Ultrafast lasers possess desirable advantages in the field of precision processing owing to their short pulse widths and high-energy peak densities. However, single-focus ultrafast laser processing technology suffers from low processing efficiencies. Parallel processing by beam splitting using a spatial light modulator (SLM) is a solution for improving the efficiency of ultrafast laser precision processing. Improving the multibeam uniformity after beam splitting is a key issue in ultrafast laser parallel processing. Two factors deteriorate the quality of multiple beams: the approximations in beam-splitting algorithms, which leads to poor calculation accuracy, and the poor beam quality of the laser in optical paths. To address these issues, a feedback GS-GA algorithm based on real-time feedback is proposed in this study.
A spatial light modulator is used for phase-only modulation by loading computer-generated holograms (CGHs). The algorithms for beam splitting and shaping are dominated by the Gerchberg-Saxton (GS) algorithm and its derivatives. The GS algorithm is highly sensitive to the initial phase value of the light source, which directly affects the quality of the reconstructed light field. However, it is difficult to obtain a suitable initial value. In this study, real-time feedback based on GS and genetic algorithms is introduced in the process of hologram generation to improve the uniformity of beam splitting. In addition, a loaded Fresnel lens phase is used to separate the zero-order beam, avoiding the aberration caused by deviations from the optical axis of a reconstructed optical field. After several iterations of the proposed algorithm, a beam array with a high uniformity is designed.
Three beam splitting methods, including the GS algorithm, GS-GA algorithm, and feedback GS-GA algorithm, are investigated in terms of the beam splitting and processing effect. Owing to a system error in the optical path, the multibeam uniformity derived from the GS algorithm is less than 80% of the theoretical value. Compared with the GS algorithm, the uniformity of the GS-GA algorithm is improved; however, the performance of the GS-GA algorithm decreases significantly when the number of split beams increases. Due to the real-time feedback from the system (Table 4 and Fig.6), the uniformity of the multi-beam energy distribution obtained by the feedback GS-GA algorithm is superior to those of the other algorithms. Compared with the conventional GS-GA algorithm, the use of the feedback GS-GA algorithm significantly improves the consistency of both the hole depth and diameter by more than 12% in the resulting hole arrays. The consistent performance for the hole depth and diameter is improved by approximately 40% compared with that of the GS algorithm (Tables 5 and 6, and Fig.7).
An optimized feedback GS-GA algorithm is proposed for pre-compensation, and its effectiveness is experimentally demonstrated. Using the results from camera feedback as the individual fitness function in a genetic algorithm, this feedback algorithm utilizes the genetic algorithm to select, exchange, and eliminate the initial value population, thereby improving the beam uniformity by approximately 94% for graphic and array multibeams. Furthermore, a laser precision punching experiment is performed using the beam arrays obtained via this feedback algorithm, which shows large aperture and hole depth uniformities of over 90%. The method proposed in this paper can achieve high-uniformity shaping for any form of designed multibeam with minimal requirements on the accuracy of the optical path system. This method obtains a multibeam distribution with beam splitting uniformity close to 90% in a certain focal depth range by dynamic zooming. The proposed shaping strategy significantly improves the beam splitting uniformity, which paves the way for the application of high-quality ultrafast laser parallel processing. The time-consuming computation of the proposed method can be improved by optimizing the solution range of the algorithm and using high-performance computer hardware to speed up the iteration of the algorithm.
