Empirical analysis of sensor type importance for data preparation of real-time operational status monitoring in fused deposition modeling 3D printers

• Published in: International journal of advanced manufacturing technology 2024-05, Vol.132 (5-6), p.2617-2630
• Main Authors: Baek, Sujeong, Kim, Byeong Su, Lee, Yebon
• Format: Article
• Language: English
• Subjects: 3-D printers; Acceleration; CAE) and Design; Complex shape objects; Computer-Aided Engineering (CAD; Data collection; Data storage; Deposition; Empirical analysis; Engineering; Euler angles; Fault detection; Fused deposition modeling; Industrial and Production Engineering; Mechanical Engineering; Media Management; Original Article; Real time operation; Remote control; Remote monitoring; Sensors; Statistical inference; Support vector machines; Temperature sensors; Three axis; Three dimensional models; Three dimensional printing
• ISSN: 0268-3768; 1433-3015
• DOI: 10.1007/s00170-024-13522-x

The fused deposition modeling (FDM)-type three-dimensional (3D) printer is a popular choice in manufacturing facilities due to its capability of printing complex-shaped objects with simple machine control. To monitor the operational state of such 3D printing systems and detect faults, analog sensor signals can be collected and analyzed in real-time. Several research works have used traditional sensor types to monitor machinery movement, such as acceleration and temperature sensor signals, and have applied dimension reduction for efficient analysis. However, since the quality of operational state monitoring easily varies depending on the sensor information obtained, identifying meaningful sensor types at the sensor installation and data preparation stage is crucial for efficient data collection and analysis, prior to performing real-time status monitoring in 3D printing systems. In this study, we analyzed the relative importance of different sensor types for improving state monitoring performance and efficiency through statistical inference. It was evident that analyzing a set of five magnetic sensor signals was more effective and efficient for support vector machine-based classification of working stages and autoencoder-based fault detection than analyzing the entire set of signals, which includes 3-axis acceleration, 3-axis Euler angle, 3-axis magnetic field, temperature, and overall current sensors. By efficiently monitoring current working stages and detecting faults, this proposed strategy not only enhances the printing speed and product quality of FDM 3D printers but also improves the efficiency of original data storage in cloud services. This facilitates the control and remote monitoring of multiple 3D printers simultaneously.