I’m building an optical profilometer. Made with a microscope, a projector, a camera, and a computer vision algorithm, it will measure the three dimensional topography of a sample by repeatedly imaging the sample at different heights, and measuring when each region of the sample crosses the microscope’s focal plane.

The idea is this:

• Begin with a fluorescence microscope, with a camera and a motorized stage. Remove the light source, and in its place, add a DLP projector, such that it can project an image onto the substrate.
• Use the projector to project a grid of crosshatches onto a tiny region of the substrate. Each one will correspond to a vertical pixel in the ultimate 3D image.
• Step the stage of the microscope up through all its possible Z values, capturing an image each time.
• Dice the image into squares, each corresponding to one of the hashmarks, a single XY pixel. Run each segment of each image through a computer vision algorithm, and assign it a score according to how in focus it is.
• For each pixel choose the Z value corresponding to the layer with the highest in-focus score. You now have X, Y, and Z values for the pixel.
• Do this process on a step-and-repeat basis over the entire area you want to image.
• Create a 3D mesh from all the pixels.

The hardware stack

The machine will be built around a metallurgical fluorescence microscope such as this one, which features a UV source, a camera, and a high-precision XYZ stage. The UV source will be replaced with an inexpensive low-resolution pico DLP projector, interface optics TBD. Since motorized microscope stages are super expensive, I’ll mechanize the stage on the microscope with NEMA 17 stepper motors, which I’ll hook up using GT2 belts and custom designed and printed pulleys. These will be controlled by a custom board, using the A4988 stepper drivers familliar to the 3D printer world.

All of this will be driven by a Tinkerboard, a souped-up RasPi clone with a bit more processing and graphics oomph. Its GPIO will drive the motors, the camera will connect by USB, and it will interact with the outside world through WiFi.

The software stack

The software will consist of three basic parts, beginning with an electron app, running on Linux on the Tinkerboard, which will do most of the work. It will handle the basic operations of the machine—graphics output, image capture, motor control—as well as higher level functionality, such as machine vision, constructing the final mesh, and executing the program that runs through the entire imaging process. A simple firebase backend will connect this to an Angular web app with which the user will control the machine via the internet.

Progress so far

• Software architecture
• Circuit board

Open source

• Tinkerboard app repo
• Frontend repo
• Circuit board—Upverter link, schematic (pdf), Gerber files (zip)
• (CAD files & bill of materials will be added in the future)