Dynamic Beam Shaping Using a Wavelength-Adaptive Diffractive Neural Network for Laser-Assisted Manufacturing

Abstract

Laser-based manufacturing has emerged as a promising alternative to conventional thermal and mechanical processing owing to its precision, versatility, and ability to work across diverse materials. In particular, tailoring the spatial intensity distribution of laser beams on the fly is pivotal for ensuring keyhole stability, minimizing defects, and enhancing processing quality. To address this need, we propose a multifunctional optical platform designed through a Diffractive Neural Network that provides wavelength adaptability for three industrially relevant wavelengths - 915 nm, 1064 nm, and 1550 nm - while dynamically generating distinct beam profiles at specified propagation planes. The proposed platform not only enables static beam shaping but also supports dynamic beam engineering, including programmable sequencing between profiles, which is highly desirable for optimal manufacturing solutions. With its multifunctionality and adaptability, the DNN-based architecture establishes a transformative pathway for next-generation laser manufacturing, aligning with the industrial revolution while unlocking opportunities in biomedical optics, free-space communications, and sensing.