Head-mounted aircraft complete coating damage visualization system

Abstract

The invention provides a head-mounted aircraft complete coating damage visualization system, which belongs to the field of stealth airplanes and comprises a cluster type uniform heating system, a head-mounted visualization system and a narrow area small heating detection component. The cluster type uniform heating system is used for carrying out cluster heating on a large-area area on the surface of an aircraft and providing a heat source for coating damage detection; the head-mounted visualization system comprises a binocular infrared imaging mirror which is used for collecting a temperature change image sequence of the airplane surface after thermal pulse excitation and realizing rapid detection and visual presentation of a large-area surface coating; the small heating detection component (hereinafter referred to as a small component) for a narrow area is used for rapid detection of corners and narrow areas of an aircraft so as to realize comprehensive and rapid detection of the whole aircraft. The aircraft stealth coating damage detection efficiency and accuracy can be improved, and the rapid inspection requirement before an aircraft task is met.

Description

Technical Field

[0001] This invention belongs to the field of stealth aircraft, specifically relating to a head-mounted aircraft coating damage visualization system. Background Technology

[0002] With the gradual deployment of various stealth aircraft models and the rapid development of detection systems, the performance requirements for aircraft are becoming increasingly stringent. The need for stricter control over aircraft performance before missions is even greater. However, currently, there is a lack of efficient pre-flight inspection methods for the performance of aircraft stealth coatings. Most methods rely on visual inspection or small-area checks of a few square centimeters, which are unsuitable for rapid pre- or post-mission testing. This lack of rapid pre-mission testing translates to a deficiency in the ability to control aircraft performance, hindering the performance testing and maintenance of high-value stealth aircraft.

[0003] Existing technologies mainly include visual inspection or deep non-destructive testing at the level of a few square centimeters. Visual inspection has poor accuracy and is greatly affected by the environment, making it a necessary but unfortunate measure due to the lack of suitable inspection methods for aircraft coatings. Deep non-destructive testing is limited by the environment and testing methods, such as ultrasonic testing, and can only perform precise inspections at the level of a few square centimeters. Over the length of an aircraft of several meters, it is time-consuming and difficult to conduct rapid testing, thus failing to meet the requirement of conducting a performance test before an aircraft is deployed. Summary of the Invention

[0004] To address the challenge of rapidly detecting the stealth performance of stealth aircraft during routine maintenance, this invention proposes a head-mounted aircraft coating damage visualization system. By using large-area cluster heating and infrared imaging, the system achieves rapid detection of the stealth performance of the entire aircraft.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A head-mounted aircraft coating damage visualization system includes a clustered uniform heating system, a head-mounted visualization system, and a small heating detection component for narrow areas. The clustered uniform heating system is used to perform clustered heating on a large area of ​​the aircraft surface, providing a heat source for coating damage detection. The head-mounted visualization system includes a binocular infrared imaging lens for acquiring image sequences of temperature changes on the aircraft surface after thermal pulse excitation, enabling rapid detection and visualization of coatings on large surfaces. The small heating detection component for narrow areas is used for rapid detection of aircraft corners and narrow areas, achieving comprehensive and rapid detection of the entire aircraft, improving the efficiency and accuracy of stealth coating damage detection, and meeting the rapid inspection requirements before aircraft missions.

[0007] Beneficial effects:

[0008] This invention enables rapid large-area detection under outdoor conditions, solving the problem of small detection areas in precise detection. It employs cluster heating to increase detection accuracy and utilizes head-mounted infrared imaging for rapid overall system inspection. Innovatively, it utilizes an improved convolutional neural network for real-time intelligent identification and type classification of damaged areas, forming a human-machine collaborative damage interpretation mechanism that significantly improves detection accuracy. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of a head-mounted aircraft coating damage visualization system according to the present invention;

[0010] Figure 2(a) is a schematic diagram of the clustered uniform heating system of the present invention;

[0011] Figure 2(b) is a schematic diagram of the wireless head-mounted infrared imaging system of the present invention;

[0012] Figure 3 The graph shows the change in brightness temperature as a function of emissivity.

[0013] Figure 4(a) shows the isotherm diagram of the target surface after thermal excitation;

[0014] Figure 4(b) shows a sample diagram of damage identification.

[0015] Figure 5 Here is a flowchart of the noise preprocessing process;

[0016] Figure 6 Network architecture design diagram;

[0017] Figure 7 This is a schematic diagram of a lightweight hybrid convolution module;

[0018] Figure 8 This is a flowchart of the channel attention module.

[0019] Figure 9 This is a schematic diagram of multi-scale feature pyramid fusion.

[0020] Figure 10 This is a schematic diagram of the dual-path attention mechanism;

[0021] Figure 11 A schematic diagram of a pyramid-shaped pooling of a hollow space;

[0022] Figure 12 To improve the schematic diagram of the SE module;

[0023] Figure 13 This is a schematic diagram of the damage segmentation head architecture;

[0024] Figure 14 This is a schematic diagram of the damage classification head architecture. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0026] like Figure 1 As shown, the present invention provides a head-mounted aircraft coating damage visualization system comprising a clustered uniform heating system and a head-mounted visualization system. The clustered uniform heating system is located next to the target under test and transmits information wirelessly to the head-mounted visualization system worn by maintenance personnel.

[0027] As shown in Figures 2(a) and 2(b), the clustered uniform heating system includes a power supply, heating components, clustered heating wave components, and an overall frame. The clustered uniform heating system is supported by the overall frame. The bottom of the system integrates a power supply system to provide stable electrical energy, while the top is equipped with heating components to convert electrical energy into heat energy. The clustered heating wave components optimize the thermal field distribution through waveguides, and the frame structure simultaneously undertakes heat dissipation protection and modular expansion functions.

[0028] The head-mounted visualization system comprises a binocular infrared imaging lens, a head-mounted component, an imaging system, and a power supply. The system uses an ergonomic head-mounted component as its main body, with a front-mounted binocular infrared imaging lens for collecting environmental thermal radiation, a rear-mounted imaging system for signal processing and image enhancement, and a compact power module on the side for continuous power supply.

[0029] When defects appear in the coating of a stealth aircraft, they can be directly observed using infrared technology. However, since the overall coating lacks a heat source, a clustered, uniform heating system is used to evenly distribute heat across the aircraft surface. During operation, for large, flat surfaces such as the upper and lower surfaces of the wings, fuselage, and tail, the clustered heating method imparts a certain temperature to the coating surface, allowing for rapid identification of defects using head-mounted imaging. For areas that are difficult to observe directly, such as the wing-fuselage junction, air intakes, and tail connection points, rapid inspection is performed using small, confined heating detection components. This enables rapid inspection of the entire aircraft, facilitating quick checks before flight missions.

[0030] The testing principle of this invention is as follows: Utilizing the stealth mechanism of the coating (with an infrared stealth layer at the outermost layer) and infrared thermometry, a large area of ​​the aircraft surface is subjected to concentrated heating. By analyzing the detailed temperature distribution of the aircraft coating, the presence of defects can be determined. The temperature sources of the aircraft stealth coating under infrared imaging mainly include the temperature from the concentrated heating. Temperature of the reflected environment The temperature inherent in the coating itself At the same temperature as the ambient temperature, while also considering the low infrared emissivity of the stealth coating. and reflectivity .

[0031] Temperature at the measured location after cluster heating for:

[0032] ;

[0033] The infrared radiation intensity at this location This includes the intensity of infrared radiation emitted by the device itself and the intensity of infrared radiation reflected from the environment.

[0034] ;

[0035] In the formula, is the Stefan-Boltzmann constant.

[0036] In infrared imaging recognition processing, the influence of the atmospheric environment must also be considered. The contribution of the atmospheric environment to the intensity of infrared radiation is as follows:

[0037] ;

[0038] In the formula, The intensity of infrared radiation in the atmosphere. The infrared emissivity of the atmosphere. It is the effective radiation temperature of the atmosphere.

[0039] An environmental compensation mechanism is employed to eliminate the atmospheric radiation term at this location through blackbody calibration. Since the coating's reflectivity is also affected by emissivity, i.e.:

[0040] ;

[0041] in, R is the infrared emissivity of the object, and R is the infrared reflectivity of the object.

[0042] Therefore, regional infrared radiation obtained by focused heating can be obtained. :

[0043] ;

[0044] In the formula, The surface temperature of the target to be measured. This refers to the temperature inherent in the coating itself.

[0045] regional infrared radiation By comparing the radiation with that of a blackbody, the temperature of the blackbody is obtained, which is the radiation temperature at that location. This temperature is then displayed on the screen as the brightness temperature at that location. :

[0046] ;

[0047] ;

[0048] The head-mounted visualization system acquires the temperature changes on the aircraft surface after thermal pulse excitation at a high frequency (100Hz), generating a set of time-temperature image sequences. , Let be the pixel coordinates and t be the time point. Perform a Fast Fourier Transform (FFT) on the temperature-time curve T(t) for each pixel to obtain the frequency domain signal. :

[0049] ;

[0050] In the formula, For frequency components (usually focusing on the fundamental frequency) , i.e., thermal excitation frequency), T(t) is the function of temperature field change with time, i is the imaginary unit, and t is the time series point of the application of thermal pulse excitation.

[0051] The result of the FFT transformation is a complex number. , can be converted to:

[0052] Where A represents the real component and B represents the imaginary component.

[0053] Amplitude: ;

[0054] Phase: ;

[0055] After the temperature field data is transmitted to the head-mounted visualization system, thermal sequence denoising is performed using temporal median filtering and spatial nonlocal mean denoising. A phase image and amplitude image are calculated for each pixel to suppress environmental noise.

[0056] Aircraft stealth coatings have low emissivity. When the coating is damaged, its emissivity will inevitably increase, and its reflectivity will decrease accordingly. The infrared emissivity of stealth coatings in normal temperature environments is generally between 0.1 and 0.3, while in high-temperature environments it is generally between 0.4 and 0.8. Taking a stealth coating in a normal temperature environment as an example, assuming an infrared emissivity of 0.2, a cluster heating area temperature of 50℃ (323K), and an ambient temperature of 15℃ (288K), the change in infrared emissivity of the stealth coating can be measured using the brightness temperature versus emissivity curve as shown below. Figure 3 As shown.

[0057] Infrared emissivity When the value is 0.2, the measured brightness temperature is 23℃. If the infrared emissivity variation of the stealth coating is required to be within the range of 0.15~0.25, the stealth coating is considered to still be functional. If the infrared emissivity variation exceeds this range, maintenance is required. In this system, this translates to a measured brightness temperature range of 21.1℃~25℃.

[0058] By combining an improved CNN network to identify damage in infrared images, damage boundaries are further defined, damage geometric parameters are automatically calculated, and maintenance warnings are triggered based on these parameters. Figures 4(a) and 4(b) show the temperature changes in different regions of the material after infrared thermal excitation. In Figure 4(a), isotherms clearly distinguish temperature regions of 23°C, 24°C, and 25°C. Typically, the damaged area forms a significant temperature difference with the surrounding area, a feature that facilitates marking the damage location using image recognition technology. As shown in Figure 4(a), the arrow points to a 25°C area and is labeled "area requiring repair," indicating potential damage requiring further repair. By optimizing the deep learning algorithm, the damage type and area of ​​this type of damaged area can be more accurately identified, as shown in Figure 4(b), where the identified damage area is 345.35 mm². 2 The damage type is scratch damage.

[0059] The preprocessing procedure first performs temporal median filtering on the original thermal imaging sequence T(x,y,t) to eliminate impulse noise; then, spatial noise is further suppressed through spatial nonlocal mean denoising. Next, a Fourier transform is performed on each pixel to extract the fundamental frequency f0 component representing the pulse wave characteristics, and its phase map φ(x,y) and amplitude map A(x,y) are calculated respectively. Finally, these two feature maps are used as dual-channel inputs and fed into an improved convolutional neural network. The process is as follows: Figure 5 As shown.

[0060] The phase and magnitude maps are input into an improved convolutional neural network (CNN) damage recognition module. This improved network architecture is a multi-scale feature pyramid fusion network (MFPN) + dual-path attention mechanism. The overall network architecture design concept is as follows: Basic features from the dual-channel input are extracted through the backbone network LHCB; cross-level feature fusion and enhancement are achieved through the multi-scale feature pyramid MFPN; subsequently, channel-spatial feature calibration is performed simultaneously through the dual-path attention mechanism DPAM, driving the damage segmentation head and classification head respectively. The segmentation head outputs pixel-level damage region detection, and the classification head determines the damage type, forming a joint learning framework that balances localization accuracy and semantic understanding. The process is as follows: Figure 6 As shown.

[0061] The backbone network uses a lightweight hybrid convolutional module (LHCB) instead of standard convolutions. The input features are first processed by a 3×3 depthwise separable convolution (DWConv) to extract spatial features, then by a 1×1 pointwise convolution (PWConv) to transform the channel dimensions. Subsequently, a channel attention mechanism (SE) is introduced to dynamically calibrate the weights of each channel, finally outputting the optimized feature architecture. The specific process is as follows: Figure 7 As shown.

[0062] DWConv is a depthwise separable convolution used for spatial feature extraction. It reduces the number of parameters to 1 / 8 of a standard convolution while preserving high-frequency details (crucially capturing micro-cracks). The formula is:

[0063] ;

[0064] In the formula, Represents the size of the convolution kernel. This represents the number of input channels.

[0065] PWConv is a pointwise convolution that fuses channel-dimensional features, allowing for the combination of damage features from different channels and enhancing cross-channel information interaction.

[0066] The SE module is a channel attention module that weights the importance of feature channels. Its workflow is as follows: Global Average Pooling (GAP) is applied to the input feature map to capture channel-level statistics; a fully connected layer (FC) compresses the channel dimension to C / 8; ReLU activation restores the original number of channels C; a Sigmoid function generates channel weight vectors; and finally, channel weighting is applied to the original features. The flowchart is shown below. Figure 8 As shown in the diagram, this module can dynamically increase the weight of damage-related channels and suppress background thermal noise channels.

[0067] The Multi-Scale Feature Pyramid Fusion (MFPN) architecture is designed as follows: the bottom path processes 1 / 4 resolution features through three LHCB modules, preserving high spatial details to capture fine structures such as microcracks; the top path, after downsampling with a stride of 2 convolutions, extracts 1 / 16 resolution global semantic features through four LHCB modules, and then fuses them with the bottom-level features after a 4x upsampling. The specific architecture is as follows: Figure 9 As shown, a dual-path parallel processing method is employed. The bottom path preserves pixel-level details of 0.1mm-level cracks, with a small receptive field suitable for local features. The top path captures the global structure of large-area damage, with a large receptive field suitable for understanding damage distribution.

[0068] The Dual Path Attention (DPAM) mechanism is designed as follows: input features are fed in parallel into the spatial attention branch and the channel attention branch, generating a spatial weight map and a channel weight vector, respectively. Feature reweighting is achieved through element-wise multiplication, as shown in the structure below. Figure 10As shown. The spatial attention path structure is dilated spatial pyramid pooling (ASPP). The process is as follows: input features are passed in parallel through three sets of dilated convolutions with dilation rates of 6 / 12 / 18 to capture short, medium, and long-range contextual information, respectively, while global pooling is performed to extract overall statistical features. After channel concatenation of the four features, cross-scale feature fusion and dimensionality compression are achieved through 1×1 convolution. Finally, the spatial attention weight map is generated by the Sigmoid function. The process is as follows. Figure 11 As shown, it serves to achieve multi-scale context awareness (adapting to damage of different sizes); highlight the spatial location of the damage area; and suppress thermal reflection artifacts. The channel attention path structure is an improved SE module process: First, global average pooling (GAP) is performed on the input features to obtain channel-level statistics. After compression to C / 8 dimensions through a fully connected layer, layer normalization (LayerNorm) is introduced to stabilize the feature distribution. After ReLU activation nonlinear transformation, the original channel dimensions are restored through a fully connected layer. Finally, channel weights are generated using Sigmoid. The structure is as follows. Figure 12 As shown.

[0069] The damage segmentation head architecture is as follows: Input features are first processed through two LHCB modules for deep feature extraction and enhancement. Then, transposed convolution is used to upsample the feature map and restore spatial resolution. Next, a 1x1 convolution is used to compress the channel dimension. Finally, a sigmoid activation is applied to generate a pixel-level binary segmentation map. The design diagram is shown below. Figure 13 As shown, a progressive upsampling strategy is adopted to avoid loss of details caused by direct upsampling, while transposed convolution (stride 2) is used to gradually restore the resolution.

[0070] The damage classification head architecture, with its top-level features, first extracts spatially independent global semantic information through global average pooling. This information is then compressed from 512 dimensions to 128 dimensions via a fully connected layer. A 50% probability Dropout layer enhances generalization ability, followed by another fully connected layer mapping to a 3-dimensional feature space. Finally, a Softmax function outputs the probability distribution of the three damage types. The process is as follows: Figure 14 As shown, a feature selection mechanism is adopted, which uses only the top-level path features while avoiding local details from interfering with classification.

[0071] The output damage segmentation map is overlaid with the original infrared image, and the potential damage locations are marked and the damage types are classified in the overlaid image.

[0072] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A head-mounted aircraft coating damage visualization system, characterized in that, The system includes a clustered uniform heating system, a head-mounted visualization system, and a small heating detection component for narrow areas. The clustered uniform heating system is used to heat a large area of ​​the aircraft surface, providing a heat source for coating damage detection. The head-mounted visualization system includes a binocular infrared imaging lens, used to acquire image sequences of temperature changes on the aircraft surface after thermal pulse excitation, enabling rapid detection and visualization of coatings on large surfaces. The small heating detection component for narrow areas is used to quickly detect aircraft corners and narrow areas, achieving comprehensive and rapid inspection of the entire aircraft, improving the efficiency and accuracy of stealth coating damage detection, and meeting the rapid inspection requirements before aircraft missions.

2. The head-mounted aircraft coating damage visualization system according to claim 1, characterized in that, The clustered uniform heating system includes a power supply, heating components, clustered heating wave components, and an overall frame.

3. The head-mounted aircraft coating damage visualization system according to claim 1, characterized in that, The head-mounted visualization system also includes a head-mounted component and an imaging system. The head-mounted component is easy for operators to wear and operate, while the imaging system works in conjunction with a binocular infrared imaging lens to further improve the image acquisition and processing effect, and enhance the convenience and accuracy of detection.

4. The head-mounted aircraft coating damage visualization system according to claim 1, characterized in that, The narrow-area miniature heating detection component includes a miniature heating element, an imaging display interface, buttons, a miniature imaging system, and a power supply. It has independent heating and imaging detection functions and can flexibly meet the detection needs of different areas of the aircraft.

5. The head-mounted aircraft coating damage visualization system according to claim 1, characterized in that, It also includes an improved convolutional neural network for real-time intelligent identification and type classification of the acquired infrared images, forming a human-machine collaborative damage judgment mechanism to achieve rapid and accurate identification and classification of coating damage.

6. The head-mounted aircraft coating damage visualization system according to claim 5, characterized in that, The improved convolutional neural network adopts an architecture that combines a multi-scale feature pyramid fusion network with a dual-path attention mechanism.

7. The head-mounted aircraft coating damage visualization system according to claim 6, characterized in that, The multi-scale feature pyramid fusion network employs dual-path parallel processing. The bottom path preserves pixel-level details of small-scale cracks, while the top path captures the global structure of large-area damage, achieving comprehensive extraction and fusion of damage features.

8. The head-mounted aircraft coating damage visualization system according to claim 6, characterized in that, The dual-path attention mechanism includes a spatial attention path and a channel attention path. The spatial attention path adopts a hollow spatial pyramid pooling structure, while the channel attention path adopts an improved SE module.

9. The head-mounted aircraft coating damage visualization system according to claim 1, characterized in that, It also includes a thermal sequence denoising module, which uses a combination of temporal median filtering and spatial nonlocal mean denoising to denoise the acquired temperature change image sequence and improve image quality.

10. The head-mounted aircraft coating damage visualization system according to claim 1, characterized in that, The damage geometry parameters are automatically calculated based on the damage identification results, and maintenance warnings are triggered based on the damage parameters.

Images