMARK_DETAILED.csv</li> <li>ML_BENCHMARK_SUMMARY.csv</li> <li>STIFFENER_MASTER_DATASET.csv</li> </ul> <p>Supervisor: Mihkel Kõrgesaar, Tallinn University of Technology  <br>Co-supervisor: Muhammed Adil Yatkın, Tallinn University of Technology</p>
title Multi-Target Machine Learning for Stiffened Panel Stress Prediction: Source Code and Dataset
topic Machine Learning
surrogate modelling
stiffened panel
finite element analysis
multi-target regression
Latin Hypercube Sampling
marine structures
Extra Trees
tabular data
FEMAP
ship structural analysis
url https://doi.org/10.5281/zenodo.19795706