Prediction of dimensional deviation and quality tolerance class in the production of AlSi10Mg alloy parts with laser powder bed fusion using machine learning algorithms

Metal-based workpieces can be produced additively with the selective laser melting (SLM) technique, which is one of the laser powder bed fusion (LPBF) methods. However, due to complex production parameters, the dimensional deviation and quality tolerance class of the produced parts cannot be predicted. This situation causes the need for secondary operations such as turning, milling, grinding, or polishing after production and even increases the amount of scrap. These processes are quite costly as they require extra time, labor, and machinery. This study aims to clarify complex manufacturing parameters, predict dimensional deviation with a high success rate before production, and categorize part quality tolerance classes with high accuracy. To examine the effect of manufacturing variables using dimensional accuracy, parts were produced at various laser power and scanning speeds, and features like length, width, thickness, and radius were measured. Dimensional changes were more significant at extreme parameter values and minimized at medium values. Additionally, extensive datasets were generated using manufacturing parameters and physics knowledge, and various machine learning models were applied and analyzed in detail. As a result of exploratory data analysis, dimensional deviation is estimated with a high success rate, while the part quality tolerance class can be categorized with a high precision rate. With the proposed method, proactive measures can be taken for possible errors before production. By effectively determining production parameters during the process planning phase, the part qualification process will become more efficient. Therefore, advantages such as reducing the cost of scrap parts, the absence of secondary processes, and saving cost and time will be achieved, and production comparable to traditional production methods will be possible.