Content area
Wire Are Additive Manufacturing (WAAM) is a technique used for 3D printing or repairing metal parts by progressively layering metal until the desired 3D structure is fully formed. This research focuses on developing realtime, in-process monitoring and quality control techniques to evaluate multiple quality indicators continuously throughout the WAAM process. It further explores methods for detecting and classifying defects, such as porosity, cracks, lack of fusion, and geometric deviations, that may occur during WAAM. To improve defect detection and streamline processes, predictive models are built using machine learning or physics-based simulations. These models analyze shifts in process parameters to anticipate potential flaws, empowering teams to make timely, data-driven corrections. 3D scanning is employed to capture the physical form of manufactured parts as point cloud data, which is processed on a computer using Visual Studio code. A complete 360° scan takes approximately two minutes, with the data processing requiring only about five seconds. The resulting point cloud data is then visualized and processed where geometrical attributes such as curvature and volume density are calculated and represented as distribution curves, aiding in clarity and precision during analysis. By examining these visualized parts, specific defects can be identified and categorized based on distinctive characteristics. Furthermore, Gaussian distribution curves are utilized to calculate the percentage of defective areas, offering quantitative insights for effective quality control and helping ensure that the final product meets desired specifications.
Abstract
Wire Are Additive Manufacturing (WAAM) is a technique used for 3D printing or repairing metal parts by progressively layering metal until the desired 3D structure is fully formed. This research focuses on developing realtime, in-process monitoring and quality control techniques to evaluate multiple quality indicators continuously throughout the WAAM process. It further explores methods for detecting and classifying defects, such as porosity, cracks, lack of fusion, and geometric deviations, that may occur during WAAM. To improve defect detection and streamline processes, predictive models are built using machine learning or physics-based simulations. These models analyze shifts in process parameters to anticipate potential flaws, empowering teams to make timely, data-driven corrections. 3D scanning is employed to capture the physical form of manufactured parts as point cloud data, which is processed on a computer using Visual Studio code. A complete 360° scan takes approximately two minutes, with the data processing requiring only about five seconds. The resulting point cloud data is then visualized and processed where geometrical attributes such as curvature and volume density are calculated and represented as distribution curves, aiding in clarity and precision during analysis. By examining these visualized parts, specific defects can be identified and categorized based on distinctive characteristics. Furthermore, Gaussian distribution curves are utilized to calculate the percentage of defective areas, offering quantitative insights for effective quality control and helping ensure that the final product meets desired specifications.
Keywords Wire Arc Additive Manufacturing, Quality Control, Manufacturing Systems, Industrial Engineering.
1. Introduction
Wire Arc Additive Manufacturing (WAAM) is an advanced additive manufacturing (AM) technique that fabricates complex, three-dimensional, near-net-shape metal components by depositing layers sequentially using welding principles. Combining arc welding with 3D printing, WAAM uses metal wire as feedstock, offering cost-effective solutions for producing large, intricate parts that are challenging or uneconomical to manufacture through conventional methods. Unlike laser- or electron beam-based AM processes, WAAM is not constrained by chamber size, enabling the creation of components with minimal dimensional limitations. However, maintaining precise control over process parameters-such as heat input, cooling rates, deposition speed, and wire feed rates are critical to ensuring consistent product quality. Deviations in these parameters can lead to defects like porosity, residual stress, distortion, or cracking, compromising structural integrity. This underscores the importance of robust quality assurance systems, as the layer-by-layer nature of WAAM amplifies the risk of cumulative defects. Detecting anomalies at each deposition stage is essential to meet stringent industry standards in sectors such as aerospace, automotive, and energy, where WAAM is increasingly adopted. Real-time monitoring systems are vital for instant feedback, allowing operators to adjust parameters mid-process to mitigate defects. Effective defect detection ensures compliance with these requirements, thereby reducing the risk of non-conformance. Since defects occurred during the welding process will form internal defects if not detected and repaired in time [1], real-time defect detection systems offer immediate feedback, enabling operators to adjust parameters on the spot. Detection methods include developing physical models (CAD models) or spatial models (point cloud models) of the process for monitoring and defect prediction, alongside utilizing existing data to create models and capturing current data for fault identification [2].
2. Problem Description
Although the mechanical properties of WAAM parts are considered moderate compared to traditionally manufactured counterparts [3], defects remain a significant challenge. These defects arise from factors such as heat input and material parameters. As shown in Fig. la, porosity is one of the most prevalent defects in WAAM, often forming due to improper control of shielding gases (e.g., Argon, Helium), welding parameters, or chamber conditions. Internal voids result from dynamic fluid behavior during deposition, affecting structural integrity. Cracks (Fig. lb) develop due to rapid heating and cooling cycles, creating thermal gradients that induce fractures. If undetected, these cracks propagate, leading to part failure [2]. Fig. 1c illustrates lack of fusion, where adjacent layers or passes fail to bond, resulting in unbonded regions and structural weaknesses. Given WAAM's layer-by-layer nature, defects in one layer affect subsequent layers, ultimately compromising the entire component [4]. Residual stress, depicted in Fig. Id, arises from uneven thermal contraction during cooling, leading to localized stress concentrations. Even after external loads are removed, these stresses cause distortions, twisting, and inaccuracies in shape and size. They can also induce fractures within layers and reduce fatigue performance in deposited structures [5]. Each of these defects significantly impact WAAM's mechanical integrity, requiring advanced detection and mitigation strategies to ensure component reliability.
3. Related Research
Current methodologies for defect detection in Wire Arc Additive Manufacturing (WAAM) emphasize real-time monitoring and machine learning integration. Clark et al. proposed a multi-modal sensor fusion system combining acoustic emission data and convolutional neural networks (CNNs) to classify defects during welding [6]. Their approach captures audio waveforms from the welding process, allowing the CNN model to analyze acoustic signatures linked to surface-level anomalies such as spatter or arc instability. By synchronizing sensor data with robotic toolpath coordinates, the system enables spatial defect tracking and achieves over 90% prediction accuracy. However, its reliance on precise sensor calibration makes it susceptible to environmental noise (e.g., electromagnetic interference). Additionally, it cannot detect subsurface flaws like internal porosity or cracks, necessitating post-process radiographic inspection.
Franke et al. developed a hybrid monitoring system that integrates deep convolutional neural networks (DCNNs) with classical image processing to detect surface defects and maintain process stability [7]. Their method employs semantic segmentation on real-time video feeds from a welding camera, focusing on critical parameters such as nozzle-toworkpiece (NtW) distance and horizontal wire positioning. The system identifies spatter formation-often caused by improper shielding gas flow or incorrect voltage-and triggers immediate parameter adjustments. This approach is particularly effective for handling complex geometries, such as overhangs or curved features, by dynamically adapting to geometric deviations. However, the computational demands of real-time DCNN processing require high (ProQuest: ... denotes formulae omitted.)performance GPUs, limiting industrial scalability. Furthermore, its focus on surface defects (e.g., spatter, geometric inaccuracies) leaves internal flaws like porosity or interlayer fusion issues undetected.
Liu et al. adopted a machine vision-based strategy using 3D point cloud analysis to evaluate geometric defects in WAAM components [8]. Their system employs high-resolution laser scanners to capture detailed surface topography after each deposition layer. Raw point cloud data undergoes noise reduction via smoothing algorithms, enhancing visibility of defects like humps, depressions, and surface porosity. The analysis combines 2D height profiles and 3D curvature metrics, comparing measured geometries against an ideal CAD model to flag deviations. For instance, abnormal local curvature or irregular normal vectors signal surface defects. While this method achieves high precision in detecting macroscopic geometric flaws and integrates with real-time workflows, its effectiveness is limited by laser scanner resolution, struggling to identify submillimeter cracks or fine porosity. Furthermore, the system is optimized for specific defect types (e.g., humps, porosity) and cannot address non-geometric flaws, such as residual stresses or microstructural inconsistencies, requiring complementary techniques for comprehensive quality assurance.
This paper seeks to bridge the current research gaps by developing a integrated multi-modal defect detection approach that enhances defect identification and incorporates geometric defect analysis using different analysis methodologies.
4. Methodology
As shown in Figure 2, the system used for this research was designed for DMS Hybrid Hero CNC machines which are suited for process development and research, or production runs of small parts. Each 3D scanning system has one 1280p resolution DLP projector and two 5.4 Mega Pixel Lucid Tritons as the spatial resolution can reach as low as 5 pm. Four sets of scanning systems are placed above and around the substrate, covering four corners of the substrate. The components for the scanning system are highly customized since there are no similar systems on the market.
A point cloud is a dense collection of data points within a three-dimensional coordinate system, often used to represent the external surface of an object in high detail. Each point in the cloud includes x, y, and z coordinates, with additional attributes like color, intensity, or normal vectors in some cases. According to Tang et al.'s analysis methods [9], an ideal plane can be generated by fitting the 3D point cloud data expressed as:
...
Where A, B, C, and D are constant plane equation parameters. The total distance for all points from the ideal plane is:
...
Where dt is the distance between each point and the ideal plane. Based on Tang et al's work [9], we improved the analysis approaches by introducing mean curvature, volume density, and arithmetic surface finish generated by 3D point cloud analysis software. The mean curvature can be expressed as:
(ProQuest: ... denotes formulae omitted.)...
Where kT represents the maximum normal curvature and k2 represents the minimum normal curvature at a given point on the surface. The volume density of at point p within a given radius r can be expressed as:
...
Where p(p) is the local volume density at point p, Nr(p) is the number of neighboring points within a radius r. The arithmetic surface finish Ra can then be calculated as:
...
Where N is the number of points. Mean curvature, volume density, and surface roughness are key factors in 3D point cloud analysis. Mean curvature measures surface bending, enabling detection of irregularities. Volume density quantifies point concentration in a 3D region, making it ideal for identifying localized defects. Surface roughness evaluates deviations from a smooth geometry, allowing recognition of surface irregularities through structural comparison. These three methods will be later applied and compared in section 5.
5. Results
Point clouds are typically created using 3D scanning technologies such as laser scanners, LiDAR, or structured light scanners, which capture an object's shape and structure by sampling points across its surfaces. The resulting data, displayed as point cloud data, offers a detailed snapshot of the scanned object's geometry. An open-source software called CloudCompare is used to analyze the point cloud data [10]. The software can generate 3D meshes from point clouds using algorithms like Poisson Surface Reconstruction, meshes can then be analyzed or exported for further processing, meshes can also be edited and refined using tools such as smoothing, hole filling, and decimation. Segmentation and data filtering are crucial steps in CloudCompare while processing and analyzing 3D point cloud data since noise reduction, data simplification, and ROI isolation are essential.
As shown in Figure 3, only the parts above the substrate were kept, which reduces the points count from 996,506 to 32,756. Defects like porosity, voids, or lack of fusion are prevalent in WAAM processes, creating regions with lower density. In WAAM, incomplete fusion between layers may lead to low-density zones, a serious defect that can cause delamination or failure when subjected to stress. Several geometric features are analyzed such as curvature, volume density, and surface roughness. The results of different approaches are shown below:
According to Figure 4, we can find out the volume density map has a more significant result with wider range and more equal distribution. Also, volume density can detect both surface and internal defects, and it has higher sensitivity to small defects as it provides more comprehensive 3D information of the object. Therefore, we will choose to analyze the volume density for defect. The defected region exhibits lower volume density compared to the surrounding areas. In the actual image, the upper portion of the scanned part shows significant porosity defects, indicated by red highlights in the volume density map. Additionally, in the actual image, the yellow-highlighted area shows substantial geometric irregularities, another form of defect. This is similarly reflected on the volume density map, where a horizontal band with reduced volume density stands out from the neighboring areas. A histogram between the counts of points and volume density can be generated , as shown in Figure 5, the average surface density is around 1.8 with a standard deviation of 1.027. By using this histogram and its gauss distribution, we can estimate the percentage of defect for the manufactured part by setting a threshold for defect characterization based on the volume density distribution which the count of points will be illustrated on the histogram to further examine its surface finish and quality.
6. Conclusions
CloudCompare is a powerful tool for defect detections WAAM with its precise processing, visualization, and analysis of 3D point cloud data. Its segmentation, filtering, and volumetric analysis features allow efficient identification of defects like porosity, crack, lack of fusion, and residual stress. This non-destructive approach allows visualization of internal structures, ensuring high-quality standards and compliance with specifications. By enabling early defect detection and WAAM quality control, part reliability and process optimization can be improved. Future advancements could expand its utility including integrating CloudCompare with deep learning frameworks could automate defect classification and enable real-time toolpath adjustments, while supporting predictive maintenance. Reducing scanning time via in-situ scanning during the cooling phases can enhance efficiency for mass production. Higher-precision scanners would improve spatial resolution, yielding finer defect detection in point cloud models. Additionally, a closed-loop control system could link real-time defect data to automated WAAM parameter adjustments, optimizing layer deposition and preventing defect propagation. These innovations would refine defect detection accuracy and enable proactive process corrections, advancing WAAM toward scalable, defect-free manufacturing.
References
[1] J. Wu, C. Huang, Z. Li, R. Li, G. Wang, and H. Zhang, "An in situ surface defect detection method based on improved you only look once algorithm for wire and arc additive manufacturing," Rapid Prototyping Journal, vol. 29, issue 5, p. 910-920, doi: https://doi.Org/10.l 108/RPJ-06-2022-0211.
[2] X. Chen et al., "A review on wire-arc additive manufacturing: typical defects, detection approaches, and multisensor data fusion-based model," The International Journal of Advanced Manufacturing Technology, vol. 117, no. 3-4, pp. 707-727, Aug. 2021, doi: https://doi.org/10.1007/s00170-021-07807-8.
[3] B. Tomar, S. Shiva, and T. Nath, "A review on wire arc additive manufacturing: Processing parameters, defects, quality improvement and recent advances," Materials Today Communications, vol. 31, p. 103739, Jun. 2022, doi: https://doi.org/10.1016/j .mtcomm.2022.103739.
[4] AT. Albannai, "Single view - International Journal of Current Science Research and Review," International Journal of Current Science Research and Review, 2020. https://ijcsrr.org/single-view/?id=8196&pid=8124.
[5] M. D. Barath Kumar and M. Manikandan, "Assessment of Process, Parameters, Residual Stress Mitigation, Post Treatments and Finite Element Analysis Simulations of Wire Arc Additive Manufacturing Technique," Metals and Materials International, vol. 28, no. 1, pp. 54-111, Aug. 2021, doi: https://doi.org/10.1007/sl2540-Q2101015-5.
[6] B. Clark, S. I. Mohsin, H. A. Poonawala, Fazleena Badurdeen, R. Brennan, and I. S. Jawahir, "Real-time in-situ defect detection via multi-modal classification of waveforms for wire arc additive manufacturing," Journal of Intelligent Manufacturing, Jan. 2025, doi: https://doi.org/10.1007/s 10845-024-02561 -6.
[7] J. Franke, F. Heinrich, and R. T. Reisch, "Vision based process monitoring in wire arc additive manufacturing (WAAM)," Journal of Intelligent Manufacturing, Mar. 2024, doi: https://doi.org/10.1007/sl0845-023-02287-x.
[8] M. Liu et al., "Detection and quantitative evaluation of surface defects in wire and arc additive manufacturing based on 3D point cloud," Virtual and Physical Prototyping, vol. 19, no. 1, Dec. 2023, doi: https://doi.org/10.1080/17452759.2023.2294336.
[9] S. Tang, G. Wang, and H. Zhang, "In situ 3D monitoring and control of geometric signatures in wire and arc additive manufacturing," Surface Topography: Metrology and Properties, vol. 7, no. 2, p. 025013, May 2019, doi: https://doi.org/10.1088/2051-672x/ablc98. [101 https://github.com/cloudcompare/cloudcompare
Copyright Institute of Industrial and Systems Engineers (IISE) 2025