BikeCAD version 20 includes a new AI feature accessible by clicking the brain icon and then the submit button in the AI dialog box.
The AI tool was developed in partnership with MIT's DeCoDE Lab. The team at the DeCoDE Lab applied machine learning techniques to a dataset obtained from the BikeCAD design archive. For these results to be accurate, we'll need to ensure that we have selected the appropriate material in the Notes tab of the Notes and title block dialog box. For now, carbon and bamboo will not be considered because the stress analysis carried out in the AI model only applies to isotropic materials. We'll also need to ensure that we've set the tube diameter and wall thickness of each tube in the Tubing dialog box. Be aware that although we now have the option to specify a butted tube profile, at this time, the AI engine assumes all tubes are straight gauge. Further, it assumes tubes are of a constant diameter and straight.
With that said, the Frame weight tab will display an approximate weight for our frame, we can also see displacement under in-plane loading, a simplified top view of the frame showing displacement under transverse loading and finally displacement and rotation under eccentric loading.
Data-driven AI models can be inaccurate, especially when shown a design that deviates significantly from the training data. This means that predicted values for unconventional or unique bikes will be much less reliable.
No safety-critical or potentially costly design decisions should be made using AI predictions without outside validation or verification.
One additional thing the AI tool will do is search the BikeCAD design archive for a model that it deems to be the closest match to the submitted model.
This is the first iteration of the BikeCAD AI tool. I'll be interested to see people using this feature and am excited about what future developments might hold.
Links to MIT's DeCoDE lab are listed below.