Laser beam diagnostics systems based on artificial intelligence algorithms
Daniel Mostowski
supervisor: Grzegorz Pastuszak
The market problem to be solved by the system developed as part of the PhD concerns is very long downtime of laser systems used in industry or medicine. In the case of a laser system failure, the repair time is close to 4 weeks. Such long breaks cause huge financial loses on the company side.
The solution to the above-mentioned problem will be to build artificial intelligence algorithms that will analyze data from laser beam profilometers. Additionally, based on the data, these algorithms will send the information to the user about the state of the laser beam. Main objectives for PhD are: creation of a dataset for AI algorithms, detection of laser beam misbehavior, time trend analysis of the laser beam.
At the moment, we have managed to generate training data using numerical scripts. We are able to successfully generate 6 classes of data. Figure 2. shows sample data generated using numerical methods. This step was very important for the project as there are no commercial datasets available on the market.
The next step was to assemble the measuring setup. With his help, we will get real data for our data set.
Figure 1. shows the assembled stand. At this point, we are training our models on 300 thousands elements dataset. We are still focused on getting more real data from different sources. This will allow for more and more accurate learning of algorithms and for the generalization of the model itself.
Fig. 1 Two Fourier ABS reconstructions in the spectral (top, binarized) and spatial (bottom) domains, for 80x80 px test targets.
Fig. 2 Examples of laser beam profiles a) perfect beam, b) beam with linear diffraction, c) beam with spherical diffraction,
d) beam with rounded aperture, e) beam with dust, f) beam with roughness.
The PhD program is realized with Perspectiva Solutions company which develops the Huaris system utilizing AI component (https://perspectivasolutions.com, http://www.laser-beam-profile.com)