Overview
This article reviews the applications and opportunities of machine learning in additive manufacturing (3D printing) materials and processes. It first introduces the background of machine learning technology in additive manufacturing, then analyzes in detail the applications of machine learning in areas such as quality control, process optimization, design optimization, microstructure analysis, and material formulation, and looks forward to future development directions, including the application of advanced machine learning models, the development of new sensors, and the application of machine learning in emerging additive manufacturing-related fields.
Background
In recent years, with technological advancements, increased data availability, and strengthened community collaboration, the application of machine learning technology in additive manufacturing has become increasingly widespread. Additive manufacturing technology creates complex three-dimensional structures by building them layer by layer, while machine learning can identify complex patterns from large amounts of data, thus optimizing decision-making in the additive manufacturing process. The article points out that the combination of machine learning and additive manufacturing is expected to fundamentally change the design and production methods of additive manufacturing parts.
Visual Analysis
Figure 1 shows the integration of machine learning in the additive manufacturing process. The left side categorizes machine learning into supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning, and introduces the emerging Transformer model. The right side details various additive manufacturing technologies. The middle part showcases the potential benefits brought by machine learning to additive manufacturing, and the bottom presents practical applications across a wide range of industries from aerospace to electronics and food. This figure emphasizes the diverse applications of machine learning in additive manufacturing, demonstrating its extensive impact across different fields.
Figure 2 is divided into four parts: a) The number of additive manufacturing publications related to machine learning over the past decade. b) The annual number of publications for each printing technology from 2013 to 2022. c) The annual number of publications for different machine learning applications in additive manufacturing research from 2013 to 2022. d) The annual number of publications on machine learning applications in various key additive manufacturing research areas from 2013 to 2022. This figure illustrates the research trends and application distribution of machine learning in the field of additive manufacturing, revealing research hotspots across different technologies and fields.
Figure 3 summarizes the application of machine learning in quality control across different additive manufacturing processes. This figure provides a comprehensive overview of machine learning in quality control, showcasing its specific applications and challenges in various additive manufacturing technologies.
Figure 4 shows a flowchart for detecting anomalies in laser powder bed fusion using machine learning. This figure details how to use machine learning technology in conjunction with sensor data to detect defects in the additive manufacturing process, demonstrating specific steps and methods in practical applications.
Figure 5 illustrates the workflow for training machine learning models using labeled X-ray CT data. This figure demonstrates how to predict defects in the additive manufacturing process by integrating various sensor data and machine learning models, emphasizing the importance of data fusion and model training.
Figure 6 presents various physical properties extracted from different sensing methods, used to train CNN architectures. The figure also shows the performance of different models in classifying laser spot sizes and pore types. This figure demonstrates how to utilize multiple sensor data and machine learning models to enhance defect detection accuracy in the additive manufacturing process, emphasizing the performance comparison of different models.
Figure 7 is divided into two parts: a) It shows a schematic diagram of the ID-CNN model used to detect spatter events during LPBF. b) It shows the model flowchart for detecting various defects in different materials. This figure illustrates how to use machine learning models in conjunction with acoustic emission signals to detect defects in the additive manufacturing process, emphasizing its application potential in different materials.
Figure 8 summarizes the variable time scales for monitoring the LPBF process using hybrid deep learning models. This figure illustrates how to achieve real-time monitoring and defect classification of the additive manufacturing process by integrating different sensor signals and deep learning models, emphasizing the model’s application effectiveness across different time scales.
Figure 9 presents the design of a convolutional neural network (CNN) for defect recognition, which uses 3D image slices from different sensor types. This figure illustrates how to utilize multiple sensor data and machine learning models to enhance defect detection accuracy in the additive manufacturing process, emphasizing the importance of data fusion.
Figure 10 illustrates a feedback flowchart for real-time parameter updates, including six key steps, and provides specific numerical ranges for parameter updates. This figure demonstrates how to achieve real-time parameter updates and error correction in the additive manufacturing process through machine learning models, emphasizing its application potential in improving print quality and reliability.
Figure 11 illustrates the real-time error detection and rapid correction capabilities of multi-head neural networks under different printing parameters. This figure demonstrates the adaptability and correction capabilities of machine learning models in different printing scenarios, emphasizing its application potential in improving print quality and reliability.
Figure 12 shows a flowchart for detecting irregular powder feed, including the design of the deposition head, model training and validation, and real-time application of pre-trained models. This figure demonstrates how to use machine learning models in conjunction with sensor data to detect powder feed issues in the additive manufacturing process, emphasizing its application potential in improving print quality.
Figure 13 provides an overview of the application of machine learning in optimizing processes across different additive manufacturing processes and various objectives related to process optimization. This figure illustrates the diverse applications of machine learning in process optimization, emphasizing its potential to improve efficiency and quality in additive manufacturing.
Figure 14 shows the different line shapes produced during powder bed fusion with different process parameters and demonstrates how to use machine learning to classify and predict the printability of parts. This figure illustrates the application of machine learning in optimizing powder bed fusion processes, emphasizing its potential to improve print quality and efficiency.
Figure 15 illustrates the workflow for multi-objective optimization using response surface methodology (RSM) to optimize conductivity and surface roughness during IPL sintering. This figure demonstrates how to optimize multi-objective problems in the additive manufacturing process by combining RSM and genetic algorithms, emphasizing its potential to improve print quality and efficiency.
Figure 16 illustrates a schematic diagram of predicting temperature fields in 3D printing processes using RNN-DNN and CNN models. This figure demonstrates how to use machine learning models to predict temperature fields in the additive manufacturing process, emphasizing its potential in optimizing printing parameters and improving print quality.
Figure 17 provides an overview of the application of machine learning in optimizing design in additive manufacturing-related applications. This figure illustrates the diverse applications of machine learning in design optimization, emphasizing its potential to improve design quality and efficiency.
Figure 18 illustrates a framework diagram for constructing design rules for additive manufacturing using machine learning and knowledge graphs. This figure demonstrates how to automate and autonomously build design rules for additive manufacturing by integrating machine learning and knowledge graphs, emphasizing its potential in improving design quality and efficiency.
Figure 19 illustrates the optimization framework for designing composite mechanical metamaterials, including sampling from random processes, training variational autoencoders, and optimizing using Bayesian optimization. This figure demonstrates how to use machine learning models to optimize the design of composite materials, emphasizing its potential to improve material performance and design efficiency.
Figure 20 illustrates a method for reverse engineering using μCT scanning and SEM images. This figure demonstrates how to use machine learning models in conjunction with imaging data to reverse engineer composite material parts, emphasizing its potential in improving design quality and efficiency.
Figure 21 provides an overview of the application of machine learning in predicting various microstructure-related properties in additive manufacturing. This figure illustrates the diverse applications of machine learning in microstructure analysis, emphasizing its potential to improve material performance and design efficiency.
Figure 22 is divided into two parts: a) It shows the workflow of machine learning techniques used to predict the mechanical properties of 3D printed metal parts. b) It shows the training workflow of machine learning models used to predict the performance of parts manufactured with new printers. This figure illustrates the application of machine learning in predicting the performance of additive manufacturing parts, emphasizing its potential to improve design quality and efficiency.
Figure 23 provides an overview of the application of machine learning in material formulation in additive manufacturing. This figure illustrates the diverse applications of machine learning in material formulation, emphasizing its potential to improve material performance and design efficiency.
Figure 24 illustrates the workflow for optimizing ink formulations using response surface methodology (RSM) to improve conductivity and line quality. This figure demonstrates how to optimize multi-objective problems in the additive manufacturing process by combining RSM and genetic algorithms, emphasizing its potential to improve print quality and efficiency.
Figure 25 provides an overview of the application of machine learning in typical bioprinting processes. This figure illustrates the diverse applications of machine learning in bioprinting, emphasizing its potential to improve print quality and efficiency.
Figure 26 illustrates a schematic diagram of 3D printing based on material extrusion and the machine learning framework for multimodal e-skin. This figure demonstrates the application of machine learning in bioelectronics printing, emphasizing its potential to improve device performance and design efficiency.
Figure 27 illustrates the application of machine learning in processing printed electrodes and sensor output signals. This figure demonstrates the application of machine learning in handling sensor data during the additive manufacturing process, emphasizing its potential to improve device performance and design efficiency.
Conclusion and Outlook
The article summarizes the current applications of machine learning in additive manufacturing, pointing out its great potential to improve the efficiency and reliability of additive manufacturing processes. Future development directions include the application of advanced machine learning models, the development of new sensors, and the application of machine learning in emerging additive manufacturing-related fields. The article emphasizes that by establishing a unified data community, the application of machine learning in additive manufacturing research can be accelerated, promoting further development of additive manufacturing technology.
Source: Frontier in Computational Chemistry, Edited by: Zhang Weiguan, Reviewed by: You Xiaoxiu
Add the editor to the additive manufacturing technology exchange group, where experts gather to communicate and learn from each other, Editor WeChat: 15801410073 (same as WeChat number), serious inquiries only,limited toadditive manufacturing-related professionals, please add friends with real-name registration (name + organization).
☞Advertising Cooperation: Sun Ha 13811718902