Why High-Tech Manufacturing Relies on Xiris Cameras

Xiris cameras have become critical in various high-tech research and industrial applications, offering real-time monitoring and precision data measurements that enhance understanding of complex manufacturing processes. In welding and additive manufacturing, the Xiris thermal and weld cameras provide researchers with insights that improve process control, system efficiency, and product quality.

Here, we dive into some of the groundbreaking research from 2024 that highlights the transformative impact of Xiris cameras across various fields.

1. Nanosecond Laser-Induced Iron Plasma Reactive Etching of CVD Diamond

In the study “Study on nanosecond laser-induced iron plasma reactive etching of single-crystal CVD diamond” [1], researchers monitored the etching of mono-crystal diamond with a nanosecond laser. For accurate real-time observation, the team chose the Xiris XIR-1800 thermal camera to track the temperature distribution of the laser-etched surface.

The camera's high thermal sensitivity and fast response time provided precise temperature measurements, which were critical in understanding the effects of the laser-induced etching process on the diamond. By employing the XIR-1800, the researchers were able to enhance the precision and repeatability of this advanced material processing technique.

Figure 1. Periodic microtextures etched on the surface of the diamond. Reprinted from [1].

2. Laser Metal Deposition and Multimodal Learning

The integration of thermal imaging with advanced machine learning models has proven especially effective in laser metal deposition processes. In the study “JEMA: A JOINT EMBEDDING FRAMEWORK FOR SCALABLE CO-LEARNING WITH MULTIMODAL ALIGNMENT” [2], researchers from the University of Porto used the Xiris XIR-1800 thermal camera to monitor powder-based Laser Metal Deposition (LMD) off-axis.

By measuring the temperature distribution in the melt pool, the team could accurately segment and measure the molten metal area. Together with the process parameters and the data from an on-axis visual camera, it created the data stream that the was used by a powerful co-learning framework called JEMA developed as a part of this research. This framework using multimodal data could predict the melt pool geometry with high accuracy, eliminating the need for extensive fine-tuning.

The integration of the XIR-1800 thermal camera was a key step in providing the most relevant real-time data to the multimodal system. It helped ensure that the machine learning model could make precise predictions for improving deposition quality.

Figure 2. Thermal image of the melt pool in laser metal deposition (LMD) obtained with Xiris XIR-1800 thermal camera as reported by [2].

3. Multi-Sensor Monitoring for Laser Hot-Wire Cladding

In “A multi-sensor based online monitoring system for laser hot-wire surface cladding process”, researchers, from the Research Center for Advanced Manufacturing, Lyle School of Engineering, Southern Methodist University, combined several sensors, including a high-speed camera, a spectrometer, and the Xiris XIR-1800, to create a thermal monitoring system for laser hot-wire cladding.

The XIR-1800 was used to track the temperature of the melt pool during the cladding process. This data was essential in calculating the cooling rate, which was directly affecting the clad microhardness. The scientists found that the monitoring of cooling rate and real-time adjustments of laser power and scanning speed make it conceivable to regulate clad microhardness in real time through an adaptive online control system

Also, Xiris is proud to introduce the first ever commercially available Cooling Time Tool that provides t8/5 measurements critical for the quality and consistency of the cladding process directly in WeldStudio™.

Download the Cooling Time Whitepaper.

Figure 3. Thermal image of the melt pool in laser hot wire cladding (LHWC) as reported by [3]. The numbered locations were used to calculate the cooling rate.

4. GMAW and LSTM U-Net for Weld Pool Reconstruction

In another innovative work “Analysis of weld pool region constituents in GMAW for dynamic reconstruction through characteristic enhancement and LSTM U-Net networks” [4], the researchers at the University of Kentucky applied a Xiris XVC-1100 camera to monitor the Gas Metal Arc Welding (GMAW) process.

The cameras provided high-resolution images of the weld pool, arc, and filler wire. The team developed a hybrid model that combined a Long Short-Term Memory (LSTM) network with the U-Net semantic segmentation architecture to provide the pixel-by-pixel segmentation of welding features. The inclusion of the LSTM component allowed the system to capture temporal changes in the weld pool's geometry, providing deeper insights into the dynamic behavior of the welding process.

Also, Xiris offers the MeltPool AI™ tool to provide accurate melt pool segmentation in GMAW. See How Artificial Intelligence Detects Melt Pool Defects, Why Weld Pool Analysis is Important for Automated Welding

Figure 4. The U-Net fusion ConvLSTM network structure developed in [4].

5. Closed-Loop Control for Laser Metal Deposition

Researchers at the Technical University of Munich have taken a proactive approach to enhancing wire-based laser metal deposition through a closed-loop control system. In the studies “Segmentation-based closed-loop layer height control for enhancing stability and dimensional accuracy in wire-based laser metal deposition” [5] and “Real-time monitoring and control of the layer height in laser metal deposition with coaxial wire feeding using optical coherence tomography” [6], the team used a laser line scanner to monitor layer height during the deposition process.

The obtained layer height measurements were fed into a MATLAB-based control algorithm. A Xiris XVC-1000 camera was used for process monitoring to evaluate the process stability and the wire-melt pool interaction in the real-time. The closed-loop system developed in this project mitigated the effect of introduced height disturbances and could restore the uniform deposit height within several layers.

Figure 5. Image sequence used in the experiment in [5]. The images were taken with a Xiris XVC-1000 weld camera.

6. Multimodal Approach for GMAW Quality Control

Researchers from Busan University of Foreign Studies in South Korea created a multimodal quality control system for Gas Metal Arc Welding (GMAW). The model and findings are described in “A Novel Multimodal Approach for Gas Metal Arc Welding Quality Control” [7]. The proposed system integrates electrical data with high-resolution welding images captured by a Xiris XVC-1000e camera.

The collected images were merged with the spectrograms of electric parameters. The joint multimodal dataset was used to train a custom model to detect defective welds. With models such as EfficientNetB2 and ResNet50 used as visual encoders for the weld images, the system achieved an accuracy rate of 97.33% and 98.67%, respectively.

When only weld images were used to test the unimodal performance, the accuracy of ResNet50 model reached 95.38%. Electric data spectrograms, when used solo, could provide the accuracy of only 88.53% on ResNet50. The results prove a critical importance of clear high-resolution weld images for defect detection that can be obtained with Xiris weld cameras.

To learn how the clarity of the weld images can be improved even more with Near-Infrared, see How the XVC-750 Redefines Melt Pool Monitoring in Welding.

Figure 6. Multimodal model architecture proposed in [7].

7. Defect Detection in Narrow Gap GTAW with synthetic images by GANs

In the field of Narrow Gap Gas Tungsten Arc Welding (NGT or NG-GTAW), the study “Narrow gap GTAW defect detection and classification based on transfer learning of generative adversarial networks” [8] investigated a novel approach for defect detection.

Lack of weld defect images was overcome by the researchers by utilizing Generative Adversarial Networks (GANs) to synthesize defect images that then were used to enhance the training dataset. By training two classification networks on images of the weld pool, the scientist were able to distinguish between good and defective welds.

Xiris offers several advanced Machine Vision and AI segmentation models and Anomaly detection tools that can be used within WeldStudio Pro™. For more information and how these can be used in your application please contact to a Xiris product expert.

Figure 7. The narrow gap GTAW defects investigated in [8]: (a) good weld, (b) contamination, (c) unevenness.

8. Anomaly Detection in Additive Manufacturing

The study “Explaining the Anomaly Detection in Additive Manufacturing via Boosting Models and Frequency Analysis” [9] examined unsupervised learning anomaly detection models in wire arc additive manufacturing of INVAR36 alloy. To address the “black box” limitations of the current methods, the researchers extracted features from electric parameters in both time and frequency domains. These features were collected for both defective and defect-free part and used to train several anomaly detection models. The performance of the models was verified based on Xiris XVC-1000 weld camera images and surface appearance. As a result, the researchers could identify key features in the electric arc parameters that help improve the models aimed for identifying potential issues in additive manufacturing early in the process.

Xiris offers several advanced Machine Vision and AI segmentation models and Anomaly detection tools that can be used within WeldStudio Pro™. For more information and how these can be used in your application please contact to a Xiris product expert

Figure 8. The experimental setup used by [9]. A Xiris XVC-1000 weld camera was used for process monitoring.

Conclusion

The wide range of applications of Xiris cameras in industrial research highlights their ability to monitor welding and AM processes with high precision. Whether in laser metal deposition, wire arc additive manufacturing, GMAW, or GTAW, the use of thermal and visual weld cameras like the Xiris XIR-1800 and XVC-1000 models helps drive the progress in automation, quality control, and process optimization. The integration of Xiris imaging solutions with machine learning is helping shape the future of manufacturing technologies enabling greater accuracy, efficiency, and productivity.

References

[1] Wen, Qiuling, Hui Wang, Xipeng Xu, Jing Lu, Hui Huang, and Feng Jiang. "Study on nanosecond laser-induced iron plasma reactive etching of single-crystal CVD diamond." Optics & Laser Technology 177 (2024): 111071.

[2] Sousa, Joao, Roya Darabi, Armando Sousa, Frank Brueckner, Luís Paulo Reis, and Ana Reis. "JEMA: A Joint Embedding Framework for Scalable Co-Learning with Multimodal Alignment." arXiv preprint arXiv:2410.23988 (2024).

[3] Yao, Mingpu, Jie Sheng, Fanrong Kong, and Wei Tong. "A multi-sensor based online monitoring system for laser hot-wire surface cladding process." Optics & Laser Technology 177 (2024): 111074.

[4] Li, Tianpu, Yue Cao, and YuMing Zhang. "Analysis of weld pool region constituents in GMAW for dynamic reconstruction through characteristic enhancement and LSTM U-Net networks." Journal of Manufacturing Processes 127 (2024): 573-588.

[5] Bernauer, Christian, Philipp Leitner, Avelino Zapata, Pawel Garkusha, Sophie Grabmann, Maximilian Schmoeller, and Michael F. Zaeh. "Segmentation-based closed-loop layer height control for enhancing stability and dimensional accuracy in wire-based laser metal deposition." Robotics and Computer-Integrated Manufacturing 86 (2024): 102683.

[6] Bernauer, Christian, Sebastian Thiem, Pawel Garkusha, Christian Geiger, and Michael F. Zaeh. "Real-time monitoring and control of the layer height in laser metal deposition with coaxial wire feeding using optical coherence tomography." Journal of Laser Applications 36, no. 4 (2024).

[7] Mustafaev, Bekhzod, Sung Won Kim, and Eung Soo Kim. "A Novel Multimodal Approach for Gas Metal Arc Welding Quality Control." In 2024 International Conference on Control, Automation and Diagnosis (ICCAD), pp. 1-6. IEEE, 2024.

[8] Yu, Zhengxiao, Ninshu Ma, Hao Lu, Hetong Yang, Weihua Liu, and Ye Li. "Narrow gap GTAW defect detection and classification based on transfer learning of generative adversarial networks." Journal of Manufacturing Processes 131 (2024): 2350-2364.

[9] Vozza, Mario, Joseph Polden, Giulio Mattera, Gianfranco Piscopo, Silvestro Vespoli, and Luigi Nele. "Explaining the Anomaly Detection in Additive Manufacturing via Boosting Models and Frequency Analysis." Mathematics 12, no. 21 (2024): 3414.