2023 Annual International Solid Freeform Fabrication Symposium (SFF Symp 2023): Wire-fed DED: Machine Learning and AI Application
Program Organizers: Joseph Beaman, University of Texas at Austin
Tuesday 8:15 AM
August 15, 2023
Room: 602
Location: Hilton Austin
Session Chair: Benjamin Bevans, Virginia Tech
8:15 AM
Monitoring of Process Stability in Laser Wire Directed Energy Deposition using Machine Vision: Anis Asad1; Benjamin Bevans1; Jakob Hamilton2; Iris Rivero2; Prahalada Rao1; 1Virginia Tech; 2Rochester Institute of Technology
The goal of this work is to mitigate flaw formation in parts made using the laser wire directed energy deposition (LW-DED) additive manufacturing process. As a step towards this goal, the objective of this work is to use real-time data from a meltpool imaging sensor to detect process instabilities. This is an important area of research, as LW-DED process tends to incessantly drift due to poorly understood thermophysical phenomena and stochastic effects. To realize the foregoing objective, we developed a machine learning model that acquires real-time imaging data, and automatically classifies the process state into one of four possible regimes: stable, dripping, stubbing, and incomplete melting. Through single track experiments conducted over 128 conditions, we show that the approach is capable of accurately classifying the process state with a statistical fidelity approaching 90% (F-score).
8:35 AM
Real-time Monitoring of Directed Energy Deposition Additive Manufacturing Process Using Multiple Sensors and Machine Learning: Shuchi Khurana1; Petros Apostolou1; Bradley Jared2; Josh Norton2; Steven Williams2; Eduardo Miramontes2; Charles Babbitt1; 1Addiguru; 2The University of Tennessee, Knoxville
Additive manufacturing of components via Directed Energy Deposition (DED) is a complex process involving melt pool dynamics, high cooling rates, power fluctuation, changes in feedstock flow, gas flow, etc. contributing to process instability and failed parts. It has been recognized that in-situ monitoring can be used to measure and control the bead shape, hence build geometry, helping save significant time and money by reducing part failures. Addiguru in collaboration with the University of Tennessee, Knoxville demonstrated the feasibility to fuse data from Short Wave Infrared (SWIR) and Long Wave Infrared (LWIR) cameras for measurement and control of bead shape.A dependency on bead size and inter-layer temperature (ILT) was observed. An experiment showed that the bead size or shape can be controlled in real-time to reduce issues that cause process instability and part failure.
8:55 AM
Thermal Imaging for Wire Arc Additive Manufacturing Using an Off-the-shelf Color Camera: Gala Solis1; Alex Myers1; Guadalupe Quirarte1; Mikhail Khrenov1; Sneha Narra1; Jonathan Malen1; 1Carnegie Mellon Univeristy
Process modeling and process monitoring are key to the adoption of additive manufacturing. For this purpose, imaging methods have attempted to measure weld pool temperature fields in wire-arc additive manufacturing (WAAM). A major challenge in estimating temperatures from monochromatic or infrared cameras is the need-to-know spectral emissivity. A novel two-color method is used to reduce sensitivity to the spectral emissivity by taking the signal ratio from two channels of an RGB color camera and correlating with the temperature predicted by Planck’s law, given the spectral sensitivity of the camera. Images are captured with varying exposure times and apertures to generate a full thermal field, along with a set of bandpass filters to reduce interference from plasma arc emission. These in-situ measurements can be used for model validation/calibration and flagging weld pool anomalies, thus furthering process parameter development, optimization, and monitoring.
9:15 AM
Toolpath Planning Approach for Parts with Multiple Revolving Features for Wire Arc Additive Manufacturing.: Wei Sheng Lim1; Gim Song Soh1; 1Singapore University of Technology and Design
In wire arc additive manufacturing, existing toolpath planner for complex 3D shapes such as propellers and turbines, with multiple revolving features radiating tend to utilize a cylindrical slicing approach. Such slicing approach are highly customized, complex, and not readily available for printing such 3D shapes. In addition, such complicated motion planning requires coordination between the print head and substrate motion to be synchronized well which can be difficult to achieve. In this paper, we propose an alternative strategy using planar slicing and adaptive width contour-based toolpath planner. To achieve this, a two-step approach is proposed with the substrate and radiating elements treated as separate features. The substrate is printed with part of the revolving feature, providing a flat surface for the second step to print from. The approach is applied for a propeller over 0.7m in diameter where a 3D scan is done to compare with the part model.
9:35 AM
Towards a Generic Deposition Model in Wire-arc Directed Energy Deposition: A Deep Learning-based Wetted Area Prediction Model: Magnus Glasder1; Maicol Fabbri2; 1ETH Zurich | IWF / AMLZ; 2ETH Zurich | IWF / AMLZ / inspire AG
Wire-arc directed energy deposition poses significant challenges in accurately predicting the geometry of weld beads, particularly regarding the overlap and stacking of multiple beads. This is due to the complex interaction between electric arc and previously deposited layers. Existing methods are inadequate in capturing this relationship for arbitrary layer geometries. A novel approach is proposed, which separates the prediction task into wetted area and shape prediction. The wetted area is predicted using a deep learning model, while shape prediction is achieved through an energy minimization technique, which places no assumptions on bead geometry. The wetted area prediction is treated as a computer vision task. 3D surface scans of the workpiece, welding parameters, and torch positions are encoded into images. A pre-trained vision network is fine-tuned on these images to predict the wetted area. The presentation emphasizes the machine learning aspect of the approach and delves into data management and pre-processing.
9:55 AM
Uncovering Fundamental Process Deficiencies in Wire-laser Directed Energy Deposition using In-situ High Speed Imaging: Jakob Hamilton1; Anis Assad2; Benjamin Bevans2; A. Cardinali1; Prahalada Rao2; Denis Cormier1; Iris Rivero1; 1Rochester Institute of Technology; 2Virginia Tech
Wire-based directed energy deposition (DED) offers key advantages over their powder counterparts but retains several major hurdles in becoming a push-button additive manufacturing technology. While powder DED provides high resolution, localized metallic deposition capabilities, it has yet to widely penetrate industrial manufacturing environments for several reasons. Powder safety, reactivity with oxygen, and low recyclability provide onerous economic and environmental barriers. Compared to atomized powder, wire requires considerably fewer safety precautions and resources during feedstock production and use. When implemented in laser DED, wire circumvents these limitations and shows promise for high-resolution, high-throughput manufacturing. This work outlines state-of-the-art limitations for wire-laser DED. Process deficiencies are captured using high-speed in-situ optical imaging and classified into several categories: incomplete melting, balling, stubbing, and stable deposition. Auxiliary deficiencies including substrate material and shielding gas coverage are also explained along with remediation strategies. Future strategies are suggested for transitioning nascent wire-laser DED into mainstream manufacturing.