An official website of the United States government
Abstract The long-term goal of this work is to predict and control microstructure evolution in metal additive manufacturing processes. As a step towards this goal, the objective of this paper is the rapid prediction of the microstructure evolution in parts made using the laser powder bed fusion (LPBF) additive manufacturing process. To realize this objective, we developed and applied an approach which combines physics-based thermal modeling with data-driven machine learning to predict two important microstructure-related characteristics in Nickel Alloy 718 LPBF-processed parts: meltpool depth and primary dendritic arm spacing (PDAS). Microstructure characteristics are critical determinants of functional physical properties, e.g., yield strength and fatigue life. Currently, the microstructure of laser powder bed fusion parts is optimized through a cumbersome and costly build-and-characterize empirical approach. This makes the development of rapid and accurate models for predicting microstructure evolution practically valuable: these models reduce process development time and enable fabrication of parts with consistent properties. Unfortunately, due to their computational complexity, existing physics-based models for predicting microstructure evolution are limited to only a few layers and are challenging to scale to practical parts. To overcome the drawbacks of current microstructure prediction techniques, this paper establishes a novel physics and data integrated modeling approach. The approach consists of two steps. First, a rapid, part-level computational thermal model was used to predict the temperature distribution and cooling rate in the entire part before it was printed. Second, the foregoing physics-based thermal history quantifiers were used as inputs to a simple machine learning model (support vector machine) trained to predict the meltpool depth and primary dendritic arm spacing based on empirical materials characterization data. As an example of its efficacy, when tested on a separate set of samples from a different build, the approach predicted the PDAS with root mean squared error ≈ 110 nm. The modeling approach was further able to predict meltpool depth with a root mean squared error of 0.012mm. This model performance was validated through the creation of 21 geometries created under 7 different process parameters. Optical and scanning electron microscopy was conducted resulting in more than 1200 primary dendritic arm spacing and meltpool depth measurements. Primary dendritic arm spacing predictions were also validated on parts of a unique geometry created in a separate work. The model was able to successfully transfer to this build without further training, indicating that this method is transferrable to other parts made with laser powder bed fusion and Nickel Alloy 718. This work thus presents an avenue for future physics-based optimization and control of microstructural evolution in laser powder bed fusion.
more »
« less
This content will become publicly available on April 1, 2027
Prediction of solidified microstructure in laser powder bed fusion of Inconel 718 using physics-aware machine learning
Accurate prediction of the solidified microstructure in laser powder bed fusion (LPBF)-processed components is critical, because, heterogeneity and spatial anisotropy in the solidified microstructure lead to variation in functional properties. This paper presents a physics-aware and real-time data-integrated machine learning approach for predicting the solidified microstructure of LPBF-processed Inconel 718 parts. Currently, the prediction of solidified microstructure in LPBF is realized primarily through three approaches: (i) empirical modeling; (ii) blackbox modeling using in-situ sensor data; and (iii) whitebox modeling leveraging physics-based process simulations. Each of these approaches has inherent limitations. Empirical and blackbox models lack generalizability because they do not account for geometry-dependent causal thermal-fluid phenomena governing microstructural evolution. Physics-based whitebox models are computationally demanding and overlook the stochasticity inherent to the process. To overcome the foregoing limitations, this work establishes a graybox modeling approach. This approach combines temperature fields predicted by a physics-based thermal model with real-time data acquired from in situ infrared thermal imaging and optical tomography sensors. The graybox model is trained to predict the following microstructural aspects of LPBF-processed Inconel 718 parts: melt pool depth; primary dendritic arm spacing; crystallographic texture, orientation, and grain aspect ratio; and microhardness. The graybox model predicted the solidified microstructure with an accuracy approaching 95% (R²). By contrast, the prediction accuracy of blackbox and physics-based whitebox models ranged between R² ∼ 60% and 85%. Thus, this work takes a critical step towards a rapid, accurate, in-situ, and non-destructive Born Qualified assessment of LPBF part quality.
more »
« less
- PAR ID:
- 10693508
- Publisher / Repository:
- Elsevier
- Date Published:
- Journal Name:
- Additive Manufacturing
- Volume:
- 122
- Issue:
- C
- ISSN:
- 2214-8604
- Page Range / eLocation ID:
- 105174
- Subject(s) / Keyword(s):
- Laser Powder Bed Fusion Solidified Microstructure Inconel 718 Thermal History In-situ Sensing Physics-aware Machine Learning
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
The long-term goal of this work is to predict and control the microstructure evolution in metal additive manufacturing processes. In pursuit of this goal, we developed and applied an approach which combines physics-based thermal modeling with data-driven machine learning to predict two important microstructure-related characteristics, namely, the meltpool depth and primary dendritic arm spacing in Nickel Alloy 718 parts made using the laser powder bed fusion (LPBF) process. Microstructure characteristics are critical determinants of functional physical properties, e.g., yield strength and fatigue life. Currently, the microstructure of LPBF parts is optimized through a cumbersome build-and-characterize empirical approach. Rapid and accurate models for predicting microstructure evolution are therefore valuable to reduce process development time and achieve consistent properties. However, owing to their computational complexity, existing physics-based models for predicting the microstructure evolution are limited to a few layers, and are challenging to scale to practical parts. This paper addresses the aforementioned research gap via a novel physics and data integrated modeling approach. The approach consists of two steps. First, a rapid, part-level computational thermal model was used to predict the temperature distribution and cooling rate in the entire part before it was printed. Second, the foregoing physics-based thermal history quantifiers were used as inputs to a simple machine learning model (support vector machine) trained to predict the meltpool depth and primary dendritic arm spacing based on empirical materials characterization data. As an example of its efficacy, when tested on a separate set of samples from a different build, the approach predicted the primary dendritic arm spacing with root mean squared error ≈ 110 nm. This work thus presents an avenue for future physics-based optimization and control of microstructure evolution in LPBF.more » « less
-
Modeling and control of the spatiotemporal temperature distribution (thermal history) in laser powder bed fusion (LPBF) is critical because the thermal history governs defects, such as porosity, poor surface finish, cracking, and deformation. This article presents a coupled physics-based computational modeling and feedforward process control framework for regulating the thermal history in LPBF-processed parts. Existing LPBF process optimization relies on empirical parameter tuning by manufacturing and testing of simple, standardized coupon geometries. Empirical coupon-based optimization inherently disregards the geometry-dependent effect of thermal history on defect formation. Consequently, process parameters optimized based on coupon studies, when used for manufacturing real-world components, often result in build failures and defects. To address this limitation, a rapid graph theory-based computational model was coupled to a feedforward control (FFC) algorithm. The approach is implemented for manufacturing a topology-optimized Inconel 718 aerospace component (GE bracket). The model-guided FFC approach maintains a constant end-of-cycle (interpass or interlayer) temperature across layers by adjusting the laser power and velocity. The processing parameters are adjusted in silico—offline and prior to manufacturing—within the thermal model. Compared to its empirically optimized counterpart, the FFC-processed GE bracket exhibited three characteristics favorable to functional integrity and production: (i) meltpool instability-induced porosity was not observed; (ii) thermal-induced deformation, dross formation, and recoater contact damage were significantly mitigated; and (iii) FFC-induced improvements enabled the part to be manufactured with 45% less support mass, resulting in a 20% reduction in the as-built part weight (with the part design unchanged). This work thus underscores the potential of physics-based control, as opposed to empirical optimization, to mitigate defects in LPBF parts and accelerate their practical deployment.more » « less
-
We developed and applied a model-based feedforward control approach to reduce temperature-induced flaw formation in the laser powder bed fusion (LPBF) additive manufacturing process. The feedforward control is built upon three basic steps. First, the thermal history of the part is rapidly predicted using a mesh-free graph theory model. Second, thermal history metrics are extracted from the model to identify regions of heat buildup, symptomatic of flaw formation. Third, process parameters are changed layer-by-layer based on insights from the thermal model. This technique was validated with two identical build plates (Inconel 718). Parts on the first build plate were made under manufacturer recommended nominal process parameters. Parts on the second build plate were made with model optimized process parameters. Results were validated with in-situ infrared thermography, and materials characterization techniques. Parts produced under controlled processing exhibited superior geometric accuracy and resolution, finer grain size, and increased microhardness.more » « less
-
This work concerns the laser powder bed fusion (LPBF) additive manufacturing process. We developed and implemented a physics-based approach for layerwise control of the thermal history of an LPBF part. Controlling the thermal history of an LPBF part during the process is crucial as it influences critical-to-quality characteristics, such as porosity, solidified microstructure, cracking, surface finish, and geometric integrity, among others. Typically, LPBF processing parameters are optimized through exhaustive empirical build-and-test procedures. However, because thermal history varies with geometry, processing parameters seldom transfer between different part shapes. Furthermore, particularly in complex parts, the thermal history can vary significantly between layers leading to both within-part and between-part variation in properties. In this work, we devised an autonomous physics-based controller to maintain the thermal history within a desired window by optimizing the processing parameters layer by layer. This approach is a form of digital feedforward model predictive control. To demonstrate the approach, five thermal history control strategies were tested on four unique part geometries (20 total parts) made from stainless steel 316L alloy. The layerwise control of the thermal history significantly reduced variations in grain size and improved geometric accuracy and surface finish. This work provides a pathway for rapid, shape-agnostic qualification of LPBF part quality through control of the causal thermal history as opposed to expensive and cumbersome trial-and-error parameter optimization.more » « less
