FRAMED

An AutoML Approach for Structural Performance Prediction of Bicycle Frames

Lyle Regenwetter1, Colin Weaver1, Faez Ahmed1

1MIT 

Explore

FRAMED is a project to explore and enable data-driven methods for performance-aware design of bicycle frames. In this work, we present a dataset of structural performance quantities for 4500 bicycles, validate our data, and explore the optimality of these designs. Get started exploring FRAMED's data, and the research directions that FRAMED enables:

Simulation Setup

FRAMED is a dataset of structural performance values for 4500 bicycle frames. We simulate each bicycle under in-plane, transverse, and eccentric loading conditions and record a variety of stresses, safety factors, and deflections in addition to model mass.

Searching for an Optimal Bicycle Frame

To explore the tradeoffs between performance parameters, we plot the performance distributions of bikes across the design space and examine scatterplots contrasting various design objectives with each other. We color code "Non-dominated" designs to identify regions of the design space with "Pareto-optimality."

This "Pareto-front" consists of all possible optimal designs from the dataset. None of these designs are outcompeted in each objective by another design in the dataset.

Citations

Chicago

Regenwetter, Lyle, Colin Weaver, and Faez Ahmed. "Framed: An Automl Approach for Structural Performance Prediction of Bicycle Frames." Available at SSRN 4132282.

Bibtex

@article{regenwetter4132282framed,
     title={Framed: An Automl Approach for Structural Performance Prediction of Bicycle Frames},
     author={Regenwetter,
     Lyle and Weaver,
     Colin and Ahmed,
     Faez},
     journal={Available at SSRN 4132282}}

ACKNOWLEDGEMENT

We would like to acknowledge BikeCAD for the original bicycle designs, Professor Daniel Frey for his input, and MathWorks for their support on this project.