Invention:
This invention is an interferometric system that captures two interferograms and utilizes a deep learning algorithm to process two interferograms. This invention enables the measurement of surface roughness and surface shape using the interferometric system and neural network to process the data. The deep learning portion of the system is in a compact form allowing for on-machine measurements and a suitable attachment to the interferometric system.
Background:
The application of high precision optical elements are vast, from smart-phone camera lenses, telescopes, in addition to optical fibers and much more. With an increase in technological development the needs for high precision optical elements, accurate and efficient fabrication process is highly demanded, placing ultrahigh requirement on the measurement tools to improve workpiece quality control and manage machining process. As a recognized accurate testing method, interferometry has been a powerful method for non-contact surface metrology of optical elements. There are two unique interferometry methods to measure surface form and roughness. Currently the commercial interferometric instruments have a separate procedure to acquire the surface form and roughness measurements. This adds additional costs, time, and fabrication errors to the interferometric process of measurements.
A University of Arizona researcher has developed a interferometric system which can take two interferogram inputs acquired from the system and apply a deep learning algorithm to output a highly accurate element surface form and roughness measurements. The deep learning aspect of the system was developed compactly, allowing for ease of use in procedure of conducting the desired measurements.
Applications:
- Metrology
- Optical testing and measurement
- Interferometric systems
Advantages:
- On-machine deep learning
- All in one process for surface form and roughness measurements
- Cost efficient, due to reduced procedure
- Less fabrication errors potential, due reduced procedure
- Compact deep learning attachment
