Elsevier

Additive Manufacturing

Volume 36, December 2020, 101435
Additive Manufacturing

Automated layerwise detection of geometrical distortions in laser powder bed fusion

https://doi.org/10.1016/j.addma.2020.101435Get rights and content

Abstract

In-situ layerwise imaging in laser powder bed fusion (L-PBF) has been implemented by many system developers to monitor the powder bed homogeneity. Increasing attention has been recently devoted to the possibility of using the same sensing approach to detect also in-plane and out-of-plane geometrical distortions of the part while it is being produced. To this aim, seminal works investigated the suitability of various image segmentation algorithms and assessed the accuracy of layerwise dimensional and geometrical measurements. Nevertheless, there is a lack of automated methods to identify, in-situ and in-process, geometrical defects and out-of-control deviations from the nominal geometry. This study presents a methodology that combines an active contours methodology for image segmentation with a statistical process monitoring approach suitable to deal with complex geometries that change layer by layer. The proposed approach enables a data-driven and automated alarm rule to detect the onset of geometrical distortions during the build by comparing the slice contour reconstruction with the nominal geometry in each layer. Moreover, by coupling edge-based and region-based segmentation techniques, the method is sufficiently robust to be applied to imaging and illumination setups that are already available on industrial L-PBF systems. The effectiveness of the proposed approach was tested on a real case study involving the L-PBF of complex Ti6Al4V parts that exhibited local geometrical distortions.

Introduction

The layerwise production paradigm in powder bed fusion processes enables the opportunity to gather a large amount of data while the part is being produced, which can be used to support qualification procedures and to anticipate the detection of defects and unstable process states [1], [2]. In laser powder bed fusion (L-PBF), it is possible to measure several quantities of interest at different levels and with different sensing methods [3]. The first level regards the acquisition of images of the entire powder bed, possibly before and after each recoating. A second level regards the measurement of so-called “process signatures” during the melting phase of each scanned track. This usually entails high-speed video imaging methods to capture fast and transient phenomena related to the cooling history or the ejection of process by-products. One last level regards the monitoring of the salient melt pool properties, which is commonly achieved by using co-axial sensing methods exploiting the optical path of the laser. This study focuses on the first level.

In this framework, increasing attention has been devoted to the use of layerwise images of the powder bed either to identify recoating errors and powder bed in-homogeneities [4], [5], [6] or to identify surface and geometrical defects in the printed area [7], [8], [9], [10], [11], [12], [13]. Some of these methods are particularly appealing for a direct industrial implementation, as in most cases they exploit in-situ sensing architectures that are either already available in industrial L-PBF systems or easy to integrate [3]. Various authors pointed out that powder bed images acquired after the laser scan of each layer could be used to detect both in-plane and out-of-plane defects. Various seminal studies focused on the identification of recoating errors and powder bed inhomogeneities by using layerwise images gathered after the powder recoating operation [4], [5], [6], and some of these methods are already implemented by most L-PBF system developers. Another stream of research regards the detection of out-of-plane irregularities in the printed area, like so-called super-elevated edges, and the measurement of the surface topography of the slice [9], [10], [14], [15], [16], [17], [18], [19]. This stream is motivated by the fact that irregular surface patterns may introduce an undesired variability in the local powder thickness, affecting the actual energy density provided to the material. Surface irregularities may also interfere with the recoating operation, producing additional defect propagations within the build area.

This study focuses on a different problem, namely the in-situ detection and characterization of in-plane geometrical deviations. In this framework, the seminal study of Foster et al. [4] demonstrated the feasibility of in-situ 3D reconstruction of the part geometry by segmenting layerwise images. As pointed out by Caltanissetta et al. [12], part dimensions and geometries measured in-situ may be not representative of the final dimensions and geometry of the as-built part, as some deviations, including shrinkage and thermal stress-induced distortions, may be not captured on a layer-by-layer basis. However, when a major departure from the nominal shape is observed in one layer, it is worth signalling as soon as possible, since it may indicate a defect that cannot be recovered as the process goes on.

Aminzadeh [20], Aminzadeh and Kurfess [11] investigated the accuracy of in-situ contour detection in L-PBF by comparing the identified contours against a manual segmentation applied to the same images. The segmentation approach proposed by the authors combined histogram-based thresholding with image pre-filtering and morphological operations. Caltanissetta et al. [12] presented an in-situ measurement performance characterization analysis based on a different family of image segmentation methods, i.e., active contours [21]. Caltanissetta et al. [12] used ex-situ optical measurements as ground truth to determine the accuracy of layerwise contour detection. They pointed out that the pure measurement error was up to one order of magnitude lower than the total measurement variability affected by part-to-part and build-to-build variations, showing that in-situ reconstructions could be adequate for macro-geometrical distortion detection.

Other authors combined contour detection with surface pattern analysis for out-of-plane defect detection and for layerwise topography reconstruction. Abdelrahman et al. [22] proposed a method to automatically detect surface anomalies in layerwise images related to uneven surface patterns within the laser printed area. They applied the active contours algorithm to register the nominal slice contour to in-situ images in a pre-processing phase. Li et al. [8] proposed the active contours methodology to identify the region of interest consisting of the printed area within the layer, and then a topography map was estimated via fringe projection coupled with stereo imaging within that region.

In all the aforementioned studies, an accurate contour detection of the printed slice was needed, either to directly determine the presence of in-plane distortions or to identify the region of interest for the following application of surface pattern analysis methods. In this framework, one open issue consists of the lack of effective and robust methods for contour detection of complex geometries in L-PBF layerwise images. Indeed, the seminal studies that investigated the accuracy of in-situ reconstructed contours involved simple and layerwise invariant shapes [12], [11]. As shown in Fig. 1, segmentation methods previously proposed in the literature [12], [22] may fail in the presence of more complicated geometries and images gathered with industrial settings. Fig. 1A shows a layerwise image acquired with the powder bed camera installed on an EOS M290 during the production of a complex shape. Fig. 1B and C show, respectively, the slice contour reconstruction with the region-based methods proposed by Abdelrahman et al. [22] and Caltanissetta et al. [12]. In both cases, the contour reconstruction is poorly representative of the real slice geometry, especially in critical features like thin walls. As a matter of fact, layerwise images in L-PBF exhibit several challenges, e.g., noisy patterns, non-homogeneous pixel intensity patterns and not well-defined edges of foreground areas, which may limit the performances of image segmentation techniques, including traditional active contours-based methods.

A further open issue regards the lack of methods to automatically identify and signal a deviation between the in-situ reconstructed contours and the nominal geometry of the slice. Although control charts have been proposed in previous studies to monitor the powder bed homogeneity [6], there is a lack of such tools for in-line monitoring of the printed geometry. The contribution of the present study is aimed at specifically filling this gap by proposing a novel methodology that combines a robust image segmentation approach with a statistical process monitoring technique for automated in-situ detection of geometrical distortions. The proposed approach relies on an active contours formulation that combines the benefits of both edge-based and region-based segmentation methods [23] instead of relying only on region-based descriptors as done in previous studies mentioned above. The inclusion of local pixel intensity gradients into the active contours energy minimization function is expected to make the image segmentation more robust to illumination conditions and machine vision equipment available in industrial systems and more suitable to deal with complicated geometries. In-situ reconstructed contours are then aligned and compared against the nominal geometry of the slice leading to a deviation map that can be synthesized into a univariate deviation metric. A control charting scheme is then applied to this metric to automatically signal any out-of-control geometrical distortion by taking into account the natural layer-by-layer variability of the deviation measurements. To this aim, an adaptive control chart is proposed to deal with geometries that continuously change from one layer to another.

The method was tested on an EOS M290 by using the powder bed imaging already available in the system. An experimental study was carried out by producing a complicated geometry with different orientation and support configurations, to induce the occurrence of geometrical errors and demonstrate their in-situ detectability. A further experimentation, involving simpler specimens, was performed to characterize the performances of the proposed approach in terms of false alarm rate.

Section 2 presents a motivating example that is used as real case study to illustrate and test the method. Section 3 presents the proposed methodology. Section 4 describes the achieved results and Section 5 concludes the paper.

Section snippets

Test case

An experimental study was carried out by producing a specimen specifically designed to include various complex geometrical features like thin walls, acute corners, massive parts with internal holes, overhang regions, etc. This kind of specimen was selected as it allowed us not only to test the suitability of the proposed approach in the presence of a complex geometry that evolves layer by layer, but also to induce distortions in the part as a consequence of the geometry itself and its

Methodology

Fig. 4 shows the scheme of the proposed methodology, which includes four major steps, i.e., image pre-processing, image segmentation, deviation map estimation and statistical monitoring of the deviation from the nominal geometry. The following sub-sections describe these four steps with examples from the case study introduced in Section 2.

Preliminary test under in-control conditions

The proposed approach was first tested by means of a build including 3 simple cylindrical specimens of 12 mm diameter and 20 mm height produced in Ti6Al4V with the EOS M290 system. Default and fixed process parameters for the Ti6Al4V powder were used. A total of 333 layers were monitored for each specimen. The powder bed camera embedded into the EOS M290 system with default illumination and image acquisition settings defined by the system developer was used (see Section 2). The aim of this

Conclusions and future developments

In-situ sensing and monitoring methodologies have been gaining a continuously increasing attention in L-PBF to anticipate the detection of defects during the process, and to support the qualification of additively produced parts. The layerwise production paradigm enables the opportunity to gather a large amount of data while the part is being produced.

This study proposed a methodology to take advantage of layerwise imaging to identify, in an automated way, geometrical distortions in terms of

Authors’ contributions

bf Luca Pagani: Conceptualization, methodology, software, investigation, validation, writing.

Marco Grasso: conceptualization, methodology, software, investigation, validation, writing.

Paul J. Scott: Supervision, funding acquisition.

Bianca M. Colosimo: conceptualization, supervision, writing – review and editing, funding acquisition.

Conflict of interest

The authors declare no conflict of interest.

Declaration of Competing Interest

The authors report no declarations of interest.

Acknowledgements

The authors are grateful to BeamIT SpA and EOS GmbH, which provided the facilities and system to carry out the experimentation. The authors are also grateful to Eng. Stefano Grulli, who designed the samples used in the real case study. LP and PJS acknowledge the UKs Engineering and Physical Sciences Research Council (EPSRC) funding the grant Ref. EP/R024162/1. MG and BMC acknowledge the Italian Ministry of Education, University and Research for the support provided through the Project

References (34)

  • B.M. Colosimo

    Modeling and monitoring methods for spatial and image data

    Qual. Eng.

    (2018)
  • M. Grasso et al.

    Process defects and in situ monitoring methods in metal powder bed fusion: a review

    Meas. Sci. Technol.

    (2017)
  • B. Foster et al.

    Optical, layerwise monitoring of powder bed fusion

  • L.T. Phuc et al.

    A high-resolution and large field-of-view scanner for in-line characterization of powder bed defects during additive manufacturing

    Mater. Des.

    (2019)
  • T. Craeghs et al.

    Online quality control of selective laser melting

  • F. Imani et al.

    Process mapping and in-process monitoring of porosity in laser powder bed fusion using layerwise optical imaging

    J. Manuf. Sci. Eng., Trans. ASME

    (2018)
  • M. Aminzadeh et al.

    Vision-based inspection system for dimensional accuracy in powder-bed additive manufacturing

    ASME 2016 11th International Manufacturing Science and Engineering Conference

    (2016)
  • Cited by (39)

    • In-situ 3D contour measurement for laser powder bed fusion based on phase guidance

      2023, Theoretical and Applied Mechanics Letters
      Citation Excerpt :

      There will be local overexposure when the camera is fixed. To reduce the effect of noise caused by local specular reflection, we also need to perform gray correction [31] on the image. Generally, the surface of the powder bed is diffusely reflective, so the gray intensity follows a Gaussian distribution (the gray intensity of this work fluctuates in the range of 60∼80).

    View all citing articles on Scopus
    View full text