Prediction of Metal Additively Manufactured Surface Roughness Using Deep Neural Network

Abstract

In recent years, manufacturing industries (e.g., medical, aerospace, and automobile) have been changing their manufacturing process to small-quantity batch production to flexibly cope with fluctuations in demand. Therefore, many companies are trying to produce products by introducing 3D printing technology into the manufacturing process. The 3D printing process is based on additive manufacturing (AM), which can fabricate complex shapes and reduce material waste and production time. Although AM has many advantages, its product quality is poor compared to conventional manufacturing systems. This study proposes a methodology to improve the quality of AM products based on data analysis. The targeted quality of AM is the surface roughness of the stacked wall. Surface roughness is one of the important quality indicators and can cause short product life and poor structure performance. To control the surface roughness, the resultant surface roughness needs to be predicted in advance depending on the process parameters. Various analysis methods such as data pre-processing and deep neural networks (DNN) combined with sensor data are used to predict surface roughness in the proposed methodology. The proposed methodology is applied to field data from operated wire + arc additive manufacturing (WAAM), and the analysis result shows its effectiveness, with a mean absolute percentage error (MAPE) of 1.93%.

Application of Artificial Intelligence for Surface Roughness Prediction of Additively Manufactured Components

“…So et al [ 11 ] developed a method to improve the quality of additively manufactured products by predicting surface roughness using data analysis techniques, including data pre-processing and DNNs combined with sensor data. The study focused on enhancing the surface roughness of the stacked wall, which is a crucial quality indicator affecting product life and structural performance.…”

“…Various methods can be employed to measure surface roughness, depending on different definitions [ 11 , 118 ]. In additive manufacturing research, the Ra parameter is commonly used to measure surface roughness due to its simplicity, though it lacks sensitivity to wavelength variations.…”

“…The main advantage of additive manufacturing is that it can fabricate complex shapes and reduces material waste and production time (for some) [ 11 , 12 , 13 , 14 ]. Despite its advantages, additively manufactured products generally have poorer quality compared to those made with conventional manufacturing systems [ 15 , 16 ], primarily due to limitations in surface integrity [ 17 ].…”

“…Surface roughness has been identified as one of the most significant factors affecting product quality, as stated in previous research [ 11 ]. For instance, Chan et al [ 70 ] conducted a study to investigate the impact of surface roughness on product life and concluded that surface roughness leads to a reduction in product life expectancy.…”

Explainable Artificial Intelligence (XAI) and Supervised Machine Learning-based Algorithms for Prediction of Surface Roughness of Additively Manufactured Polylactic Acid (PLA) Specimens

“…The study aimed to develop an accurate and reliable predictive model that can assist in controlling the quality of the final FDM products. So et al [27] developed a methodology to enhance the quality of additive manufacturing (AM) products based on data analysis. The study utilized various data analysis techniques, including data pre-processing and Deep Neural Networks (DNNs), combined with sensor data to predict surface roughness.…”