OpenCV for real-time quality control (CV = computer vision)

I think the future of high-quality 3D printing is real-time quality control.

Self-calibration, temperature control, error detection/correction, and positional feedback (STEP?) are essential for achieving that next big leap in real-world utility. If the goal is multiple materials and processes, maximum print speed while attaining a desired quality (e.g. dynamic 'draft' vs 'production' profiles) and ease of setup/maintenance, then we need feedback in our systems.

This is where libraries like OpenCV come into play. OpenCV interfaces with cameras/Kinects and provides an API for turning that real-world data back into information that we can use.

Yes, this would unfortunately require slaving our devices to more powerful machines in the short term. But how many more times do you really want to calibrate Z and level that bed?

Let's discuss other ideas for feedback/quality improvement scenarios that could serve as goals for such a project..

For example,

- a mic input might let us listen for stepper whine and calibrate current or detect vibration and accept voice commands to pause a print or change colors
- color detection could be used to generate a color palette profile from multiple feed colors by printing a 'color wheel'
- OpenCV's point detection could allow us to detect and calibrate the printable area, extraneous materials on the bed surface, improve bridging, print onto an existing object, identify trace trenches
- incorporating a scale into the bed could allow us to calibrate extrusion rates, and incorporated with point detection data, could be used to resume a cold print or evaluate evaporation rates

What ideas do you have?

Some OpenCV videos:

[youtu.be]
[youtu.be]
[youtu.be]
[youtu.be]

This has been discussed before.

The main problem is that sensors with high enough accuracy to be useful are expensive. Open-loop position control with stepper motors is extremely accurate and easy to implement. A properly-tuned system will basically never skip steps.

Cost is a silly problem to stall at. Expensive sensors were 10x more expensive a few years ago and will be 10x less expensive in a few more years. But there are people today that can afford them, and are in a position to test or advance the state of the art.

jason.fisher Wrote:
-------------------------------------------------------
> Cost is a silly problem to stall at. Expensive
> sensors were 10x more expensive a few years ago
> and will be 10x less expensive in a few more
> years. But there are people today that can afford
> them, and are in a position to test or advance the
> state of the art.

If that were true, that would be great, but I am not so convinced. The cost of inaccurate sensors has come down, accurate ones not so much. I suspect this is as much to do with volume of supply as technology.

The thing about RepRap is the KISS approach, and there is not much simpler than a stepper motor. A lot of the things you mention (usability issues) could be addressed by standard technology, it just takes some time and effort to implement them. There are rarely any technology advances that solve a lot of problems easily - IME they usually introduce a new can of worms.

But, if there are specific suggestions for new ideas, there is no harm in exploring them. The magnetic sensors in cheap digital calipers seem like they may be useful.