Improving ceramic additive manufacturing via machine learning‐enabled closed‐loop control

Abstract

Advanced ceramic products are widely used in aerospace, automotive, electronic, laboratory equipment, and other industries. To achieve the geometric complexity and desirable properties that are difficult to obtain by conventional manufacturing methods, ceramic additive manufacturing (AM) methods have been studied intensively in recent years. However, the adaptive control with feedback is not currently implemented in any commercially available ceramic threedimensional printer. Robocasting is one of the most widely utilized constantvolumetric-flow AM processes for creating various ceramic materials at a low cost. This study employed robocasting as a model to implement an adaptive control with a feedback loop in the ceramic AM process. In this research, processing load that was proportional to the processing pressure, width of the print, and length of extrusion were selected to be representative of process signal, quality signal, and control parameter, respectively. First, a database of the load and extrusion length was established. An artificial neural network model was created using that established database. The data-driven, closed-loop control was integrated into the robocasting process. Finally, the improvement was validated by comparing the quality of the prints produced by both the closed-loop process with the adaptive control and the open-loop process without the adaptive control.

3D printing in materials manufacturing industry: A realm of Industry 4.0

“…The study integrates a NN with a proportion-integration-derivation control and determine the drive voltage. Zhang et al [182] proposed a feedback loop by using adaptive control in the ceramic AM process to improve print quality. A NN model is developed to create the relationship between the processing and control parameters, forming a closed-loop process.…”

Future directions in ceramic additive manufacturing: Fiber reinforcements and artificial intelligence

“…Poor identification and quantification of defects and limited standardization and speed of advancements the processes (feedstock development, processing parameters, postprocessing steps, etc.) 1,2,23,70,76,94,192,222,223 With these challenges in mind, the focus in this perspective will be on the following two future directions for ceramic AM: (1) incorporation of fiber reinforcement during printing and (2) approaches utilizing artificial intelligence and machine learning (AI/ML).…”

“…Poor identification and quantification of defects and limited standardization and speed of advancements the processes (feedstock development, processing parameters, postprocessing steps, etc. ) 1,2,23,70,76,94,192,222,223 …”

“…Recently, there has been progress toward bringing meaningful AI/ML to defect detection and/or print correction in MEX techniques, mainly in the polymer field. This work has included autonomous corrections to printed feature dimension, 227 feature spacing, 255 line stability, 2 and start/stop defects 256 . Although these all represent important steps toward integrating new digital technologies into AM processes, the corrections made were fairly intuitive (e.g., slowing nozzle motion to increase extruded features size).…”

“…In fact, the first instance of a patent related to AM techniques occurred in the 1980s, but the adaption to ceramics did not occur until the 1990s. 1 Since then there have been thousands of publications covering ceramic AM in various techniques and compositions, 2 providing evidence of buy-in from the research community. Even still, limited commercial off-the-shelf equipment exists tailored to ceramic AM 3,4 with few certified parts in service due to the remaining challenges and obstacles to overcome.…”