Multivariate prediction of laser deposited geometry based on machine vision

Abstract

Laser Metal Deposition is a widely used technology allowing metal component regeneration, rapid prototyping and functional coating deposition. Its main limitations are low stability and repeatability making it necessary to monitor and often control the process in real time. This paper presents a study on modelling the influence of process parameters on clad geometry and assessment if including sensor data can help model the outcome properly. For this purpose 9 sets of parameters were used with multiple repetitions. During the process a profile of deposited geometry, temperature and images in visible range were acquired. Multivariate analysis were performed to investigate correlations. Multiple prediction models were constructed including signals’ non-linearity correction. The results prove that process parameters have main influence on process outcome but cannot explain natural process variation whereas models including sensor data allow prediction of those deviations.