Automated laser beam characterization using artificial intelligence (AI) for the predictive maintenance of lasers

Abstract

A machine learning approach was used to evaluate the laser beam quality. Nowadays, the beam characterization is performed in a manual manner and with the use of custom setups typically built by the laser owners. The effort, knowledge requirement and the fact that the evaluation process must be conducted by a skillful man prevent laser beam diagnostics to be implemented on a mass scale in industrial, medical and scientific applications. This paper investigates six popular models of classic neural network classifiers. The final results of experiments show that automatic detection of various kinds of diffraction and speckles in the beam footprint can be performed by trained AI algorithms with the accuracy exceeding 99%. During the research, the dataset intended for those experiments was created. The dataset contains six classes that describe the most often occurring problems with the laser systems. The extremely short data processing time at the level of tens of milliseconds allows real-time monitoring of a laser beam quality and the development of a tool allowing implementation of the concept of predictive maintenance in the laser field.