Online melt pool depth estimation during directed energy deposition using coaxial infrared camera, laser line scanner, and artificial neural network

Jeon, Ikgeun; Yang, Liu; Ryu, Kwangnam; Sohn, Hoon

doi:10.1016/j.addma.2021.102295

Cited by 33 publications

(11 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The deposition occurs by means of the melt pool and is related with all those fundamental parameters governing the L-DED process [13], so it is reasonable to expect that its state represents a great portion of the quality of the built part. Monitoring performed in real time using machine learning (ML) algorithms and neural networks (NN) present a model-free monitoring approach, which is suitable for complex systems that require a multiphysical modelling involving a great number of variables.…”

Section: Introductionmentioning

confidence: 99%

“…Recently, there has been some research inclined towards an establishment of melt pool monitoring through imaging [17,18] with the aid of artificial intelligence methods [7,15]. They can also be linked to the rapid growth and improvement of image processing techniques, especially those using modern ML techniques, such as Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNNs).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A hybrid machine learning model for in-process estimation of printing distance in laser Directed Energy Deposition

Ribeiro

Núñez²,

Venter³

et al. 2023

Preprint

View full text Add to dashboard Cite

There are several parameters that highly influence material quality and printed shape in laser Directed Energy Deposition (L-DED) operations. These parameters are usually defined for an optimal combination of energy input (laser power, scanning speed) and material feed rate, providing ideal bead geometry and layer height to the printing setup. However, during printing, layer height can vary. Such variation affects the upcoming layers by changing the printing distance, inducing printing to occur in defocus zone then cumulatively increasing shape deviation. In order to address such issue, this paper proposes a novel intelligent hybrid method for in-process estimating the printing distance ( $Z_s$ ) from melt pool images acquired during L-DED. The proposed hybrid method uses transfer learning to combine pre-trained Convolutional Neural Network (CNN) and Support Vector Regression (SVR) for an accurate yet computationally fast methodology. A dataset with $2,700$ melt pool images was generated from the deposition of lines, at $60$ different values of $Z_s$, and used for training. The best hybrid algorithm trained performed with a Mean Average Error (MAE) of $0.266$ , which indicates an average target error of $6.7%$ . The deployment of this algorithm in an application dataset allowed the printing distance to be estimated and the final part geometry to be inferred from the data. Thus, the present method can aid on-line feedback control on the Z-axis increment, to regulate layer height, improving 3D shape geometry in L-DED.

show abstract

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

A hybrid machine learning model for in-process estimation of printing distance in laser Directed Energy Deposition

Ribeiro

Núñez²,

Venter³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Various monitoring techniques have been employed to capture process parameters, e.g. laser power, powder feed rate and feed speed and position [14], and indirect information on the process, such as temperature [15,16], melt pool behavior and characteristics [17][18][19], etc. Infrared cameras and pyrometers are commonly used to measure the temperature distribution and monitor the melt pool dynamics; and high-speed cameras capture the deposition process in real-time, allowing for visual inspection of the bead formation and detection of defects such as lack of fusion or porosity [20].…”

Section: Introductionmentioning

confidence: 99%

Real-time prediction of deposited bead width in L-DED using Semi-Supervised Transfer Learning

Mochi,

Núñez,

Ribeiro

et al. 2023

Preprint

View full text Add to dashboard Cite

Laser Directed Energy Deposition (L-DED) is an additive manufacturing (AM) technology that offers unique advantages for creating and repairing metallic parts. The quality of the final print is highly dependent on the consistency of the printing parameters, and as a result, the deposited bead geometry along the printing pathway. Inconsistencies, particularly in the bead width, can produce defects between adjacent beads, increasing the material susceptibility to failure. Therefore, it is crucial to monitor and predict the printing parameters in real-time during the process. The objective of this study is to develop a machine learning model that accurately predicts the width of the deposited track in real-time. To achieve this, a dataset with approximately 12,379 melt pool images was used to train the algorithm and predict the width of the deposited track. Each one of the deposited track was measured by a profilometer at 10 points along the pathway, and intermediate labels were predicted using a spline curve, making the model semi-supervised. To implement the machine learning model, pre-trained frozen networks based on Convolutional Neural Networks (CNN) architectures VGG, ResNet, and DenseNet were employed, using transfer learning principles. These networks were integrated into a dense, fully connected network containing trainable parameters. The results demonstrate a good model fit, with a mean absolute error of 4.5% and a mean absolute error of 0.0358 mm. Moreover, the processing frequency of the best model, at 55 Hz, enables real-time control of the L-DED manufacturing process. It is important to highlight, however, that the VGG based model shows a frequency of 250 Hz and similar results. Thus, accurately predicting the width of the deposited bead has the potential to significantly improve the quality of the final part by reducing overall defects, such as lack of fusion and porosity, whenever this data is used as input to a feed-back control system. This is particularly valuable in industries where mechanical properties and fatigue life are critical, such as automotive, aerospace, and biomedical sectors. The models developed in this study offer a promising approach to effectively provide data to control the L-DED manufacturing process in real-time, enabling the reliable production of high-quality components associated with reduced manufacturing costs.

show abstract

“…These signals are taken in situ and follow the melt pool as it scans across the build surface. [17][18][19][20] This method of monitoring aims to match features of the melt pool images to process dynamics. Seifi et al 21 used these melt pool images to detect anomalies layer-wise.…”

Section: Introductionmentioning

confidence: 99%

An unsupervised online anomaly detection method for metal additive manufacturing processes via a statistical time-frequency domain algorithm

Chen,

Kopsaftopoulos,

Mishra

2023

Structural Health Monitoring

View full text Add to dashboard Cite

Anomalies often occur in metal additive manufacturing from processing inconsistencies and uncertainty. A robust fault detection system that uses sensor measurements such as melt pool imaging has the potential to improve part quality and save production time by anticipating print failure. Toward this goal, we develop and validate a fault detection technique using melt pool geometry-related measurements from an in situ near-infrared optical camera. This method is unsupervised and is trained on a small dataset, mitigating human error in classifying fault types, and reducing lead times for preparing training datasets. Furthermore, this method uses learned geometry-informed nominal behavior of the melt pool signal to make informed decisions on the process health. There are spatial-temporal characteristics embedded in the melt pool images, caused by the periodicity in the geometry-dependent raster pattern. These characteristics can be captured in the frequency domain using the signal spectrogram, a representation of the frequency content over time. Defects will appear in the spectrogram, disrupting the healthy spectral response. To quantify healthy spectrograms, we use principal component (PC) decomposition to extract the features of these spectrograms as a set of nominal basis vectors. Anomaly detection is then performed by calculating the error between the original and reconstructed spectrogram vector by projection of the spectrogram PCs onto the nominal basis. The reconstruction error for anomalous signals is larger than that from healthy signals, which is then used for fault detection. A one-tailed statistical test is used to determine the fault detection threshold for the reconstruction error signal. This method is tested on three raster patterns and performs better than a comparative time-series thresholding method. We demonstrate that this time-frequency algorithm can detect both temporal faults (which occur at a single time instant) and spatial faults (such as those introduced by an improper sintering), differentiating them from nominal operation.

show abstract

Online melt pool depth estimation during directed energy deposition using coaxial infrared camera, laser line scanner, and artificial neural network

Cited by 33 publications

References 31 publications

A hybrid machine learning model for in-process estimation of printing distance in laser Directed Energy Deposition

A hybrid machine learning model for in-process estimation of printing distance in laser Directed Energy Deposition

Real-time prediction of deposited bead width in L-DED using Semi-Supervised Transfer Learning

An unsupervised online anomaly detection method for metal additive manufacturing processes via a statistical time-frequency domain algorithm

Contact Info

Product

Resources

About