Introduction

The accumulated cracking damage mostly appears on the surface of the in-service irrigation structure. It is important to evaluate that for maintaining each facility’s gate function for water utilization. Effective inspection and evaluation methods are required due to the long-term serving and the large amount of agro-infrastructure. In this research, the evaluation method for the characteristics of concrete surfaces by laser scanning parameters in service agricultural structure is proposed.

In the civil engineering fields, non-contact sensors have been developed and used to monitor structures. Acoustic-based sensor, an active technique, can detect the surface condition of structures by the reflection of transmitted signals from a target. It is considered to be difficult to introduce in a generalized setting because these devices are limited to a specific objective (Mutlib et al. 2016). The digital camera was used for monitoring large-size structures with image processing as a passive technique of the non-contact sensor (Fukuda et al. 2010). The damage detection methods, which are conducted an important role in a monitoring system, have evolved much by image-based sensing and machine learning method (Shimamoto et al. 2020; Dung 2019). In our recent research, the development of damage estimation methods for in-service concrete structures by using elastic wave and X-ray CT methods is conducted (Morozova et al. 2022; Suzuki et al. 2020). Thus, the accumulated damage in the concrete structure appears on the surface part, which is detected by the digital image with UAV (Shibano et al. 2022). In recent years, light detection and ranging (LiDAR) and structure from motion (SfM) techniques (DJI 2022; Zhang and Elaksher 2012) provide high-quality point cloud (Lichti and Jamtsho 2006; Jutzi and Gross 2009).

LiDAR technique has been improved and discussed to generate point cloud and modeling for several decades (Kaartinen et al. 2022). The advantage of the terrestrial laser scanner is higher accuracy than dynamic measurement methods, such as mobile laser scanner (MLS) and airborne laser scanner (ALS) (Guo et al. 2011; Che et al. 2019; Mechelke et al. 2007). The resolution of point cloud is restricted by sampling interval and beamwidth (Lichti and Jamtsho 2006). The fundamental principle and characteristics of point cloud by the laser scanning method were investigated in laboratory experiments and simulations (Pfeifer et al. 2008). The accuracy of intensity parameter was validated by the comparison of different types of rangefinders (Suchocki 2020). The effects of surface irregularities, color and incident angle on intensity parameter were confirmed (Nicodemus 1965; Pesci and Teza 2008). The high accuracy of primitive detection for incomplete point clouds was maintained by a reconstructed optimal strategy (Schnabel et al. 2009). From these fundamental results, the management with three-dimensional point clouds was applied in terms of efficiency and upgrading in the construction fields (Mill et al. 2013). The laser scanning method was used to monitor civil infrastructures in many practical cases, such as Moon et al. (2019) and Chang et al. (2005).

The laser scanning method is different from visual images because of acquiring geometric information and intensity parameter (Stałowska et al. 2022). The purpose of the laser scanning method is mainly divided into deformation monitoring, surface defects detection and moisture detection for the maintenance fields. De Blasiis et al. identified and quantified pavement surface defects using geometric information acquired by a mobile laser scanner (De Blasiss et al. 2020). For arch bridges, the structural faults were detected based on shape parameters obtained by laser scanning data (Sánchez‐Rodríguez et al. 2018). Using the intensity parameter for evaluation of the surface, Suchocki et al. detected the moisture condition of building materials through the comparison of color, roughness, and saturation (Suchocki and Katzer 2018). For other structures, surface defects, leaks (Tan et al. 2016) and vegetation were detected by classifying the change of surface (Law et al. 2018; Suchocki et al. 2018; Leronas et al. 2016). In the previous research, the effects of moisture condition on the surface of construction material were considered adequately. The laser reflection affected by water should be taken into account in treating the agricultural structure. Therefore, the possible area to measure and the difference in ranging accuracy should be recognized (Lichti 2007).

In this study, evaluation of surface damage is conducted by the laser scanning parameters, geometric information and intensity parameter, revealing the advantages and limitations of each parameter. Deteriorated concrete headworks are selected as a testing structure for a 3D model constructed by the terrestrial laser scanning method. Several point cloud processing, primitive detection (Hackel et al. 2017), RANSAC algorithm, local geometric feature (Weinmann 2013; Jérôme et al. 2011) and intensity analysis (Höfle and Pfeifer 2007) are implemented to evaluate the characteristics of surface damage.

Materials and methods

Monitoring structure

The laser scanning measurement was conducted on in-service deteriorated concrete headworks in Niigata, Japan, which was constructed in 1976 in Fig. 1a. Concrete headworks are one of the main agricultural infrastructures to intake water for irrigation. For the monitoring structure, the span length is approximately 15.0 m and the height is 7.9 m. Surface damage has accumulated due to the long operation period. Cracking situations, efflorescence and surface defects are concentrated in the left side gate pillar in Fig. 1b. This surface is used for the evaluation of this research paper.

Fig. 1
figure 1

Testing concrete headworks. Overview of monitoring structure e (a), surface damage condition at left side gate pillar (b)

Full size image

Laser scanning method

Laser scanning was conducted to take point clouds of the structure with a FARO® Focus S150 laser scanner in Fig. 2. The international standard on instruments and measurement procedure was developed based on ISO Optics and optical instruments—field procedures for testing geodetic and surveying instruments—2022. The measurement was conducted following ISO 17123: Optics and optical instruments—field procedures for testing geodetic and surveying instruments. The measurement of the laser scanner is performed according to this procedure. An infrared laser beam is transmitted into the center of a rotating mirror. The mirror deflects the laser beams vertically with rotation. The scanner detects scattered light that is deflected from surrounding objects. The x, y, z coordinates of each point are then calculated by using angle encoders, which measure the mirror rotation and the horizontal rotation of the scanner. These angles are encoded with the distance measurement. The type of rangefinder is a phase-shift method in Fig. 3.

Fig. 2
figure 2

Setup for a laser scanner. Installing a laser scanner (a) and FARO Focus Laser Scanner S150 (b)

Full size image
Fig. 3
figure 3

Phase-shift method

Full size image

The relationship between the phase shift in the wave of the infrared light and distances R is shown in the following equation (Eq. 1).

$$R = \frac{{{\text{ct}}}}{2}, \,\,t = \frac{\theta }{2\pi f},$$
(1)

Here, \(c\) is the speed of light, \(t\) is the time, \(\theta\) is the phase shift, and \(f\) is the frequency.

The intensity parameter is registered by the reflectivity of the captured surfaces, measuring the intensity of the received laser beam. The received beam power is described by the following equation (Eq. 2) (Stałowska et al. 2022).

$$P_{{\text{R}}} = \frac{{\pi P_{{\text{T}}} \rho }}{{4R^{2} }}\eta_{{{\text{Atm}}}} \eta_{{{\text{Sys}}}} \cos \alpha$$
(2)

where \({P}_{{\text{R}}}\) is the received signal power, \({P}_{{\text{T}}}\) is the transmitted signal power, \(\mathrm{\alpha }\) is the angle of incident, \(\rho\) is the reflectance of the material, \({\eta }_{{\text{Atm}}}\) and \({\eta }_{{\text{Sys}}}\) are the atmospheric and system transmission factors and \(R\) is the distance.

Technical specifications of the laser scanner are shown in Table 1. Aligned point clouds were registered using target matching in FARO SCENE software (FARO Technologies 2022). The reference targets were placed to register and merge scans automatically.

Table 1 Technical specifications for the FARO® Focus S150 laser scanner and monitoring conditions
Full size table

Point clouds processing

Point clouds are analyzed using open-source CloudCompare software (Cloudcompare.org 2022a). The analysis surface for evaluation is set in the left side gate pillar surface because of the variety of surface conditions. For complete point clouds of structures, several laser scannings are needed to be conducted from multiple angles. Obtained each point clouds are aligned to get point clouds of the whole structure. In the aligned point clouds, noise is included because of point clouds generated from multiple scans with different accuracy. That is why, in this research, point clouds derived from a single scan are used for quantitative evaluation, while the aligned point clouds are displayed for primitive consideration of the characteristics of point clouds. The noisy data influence result of post-processing consequently (Rusu et al. 2008). The noise is removed by Statistical Outlier Removal (SOR) filter before the post-processing and evaluation (Eq. 3).

$$D_{\max } = D_{{{\text{ave}}}} + n{\text{Sigma}}\, \times \,{{sd}}$$
(3)

where \({D}_{{\text{max}}}\) is the threshold of distance for the removal, \({D}_{{\text{ave}}}\) is the average distance between k nearest neighbors, \(n{\text{Sigma}}\) is the standard deviation multiplier threshold, and \({{sd}}\) is the standard deviation. Point clouds derived from a single scan are selected as an analytical part to evaluate the damage. This is because the error derived from the alignment processing is larger than the errors removed by SOR. The selection of whether to use multiple scans or a single scan should be decided according to the objective and target (Kersten et al. 2005).

The characteristics of surface damage are evaluated based on coordinate and intensity parameters. In this study, surface damage is defined as visible cracks and cracks with efflorescence in the surface field of concrete structure. The distance between the fitted plane estimated from the coordinate and point clouds is calculated and used to evaluate the efflorescence and cracks quantitatively. The motivation for this approach is enabled to detect each designed element for man-made structures (Grilli et al. 2017). The fitted plane is calculated from point clouds of the analytical surface by a least-square method based on RMS error for the conventional method (Jérôme et al. 2011). The coordinate of point cloud used as an argument; therefore, the fitting plane is affected by noise, such as incidental equipment and deposits (Weinmann 2013). RANdom SAmple Consensus (RANSAC) algorithm is introduced to improve the conventional fitting method (Schnabel et al. 2007; Gönültaş et al. 2020). RANSAC algorithm has advantages in terms of robustness, generality, low memory consumption and simplicity. The processing of RANSAC algorithm consists of three parts in Fig. 4. First, minimal point sets from any point clouds are selected (Selected point) randomly, and primitive (Fitting line) is estimated (Fig. 4a). A minimal set is the smallest number of points required to decide a primitive uniquely. The larger the minimal point sets are set, the fewer the number of extracted planes is become, since the small planes are ignored. Next, the score of the shape is calculated by counting surrounding points (Optimal fitting point) in the range of threshold (Threshold for good fit) (Fig. 4b). After any iteration, the shape (Optimal fitting line) is selected when the score of the shape is maximum (Fig. 4c). The distance between the selected plane and point clouds is calculated to evaluate the amount of efflorescence quantitatively (Fig. 4d).

Fig. 4
figure 4

Evaluation procedure by RANSAC algorithm in 2D. Random sampling and fitting (a), calculating scores (b), optimal fitting (c) and evaluation damage by distance between fitting line and point (d)

Full size image

The local geometric features are calculated from the neighbors of points to detect cracks. Many shape signatures were proposed regarding curvature (Douros and Buxton 2002; Teza et al. 2009), normal and local distribution (Dimitrov and Golparvar-Fard 2015). The maximum height of valleys and the maximum height of peaks are defined as roughness according to ISO 25178: geometrical product specifications (GPS)—surface texture: areal (ISO 2021). The distance between points and the best-fitted local plane computed on its nearest neighbors is defined as roughness in the point cloud (Cloudcompare.org 2022b). In this study, roughness is proposed. This value varies depending on the radius defined as the neighborhoods.

In this research, the characteristics of intensity parameter in wet condition are investigated, and then, the characteristics of surface damage are evaluated by geometric and intensity parameters.

Results and discussion

Generation of point clouds

The laser scanning measurement is conducted with the scanner being placed around headworks 20 times to support the completeness of point clouds. The measurement time is less than about 30 min per scan. Regarding the weather condition, the daily mean temperature was 23.9°, the daily mean humidity was 83% and the daily precipitation was 0.0 mm as shown in Table 1 (Japan Meteorological Agency 2022). The scanning points are shown in Fig. 5. The number in each circle is in scanning order. Most scanners are located about 10 m from the closest surface. If the surface is scanned at the range of 10 m, the range noise is estimated to be 0.3 mm, 0.4 mm and 1.3 mm at a reflectivity of material of 90% (white), 10% (dark gray) and 2% (black), respectively (see Table 1). Range noise is defined as a standard deviation of values about the best-fit plane for a measurement speed of 122,000 points/sec.

Fig. 5
figure 5

Scanning points

Full size image

Aligned point clouds are shown in Fig. 6. Analytical result of point clouds is given in Fig. 6a. These point clouds contain constructed elements (e.g., headworks) and unconstructed features (e.g., vegetation and gravel). The point clouds corresponding to headworks are segmented by the clipping box to decrease data volume. The data holes are observed in two areas due to the reflectance characteristics of the object surface and the scanner’s constraints. The low-density parts of point clouds around the water are shown in Fig. 6b. A similar phenomenon was observed on point cloud data at the riverbank (Dong et al. 2020). Most of the laser beams are absorbed by the water; therefore, point clouds are not generated. Some laser beams are reflected by water to generate incorrect points as noise (Suchocki and Katzer 2018; Nicodemus 1965). From these results, the detected area of point clouds is affected by the intake season and climate in the case of irrigation facilities. The data holes are observed under the tripods of the scanner in Fig. 6c because of deflection unit specification. The field of view in the vertical direction of this scanner is 300° (see Table 1). The number of neighbors is shown in Fig. 6d. In general, the density of point clouds is represented by the number of neighbors within the sphere of any radius r. The radius r is not decided to be a unique value because it depends on purpose and object. In this analysis, r is set to 1.0 cm in order to observe the detailed distribution of density in concrete headwork because of the shape and color of the object surface. The dense parts of point clouds are confirmed at the flat face (e.g., the side face of the left and right piers), while the lower-density parts are confirmed at the corner or edge of the object (e.g., the pipeline and the front side of the weir pillar). Received signals are decreased at low-density parts due to the edge effect. The edge effect refers to the distance averaged by the scanned surface (Stałowska et al. 2022). Point cloud is unable to be acquired without adequate intensity (Soudarissanane et al. 2011). The reasons for varying density are not identified due to relating to many factors (e.g., The number of used scans, distance, scanning time, angle of incident and object characteristics) (Mazalova et al. 2010; Voisin et al. 2007; Costantino and Angelini 2013; Suchocki and Katzer 2018).

Fig. 6
figure 6

Generated point clouds. Analytical result of point clouds (a), the low-density parts of point clouds around the water (b), data hole part (Noise) (c) and the number of neighbors (d)

Full size image

Evaluation of cracks and efflorescence by geometric parameters

The point clouds at the left side gate pillar are adapted as the analytical surface in Fig. 7a. Point clouds, shown in Fig. 7, are aligned from the scan data in the previous section. The side view of Fig. 7a is shown in Fig. 7b, c. The rectangles are shown as enlarged views corresponding to each dotted rectangle. Errors from the alignment of multiple scans are shown in the lower rectangle of Fig. 7b, c. The multiple layer is one of the characteristics of errors by alignment. The outliers are shown in the upper rectangle and circle of Fig. 7b, c. The outliers are point clouds of concrete surface failed ranging. These point clouds, however, include outliers and multilayers since the accuracy of scanning depends on the distance between the scanner and the object surface in Fig. 7b. SOR is performed for whole point clouds in Fig. 7c. Isolated points are removed in the upper rectangle of Fig. 7c, while errors from the alignment of multiple scans remain in the lower rectangle of Fig. 7c. To evaluate geometric features, aligned point clouds are inappropriate. Point clouds generated from a single scan are used in Figs. 8c, d, e, 9 and 11.

Fig. 7
figure 7

The result of noise removal processing. Analysis surface at the left side on the left gate pillar (a), non-treated point clouds with outliers and multilayers from the side view (b) and SOR-treated point clouds without isolated points from the side view (c)

Full size image
Fig. 8
figure 8

Detection of damages by primitive detection. Detected planes by conventional plane fitting method (a), detected planes by RANSAC algorithm (The color of the point cloud depends on the detected plane.) (b), distance between fitting plane and clouds (c), efflorescence part (d) and crack part (e)

Full size image
Fig. 9
figure 9

Roughness at crack part. r = 0.01 (Rate of concordance = 0.604) (a), r = 0.05 (Rate of concordance = 0.643) (b), r = 0.1 (Rate of concordance = 0.630), (c) and r = 0.5 (Rate of concordance = 0.532) (d)

Full size image

Surface damage is evaluated by two plane fitting methods. Examples of plane fitting using full point clouds by conventional method and RANSAC algorithm are depicted in Fig. 8a, b. The whole red plane is detected by the conventional method in Fig. 8a. The color of point clouds depends on each detected surface in Fig. 8b. The different color of planes shows the detected surface corresponding to each face of structural elements. The difference between the two methods is whether they fit a plane locally or globally. It is necessary to process point clouds without noise since the fitted plane is affected by outliers when using the conventional method (Schnabel et al. 2007). The significance of primitive detection is to construct designed structures from unconstructed point clouds (Hackel et al. 2017). In these results, the RANSAC algorithm seems optimal for detecting primitive conceptually. The optimal parameters of RANSAC algorithm are determined from the relationship between the number of detected planes and the detection rate of the target. Optimal minimum support points per primitive are set to seven cases, ten thousand, a hundred thousand and one to five million and investigated. The relationship between the number of detected planes, the detection rate of the analytical surface and the minimum support points per primitive are shown in Table 2. The number of detected planes tends to decrease if the minimum support points per primitive are larger. Detection of the analytical surface failed for the minimum support points per primitive is 4,000,000. In the case that the minimum support points per primitive are lower, the detected planes are likely to be fitted locally and include noise elements (Schnabel et al. 2007). Considering this relationship, the number of minimum support points per primitive is set to 1,000,000 for a single scan according to the balance between detectivity of specific surface and accuracy. The distance between the fitting plane by RANSAC algorithm and points is calculated for the evaluation of the surface condition. The distance between the fitting plane and points at the left side gate pillar is shown in Fig. 8c. In the top left part, a larger distance is observed than in the other part ((d) of Fig. 8c). Areas with large distances match the position of efflorescence which is verified by visual inspection in Fig. 8d. The upper and lower figures show RGB point clouds and the distance between the fitting plane and points. Efflorescence indicates the existence of cracks and waterways. The efflorescence part is illustrated by white points in RGB point clouds. The distance between the fitting plane and points at efflorescence is higher than one of the others. The highest point indicates 0.031 m efflorescence is deposited. From these results combined with recognition by image and intensity parameter, efflorescence is able to be quantitatively evaluated by primitive detection and distance between the fitted plane and points. Focusing on the crack part in lower area of analysis surface ((e) of Fig. 8c), crack is not clearly observed by this parameter due to the relative magnitude relationship to efflorescence in Fig. 8e.

Table 2 The effects of the minimum support points per primitive on detection of the tested surface by RANSAC
Full size table

The possibilities of local geometric features for crack detection are considered. The asperity of the concrete surface is quantified by geometric features. r is also set to define the neighbors. Roughness which is the local plane fitting method is calculated in Fig. 9. The point clouds shown in Fig. 9 are the same as in Fig. 8e. The optimal r which defines neighbors is chosen from four cases, r = 0.01, 0.05, 0.1 and 0.5 m by the rate of concordance to the RGB image. The calculation procedure is mentioned below. The point clouds are converted into an image. Binary images are generated by Otsu method (histogram-based binarization). The rate of concordance for binary images of RGB images and roughness values are calculated, comparing allocated classes. The rate of concordance is 0.604 (r = 0.01), 0.643 (r = 0.05), 0.630 (r = 0.1) and 0.532 (r = 0.5). r = 0.05 is adopted to calculate roughness. The concordance rate shows a contour matching from the RGB image and inflection line of roughness values. This means the irregular shape of the crack is successfully detected for the average level within the sphere of radius. Therefore, roughness is the most effective parameter to evaluate surface damage in three plane fitting methods.

Detection of surface damage by intensity parameter

Characteristics of intensity parameter need to be revealed since this is affected by measurement environment, laser scanner and material condition. The transmitted signal power, system transmission factor and atmospheric transmission factor related to intensity can be considered constant since scanning is conducted by the same scanner during a short time (Stałowska et al. 2022). Range and angle of incidence are assumed to be the same because of using point clouds generated from a single scan. The reflectivity of a target depends on the roughness, color and saturation of the target surface, especially. In this section, the effects of water, saturation and surface roughness on intensity parameter are focused on. An image of aligned point clouds with RGB and intensity at the bottom part of the weir in wet condition is shown in Fig. 10a, b and c. Point clouds in wet condition are identified by both RGB and intensity (Fig. 10b, c). The intensity parameter does not dominate over the RGB image for the evaluation of surface damage in wet conditions. Water effects on the reflectance of the laser beam at roughness are based on two phenomena (Suchocki and Katzer 2018). Ordinary, reflection of a laser beam is diffused at a rough concrete surface. A part of an emitted laser beam is absorbed in the interface between water and air. That is why wet concrete surfaces darken and intensity decreases. Environment and weather condition restricts the result of laser scanning.

Fig. 10
figure 10

The characteristics of intensity parameter in wet condition. The bottom part of the weir in the wet condition (a), point clouds (RGB) (b) and point clouds (Intensity) (c)

Full size image

Visual image and intensity point clouds at the analytical surface are shown in Fig. 11a, b. A lower intensity value (0–200) is allocated at the color change part and nameplate (black), while a higher intensity value (200–241) is allocated at the efflorescence part (white). Intensity values are affected by color (Höfle and Pfeifer 2007). The contour of the crack part is observed by intensity. The intensity value at the crack part is lower than the surroundings. Comparing RGB image, all crack parts are unable to be detected by this method. To reveal the limitation and get a reliable result, the relationship between roughness, intensity and surface condition of the material is analyzed. Tested points are shown by a dotted line in Fig. 11a, b. The relationship between roughness, intensity and surface condition is presented in Fig. 11c. Value deviations on roughness at several parts of efflorescence, framework and cracks are observed. Meanwhile, the detection of cracks with a narrow width is inconclusive. Roughness at the change color part is constant. The intensity value at efflorescence is higher than other parts. The intensity value at the change color part is not constant because of irregular dirt. At the crack part, a decrement in intensity is not observed. Stałowska et al. reported that the maximum crack width detected by geometric information is 2 mm and by intensity is 1 mm at a range of 10 m and an angle of incidence 0 gon (Stałowska et al. 2022). According to this research, a crack whose width is less than 1 mm is impossible to be detected. Crack detection is not simple due to in situ scanning. Intensity shows relative differences depending on surface damage types. The geometric parameter has limitations regarding detecting crack width.

Fig. 11
figure 11

The evaluation of surface damage types by the intensity and geometric parameters. Visual image (a), point clouds (intensity) (b) and the relationship between roughness and intensity depending on surface condition (c)

Full size image

The conceptual diagram of evaluation by roughness and intensity depending on the characteristics of the concrete surface is shown in Fig. 12. This relationship is revealed by these results for point clouds derived from a single scan. The roughness and intensity parameters at the efflorescence part are higher than one at the other surface. The roughness of the change color part is constant, while the intensity is lower than others. The roughness at the surface defects part such as a boundary between frameworks and cracks is high values depending on scale, as mentioned. The intensity parameter at the surface defects is lower than one of a non-damaged part. The framework part and crack part have similar characteristics of geometric and intensity parameters. Therefore, the surface damage and condition are characterized by roughness and intensity parameters.

Fig. 12
figure 12

The conceptual diagram of evaluation by roughness and intensity depending on the characteristics of the concrete surface

Full size image

Conclusions

The laser scanning method is applied to evaluate the characteristics of surface damage for agricultural concrete headworks. The evaluation is based on geometric and intensity parameters of point clouds which are generated from laser scanning measurement. Three types of plane fitting methods using coordinate information are attempted for the detection of cracks and efflorescence quantitatively. The characteristics of intensity parameter in wet condition are confirmed. The damage is identified by both geometric and intensity parameters. Conclusions are summarized below:

  1. (1)

    Point clouds around the water area and under the tripods of the scanner are not detected because of the reflectivity characteristics of water and restraint of the deflection unit.

  2. (2)

    Aligned point clouds from multiple scans have errors since each scan is measured from a different range. The scale of these errors is larger than the outliers which can be removed by SOR processing. A single scan is used for analytical and evaluation processes.

  3. (3)

    The amount of efflorescence is evaluated by RANSAC algorithm, while crack parts are not evaluated. The roughness parameter which is calculated from local neighbors enables to detection of cracks quantitatively.

  4. (4)

    The reliability of intensity parameters in wet condition is lost by the reflectivity of laser beams on water and rough surface. Weather condition and water level affect the limitation of acquiring point clouds and intensity parameter.

  5. (5)

    The roughness and intensity at the efflorescence part are higher than one of the other parts. The roughness at the crack part is higher than one of the normal parts, while intensity is not changed depending on the crack. The variation of intensity increases at change color parts, while roughness is not changed depending on color.