21 March 1989 Development of a Neural Network Based Real Time Control for Laser Welding
Author Affiliations +
Abstract
This paper describes the development of an adaptive control scheme for automated laser fusion welding applications, based on a software module capable of quickly recognizing weld cavity features indicated by sensor data without need for computation. Sensor data arrives to system in a high bandwidth stream, necessitating a quick recognition method in order to attain real time control. The method achieves speed through rigorous offline work within the actual system development. An iterative training procedure is exeuted in order to produce a software module capable of reliably recognizing prescribed weld puddle conditions within complex sensor data. The system is trained to mimic the classifications made by a metallurgical engineer between infrared sensor status and defect presence. Research will thereby attempt to automate existing metallurgical engineering knowledge relating causal relationships between particular states of sensor parameters and a limited set of defect types. The ability of these relationships to support a viable industrial control scheme will be assessed. Also, the required performance (speed, reliability) of the controllers primary component (the neural network) will be assessed. Research leaves the development of a computed reaction to such assessed condition to other research (some of which has been performed earlier at IITRI)
© (1989) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Nicholas E. Longinow, Edmund R. Bangs, "Development of a Neural Network Based Real Time Control for Laser Welding", Proc. SPIE 1094, Thermosense XI: Intl Conf on Thermal Infrared Sensing for Diagnostics and Control, (21 March 1989); doi: 10.1117/12.953391; https://doi.org/10.1117/12.953391
PROCEEDINGS
10 PAGES


SHARE
Back to Top