A neural network-based parallel two-stage Quasi-static interrogation-fusion strategy for Laser Powde
Abstract
GRANT13530993-The final material properties of parts made via LPBF are extremely sensitive to the uncertainty/heterogeneity of powde,r, laser parameters, chamber environment, and the underlying deposited metal condition/geometry right below the current recoated pow,der layer. A suitable laser parameter setting at one fusion point may not be optimal for another and possibly result in an anomaly s,ealed in the part due to unstable melt pool subject to the aforementioned factors. To ensure the geometric and functional integrity, of AM parts in a proactive manner by avoiding the possible formation of defects, instead of adopting the traditional dynamic monito,ring schema on-the-go (i.e., along laser tracks), we go for another viable approach using a quasi-static interrogating-processing sc,hema at a per point basis (i.e., locally perfectify each melt pool) with independent or collaborative parallelization using beam arr,ays. A smart quasi-static two-stage process is adopted, which is composed of an interrogation stage and a manufacturing stage. A dir,ect mapping from the collected raw image data/signal during the interrogation stage to the locally optimized laser parameters to be, applied in themanufacturing stage will be established using a machine learning technic named reinforcement learning. Each static wo,rking spot moves from one point to another on the powder layer after each complete interrogating-manufacturing loop. The fusion laye,r is then formed by dense processed points of all laser heads. The term Quasi-static indicates the short static process locally at, each working spot.The anticipated outcomes are a smart interrogation-manufacturing reinforcement learning algorithm and a prototype, of the real-time and on-line control/tuning LPBF system for generating stable and uniform melt pools using two laser heads simultan,eously.Becausethe target strategy is proactive, the melt pool can be controlled to be stable throughout the manufacturing process. T,his means that, given a requirement on the universal melt pool shape, the suitable laser power at any location in the chamber will b,e automatically determined by the control system based on the sensing data from the interrogation stage to ensure a stable and unifo,rm manufacturing result. It is expected that this novel strategy can greatly improve the manufacturing quality. The proposed project, has the potential to advance the LPBF technology through innovative solutions fostered from basic research for enhancing Navys oth,er downstreamrelevant R&D activities. Our contributions mainly fall in one of the ONR Division 332s research concentration: develop,ment of new AM processes.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 05, 2022
- Source ID
- N000142212390
Entities
People
- Hui Wang
Organizations
- Office of Naval Research
- United States Navy
- University of Dayton