**4. Experiments**

In this section, we present the experiments for testing the proposed approach to computer vision-based assessment of surface cracks in concrete structural elements during laboratory experiments. The first part of the assessment consists of monitoring deformation of the element and analyzing development of crack width formed on the surface of the beam to determine the moment in which each of them was created. This part was based on optical measurements and DIC method. In the second part, crack detection and localization using Faster R-CNN model was carried out.

In this work, the proposed methodology is verified by applying it to the laboratory assessment of post-tensioned, precast crane runway beams after more than 50 years of exploitation. They were produced between 1962 and 1963 and were disassembled from the structure at the industrial hall [1]. The three-point bending tests were conducted on girders made of two precast segments, with the total length of the span L = 2 × 290 cm + 20 cm = 600 cm. The girder has an I-section with height H = 80 cm. The segments were connected due to the action of the force and welding of steel sheets to steel marks. The joint was filled by applying fine-grained concrete.

The static three-point bending test yields the maximum bending moment at the joint of the segments. The loading cycle consisted of two stages. In the first stage a relatively small load value was applied and then the beam was unloaded. In the second stage the loading was carried out to the complete collapse of the beam.

The experiments were conducted in the Research Laboratory for Building Materials and Structures at the Cracow University of Technology. The deformation of the beam was monitored by applying optical measurements carried out using 3 synchronized DSLR cameras. The cameras were located at the center of the beam. Before experiments, the side surface of the beam was prepared for vision-based measurements by adding random distribution of black dots with spray paint on whitewashed surface of the beam. Figure 9 shows the test stand prepared for this research with the numbering of acquisition devices.

The CivEng Vision system, developed at Cracow University of Technology (CUT), was used for acquisition, storing and processing the images [38,39]. The images were then processed using DIC method for computing deformation fields and crack width visible on the surface of the beam. Finally, the images were processed by trained R-CNN model to automatically detect and localize the cracks on the surface. Figure 10 shows four images of the side surface of the beam, taken using the CivEng Vision system, during the three-point bending test.

**Figure 9.** Test stand prepared for monitoring post-tensioned, precast crane runway beam during the three-point bending test using the CivEng Vision system.

**Figure 10.** Images of the side surface of the beam, taken using the CivEng Vision system, during the three-point bending test.

Figure 11 shows grid of subsets for monitoring deformation of the surface of the tested crane runway beam using DIC.

**Figure 11.** Grid of subsets placed on the surface of the tested crane runway beam.
