Keywords

1 Introduction

At present, under the support of a series of domestic policies, the practice of assembly construction has emerged as a pivotal driver for propelling the metamorphosis and elevation of the construction sector towards standardization, industrialization, and ecological sustainability [1]. In contrast to conventional cast-in-place concrete structures, assembly bridges possess a distinctive advantage in terms of their heightened industrialization. This notable attribute bears profound significance for the advancement of the bridge industry, as it stems from the prefabrication and fabrication of structural components within dedicated girder yards. Consequently, this approach engenders construction processes that are both remarkably efficient and environmentally sustainable [2]. However, due to factors such as uneven professional ability of construction personnel, immature construction technology, and non-uniform acceptance standards, assembly prefabricated girders often have a series of component production deviation problems in the production process [3].

Three-dimensional laser scanning technology encompasses an automated, all-encompassing, and exquisitely precise method of capturing spatial data [4]. This technology employs a state-of-the-art three-dimensional laser scanner to delicately capture the essence of the target object, entirely devoid of any physical contact. By meticulously acquiring the spatial coordinates of each point gracing the surface of this remarkable entity, it deftly constructs an intricate three-dimensional point cloud model, harnessing the power of its immaculate three-dimensional spatial point coordinate data [5]. This innovative approach transcends conventional surveying and mapping methods, which rely solely on single-point acquisition. Instead, it enables the rapid and high-precision acquisition and presentation of three-dimensional data pertaining to target object surfaces. Ultimately, it culminates in the construction of a comprehensive three-dimensional point cloud model. By utilizing a cutting-edge three-dimensional laser scanner, wide-ranging areas can be scanned with exceptional accuracy and speed, regardless of weather conditions. This direct and high-precision technique effectively circumvents the limitations inherent in data analysis that rely on unilateral perspectives and localized approaches. Distinguished scholars both domestically and overseas have undertaken profound research endeavors grounded in the realm of 3D laser scanning technology, particularly concerning comprehensive deformation monitoring of landslides [6], full-section deformation measurement of tunnels [7], and feature extraction of buildings [8]. However, the practical utility of 3D laser scanning technology as a means for acquiring quality inspection data for assembled bridges is limited.

In accordance with this, we put forth a methodology for quality control in the production of precast concrete girders, utilizing the innovative technology of 3D laser scanning. Drawing upon the context of the precast girder yard within the Hangzhou-Ningbo Expressway Duplex, we present a comprehensive exposition on the application of 3D laser scanning technology in the process of precast girder fabrication. This entails precise evaluation of reinforcement distribution distances, dimensional accuracy of concrete components, and surface uniformity. Furthermore, it enables swift rectification of any quality concerns that may arise during precast component manufacturing, thereby affirming and substantiating the dependability and trustworthiness of 3D laser scanning technology.

2 3D Laser Scanning Technology

2.1 Point Cloud Data Segmentation

To extract intricate geometric details of voluminous prefabricated T-beams, the initial endeavor in unraveling the three-dimensional laser scan data entails point cloud segmentation. By scrutinizing the attribute label information, feature planes and crucial geometric points are ascertained and isolated from the point cloud data. To tackle the diverse spectrum of large-scale prefabricated structures characterized by varying degrees of geometric intricacy and magnitude, scholars both domestically and internationally have devised an array of algorithms for extracting geometric information.

For components with relatively simple geometry, Castillo et al. [9] proposed a point cloud segmentation algorithm based on surface normal estimation and local point connectivity, which can deal with unstructured point cloud data and stably detect corner points and edge data from point cloud data. Pu [10] used a plane growth method to delineate and identify architectural features. In the realm of substantial preassembled constituents, Tan et al. [11] ingeniously amalgamated the region germination algorithm with the Random Sampling Consensus (RANSAC) algorithm to ascertain and delineate the distinctive planar facets of the prefabricated components. Furthermore, they adroitly employed the RANSAC algorithm to deftly expunge any extraneous perturbations that may have occurred in the z-axis. However, this technique proves to be susceptible to noise interference when conducting line searches and lacks the ability to address minute structural elements. Previous investigations have primarily concentrated on sizeable components boasting uncomplicated geometries within pristine indoor settings, thereby leaving a significant research gap regarding the intricate modeling of expansive institutional elements featuring multi-unit amalgamations. Consequently, this study presents an innovative approach for noise reduction and seamless integration of three-dimensional point clouds pertaining to extensive prefabricated T-beam members.

2.2 Point Cloud Data Noise Reduction

The acquisition of point cloud data through 3D laser scanning is, unfortunately, prone to the intrusion of noise. This unwelcome addition stems from a confluence of factors: the precision of the scanner itself, the irregularity in operator technique, the susceptibility of measured surfaces to reflection, and the influence of environmental conditions. Furthermore, variables such as the angle between laser and surface, object movement during scanning, material composition, ambient light levels, temperature fluctuations, and humidity further encroach upon the accuracy of three-dimensional laser scanners. Consequently, discernible noise manifests within the gathered point cloud data. Regrettably, this presence of noise obfuscates key features within said data set; thus compromising not only algorithmic efficiency but also result fidelity.

In this paper, the K-Nearest Neighbors (KNN) algorithm is employed for the purpose of mitigating noise in point clouds. This nonparametric method, encompassing both classification and regression tasks, operates by evaluating the distances between distinct samples within the feature space. The computation of these distances adheres to Eq. (1).

$$ d(x,y) = \sqrt {\sum\limits_{k = 1}^{n} - \left( {x_{k} - y_{k} } \right)^{2} } $$
(1)

In Eq. (1), the distance metric, d(x, y), is computed using the K-Nearest Neighbors (KNN) algorithm to determine the dissimilarity between various samples. The essence of the KNN algorithm lies in its ability to discern patterns within the feature space. If a sample under classification has most of its k-nearest neighbors (i.e., those with the smallest distances) belonging to a particular category, it can be inferred that this sample also belongs to said category and shares similar characteristics with others in that category. In the context of denoising with the KNN algorithm, if the average distance between a point and its k nearest neighbors exceeds a predetermined threshold, it is deemed as noise and subsequently discarded from further analysis.

2.3 Point Cloud Data Stitching

The technique of multi-site cloud splicing has a profound impact on the precision of the point cloud model. The comprehensive point cloud representation of the precast concrete element is encompassed by the diagonal scanning performed at two different stations, resulting in a considerable overlap in point cloud coverage. This allows for automatic splicing, and the usability of the point cloud model is assessed through an analysis of the accuracy of data fusion in a corresponding report. On the other hand, generating a unified point cloud for reinforced members poses greater complexity. To address various working conditions simultaneously, a target sphere is employed as a reference to seamlessly integrate distinct station-based point clouds. The goal is to maintain an average splicing error within 1mm for the target sphere, thus producing an intricate 3D model capturing the section of the prefabricated girder. Figure 1 visually presents representations of both the beam body model and reinforcement skeleton model following their meticulous scanning and seamless integration process.

Fig. 1.
figure 1

Model of reinforcement and beam body under point cloud.

Full size image

3 Project Overview

The background of this study is the Hangzhou-Ningbo Expressway Duplicate Line, with a total of 6,499 prefabricated T-beams, of which 5,291 are 30 m span T-beams. The parameters for quality control of the final precast T-beams primarily encompass the robustness of the concrete composition, accuracy in the deployment of steel reinforcement, precision in the installation of prestressing pipes, and adherence to other stipulated construction standards. Table 1 detailedly lists installation requirements for concrete. Table 2 detailedly lists installation requirements for rebar.

Table 1. Installation requirements for concrete.
Full size table
Table 2. Installation requirements for rebar.
Full size table

Furthermore, the examination of the beam’s finished product necessitates an assessment of the concrete’s surface evenness, ensuring a uniform hue, absence of conspicuous construction joints, and the avoidance of honeycomb or pockmarked blemishes. Moreover, the sealing anchor concrete must display impeccable density and a flawlessly level finish. While assessing concrete strength through rebound meter tests remains crucial, most quality control measures pertain to dimensional checks. To expedite the inspection process for concrete T-beams and enhance production line capacity, we have embraced three-dimensional laser scanning technology to monitor precast girder plate quality. Empirical evidence substantiates that the comprehensive 3D model scanning report can supplant traditional quality inspection records for precast girder plates.

4 Data Acquisition and Processing

4.1 3D Laser Scanner

The project utilizes the FARO S350 Plus 350 3D laser scanner, a sophisticated and fully automated device renowned for its outstanding precision. With a weight of 4.2 kg, this portable scanner boasts an omni-directional capability, rendering it suitable for a diverse range of measurement tasks within a distance range extending from 0.6 m to an impressive 350 m. The device’s remarkable distance accuracy, which can achieve up to 1 mm precision, enables it to adapt effectively to various lighting conditions while achieving scanning speeds of up to an astonishing 970,000 points per second.

Employing advanced non-contact 3D laser scanning technology, this equipment is capable of rapidly capturing highly precise three-dimensional information across expansive areas. Particularly in the realm of quality control for precast concrete beams, the utilization of this scanner significantly enhances inspection efficiency and accuracy. By configuring the scanning resolution at half its maximum capacity and employing a triple factor increase in sweeping quality, the resulting point spacing reaches an impressive level of precision at 3.1 mm per every ten meters scanned.

Fig. 2.
figure 2

A three-dimensional laser scanner scan

Full size image

4.2 Measurement Point Arrangement

Given the vast multitude of reinforcement bars and the intricate interlocking phenomenon they exhibit, a solitary measurement often fails to capture the full extent of their three-dimensional characteristics. To mitigate any potential obstruction of the camera’s field of view, minimize measurement redundancy, and expedite field data processing speed, a total of twelve meticulously crafted T-beam/reinforcing steel skeleton scanners were strategically positioned, as illustrated in Fig. 3. These scanners were distributed amongst adjacent partitions (two at either end, three in the middle, and four on each side), with an additional two stationed at the extremities of the girder. Furthermore, to ensure utmost precision in point cloud data alignment, numerous reference spheres were thoughtfully installed within the shared visual range of these stations, as depicted in Fig. 2(b).

Fig. 3.
figure 3

A three-dimensional laser scanner scans the steel skeleton

Full size image

5 Data Analysis

5.1 Rebar Size

The 3D point cloud model that was generated underwent slicing at appropriate locations in order to accommodate the rebar sections. This slicing process utilized the Randomized Sampling Consistent (RANSAC) algorithm, which accurately calculated the spacing of the rebars. A total of 120 hoop spacings were randomly selected from various positions within 30 T-beams and compared with the designated value of 150 mm. The final results of this comparison can be observed in Fig. 4. It is evident that the majority of the hoop spacings fall within an acceptable margin of error for rebar installation. Nonetheless, there are a few instances where certain hoop spacings exceed this limit. These discrepancies were identified through a meticulous analysis involving three-dimensional laser scanning and precise calculations regarding the dimensions of the reinforcement cage. Subsequently, prompt adjustments were made to those beams exhibiting significant deviations in their hoop spacing.

Furthermore, a probability density function curve was plotted using the data obtained from these observations. Upon fitting it with a normal distribution, it became apparent that its mean value slightly deviates from the prescribed design requirements. However, it is noteworthy that there is a substantial density of hoop reinforcement, thereby satisfying said requirements.

Fig. 4.
figure 4

Random sampling of 120 stirrup spacing

Full size image

5.2 Concrete Size

The spatial configuration of prefabricated girders is assessed employing the slice fitting technique. Initially, the three-dimensional point cloud representation of the completed precast T-beam is sliced along its width, length, and height. For each slice, an automated boundary fitting and extraction of corner points are conducted utilizing the Random Sampling Algorithm for Consistency (RANSAC). This approach allows for the determination of the spatial geometric properties of the precast girders within each slice. Figure 5 presents a plot depicting the girder length, height, and roof width of 30 scanned girders using 3D laser technology. Remarkably, the structural dimensions consistently fall within acceptable construction tolerances.

Fig. 5.
figure 5

30 pieces of concrete beams and beams long structural dimensions

Full size image

5.3 Concrete Leveling

In the realm of evaluating the evenness of concrete surfaces, this project employs the method of optimal plane fitting to assess the flatness of precast T-beams. The desired plane is derived from an analysis of various factors including inspection criteria, geometric attributes of the components, normal vectors assigned to each point, and other pertinent characteristics. Subsequently, utilizing the segmented point cloud data, we employ an “adaptive sliding plane method” to determine the optimal reference plane for the surface. Finally, by calculating the distance between each point and this reference plane, we are able to generate a visual representation of flatness distribution.

To exemplify this process, a single prefabricated T-beam has been randomly selected and its web surface flatness is depicted in Fig. 6 using the adaptive sliding plane technique. From this illustration, it is evident that the maximum variation in unevenness on the web surface measures less than 5 mm, thereby satisfying construction requirements. Additionally, in order to address any potential widespread inconsistency in maximum unevenness across all T-beams within a given batch, diligent inspections and prompt cleaning of concrete outer formwork are carried out to ensure uniformity and flatness during pouring procedures.

Fig. 6.
figure 6

Surface flatness of concrete

Full size image

6 Conclusion

This research paper is centered around the implementation of three-dimensional laser scanning technology for the purpose of assessing and managing the production quality of precast concrete girders. The project draws from the Hangzhou-Ningbo Expressway Duplicate Line Ningbo Section Phase II Project as its contextual backdrop and achieves intelligent detection in various aspects, including reinforcement bar dimensions, concrete appearance, and flatness. By employing 3D laser scanning as a means to inspect precast concrete beams, it is possible to entirely supplant conventional quality inspection methods. This innovative approach allows for prompt rectification of construction errors, markedly enhancing efficiency in the quality inspection process while reducing labor costs. As a result, it holds immense potential for widespread application within engineering projects.