This paper presents a data-driven analysis of the structural performance of 4500 community-designed bicycle frames. We introduce FRAMED – a parametric dataset of bicycle frames based on bicycles designed by bicycle practitioners from across the world. To support our data-driven approach, we also provide a dataset of structural performance values such as weight, displacements under load, and safety factors for all the bicycle frame designs. Our structural simulations are validated against results from physical experiments on real bicycle frames. By exploring a diverse design space of frame design parameters and a set of ten competing design objectives, we present a data-driven approach to analyze the structural performance of bicycle frames. Through our analysis, we highlight overall trends in bicycle frame designs created by community members and study several bicycle frames under different loading conditions. We then undertake a systematic search for optimal performance and feasibility-predictive Machine Learning models, applying a state-of-the-art Automated Machine Learning framework. We demonstrate that the proposed AutoML models outperform commonly used models such as Neural Networks and XGBoost, which we tune using Bayesian hyperparameter optimization. This work aims to simultaneously serve researchers focusing on bicycle design as well as researchers focusing on the development of data-driven design algorithms, such as surrogate models and Deep Generative Models. The dataset and code are provided at http://decode.mit.edu/projects/framed/ . • We introduce a dataset of 4500 bicycle frames inspired by bicycles designed by community members using the BikeCAD software. For each frame, we provide ten structural performance indicators evaluating the frame's performance under three load cases (in-plane, transverse, and eccentric loading). Indicators consist of seven deflections, two safety factors, and a weight value and are calculated through Finite Element Analysis. • We validate our Finite Element Analysis framework through a mesh convergence study and verify the accuracy of our simulation results against physical testing of bicycle frames. • We identify optimal surrogate models which predict the performance and feasibility of frames. Surrogates are selected using an AutoML framework which automates the selection of algorithms, architectures, hyperparameters, and instantiations for optimal model performance. AutoML models achieved a coefficient of determination of 0.605 in structural performance prediction and an F1 score of 0.915 in feasibility classification. • We validate our proposed AutoML framework against several common models (Neural Networks, XGBoost, etc.) which we optimize using Bayesian hyperparameter tuning. The proposed AutoML models attain the best coefficient of determination and mean absolute error among all methods tested in regression and the best F1 score, precision, recall, accuracy, and ROC AUC in classification.
Read full abstract
May 1, 2023
P Suthahar + 4
Open Access