Development of Machine-Learning Algorithm for Reducing Defects in Automated Fiber Placement Process
Abstract
In order to reduce the acquisition and sustainment cost for composite structures, innovative design tools and low-cost rapid manufacturing techniques must be explored. Manufacturers are researching on innovative technologies to increase the production rates, while creating a data-driven factory of the future for managing flexible rates. Automated fiber placement (AFP) technologies considerably increase the production rates and reduce the part count with advanced joining technologies. As the tows from multiple spools arelaid down, there is a wide verity of defects such as gaps, overlaps, missing tows, twisted tows, puckers or wrinkles, and fiber bridging that needs to be inspected for during quality assurance of AFP parts. Since these defects can have a significant impact on the design allowables or structural margin of safety, it is important to detect and repair such defects. Quality assurance through inspections and process controls are essential to ensure that material is laidup and processed according to specification with appropriate consolidation and with no process-induced defects. Currently, AFP processes are interrupted after each layer so that the layupcan be manually inspected for defects. This manual inspection process that can consume 20-70 percent of the production time diminishes the benefits of automation to improve the production rate. In addition, manual inspection processes have deficiencies such as operator/training/environment dependency and inconsistencies. Main goal of this research is to develop and implement a machine-learning algorithm (MLA) for an in-process automated manufacturing inspection system (IAMIS) for reducing defects in automated fiber placement process. The development and implementation of MLA has two major phases; (a) the development of MLA for detecting AFP manufacturing defects and integration of the inspection system to AFP process in order to eliminate interruption to the manufacturing processand (b) the use oacturing rate on areas where possible.MLA developed under this program will be integrated to an in-process inspection system that will be attached to the layup head for automatically detect defects/features without interrupting the manufacturing process. MLA willcontinuously interrogate such data against a database that consists of allowable defect limits for a wide variety of defects as well as specification limits for AFP features. MLA will provide instant feedback to the operator with highlighting findings that are outside of the specification limits used for determining whether to interrupt the AFP process to repair or continue, based on the severity or repairability of the defect. In-process inspection system enables storing the defects information along with associated procontinuity Digital Thread, because the digital information collected here can be used for carrying out repair design and global-local analyses during sustainment accounting the allowable defects present during manufacturing of the part. In addition, the digital twin defect information can be used for repairing structures using the same AFP system without interrupting the manufacturing process (ex, adding a missing tow prior to the next layer). Probability of detection (POD) and analysis validation exercises will be carriedout to compare the fidelity of automated inspection data compared to traditional post-manufacturing inspection methods (manual and automated) used by part manufacturers. Validation exercise will then be expanded from panel-levels to complex components manufactured using AFP technologies with advanced materials.**Approved for Public Release**
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 05, 2021
- Source ID
- N000142112678
Entities
People
- Waruna Seneviratne
Organizations
- Office of Naval Research
- United States Navy
- Wichita State University