A Novel Debonding Damage Identification Approach of Hidden Frame-Supported Glass Curtain Walls Based on UAV-LDV System

Abstract

This study introduces a novel Unmanned Aerial Vehicle-mounted (UAV-mounted) Laser Doppler Vibrometer (LDV) system for detecting debonding damage in Hidden Frame-Supported Glass Curtain Walls (HFSGCW). The established system enables UAVs to transport the LDV to high altitudes for operation. The vibration signals acquired by the UAV-LDV system are decomposed into different energy bands by wavelet packet analysis, and then the occurrence and location of the damage are identified by the Sum of Squared Differences (SSD) of the wavelet packet bands’ energy. This paper investigates the potential factors affecting the performance of the Unmanned Aerial Vehicle-Laser Doppler Vibrometer (UAV-LDV) system, including the arrangement of measuring points, measuring distance, noise level, and wind speed through the first-order natural frequency, the normalized frequency response functions, and the SSD indicator. Experimental and simulation results confirm the effectiveness of the UAV-LDV system, highlighting its advantages over traditional methods by offering remote, non-contact, and efficient debonding detection. This method not only indicates the presence of the damage, as traditional indicators do, but also pinpoints the exact location of it, ensuring safety and cost-effectiveness in high-rise inspections. The proposed method and indicator offer advantages in terms of convenience, visualization, and efficiency. The study discusses the impact of measurement point arrangement, measuring distance, noise levels, and wind speed on the system’s performance. The findings demonstrate that while the UAV-LDV system introduces new capabilities in rapid and reliable structural damage assessment, operational challenges such as wind and noise levels significantly influence its accuracy.

1. Introduction

The glass curtain wall is an aesthetically innovative method of architectural wall decoration and a distinctive feature of modernist high-rise buildings [1]. However, the extended life of glass curtain walls may result in sudden and dangerous detachment which can cause serious injury or property damage [2]. For instance, structural adhesives can suffer from deterioration and debonding problems. Debonding damage to existing glass curtain walls is mainly identified by visual inspection [3], manual inspection [4], contact sensors [5], numerical simulation [6], health monitoring systems [7], etc. These methods rely on experienced workers or contact sensors for stationary testing, which are insufficient to cope with the increasing height of skyscrapers, the emerging variety of curtain walls and the various extreme working conditions in modern buildings. Since these conventional stationary inspection approaches are inefficient [8], hazardous [9], complicated [10], and hard to employ [11], despite being very effective at detecting debonding damage of glass curtain walls, they cannot be widely used. In light of the rapid development of Unmanned Aerial Vehicle (UAV) and new laser vibrometry technologies, mobile inspection approaches show promising advantages in application [12] and further investigations.

The increasing demand for structural inspections, focusing on safety [13], durability [14,15], and comfort [16,17], makes the intelligent inspection of UAVs an inevitable trend in health monitoring processes [18]. The development of UAV technology has provided formidable technical support for this demand, including improved stability [19], enhanced wind resistance [20], increased load capacity [21], extended endurance [22], strengthened obstacle avoidance capability [23], etc. For example, the DJI Mavic 3 is a powerful and efficient drone available to the consumer designed by SZ DJI Technology Co., Ltd. (Suzhou, China). It can bear strong winds up to 12 m/s (strong breeze) and has a vertical hovering accuracy of ±0.1 m and a horizontal hovering accuracy of ±0.3 m when the visual positioning system is working properly [24]. These outstanding properties make such UAVs reliable for inspecting structural health in windy scenarios, such as high-altitude inspections of curtain wall structural damage, seismic performance of bridges [25], blade damage of wind turbines [26], etc.

Additionally, UAVs have been involved in numerous civil engineering applications for structural health inspections. They have been used for detecting bridge cracks [27], identifying dynamic parameters of bridge components [28], measuring geometry shapes of bridge components [29], and the 3D reconstruction of bridges [30]. The utilization of UAVs under certain weather conditions can improve efficiency and ensure the safety of the inspection personnel compared to manual inspections [31]. Similarly, wind turbines are often constructed in remote locations, such as offshore [32], in deserts [33], on mountaintops [34], and other areas where wind energy is abundant but difficult for humans to access. The application of UAVs replaces manual detection and shutdown maintenance, making regular inspections of wind turbines possible [35]. UAVs are also increasingly employed in other inspection processes, such as power transmission lines [36], pipelines [37], dams [38], and water towers [39]. This technology has become a promising solution for labor-intensive, accessibility-challenging, and sometimes dangerous inspection tasks in civil engineering [40].

There are a variety of practical difficulties associated with detecting debonding damage on glass curtain walls using traditional methods. Manual inspections on the facade of an office building can be disruptive to the normal work of the building’s occupants, challenging for the inspector [4], and costly for the maintenance company. Wall-climbing robots struggle with the diverse shapes of glass curtain walls and often require manual assistance [41]. Health monitoring systems are not commonly used for checking the condition of glass curtain walls due to their high cost and difficult installation [42]. Moreover, not all damage conditions result in large areas of peeling and stripping of the structural adhesive. In many cases, the structural adhesive may appear to be intact from the outside but is already failing or delaminating from the inside. This issue hinders identification using traditional vision-based methods. Conventional methods often rely on periodic inspections, and the time between inspections can lead to prolonged damage and delayed repairs. Therefore, the detection method of debonding damage on glass curtain walls should be efficient, immediate, safe, and cost-effective. These problems can be avoided by identifying damage using the vibration characteristics of the curtain walls [13]. Laser technology provides benefits of range, accuracy and efficiency. However, the power of a laser is limited, which makes ground-based damage assessment an impossible task.

The combination of UAV and laser vibration measurement technology effectively addresses the challenge of detecting debonding damage on glass curtain walls. Remote, non-contact, and low-error vibration measurements can be achieved by attaching the Laser Doppler Vibrometer (LDV) to a UAV. Some researchers have identified hidden vehicles in the forest by using LDVs placed at high altitudes to detect the vibrations of leaves [43]. Although the effect of wind turbulence on leaf vibration was considered, the interference with the LDV signals was not. Previous research also proposed the concept that LDVs could be transported in aerial platforms for testing, but their experiment still used a lift vehicle, which is no different from a stationary LDV. Other scholars claimed that they have built a UAV-LDV system to achieve aerial testing. However, the utilization of the UAVs in these systems is limited to the placement of reflective targets [44], for the enhancement of the reflected laser intensity or the carriage of acoustic exciters [45], for the replacement of manual tapping, or carrying mirrors to reflect the laser light from stationary LDVs [46,47]. Their actual tests still rely on stationary LDVs, and the purported UAV-LDV systems have not been fully realized. The UAV-LDV system is susceptible to factors that can cause interference, which in turn compromises the accuracy of the measurements. It is only used for vibration monitoring in situations with a significant degree of displacement, such as dynamic lateral displacements in railways [48]. While some studies have optimized UAV-LDV systems based on the assumption that the target remains in a steady state of vibration [49] and that the main source of error is induced by vibrations from the UAV [50], it is crucial to recognize that the UAV-LDV system is notably prone to inaccuracies resulting from the UAV’s aerial instability, and more damage detection methods are based on transient vibration. Specifically, the vibration monitoring of LDVs relies on detecting differences in the optical path, which is significantly affected by the deflection of the UAV-LDV system during flight. Consequently, the most substantial errors in these systems typically stem from strong winds and difficulties in the attitude adjustment of UAVs during operation.

The dynamic response of the curtain wall can be obtained by the subsequent analysis of the filtered vibration signals, which can ultimately reveal the debonding damage of glass curtain walls. This study investigates multiple sets of test conditions that differed in the location and extent of damage to the structural adhesive on curtain walls. The obtained signals were analyzed using the first-order of natural frequency, frequency response function, and the Sum of Squared Differences (SSD) of wavelet packet band energy as indicators. The reliability of the vibration signals obtained from the UAV-LDV system is verified by comparing them with those obtained from the fixed LDV. The results of the study demonstrate that the proposed SSD indicator based on the UAV-LDV system provides a more accurate identification of debonding damage to curtain walls in transient vibration. The study also explores the factors affecting the accuracy of the UAV-LDV system and their respective degrees of influence, including distance, noise, number of measuring points, and wind strength. The research provides a useful reference for the combination of UAV and laser vibration measurement technology for debonding damage detection on Hidden Frame-Supported Glass Curtain Walls (HFSGCW).

2. UAV and LDV Selection

There are two main types of UAVs: fixed-wing and multi-rotor. Fixed-wing UAVs are mainly suitable for long-duration, long-range, and wide-area flight operations [51]. They have high-speed flight capability, longer endurance, and larger load capacity, making them suitable for agricultural and environmental missions. Multi-rotor UAVs are designed for tasks that require small-scale, aerial hovering, and minor space shuttling capabilities [52]. They consist of multiple rotor blades, enabling vertical takeoff and landing, and hovering in the air. These characteristics enable the systems to be suitable for energy surveying, electrical power inspection, etc. Consequently, they are more commonly employed in detailed image acquisition and inspection of stationary structures, such as power lines and wind turbines.

Multi-rotor UAVs are preferred in this study owing to their excellent weight-carrying capability for carrying equipment such as LDVs. Additionally, they can hover steadily and align themselves with the measuring points to obtain the acceleration time history during the test. The advanced DJI Mavic 3 was then chosen for testing due to its potential use in curtain wall inspection.

Table 1. DJI Mavic 3 specifications.

Specific Items	Specifications
Bare Weight	895 g
Dimensions (Folded/Unfolded)	Folded (without propellers) 221 × 96.3 × 90.3 mm (Length × Width × Height) Unfolded (without propellers) 347.5 × 283 × 107.7 mm (Length × Width × Height)
Max Flight Time (no wind)	46 min
Max Hovering Time (no wind)	40 min
Max Wind Speed Resistance	12 m/s
Operating Temperature	−10° to 40° C (14° to 104° F)
GNSS	GPS + Galileo + BeiDou
Hovering Accuracy Range	Vertical: ± 0.1 m (with Vision Positioning); ± 0.5 m (with GNSS Positioning) Horizontal: ± 0.3 m (with Vision Positioning); ± 0.5 m (with High-Precision Positioning System)
Maximum Take-off Weight with Payload	1200 g

shows the specifications, and Figure 1a,b illustrate its appearance, respectively. The vibration signal of HFSGCW is measured using the LDV (MV-H100) from OMNISENSING PHOTONICS in China. Figure 1c depicts the appearance of the LDV and

Table 2. LDV specifications and weight of the system.

Frequency Range	Laser Wavelength	Displacement Resolution	Maximum Speed Range	Displacement Repeatability	Output Signal	Weight
DC~2.5 MHz	1310 mm	1.28 nm	1500 mm/s	10 nm	Digital	213.7 g
Weight of the bracket			Weight of the system
90.2 g			1198.9 g

shows the specifications of it.

Figure 2 shows the constructed testing system. The LDV is first attached to the bracket using a carbon fiber plate and stainless-steel screws. Then, the bracket is fitted onto the UAV’s fuselage. A system for acquiring acceleration time history is constructed by combining the selected UAV and LDV. The LDV directly captures and transmits signals to a PC for analysis, enabling efficient detection of acceleration time history at high altitudes without an additional data acquisition system, as it is an integrated device.

3. Influence Factors of the UAV-LDV System

The number and distribution of the measuring points, as well as the testing distance, are first considered when using the LDV to acquire the acceleration of the HSFGCW. It is important to compare the results of the mobile test with the stationary test to verify its accuracy, as the stationary test has shown great efficiency and reliability [53]. However, since the mobile test is susceptible to noise and wind disturbances, it is essential to assess their impact.

3.1. Selection of Assessment Indicators

Modal analysis can be used to detect the safety state of glass panels due to the significant impact of sealant failure on modal parameters. The first-order modal frequency is considered as the safety condition evaluation indicator because of its stability and accessibility [54]; therefore, this indicator is used in this paper to evaluate the impact of each disturbance. The inspection method based on the accumulated change rate of the frequency response function (FRF) can efficiently, accurately, and expeditiously identify the sealant failure of HFSGCW. This study employs the normalized FRF at different locations to identify the number and distribution of measuring points, according to the literature [55].

Assume that the natural frequency and mode vector of the glass panel system with $n$ degrees of freedom are ${\theta }_r$ and ${\left\{\phi \right\}}_r(r=1,2,\cdots ,n)$, where the ${\left\{\phi \right\}}_r$ can be expressed as $\{{\phi }_1^r {\phi }_2^r\cdots {\phi }_n^r{\}}^T$. Thus, in the p-th degree of freedom, the acceleration FRF $H_{\mathrm{pp}}(\omega )$ is [56]

$$H_{\mathrm{pp}}(\omega )=\frac{A_{pp}(\omega )}{Q_{pp}(\omega )}={\displaystyle \sum _{r=1}^n\frac{-{\omega }^2{\phi }_p^{r2}}{-{\omega }^2{m}_r+i\omega c_r+k_r}}$$

(1)

where $A_{pp}(\omega )$ and $Q_{pp}(\omega )$ are the Fourier transforms of the displacement and force, respectively, at frequency $\omega$; the modal mass is $m_r={\left\{\phi \right\}}_r^T[M]{\left\{\phi \right\}}_r$, the stiffness coefficient is $k_r={\left\{\phi \right\}}_r^T[K]{\left\{\phi \right\}}_r$, and the modal damping coefficient is $c_r={\left\{\phi \right\}}_r^T[C]{\left\{\phi \right\}}_r$.

The wavelet packet of the structural dynamic response under vibration excitation can be decomposed into a number of independent frequency bands. Then the changes in the energy spectrum of the wavelet packet bands can indicate the occurrence of structural failure. This energy change can be identified by the Sum of Squared Differences (SSD) of the wavelet packet band energy, which can be defined as [57]

$$SSD={\displaystyle \sum _{i=1}^m}{\left[{\left(E_i\right)}_b-{\left(E_i\right)}_a\right]}^2$$

(2)

where m is the number of the main frequency bands, ${\left(E_i\right)}_a$ and ${\left(E_i\right)}_b$ represent the signal energy from the intact and damaged structure, respectively. The SSD indicator has been demonstrated to be effective in detecting structural sealant failure [13]. In this paper, the reliability of the UAV-LDV system and the factors affecting its operation are also verified by this indicator.

3.2. Similarity Relationship

The number and distribution of the measuring points are related to the location of the excitation and structural sealant failure during the test. Theoretically, the number of measuring points should be as high as possible to accurately localize the occurrence of structural adhesive damage. However, a large number of measuring points reduce detection efficiency and increase the difficulty of analyzing the acquired data. This acts against the original purpose of developing a UAV-LDV system for rapid and mobile inspection. Due to the difference in the frequency response function at different locations of the glass curtain wall panel system [55], this section examines the effect of the measuring point location on the identification.

As depicted in Figure 3, twelve measuring points were arranged diagonally across the panel. The method of excitation at the measuring points is employed to obtain a driving point Frequency Response Function (FRF). Point A is the hammering excitation point which is located at the geometric center of the glass panel to obtain the average first-order natural frequency. In the area where measurement points 2, 4, and 6 are located, a simulated excitation experiment was performed according to the literature [13]. The first- to fifth-order modal frequencies of the glass panel obtained by modal analysis and LDV tests are shown in

Table 3. Glass panel frequency of the first five-step mode.

Order	Frequency/Hz
	Numerical Simulation [13]	Point 2 Test	Point 4 Test	Point 6 Test
1st	101.0	99.2	95.7	93.1
2nd	234.3	/	227.1	228.9
3rd	354.2	329.6	330.8	/
4th	364.3	378.4	375.9	/
5th	449.8	436.4	435.2	433.3

.

After testing and analysis, the normalized frequency response functions of the acceleration at points 2, 4, and 6 are shown in Figure 4. It shows that point 4 has five resonance peaks between 0 and 500 Hz, while point 2 has four resonance peaks, and point 6 only has three resonance peaks. Table 3 shows that in the first fifth-order frequency range, the resonance peak of the frequency response function does not include the second-order modal frequency at point 2, and only includes the first-, second- and fifth-order modal frequencies at point 6. As the acceleration time history should be recorded at a location where more modal frequencies can be obtained [55], point 4 is selected as a relatively preferred location.

Similarly, the optimal locations for the measuring points’ arrangement are selected at points 3, 4, 9, and 10. In order to complete the damage assessment mapping based on the SSD indicator, the number of measuring points was finalized to 9, arranged in a square (similar to the geometry of the glass curtain wall). Points 3, 4, 9, and 10 are the corners of the square as shown in Figure 5.

3.3. Testing Distance

Remote testing allows for a wider field of view and more comprehensive coverage, which enables effective detection of a wide range of injuries. However, proximity testing is also indispensable for specific applications, such as damage detection based on mobile platforms such as UAVs or wall-climbing robots. Therefore, it is necessary to investigate the effect of testing distance on the system.

Comparative experiments were conducted for typical conditions of intact, unilateral, bilateral, and multilateral debonding damage, as illustrated in Figure 6. The diagram shows the glass panel (dotted shadow), the border of the specimen (blue area), and the length of the structural adhesive where damage occurred (yellow area). The four types of structural adhesive damage conditions are as follows: intact condition, complete damage on one side, half of the damage on each of the two sides, and complete damage on all three sides. The distances tested were 1 m and 10 m from the specimen, respectively.

The first-order natural frequencies of samples under each condition were measured using the nine-point method identified in Section 3.2. Photographs of the laboratory tests are shown in Figure 7. Manual tapping is employed to create an incentive for the curtain wall. As the natural frequency is an inherent property of the structure which is independent of the magnitude of the excitation force, there is no need to deliberately ensure the same force’s magnitude.

Table 4. Comparison of first-order natural frequency results.

Conditions	First-Order Natural Frequency (Hz)
	Numerical Simulation	Fixed Test				Mobile Test
		1 m	Error	10 m	Error	1 m	Error	10 m	Error
Condition 1	101.00	101.43	0.43%	101.27	0.27%	104.34	3.30%	106.15	5.10%
Condition 2	66.11	65.13	−1.48%	64.47	−2.48%	69.64	5.34%	70.48	6.61%
Condition 3	75.37	72.04	−4.42%	71.86	−4.66%	74.07	−1.72%	80.67	7.03%
Condition 4	15.34	15.10	−1.56%	14.09	−8.15%	15.92	3.78%	13.36	−12.91%

compares the average first-order natural frequencies obtained from the numerical simulations with the experimental results of 1 m and 10 m ranging. The four identical panels were subjected to the same experimental procedure. The findings demonstrate that the fixed test conducted at a closer range exhibits a smaller error than the one conducted at a longer range under greater damage conditions. This is attributed to the shorter laser path, which results in a reduction in the power loss of the laser. Meanwhile, under all conditions, the mobile test consistently has a greater error than the fixed test. However, the maximum measuring error for the mobile tests was 12.91% (10 m ranging test under Condition 4). Since most curtain wall boundaries have failed, the sample is easily disrupted by external factors, resulting in unstable results. The remaining error is no more than 10%. This indicates the reliability of the UAV-LDV system for first-order natural frequency testing. The system has acceptable testing errors at various distances, but tests should be performed as closely as possible for the stability of results.

The effectiveness of the UAV-LDV system in identifying the debonding damage on HFSCGCW also requires verification. The acceleration time history under four conditions is tested and analyzed using the nine-point method identified in Section 3.2. In Figure 8, the results obtained by the UAV-LDV system for detecting debonding damage on HFSGCW closely align with the results predicted by numerical simulation and those obtained through fixed LDV testing. This alignment mainly refers to the intuitive indication of where the damage occurred and the clear reflection of the extent of the damage. It demonstrates the system’s capability to effectively identify the occurrence and location of debonding damage. The testing distance did not significantly affect the system when using the SSD indicator to detect debonding damage.

3.4. Noise Level

The LDV exhibits self-vibration caused by machinery, transportation, passersby, wind, etc. These natural noises can be mitigated by incorporating vibration sensors on the LDV and analyzing the combined data from them, or alternatively by using two LDVs together [54]. However, random noise from the environment affects the measuring accuracy, which necessitates further investigation. Four unilateral damaged conditions are used as examples to test the performance of the UAV-LDV system under different noise levels, as illustrated in Figure 9. The damage all occurs unilaterally, which distinguishes it from the four conditions depicted in Figure 6. In the case of unilateral damage, most of the structural sealants work properly, and the overall structure remains relatively stable. If noise levels significantly interfere with structural damage detection in this situation, more severely damaged conditions will further deteriorate. Therefore, the unilateral damaged conditions are chosen for testing. The damage lengths of the four conditions are intact, 1/4 length damaged, 1/2 length damaged, and completely damaged on one side. The measuring points are arranged according to the nine-point method presented in Section 3.2. Based on the results of the study in Section 3.3, the measuring distance was set to 1 m.

During the test, to simulate the actual environment, a stereo system is established using two Apple HomePod minis to play the recorded construction noises with a frequency distribution range of 0–500 Hz, as shown in Figure 10. They are located 2 m apart and symmetrically distributed along the central axis of the specimen, 0.5 m away from it. This allows them to interfere with both the curtain wall vibration and the LDV test results, resulting in a more realistic simulation of ambient noise. A decibel meter is used to measure the ambient noise level, and the test is performed when the noise level reaches 20 dB, 30 dB, 40 dB, 50 dB, 70 dB, and 100 dB, respectively.

Table 5. Comparison of first-order natural frequency results.

Conditions	Noise Level (dB)	First-Order Natural Frequency (Hz)
		Numerical Simulation	Fixed Test	Error	Mobile Test	Error
Condition 1	20	101.00	101.73	0.72%	105.97	4.17%
	30		101.92	0.91%	105.92	3.92%
	40		102.1	1.09%	92.48	−9.42%
	50		103.21	2.19%	93.28	−9.62%
	70		98.43	−2.54%	92.65	−5.87%
	100		96.62	−4.34%	86.26	−10.72%
Condition 2	20	91.09	91.61	0.57%	96.37	5.80%
	30		90.63	−0.50%	88.67	−2.66%
	40		92.83	1.91%	92.48	1.53%
	50		91.25	0.18%	94.34	3.57%
	70		94.12	3.33%	84.05	−7.73%
	100		88.24	−3.13%	101.04	10.92%
Condition 3	20	88.52	88.81	0.33%	80.32	−9.26%
	30		88.1	−0.47%	86.58	−2.19%
	40		89.29	0.87%	92.72	4.74%
	50		90.79	2.56%	95.31	7.67%
	70		86.02	−2.82%	94.68	6.96%
	100		85.31	−3.63%	96.95	9.52%
Condition 4	20	66.11	66.28	0.26%	64.1	−3.04%
	30		66.9	1.19%	64.47	−2.48%
	40		66.1	−0.02%	61.94	−6.31%
	50		63.04	−4.64%	62.88	−4.89%
	70		68.77	4.02%	71.94	8.82%
	100		70.12	6.07%	73.67	11.44%

compares the average first-order natural frequencies obtained from the numerical simulations with the tests. The same panel was tested under different conditions with the same technique. The results indicate that the noise level has a significant impact on the reliability of the UAV-LDV system. Even in the case of unilateral damage, the error in the first-order natural frequency always exceeds 5% or even 10%, and it will be larger for the cases with bilateral and multilateral damage where boundary failure is more severe. Consequently, to acquire first-order natural frequencies, it is recommended to perform the test during night-time or at low noise levels to minimize the effect of noises on accuracy. Photos of the tests under different noise levels are shown in Figure 11.

The study also verified the effectiveness of the UAV-LDV system in identifying debonding damage to the HFSCGCW at noise exposure, and the results are shown in Figure 12. The research findings indicate that noise levels do affect the test results, as the outcomes at different noise levels vary significantly from those predicted by the numerical simulation. The results show that the damage detection capability of the UAV-LDV system is significantly affected when the noise levels are 20, 30, and 40 dB. The assessment diagrams do not accurately reflect the location and severity of the damage. This is attributed to the excitation energy level at this time, which is similar to that of the ambient noise, causing the energy level of the structure to fluctuate more and have a larger absolute value. As the background noise level continues to rise, it eventually reaches the threshold for disturbing the curtain wall vibration. At this stage, the SSD index is more sensitive to the vibration characteristics of the structure, and the system’s recognition ability is gradually recovering. However, if the noise level is smaller than 20 dB or larger than 40 dB, either excitation or noise will dominate while others attenuate this tendency. As a result, the absolute value of the energy level will be smaller, leading to a smaller range of overall energy variations. Therefore, it is advisable to perform HFSGCW damage detection in a quiet environment.

3.5. Wind Speed

Controlling the attitude and stability of UAVs in windy conditions has been a significant technical challenge [58]. The UAV-LDV system is designed to inspect curtain walls, which requires long-term operation at high altitudes, making it subject to wind interference. This section conducts a wind resistance test to investigate the impact of wind speed on the system’s performance.

The tests of the first-order natural frequency are conducted for Condition 3 in Section 3.4 under the influence of wind speed. This condition is primarily selected due to the shorter damaged length and the fact that most of the structural adhesive is still in use. The glass panel will perform better in high winds during this time, while the wind speed mainly affects the performance of the UAV-LDV system. If there were significant errors in the results under this condition, the conditions of greater severity of damage would have larger errors. The measuring points were arranged and the distances were measured in accordance with the provisions of Section 3.4 (1 m). The working UAV-LDV system was subjected to lateral wind at three different speeds: 1.5 m/s, 3.0 m/s, and 5.0 m/s (produced by an electric fan). Figure 13 shows the arrangement of the tests. Photos of the tests in the wind are shown in Figure 14.

Table 6. Comparison of first-order natural frequency results of Condition 3.

Average of 1st Order Natural Frequency (Hz)
Numerical Simulation	Wind Speed (m/s)	Fixed Test	Error	Mobile Test	Error
88.52	1.5	88.68	0.18%	85.41	−3.51%
	3.0	88.97	0.51%	83.48	−5.69%
	5.0	88.13	−0.44%	96.13	8.60%

presents the average first-order natural frequencies obtained from the nine measuring points. The results indicate that the wind speed has a significant impact on the reliability of the UAV-LDV system. Therefore, it is recommended to acquire the first-order natural frequency with this system in a windless environment or use an error correction algorithm to minimize the effect of wind speed on accuracy.

The study verifies the effectiveness of the UAV-LDV system in identifying debonding damage to the HFSCGCW under the influence of wind. Figure 15 validates the results of different wind speeds through numerical simulation and fixed tests. The assessment chart can give a general indication of the locations of damaged structural adhesives, but it is not very effective in pinpointing the exact location. When the wind speed reaches 5.0 m/s, it is difficult to provide useful information about the damaged side. The results indicate that the wind greatly affects the detection capability of the UAV-LDV system. The SSD indicator has a large numerical error in windy conditions and does not clearly indicate the exact location of the damage, even in the case of lighter damage. During the test, high winds interfered with the hovering and stabilization of the UAV. Even aligning the measuring points was difficult at times. To maintain alignment in strong winds, the drone continuously maneuvers through turns and side shifts. In contrast to the drone’s camera, which is stabilized by a gimbal, the LDV is fixed directly to the fuselage. This configuration leads to the laser beam deviating from the target as the orientation of the fuselage changes, and passing through transparent glass. Consequently, the laser intensity that reflects the LDV is insufficient for stable data acquisition, resulting in failed assessments. As demonstrated in Figure 15, at a wind speed of 5 m/s, the processed results suggest extensive and severe debonding damage along both boundaries, even when the actual damage is minor. This contradicts the actual conditions. In subsequent research, it is recommended that a gimbal structure be developed to stabilize the LDV device, in order to minimize the errors in optical path differences caused by attitude adjustments. This approach is simpler and more efficient than developing complex algorithms. As a result, it is not recommended to perform tests in windy weather now.

4. Conclusions

In this paper, based on previous research, an identification method of debonding damage of HFSGCW based on a UAV-LDV system is proposed. The established system enables UAVs to transport the LDV to high altitudes for operation rather than utilizing a fixed LDV which works with the assistance of a UAV. It solves the previous problem of relying on manual inspection for debonding damage, which was inefficient, costly, and risky. The new assessment indicator (SSD) based on wavelet packet band energy decomposition has the additional advantage of providing a visual representation of the precise location of the damage, which is a superior feature compared to traditional indicators such as first-order natural frequency. In addition, potential factors affecting the performance of the UAV-LDV system were investigated, including the arrangement of measuring points, measuring distance, noise level, and wind speed. The study demonstrated that a wider distribution of measurement points and closer proximity of the UAV to the curtain wall enhance the detection efficiency by increasing data density in assessment diagrams and reducing laser energy loss. The results demonstrate that despite the influence of ambient noise and wind speed on the system performance, the system is still able to provide accurate and reliable detection results under reasonable operating conditions. Compared with traditional methods, the UAV-LDV system has significant advantages. Firstly, the system is able to complete large-area inspection tasks in a shorter time, which greatly improves inspection efficiency. Secondly, the system realizes non-contact detection, avoiding the safety hazards in the traditional method. Finally, the SSD indicator is able to accurately identify the damage location, providing more intuitive damage detection results.

The main innovations of the study include the introduction of a novel UAV-LDV system for damage detection in Hidden Frame-Supported Glass Curtain Walls, the development of a new SSD indicator that offers visualization benefits and outperforms conventional indicators, the investigation of potential factors affecting the UAV-LDV system’s performance, the conduction of an indoor full-scale experiment to examine the efficiency of the novel approach and the provision of valuable guidelines for utilizing the UAV-LDV system to inspect glass curtain walls.

The establishment of the UAV-LDV system provides a low-cost, remote method for the early warning of HFSGCW, similar to the health monitoring of bridges and building structures. The results of our study indicate that the proposed nine-point measuring arrangement method should be adopted. It is inadvisable to utilize the UAV-LDV system for the identification of damage in windy or noisy environments, given the potential for a reduction in accuracy. In the future, we will further optimize the wind resistance of the UAV-LDV system and methods to reduce noise interference and will integrate the system with other technologies, such as incorporating machine learning methods to further improve inspection accuracy and efficiency. Finally, more field tests will be conducted to verify the stability and reliability of the system under different environmental conditions.