A group of researchers from University of Nebraska-Lincoln, Drexel University, Navajo Technical University, and SigmaLabs have developed a new process for detecting defects in 3D printed parts by laser powder bed fusion (LPBF) that use digital twins.
In a new article, the team describes a digital twin strategy that integrates physics and data to provide real-time detection of defects as they form during the LPBF printing process. By combining in situ weld puddle temperature measurements with computer predictions, researchers were able to detect and identify three different types of defects in stainless steel impellers.
The aim of the study was to address concerns about the tendency of the process to create faults in order to make it suitable for precision industries such as aerospace and biomedicine, while protecting against cybersecurity threats such as tampering with processes.
Defect formation during LPBF
Despite the geometric freedoms and the considerable time and cost savings that LPBF 3D printing can achieve, high-precision industries such as aerospace and medical have so far been reluctant to adopt the technology to manufacture critical parts. for safety, because of its tendency to create faults.
Cyber security risks have also become another emerging concern not only within LPBF but also in other 3D printing processes, with malicious parties potentially able to tamper with process parameters and plant faults to inside a room to compromise its performance.
Increasing research is being undertaken to address these issues and reduce the risk of faults in the LPBF process. The causes of microcracking in certain metals have been studied in order to improve the process, as well as the effects of beam shaping.
Texas A&M, in particular, has worked extensively in this area, having worked with Argonne National Laboratory to deploy machine learning to predict defects in 3D printed parts, and also by setting an LPBF “speed limit” at which defects such as J-shaped bubbles are less likely to form on printed parts In 3D.
Last month, scientists at Texas A&M demonstrated a universal LPBF 3D printing method of defect-free metal parts based on single-lane print data and machine learning. The team says their method is cheaper, longer, and simpler than existing methods of parameter optimization, making it well suited for aerospace, automotive and defense applications.
The digital twin approach
Defects tend to form during LPBF processes due to thermal occurrences during melting, cooling, solidification and reflow of the powder by the laser. On a microscopic scale, the melting of the powder creates a wake of molten material, called a weld pool, in which the temperature distribution, flow and spatter influence the microstructure, porosity and cracking of the part. .
On a macro scale, the rapid scanning action of the laser and the continuous melting of the material at high temperature causes heating and cooling cycles which can lead to residual stresses and deformation of parts.
To address this problem, the latest study aims to develop and apply an integrated data and physics strategy for on-line monitoring and detection of defect formation in LPBF parts. To do this, the team combined in situ molten bath temperature measurements with a thermal simulation model that quickly predicts the temperature distribution in a room.
The novelty of their approach, according to the researchers, is the temperature distribution predictions provided by the model, which are updated layer by layer with in situ weld puddle temperature measurements. As such, scientists call their method the “digital twin” approach to detecting defect formation.
The digital twin strategy is able to provide feedback to correct anomalies in parts, thereby reducing waste from construction failures. The researchers propose their strategy as an alternative to pure data-driven process monitoring techniques to overcome the drawbacks of these processes, namely detection delays, poor generalization of data-based models to part shapes, as well as resource intensive expenditure and resources. nature of data acquisition.
Additionally, as the digital twin incorporates both the macro-scale effect of part shape on thermal history and the micro-scale effect of laser-material interaction in the form of temperature of the molten pool, it can encapsulate the effect of different processing parameters, such as scanning pattern, hatch spacing, laser power and speed.
Digital twin test
To test their method, the team 3D printed four stainless steel wheel-shaped parts using an EOS M290 LPBF system that exhibited different types of defects covering process drifts, lens delamination, and cyber intrusions. To create the defects, the researchers altered processing parameters, caused machine-related malfunctions, and deliberately altered the process to create defects inside the part.
The team chose to print turbine parts to demonstrate their digital twin, as it is divisible into three distinct regions along the construction direction – the base, middle, and fin sections. Each of these sections includes complex features that are difficult to print, such as an internal teardrop-shaped cooling channel, which resulted in variable cooling time between layers and, therefore, a complex thermal history.
During construction, the process was continuously monitored using an array of three coaxial photodetectors integrated into the laser path. Signals obtained from the array of sensors were processed to create two types of measurements, namely thermal energy planck (TEP) and thermal energy density (TED). The TEP signature correlated with the temperature of the molten bath, while TED captured the radiation from the broadband chamber.
These signatures were then incorporated into the graph theory model to continuously update it with the micro-scale activity of the melting basin throughout the process.
The digital twin was able to detect all three types of defects in the 3D printed turbine parts during the LPBF process. According to the researchers, the results demonstrated that the method allowed precise and interpretable detection of defect formation as opposed to using sensor data alone. To this end, the digital twin approach overcomes the need to transfer signatures from the sensors to a separate data analysis algorithm and thus avoids delays in fault detection.
In the future, the team will look to expand the capabilities of its digital twin to detect other types of defects, such as distortion. They will also test the approach with different processing parameters, scanning strategies and part shapes.
Further information on the study can be found in the document titled: “Digitally twinned additive manufacturing: detection of defects in powder bed laser fusion by combining thermal simulations with in situ fusion sensor data”, published in the journal Materials and Design. The study is co-authored by R. Yavari, A. Riensche, E. Tekerek, L. Jacquemetton, H. Halliday, M. Vandever, A. Tenequer, V. Perumal, A. Kontsos, Z. Smoqi, K. Cole, and P. Rao.
Looking for a career in additive manufacturing? Visit 3D printing works for a selection of roles in the industry.
Subscribe to our Youtube channel for the latest 3D printing video shorts, reviews and webinar reruns.
Featured Image Shows the integrated digital twin strategy of physics and data researchers. Image via Materials and Design.