Keywords

1 Introduction

There are a variety of technologies for digitally acquiring the shape of 3D objects. The techniques work with many sensor types including optical, acoustic, laser scanning, radar, thermal, and seismic. A well-established classification divides them into two types: contact and non-contact [1]. Non-contact solutions can be further divided into two main categories, active and passive. There are a variety of typologies that fall under each of these categories.

The usual result obtained from measurements based on 3D scanning is represented by a considerable number of points, called in the literature point cloud [2]. The point cloud is a collection of points, defined as position by XYZ coordinates in a common system of reference, which reveals to the observer information about the spatial shape, position, and distribution of an object or group of objects. Other types of data obtained directly from the scanning process or based on the point cloud processing are polygon mesh models, surface models and solid CAD models.

2 Current Applications and Directions

The potential use of 3D point clouds remains an emergent topic among researchers for a vast variety of applications in the field of civil engineering and construction industry. For example, Vacca et al. [3] presents an interesting approach to use the terrestrial laser scanner for monitoring the deformations and the damage of buildings.

A highly relevant contribution to this field is proposed by Funari et al. [4], developing a framework for digital twin generation of historic masonry structures based on point clouds. The paper has shown the benefits of being able to monitor in real-time the evolution of the behaviour of existing structures. The proposed procedure exploits the capabilities of Generative Programming, in which the user can interact with the code by modifying and/or implementing new capabilities with the aim of obtaining a solid model based on point clouds. Pepe et al. [5] presents a comprehensive comparison of latest-generation 3D scanners in reconstruction processes of elements belonging to Cultural Heritage. Angjeliu et al. [6] developed a simulation model for Digital Twin applications in historical masonry buildings and presented the integration between numerical and experimental data. For this scope the finite element structural model of the Cathedral of Milan was obtained based on photogrammetric geometric measurements, extensive in situ 3D scans (LIDAR) and other archive data. The analysis provided information on the distribution of the internal forces, displacements, and damage. Guarnieri et al. [7] proposed a 3D scan model-based structural analysis on a structure belonging to Cultural Heritage that was subjected to three different effects: self-weight, wind load and out off-plumb effect. Herban et al. [8] investigated an innovative approach to obtain the 3D model of a structure belonging to Cultural Heritage, based on spherical photogrammetry.

Hu et al. [9] proposed numerical modelling for landslide analysis based on scans with LIDAR. The paper extends literature on manual and automatic approaches for converting data from point clouds to FEM numerical models. Castellazzi et al. [10] proposed a new semi-automatic procedure, called CLOUD2FEM, to transform point clouds of complex objects to finite element models.

Xie et al. [11] developed a 3D modelling algorithm for tunnel deformation monitoring based also on terrestrial laser scanning (LIDAR). Cui et al. [12] developed another interesting approach for generating finite element mesh from 3D laser point clouds for tunnel structures. The point cloud model of the tunnel was acquired using terrestrial laser scanning and mobile laser scanning. Fernandez et al. [13] suggests numerical model development from 3D scans applied to corroded steel bars, in order to investigate the failure process and local effects on the pits.

Zhao et al. [14] presents a procedure for processing 3D point clouds that are generated from laser-based scanning of a cold-formed steel member into useful measurements of cross-section dimensions, as well as for use in finite element simulations of the as-measured geometry. The paper demonstrated the use of laser scans for dimension control providing comparisons to manual measurements and nominal manufactured specifications, in case of cold formed elements.

3 3D Scanning of Steel Links

3.1 Research Framework

The finite element method (FEM) is commonly adopted for numerical investigation of the behaviour of various structural components. The most refined analysis is currently considered the geometrically and materially nonlinear analysis with imperfections (GMNIA). As it is clearly stated by its name, the analysis method requires modelling of geometrical imperfections, which is particularly important for numerical verification of buckling resistance. The conventional FE modelling of geometrical imperfections of an experimental specimen involves:

  1. 1)

    Modelling the “perfect” shape of the specimen using nominal or measured cross-section dimensions.

  2. 2)

    Modelling of the geometrical imperfections by applying an appropriately scaled deviation obtained from buckling analysis. As imperfections are difficult to measure, they are usually obtained from literature or specific codes (e.g. EN 1993-1-5).

The use of 3D scanning can simplify the process of FE modelling due to the fact that the point cloud of the specimen includes the geometry and also the initial imperfections that can be used directly in the numerical analysis, if steps are taken to post-process the point cloud into a model supported by the FE analysis software.

3.2 Experimental Specimens

In the framework of the HYLINK research project [15], and for the scope of this paper, it was proposed to measure the geometry and initial imperfections of replaceable steel links using 3D scanning technology and to develop a numerical model for validation of the experimental tests. The test specimens that were investigated are presented in Table 1 and Fig. 1. A total number of three steel links, with and without stiffeners were subjected to cyclic tests. A schematic and a typical longitudinal section view of a link is illustrated below.

Table 1. Characteristics of experimental link specimen.
Full size table
Fig. 1.
figure 1

Geometry of the link specimens.

Full size image

3.3 3D Scanning of the Steel Links

For the aim of measuring the geometry and initial imperfections of the steel links, 5 different scanners (Table 2) based on 3 scanning principles (LiDAR, blue laser, and white structured light) were selected and applied to compare their workflow, accuracy, and time demand for obtaining the measurements (Table 3).

Table 2. Technical specifications of the used scanners.
Full size table
Table 3. Comparison between the five scanner systems.
Full size table

As shown above each technology has positive and negative aspects that can influence the final raw product of the scans – the point cloud. In case of three (Z+F, Hexagon, Iphone) out of the five scanners, the technology implies that the measured specimen keeps its initial position, which leads to unscanned areas. The measurement time varies between 15–20 min in the case of the smartphone and up to 90 min for the structured light scanner. Most technologies require manual pre-measurement calibration, excepting the terrestrial laser scanner (TLS) Z+F IMAGER and the smartphone. Reflective targets are also demanded excepting the Hexagon Absolute Arm and Iphone. Sensitivity to reflective surfaces is an aspect observed during the scans just in the case of the structured light scanner and smartphone.

4 Post-processing of the Scanned Data and Finite Element Modelling

4.1 General

From the structural perspective, point clouds cannot be used directly for 3D modelling and for numerical analyses because they are formed by many discrete points defined by three-dimensional coordinates. To effectively use the geometric data derived from the 3D scanning measurements, it is necessary to perform operations that transform the point cloud into a 3D model. The procedure to obtain a solid model compatible with a FEA software (for example Abaqus) from point clouds could be sometimes sophisticated (Fig. 2).

Fig. 2.
figure 2

Workflow of post-processing point clouds and generating solid models

Full size image

Fortunately, the current features of most of the post-processing software bundled with 3D scanners, include the possibility to convert in a semi-automatic manner the point cloud into a mesh, reducing the time spent for processing purposes. However, for the moment LiDAR terrestrial laser scanners do not offer a converter software. In comparison with the other selected scanners, the Z+F LaserControl software facilitates just the point cloud alignment, while the exported formats are also point clouds. Other third-party software, such as CloudCompare need to be used for specific post-processing steps to obtain a mesh model, and Autodesk Fusion 360 to obtain the final solid model.

4.2 Manual Mesh Generation from Point Clouds

In case of the point clouds obtained from the TLS, several steps were considered to obtain a mesh model. First, the aligned point cloud was exported from Z+F LaserControl as e57 file format and imported in the open-source software CloudCompare for post-processing. The first step consisted in identifying the measured specimen in the dense point cloud of each of the six scanning stations. For redundant data removal, segmentation was used in this case, by defining 2D polygons in the point cloud. Next, a single entity was generated. Since the measurements took place in laboratory conditions, and the scanned specimens were relatively small, a manual noise reduction through further segmentation was chosen instead of other removal procedures.

In order to obtain the mesh model of the link, the next step was to compute ÿ normal on the point cloud. The normal vector to a surface is a vector which is perpendicular to the surface at a given point. When normal vectors are considered on closed surfaces as in this case, the inward-pointing normal (pointing towards the interior of the surface) and outward-pointing normal need to be distinguished. By considering the normal vectors to the surface, the “Poisson surface reconstruction” method [16] was used to create the mesh model (Fig. 3). The final mesh model can be then exported to other programs such as Meshlab, FreeCad, Autodesk Fusion 360, and even Abaqus, as an unstructured triangulated surface (stereolithography).

Fig. 3.
figure 3

Obtaining the mesh model of the link in CloudCompare

Full size image

The four other used scanning devices (EinScan, Creaform, Hexagon Absolute Arm, iPhone 14 PRO) enabled a more straight-forward procedure to obtain the.stl file format and could directly export the mesh model from the embedded software (Vxelements, Polycam, Solid Edge, Inspire, respectively).

As a further step, it was investigated if the stereolithography data (.stl file) exported from these programs can be imported in the FEM software (Abaqus) and used in a numerical analysis. The stereolithography data is imported into Abaqus as an orphan mesh, and it is not thus editable. An orphan mesh part contains no feature information and is basically a collection of nodes, elements, surfaces, and sets with no associated geometry. Abaqus/CAE enables the conversion of an orphan mesh to editable geometry, but during the conversion process, it was observed that the models were missing elements and were not complete. This issue is related to the fact that the raw stereolithography data contain mesh models that are not closed properly.

4.3 Generating Solid Models from 3D Surface Models

To use data obtained from 3D scanning in Abaqus CAE it is important that the imported model to be a solid that can be assigned volume and weight. These solids are called manifold solids. The 3D models, directly constructed by the programs (Vxelements, Solid Edge, Polycam) that are offered for post-processing with the presented scanners, are only individual triangular 3D surfaces. Thus, it was necessary to identify a third-party software in which the 3D surface models (mesh models) can be imported for repairing and improvement of the mesh and eventually to obtain the solid model.

A 3D surface model or simply stated as mesh model is generally a collection of multiple individual triangular planes formed by connecting points in the point cloud. As the number of points increases, the point cloud is more likely to have individual planes that are intersecting each other, or holes which can cause issues for the later processes. These surfaces are defined as non-manifold surfaces. A manifold 3D surface is defined as a 3D surface that encloses itself and does not have any gaps and self-intersections. Therefore, to ensure that the 3D surface model can be transformed into solid, any holes in the 3D surface model need to be closed. Repair of non-manifold 3D models is important because it will directly affect whether or not the model is able to be used in FEM software packages.

The capabilities of three open-source programs that are supporting semi-automatic repair solution for non-manifold 3D models were investigated: Autodesk Fusion 360, FreeCad and Meshlab. The three programs were all successful in repairing the models, but the solid conversion was only managed in Autodesk Fusion 360. After importing the stereolithography data, the non-manifold surfaces were repaired successfully, and the mesh body was entirely rebuilt while preserving sharp edges (Fig. 4).

The repaired manifold 3D surface was then scaled to the actual size of the specimen. Due to the high complexity of the obtained manifold 3D surface, it was necessary to reduce the mesh size. Autodesk Fusion 360 enables a semi-automated conversion procedure of the mesh model to a solid model. From the three methods available, the “Faceted method” has shown the more reliable results. The final output is a solid model with virtual topology that can be exported as a common file format for 3D models (.step file) to the FEM software (Abaqus). In this manner the measured geometry and initial geometric imperfections of the specimens are readily embedded in the numerical model for FE analysis.

Fig. 4.
figure 4

Rebuild mesh body function.

Full size image

4.4 FE Model

After importing the model in Abaqus, it was observed that for discretizing the model with flexible and even sizes of elements, the virtual topology needs to be combined before the discretization by using the tool Virtual Topology – Automatic create. Since boundary conditions and loads can only be assigned at vertexes and edges, some of them might need to be restored using the tool Virtual Topology – Restore entities. The vertexes and edges are also the baselines for discretization.

Free meshing with tetrahedral elements was used in this application and the FE mesh model was successfully obtained as shown in Fig. 5.

Fig. 5.
figure 5

The FE mesh of the model.

Full size image

5 Conclusions

FE modelling based on data obtained from 3D scanning is more versatile in comparison to conventional FE modelling and can reflect more accurately the real magnitude and shape of the geometric imperfections.

Further research is still needed to validate the latter assumption, by using the FE model obtained from 3D scanning in a numerical analysis and comparing the results with experimental data.

The advance of 3D scanning technologies and post-processing programs in the future may also enhance its application in FEM. The development of software and computational ability of PC would also benefit the accuracy and efficiency in reconstructing 3D models of complicated structures. All the procedures of point cloud alignment, 3D surface model construction, repair of models, and conversion in a solid model may be integrated into a single program in the future to make the post-processing stage fully automatic.

Terrestrial laser scanning technology (LiDAR) has proven highly accurate and independent of the lighting conditions of the surroundings and surface reflectiveness of the scanned object. However, it is not the most reliable option from an economic point of view and has the disadvantage of the need of a third-party program to obtain the 3D surface model (initial mesh model) from the point cloud.

Two different blue laser scanners were investigated within this work: with measuring arm and handheld. The blue laser scanner with measuring arm does not imply the use of targets on the measured specimen, but his scanning range is conditioned by the length of the measuring range, and the costs of this technology rise to six-figure values. Furthermore, the 3D surface models obtained directly from the embedded software have shown high quality in terms of measured dimensional values and imperfections. The handheld blue laser scanner was more flexible in use, being able to scan the whole surface of the specimens (allowing to change the initial position of the specimen). Also, the obtained 3D surface models were highly accurate, and the technology is more economically reliable. However, a major drawback is the necessity to use reflective targets on the measured specimen that can be time-consuming.

The structured light scanner had a similar accuracy to the blue laser scanners and the TLS, but the obtained 3D surface model is less qualitative, this technology being more sensitive to the lighting conditions and surface reflectiveness. The actual measuring time was longer and the number of used targets on the specimens was higher. The quality-price ratio of such a device is still satisfactory.

The smartphone LiDAR scanner is still in a continuous development stage and has proven because of this reason a much lower accuracy than the other devices. Hopefully, in the near future it will be considered a reliable alternative for large-scale use due to its low price and versatility in obtaining the 3D surface model directly.