A machine learning component is needed so that the modified 3D printer will know where to apply the soldering iron on the board.

1. The modified printer will take in a video stream of the board.
2. The machine learning algorithm will identify solder blobs.
3. The center of each identified blob will be sent to the 3d printer which will adjust the soldering iron.

This model was made using Edge Impulse. Julia collected roughly fifty images of PCBs with solder on them. She used bounding boxes to label the solder and the board in the training set. This training sample shows the image and labels which were used to train the model to identify the solder and the board.

The testing samples show where the model was able to identify solder on the board.

The model doesn’t always detect all of the solder on the board, but it always correctly identifies solder. As we move forward on constructing this, we’ll re-train the model based on the images coming from the camera. Having more images and a more consistent view should help us increase the model’s accuracy.

If you want to see the model itself, here is where we’re working on it.