Track: Machine Learning

Abstract

Manufacturing sheet metal parts with laser cutting machines involves placing several parts of various geometries from multiple customers on one sheet blank (e.g., 6’ x 12’). While this practice is common, it can cause complications in accurately calculating fabrication time and subsequent cost of parts for each customer. The reason for this is that laser cutting machines display only the total cutting time per entire sheet blank, and the cutting order within and between parts on sheet blanks may not be sequential. To resolve this long-standing machine time distribution problem, this paper proposes a Machine Learning approach based on the geometric characteristics of parts. To test this approach, 348 sheet blanks with parts of various shapes and sizes were processed on a laser cut machine to collect the machine's cutting time for each sheet blank. The parts on the sheet blank were broken down into their component geometric characteristics (e.g., line length, number of vertices, arc length, number of piercing, etc.), which were used as features for the Machine Learning model. Data from 338 sheet blanks were used for training, and data from the remaining 10 were used to validate the trained model. Three Machine Learning algorithms—Linear Regression, Ridge Regression, and Lasso Regression—were selected and compared. The results show that the Machine Learning approach, based on parts' geometric characteristics, is a viable method for designating cutting time for parts on sheet blanks. With the real-time data collection of parts' geometry and shop floor machine run time in the era of Industry 4.0, this approach can be implemented on any laser cut machines to automatically determine the cutting time (and therefore manufacturing cost) of each part.