A method for rapid defect detection of pressure-bearing equipment in service using organic mechanoluminescent materials

Abstract

The present invention is a method for quickly detecting defects of pressure-bearing equipment in service by using organic mechanoluminescent materials. An organic mechanoluminescent material is selected to prepare a solution, and an organic mechanoluminescent coating is constructed on the surface of the component. Through a tensile test of a standard sample containing defects, mechanoluminescent images under different loads and different defect depths and sizes are obtained respectively. The relationship between the characteristic value of the mechanoluminescent area and the load and defect size is obtained through image processing, and an image database of mechanoluminescent images induced by defects in pressure-bearing equipment is established. An organic mechanoluminescent coating is applied to the surface of the pressure-bearing equipment, and the response of the mechanoluminescence on the surface of the pressure-bearing equipment is monitored online during service. The position of the internal defect is quickly located by using the position where the mechanoluminescence appears; computer vision and artificial intelligence technology are used to identify the mechanoluminescent images in the database, train the data and establish a training model; the mechanoluminescent images collected in real time are brought into the model to quantitatively evaluate the defect size and depth of the pressure-bearing equipment.

Description

Technical Field

[0001] The invention relates to a detection method for rapidly identifying and quantitatively evaluating defects of pressure-bearing equipment in service based on organic mechanoluminescence. Background Art

[0002] Pressure equipment is a key equipment in the field of petrochemical and nuclear power, and is often exposed to flammable, explosive and corrosive media during use. During long-term service, the container will suffer from corrosion and cracks in the weld, which will pose a major risk to the safe service of the pressure equipment. During service, it is difficult to monitor in real time the location and size of internal or external defects of pressure equipment caused by corrosion and cracking of welds. Therefore, it is crucial to develop a detection method for rapid identification and evaluation of defects in pressure equipment in service for safety monitoring and early warning of pressure equipment in service during production.

[0003] At present, the detection methods for defects in pressure-bearing equipment are mainly concentrated in ultrasonic testing, radiographic testing, eddy current testing, magnetic particle testing and penetration testing. Ultrasonic testing relies on its super strong penetration and has high sensitivity, but the inspection results for complex components are not satisfactory. Radiographic testing can intuitively determine the size and location of internal defects in pressure-bearing equipment, but radiographic testing is more harmful to the inspectors and difficult to detect pressure-bearing equipment in service. Eddy current testing can quickly obtain the signal of internal defects, but the specific location of the defects needs to be re-inspected by combining multiple detection methods to determine. Although magnetic particle testing is easy to operate and the defect expression is intuitive, it is limited by the materials of pressure-bearing equipment and can only be applied to magnetic conductive materials. Penetration testing can only obtain the shape and scale of surface or near-surface defects, but cannot obtain the depth of defects, and cannot detect internal defects. Operators need to enter the pressure-bearing equipment to carry out inspections. The detection area of ​​the sensors of the above detection methods is limited, and large-scale rapid detection is difficult. It is often only carried out during regular maintenance of pressure-bearing equipment, and it is difficult to conduct large-scale and rapid screening of internal defects of pressure-bearing equipment in service. Therefore, it is necessary to develop a large-scale, rapid and intuitive visual detection method for the location, length and depth of defects in service pressure-bearing equipment.

[0004] Fluorescence is widely used in optoelectronic devices, biological and hazardous material detection due to its high sensitivity, real-time response and visualization. Real-time visualization detection of deformation and damage of engineering components using mechanofluorescence is an important development direction in the field of structural health monitoring. The Chinese invention patent "A method for detecting and monitoring cracks in mechanical components using fluorescent quantum dots" (invention patent application number: CN 201210586280 Publication number: CN 103901003 A) discloses a method for detecting fatigue cracks in mechanical components based on the fluorescence response of inorganic quantum dots. The Chinese invention patent "A method for dynamically monitoring stress intensity factor at crack tip" (invention patent application number: CN202110093811.0 Publication number: CN113008669A) discloses a method for dynamically monitoring the fracture toughness of cracks on the surface of components using organic mechanofluorescence. In summary, there are only individual reports on the use of mechanofluorescent materials to monitor damage to mechanical components. The above method is mainly used for detecting the position and length of surface cracks in engineering components. There are no reports on using force-induced fluorescence to quickly detect the depth of defects in pressure-bearing equipment, especially internal defects.

[0005] Tetranitro-tetraphenylethylene (TPE-4N) has the advantages of good film-forming properties on metal surfaces and rapid mechanoluminescent response. The Chinese invention patent "A method for detecting the mechanical response of mechanical parts using organic mechanoluminescent materials" (invention patent application number: CN201810258759.8 Publication number: CN108680288A) discloses a method for detecting stress-strain response and predicting fatigue crack propagation using tetranitro-tetraphenylethylene (TPE-4N) materials. This makes it possible to monitor defects in pressure-bearing equipment using organic mechanoluminescent materials. The above method only detects the overall stress-strain response of the component and the path and length of the initiation and propagation of fatigue cracks on the surface in real time. It is impossible to monitor and evaluate equipment defects or crack depths. There is no report on a method for large-scale, rapid and visual monitoring of internal defects of pressure-bearing equipment based on organic mechanoluminescence. Summary of the invention

[0006] In view of the above problems, the object of the present invention is to provide a large-scale, rapid and visual detection method for pressure-bearing equipment defects based on organic mechanoluminescence.

[0007] The technical solution of the present invention:

[0008] A method for quickly detecting defects of pressure-bearing equipment in service using organic mechanoluminescent materials comprises the following steps:

[0009] (1) Establishment of a database of pressure-bearing equipment defect-induced force-induced fluorescence images:

[0010] (1.1) Prepare standard tensile specimens with internal or external defects of different sizes and depths;

[0011] (1.2) selecting an organic mechanoluminescent material as a material and preparing an organic mechanoluminescent material solution; using a brush to evenly apply the organic mechanoluminescent material solution on the front and back surfaces of a standard tensile specimen containing defects, and heating the solution with a heating tool to form a film, thereby preparing a uniform organic mechanoluminescent coating;

[0012] (1.3) A tensile test was conducted on a defective standard tensile specimen coated with organic mechanoluminescent material. An ultraviolet light source was used as the excitation light source. A CCD fluorescence acquisition system was used to record in real time the mechanoluminescent images of both sides of the specimen under different loads, defect lengths and depths to simulate the mechanoluminescent response caused by surface defects and internal defects of pressure-bearing equipment.

[0013] (1.4) Processing of the mechanofluorescence image: The mechanofluorescence image is processed to obtain the characteristic value of the mechanofluorescence image. The mechanofluorescence image area is extracted through noise reduction and binarization processing, and the area and distance of the fluorescence in the binary image are calculated;

[0014] (1.5) Establish a defect-induced force-induced fluorescence image database for pressure-bearing equipment: As the load and the length and depth of the defect increase, the area and distance of the force-induced fluorescence image on the sample surface gradually increase; Analyze the results obtained in the test (1.3) in detail, establish the relationship between the area and distance of the force-induced fluorescence image under different loads and defect sizes, and establish a defect-induced force-induced fluorescence image database;

[0015] (2) Online monitoring method for defects of pressure-bearing equipment in service:

[0016] (2.1) Selecting an organic mechanoluminescent material as a material and preparing an organic mechanoluminescent material solution; using a brush to evenly apply the organic mechanoluminescent material solution on the surface of the pressure-bearing equipment, and heating it with a heating tool to form a film, thereby preparing a uniform organic mechanoluminescent coating;

[0017] (2.2) Use the CCD fluorescence acquisition system to scan and monitor the force-induced fluorescence signal on the surface of the pressure-bearing equipment in service in real time, and quickly locate the internal or external defects of the pressure-bearing equipment according to the location of the force-induced fluorescence;

[0018] (2.3) Combining computer vision and artificial intelligence technology (machine learning or deep learning) to build a residual attention network for the force-induced fluorescence images collected in the standard database, and extract the deep features of the fluorescence images with the deep network of the attention mechanism;

[0019] (2.4) Extract the feature values ​​of the induced fluorescence image based on image processing technology, and enhance the feature representation of the fluorescence image through feature fusion; train and verify the constructed residual attention classification network to obtain a training model;

[0020] (2.5) The force-induced fluorescence image results of the pressure-bearing equipment surface collected in real time online are input into the training model. The classification model will output the classification results of the fluorescence image and calculate the load and size of the force-induced fluorescence image induced by the defects of the pressure-bearing equipment according to the pre-defined category labels.

[0021] The organic mechanoluminescent material is tetranitro-tetraphenylethylene (TPE-4N), and the concentration of the TPE-4N solution is 0.01-0.05 g / mL; the organic solvent is chloroform.

[0022] The heating tool uses a hot air gun, a heating furnace or a heating sleeve; the heating temperature is 80° C. to 300° C., and the heating time is about 1 to 30 minutes.

[0023] The fluorescence detection device adopts a CCD fluorescence acquisition system to collect real-time force-induced fluorescence images of defective samples and service pressure-bearing equipment on the surface after being subjected to force.

[0024] The CCD fluorescence collection system includes a CCD camera, a lens, an ultraviolet light source and a computer data analysis system. The fluorescence collection system is installed outside the standard sample and the service pressure container, and the collected force-induced fluorescence characteristic image is transmitted back to the computer data analysis system in real time.

[0025] The image processing method is to process the recorded force-induced fluorescence image of the defective sample and apply it to fluorescence analysis. The force-induced fluorescence image of the surface of the standard sample during deformation is selected, and a 1-50mm×1-50mm area is selected as the analysis object, which is converted into a grayscale image through image processing, and then the image is subjected to noise reduction, median filtering, smoothing and binarization processing to obtain the force-induced fluorescence area, and the area and distance occupied by the force-induced fluorescence area in the selected area are calculated.

[0026] The defect-induced force-induced fluorescence image database is established by conducting tensile tests on standard tensile specimens with internal and external defects of different sizes and depths to obtain force-induced fluorescence images. The relationship between the area and distance of force-induced fluorescence and the change of defect size, depth and load is established, and the defect-induced force-induced fluorescence image database of pressure-bearing equipment is established based on the force-induced fluorescence response test results of a series of defective specimens under different tensile loads.

[0027] The rapid defect identification of the pressure-bearing equipment in service is to apply an organic mechanoluminescent coating on the surface of the pressure-bearing equipment, use a CCD fluorescence acquisition system to detect the response area of ​​mechanoluminescence online, and use the appearance position of mechanoluminescence on the surface of the component to quickly locate the location of the defect inside the component.

[0028] The quantitative evaluation of defects of the pressure-bearing equipment in service is to combine computer vision and artificial intelligence technology (machine learning or deep learning) to perform eigenvalue analysis on images in the database and establish a training model; the eigenvalues ​​of the force-induced fluorescence image on the surface of the pressure-bearing equipment collected online in real time are input into the training model to quantitatively obtain the depth and length of the defects of the pressure-bearing equipment in service.

[0029] The specific instructions are as follows:

[0030] A detection method for rapidly identifying and quantitatively evaluating defects of pressure-bearing equipment in service based on organic mechanoluminescence of the present invention comprises the following steps:

[0031] (1) selecting an organic fluorescent material as a material and preparing an organic fluorescent material solution;

[0032] (2) Establishment of a database of pressure-bearing equipment defect-induced force-induced fluorescence images:

[0033] (2.1) Standard tensile specimens with defects of different sizes (defect length or diameter is 1 to 30 mm) and depths (depth range is 10% to 90%) were prepared.

[0034] (2.2) An organic mechanoluminescent coating is prepared on both sides of the sample, evenly applied to the sample surface by a brush, and heated to form a film.

[0035] (2.3) A tensile test was performed on a defective standard tensile specimen coated with organic mechanoluminescent material. The mechanoluminescent images of both sides of the specimen under different loads (load 50-300 MPa) were recorded in real time using a CCD fluorescence acquisition system to simulate the mechanoluminescent response of the specimen surface caused by internal and external defects. A tensile test was performed on each specimen and its mechanoluminescent image was recorded.

[0036] (2.4) Processing of the mechanofluorescent image: First, the mechanofluorescent image is processed to extract the characteristic value of the mechanofluorescent area. The background is processed first, and then the threshold is selected. After the threshold is selected, the image is binarized so that the fluorescent area in the middle of the image is white and the other areas are black. The area and distance of the fluorescence (white part) in the binary image are calculated.

[0037] (2.5) Establishment of a database of force-induced fluorescence images induced by defects in pressure-bearing equipment: Perform eigenvalue analysis on the force-induced fluorescence response images collected in (2.3) to obtain the area and distance of force-induced fluorescence caused by internal or external defects. Establish the relationship between the eigenvalue of the force-induced fluorescence image and the change of defect size and tensile load; Based on the eigenvalues ​​of the force-induced fluorescence images of a series of defective specimens under different tensile loads, establish a database of force-induced fluorescence images induced by defects in pressure-bearing equipment.

[0038] (3) Rapid identification of internal defects of pressure-bearing equipment in service:

[0039] (3.1) An organic mechanoluminescent coating is prepared on the surface of the pressure-bearing equipment, evenly applied to the surface of the equipment by a brush, and heated to form a film.

[0040] (3.2) The CCD fluorescence acquisition system is used to monitor the force-induced fluorescence signal on the surface of the pressure-bearing equipment in service in real time, and the position of the force-induced fluorescence signal is used to quickly locate the location of the defect.

[0041] (3.3) Combining computer vision and artificial intelligence technologies (such as machine learning or deep learning) to construct a residual attention network for the force-induced fluorescence images collected in the standard database, and combining the deep network with the attention mechanism to extract the deep features of the fluorescence images.

[0042] (3.4) According to the image processing technology, the eigenvalues ​​of the induced fluorescence image are extracted, and the feature representation of the fluorescence image is enhanced through feature fusion. The constructed residual attention classification network is trained and verified to obtain the training model.

[0043] (3.5) The force-induced fluorescence images collected in real time are quantitatively analyzed and the force-induced fluorescence image results are input into the training model. The classification model outputs the classification results of the fluorescence images and calculates the tensile load and depth level of the force-induced fluorescence images induced by the defects of the pressure-bearing equipment according to the pre-defined category labels.

[0044] The present invention is a detection method for dynamically and visually monitoring defects of pressure-bearing equipment in service based on organic mechanoluminescence, comprising selecting an organic mechanoluminescent material, preparing an organic mechanoluminescent material solution, and applying the mechanoluminescent material to the surface of the pressure-bearing equipment over a large area; monitoring the mechanoluminescence generated on the surface of the pressure-bearing equipment in service in real time by direct observation or by a CCD fluorescence acquisition system, and quickly locating the position of internal defects; and correlating the characteristic value of the mechanoluminescence image with the load and defect size by computer vision and artificial intelligence technology to establish a training model; and collecting and analyzing the organic mechanoluminescence image of the surface of a standard tensile specimen containing defects. Information, quantitatively analyze the effects of load, defect depth and size on the morphology and luminescence range of force-induced fluorescence images, and establish a defect-induced force-induced fluorescence image database; construct an organic force-induced fluorescence coating on the surface of pressure-bearing equipment on a large scale, monitor the force-induced fluorescence response of the pressure-bearing equipment surface online during service, indicate the location of the defect according to the location of the force-induced fluorescence, identify the force-induced fluorescence images in the database according to the force-induced fluorescence images and loads, train the data and establish a training model, and bring the real-time collected force-induced fluorescence images into the model to quantitatively evaluate the defect size and depth of the pressure-bearing equipment. This method is simple to operate and install, and the force-induced fluorescence is visible to the naked eye. It can locate and evaluate the internal or external defects of pressure-bearing equipment in service on a large scale and quickly.

[0045] The advantages of the present invention are:

[0046] (1) A method for rapid detection of defects in pressure-bearing equipment based on organic mechanofluorescence is developed. The organic mechanofluorescence response of the surface of a standard sample with defects after being subjected to stress is detected in real time. The characteristic values ​​of the fluorescence image are obtained through a series of image processing and analysis. The basic law between the area and distance of the mechanofluorescence image and the defect size and load size is obtained, and a mechanofluorescence database of defects in pressure-bearing equipment is established. By online monitoring of the mechanofluorescence information on the surface of pressure-bearing equipment in service, a detection method for rapid positioning and evaluation of the position and size of internal and external defects of pressure-bearing equipment in service is provided through computer vision and artificial intelligence technology.

[0047] (2) A method for rapidly detecting defects in pressure-bearing equipment using organic mechanoluminescent materials. Tetranitro-tetraphenylethylene can be excited under ultraviolet light after being deformed by force, and the mechanoluminescent material is visible to the naked eye. The response is long-lasting and can intuitively reflect the internal and external position and degree of damage over a large range. It is simple, easy, low-cost, non-toxic, and has a wide range of applications.

[0048] (3) The method of using organic mechanoluminescent materials to quickly detect defects in pressure-bearing equipment is not limited by the geometry of the mechanical structure and the complexity of the load. Through computer vision and artificial intelligence technology, the mechanoluminescent images generated on the surface of the pressure-bearing equipment in service can be analyzed to quantitatively obtain the size and depth of internal and external defects. This provides an important reference for the structural health monitoring of the pressure-bearing equipment in service. Therefore, it has more advantages than other methods for online monitoring of pressure-bearing equipment in service.

[0049] (4) A method for rapid detection of defects in pressure-bearing equipment using mechanically induced fluorescent materials was established to establish a defect-induced mechanically induced fluorescent image database for pressure-bearing equipment. This method can be connected to the collaborative management and control system of pressure-bearing equipment to achieve safety monitoring and early warning of the production process of pressure-bearing equipment through the Internet of Things. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 :Technical route flow chart.

[0051] Figure 2 A: Front view of the component with internal notch;

[0052] Figure 2 B: Defect view of the specimen and the shape of the prefabricated defect.

[0053] Figure 3 : Force-induced fluorescence images of the sample surface of standard tensile specimens with defects of different depths under tensile load.

[0054] Figure 4 A: Force-induced fluorescence image of the sample surface obtained by typical internal defect induction;

[0055] Figure 4 B: Grayscale distribution of force-induced fluorescence image;

[0056] Figure 4 C: Binary image of the force-induced fluorescence image.

[0057] Figure 5 : Characteristic images of force-induced fluorescence induced by internal defects at different depths.

[0058] Figure 6 : Mechanofluorescence image, grayscale distribution, DIC method and finite element simulation of von Mises equivalent strain distribution of horizontal back notched specimen under tensile load.

[0059] Figure 7 : Schematic diagram of the online detection system for pressure-bearing equipment. DETAILED DESCRIPTION

[0060] The specific implementation route of this method is as follows Figure 1 shown.

[0061] (1) Selecting the organic mechanoluminescent material tetranitro-tetraphenylethylene (TPE-4N) as the material; or synthesizing it by chemical method. The synthesis method is shown in the Chinese invention patent "Preparation and application of polynitro-substituted tetraphenylethylene compounds" (invention patent application number: CN 201310057959 publication number: CN104003886 A).

[0062] (2) Prepare TPE-4N solution, coat the surface of standard tensile specimens with defects and pressure-bearing equipment in service, and heat to form a film.

[0063] Sample 1: TPE-4N was dissolved in chloroform to prepare a TPE-4N solution with a concentration of 0.01 g / mL. The solution was applied to the surface of the pressure-bearing component with a brush and heated at 80°C for 30 minutes with a hot air gun to form a film.

[0064] Sample 2: TPE-4N was dissolved in chloroform to prepare a TPE-4N solution with a concentration of 0.03 g / mL. The solution was applied to the surface of the pressure-bearing component with a brush and heated at 150°C for 2 minutes with a hot air gun to form a film.

[0065] Sample 3: TPE-4N was dissolved in chloroform to prepare a TPE-4N solution with a concentration of 0.05 g / mL. The solution was applied to the surface of the pressure-bearing component with a brush and heated at 300°C for 1 minute with a hot air gun to form a film. The heating tool can be a hot air gun, a heating furnace, a heating jacket, etc.

[0066] (3) The results of the samples prepared above with TPE-4N film on the surface are consistent. The TPE-4N solution is stored in a dark and low temperature place.

[0067] (4) Design of defective standard tensile specimens. Defective standard specimens are used to simulate the damage of pressure-bearing equipment and design defective standard tensile specimens required for the pressure-bearing equipment defect-induced force-induced fluorescence image database. The internal defects include: circular holes and cracks with a defect depth range of 10% to 90% and a defect length of 1 to 30 mm, and a tensile load of 50 to 300 MPa. Preparation is done on the surface of defective standard tensile specimens.

[0068] (5) Acquisition of force-induced fluorescence images of defective standard specimens. The morphology of defective 316L stainless steel components is shown in Figure 2. Figure 2 As shown, Figure 2 A is the front view of the component. Figure 2 B is the morphology of the defect on the back. A uniaxial tensile test was carried out on a defective sample coated with TPE-4N. During the test, a CCD fluorescence acquisition system was used to record the force-induced fluorescence image of the TPE-4N coating on the sample surface in real time, and a series of force-induced fluorescence images of the defective sample after being subjected to force were obtained. Figure 3As shown, ultraviolet light source excitation and CCD camera are used to collect front-side force-induced fluorescence images of samples without notches on the back and defective samples with defect depths of 60%, 75%, and 90% under various loads.

[0069] (6) Force-induced fluorescence image processing. The recorded fluorescence image of the back-notched sample is processed and applied to analyze the characteristic values ​​of the force-induced fluorescence image on the sample surface. Figure 4 As shown in the figure, the surface fluorescence response image during deformation is selected, and the 24mm×4mm area on the front of the specimen except the fixture part is taken as the analysis object, and it is converted into a grayscale image using Matlab software. The force-induced fluorescence ( Figure 4 A) Graphic processing to obtain the grayscale distribution of mechanoluminescence ( Figure 4 B). Add salt and pepper noise to the grayscale image, then perform median filtering on the image to minimize the impact of the noise and smooth it. Select an appropriate threshold (0.01-0.1) for the processed image and binarize it. After binarization, delete the part with an area less than 30 ( Figure 4 C). The binary connected regions are then marked, the area of ​​the connected regions is calculated, and the proportion of the area occupied by the force-induced fluorescence is calculated.

[0070] (7) Establishment of a database of force-induced fluorescence images of pressure-bearing equipment defects. By analyzing the characteristic values ​​of force-induced fluorescence images of different defective specimens under load, the corresponding force-induced fluorescence area and distance are obtained. Figure 5 As shown in Figure 1, the force-induced fluorescence response area changes with the depth of the defect on the back of the specimen and the tensile load. As shown in Table 1, as the tensile load continues to increase, the proportion of the force-induced fluorescence luminescence area gradually increases. When the tensile load is fixed, the increase in defect depth will lead to an increase in the force-induced fluorescence luminescence area. A series of characteristic values ​​such as area and distance of the force-induced fluorescence image under the action of defect size and tensile load are obtained, and a force-induced fluorescence image database induced by defects of pressure-bearing equipment is established.

[0071] Table 1 Relationship between the characteristic values ​​of force-induced fluorescence image and load and defect size

[0072]

[0073] (8) Verification of the results of defect-induced force-induced fluorescence detection of pressure-bearing equipment. In order to verify the accuracy of this method, the strain distribution of the same defective 316L stainless steel specimen under tensile load was analyzed using the digital image correlation (DIC) method and the finite element simulation method. Figure 6 As shown in the figure, due to the existence of defects, local strain is generated on the surface of the specimen when subjected to tensile load, and an X-shaped strain distribution cloud diagram is formed. Figure 6The results of three methods of testing a sample with 75% depth horizontal defects during the tensile process are listed in the paper. The results obtained by the three methods are compared. The force-induced fluorescence image obtained by the organic force-induced fluorescence detection method is basically consistent with the strain distribution analyzed by DIC and ANSYS. The results show that it is feasible to use organic force-induced fluorescence to characterize the local deformation of the surface caused by defects.

[0074] (9) A method for quickly locating defects in pressure-bearing equipment in service. Figure 7 As shown in the figure, a TPE-4N coating is constructed on the outer surface of the pressure-bearing equipment in a large area. Under the excitation of ultraviolet light, a CCD camera is used to quickly detect the surface of the pressure-bearing equipment in service in a large area, and the force-induced fluorescence image of the equipment surface is obtained in real time and transmitted back to the damage assessment system. The location of the internal or external defects can be quickly located by using the location where the force-induced fluorescence appears.

[0075] (10) Force-induced fluorescence image recognition based on computer vision and artificial intelligence technology. The problem of predicting the depth level of tensile loads in a known force-induced fluorescence image database is transformed into an image classification problem containing 12 categories by using a deep neural network method. The initial data includes fluorescence images used for training, verification and testing, binary images of damage, and label information of the corresponding tensile load size and depth level. The category labels of the tensile load size are {1800, 2000, 2300, 26000}, and the category labels of the depth level are {60%, 75%, 90%}. For simple processing, {1800, 60%}, {1800, 75%}, {1800, 90%}, {2000, 60%}, {2000, 75%}, {2000, 90%}, {2300, 60%}, {2300, 75%}, {2300, 90%}, {2600, 60%}, {2600, 75%}, and {2600, 90%} are represented by 1 to 12 respectively and used as the category labels of each fluorescence image.

[0076] Constructing a residual attention network: Based on the idea of ​​residual learning and attention mechanism, a 56-layer residual module combined with a deep network of attention mechanism is constructed to extract deep features of fluorescence images. The feature extraction branch selects a residual module of a neural network pre-trained in large-scale image data as the basic network unit of this branch to extract deep features of the input image. Among them, each residual unit consists of two 1×1 convolutions and one 3×3 convolution with batch normalization and rectified linear units and residual connections. Assume that the input fluorescence image is x, and the residual unit is The network parameter in the residual unit is W. By adding the output information calculated by the cascaded convolutional layer and the input information transmitted by the residual connection, the deep features of the fluorescence image can be calculated. The formula is as follows:

[0077]

[0078] The attention branch adopts a simple structure similar to a fully convolutional network. It is mainly composed of multiple basic units consisting of a maximum pooling layer and a residual unit, a linear interpolation, a 1×1 convolution, and a sigmoid layer. In this way, it is possible to increase the receptive field while performing feature extraction and quickly obtain the global feature information of the image. For the input fluorescence image x, the deep feature output of the feature extraction branch is F d , the attention weight map output by the attention branch is Similar to the residual learning method, the attention weight map obtained With the deep feature F d Perform pixel-level addition processing to calculate the deep features of the enhanced fluorescence image. As a feature selector, F d The effective feature enhancement in is described as follows:

[0079]

[0080] According to the image processing technology, the area of ​​the white area and the maximum coverage in the x-axis and y-axis directions of each binary image produced by the damage in the standard database (training set, validation set, test set) are calculated, and these two image characteristics are used as the manual features H of each fluorescence image. t In order to further improve the classification performance of the deep residual attention network, the deep feature F d and handcrafted visual features H t Fusion is performed to enhance the feature representation of the fluorescence image. First, for the deep feature F extracted in step 2) d , send it to an average pooling layer to adjust the deep feature dimension, keep the deep classification network with rotation invariance and translation invariance when classifying images, and calculate the new feature F d ′; Secondly, the feature F d ′ and handcrafted features H t Connect and add to the linear layer L of the classification network; finally, the prediction probability p calculated by the penultimate layer is sent to the sigmoid layer to be normalized to [0, 1], and the category c with the largest prediction probability is selected as the category label of the input fluorescence image, and the tensile load and depth level of the fluorescence image can be obtained. The specific formula description of feature fusion is as follows:

[0081]

[0082] p=L(F d ′,H t ,θ)

[0083]

[0084] In the above formula, R represents the characteristic F d h×w area, θ represents the network parameters of the linear layer L, and argmax represents the maximum value index function.

[0085] (11) Perform data training and verification. According to the pre-prepared data, the constructed residual attention classification network is trained and verified. Use the small batch training mode and use the average value of the cross entropy as the loss function to measure the difference between the classification result and the actual category result, optimize the residual attention network until the network converges and stop training. The formula of the cross entropy loss function is as follows:

[0086]

[0087] In the above formula, L(·) represents the loss function value, n represents the number of samples participating in the training in the batch, n is 12, x represents the data matrix participating in the training in the batch, ∑ is the summation operator, y′ represents the actual category label matrix corresponding to x, log(·) represents the logarithmic operation, and y represents the category label matrix of x obtained by network classification.

[0088] When the residual attention network has learned all the images in the training set, the classification model will switch from the training model to the evaluation mode, predicting and classifying all the images in the validation set, and evaluating the performance of the currently trained model based on the classification index calculated from the prediction results and the true labels of the images. Finally, the model with the highest evaluation index of the validation set will be saved.

[0089] (12) Quantitative assessment of defects in pressure-bearing equipment in service. The feature values ​​of the force-induced fluorescence image of the pressure-bearing equipment surface obtained in real time due to damage are input into the saved model. The classification model will output the classification results of the fluorescence image and calculate the quantitative assessment of the tensile load and depth level of the fluorescence image according to the pre-defined category label representation.

[0090] The present invention is a rapid detection method for defects of pressure-bearing equipment. An organic mechanoluminescent material is selected to prepare a solution, and an organic mechanoluminescent coating is constructed on the surface of a component. Through a tensile test of a standard sample containing defects, mechanoluminescent images under different loads and different defect depths and sizes are obtained respectively. The relationship between the characteristic value of the mechanoluminescent region and the load and defect size is obtained through image processing, and an image database of mechanoluminescent images induced by defects of pressure-bearing equipment is established. An organic mechanoluminescent coating is applied to the surface of the pressure-bearing equipment, and the response of the mechanoluminescence on the surface of the pressure-bearing equipment is monitored online during service. The position of the internal defect is quickly located using the position where the mechanoluminescence appears; the mechanoluminescent images in the database are identified, data trained, and a training model is established using computer vision and artificial intelligence technology; the mechanoluminescent images collected in real time are brought into the model to quantitatively evaluate the defect size and depth of the pressure-bearing equipment. The method is simple in operation and device, and the detection results of equipment defects are visible to the naked eye. It can quickly screen defects generated on the surface and content of pressure-bearing equipment in service and carry out damage assessment on a large scale, and has important reference value for structural health monitoring and maintenance plan formulation of pressure-bearing equipment.

[0091] The method for visually monitoring defects of pressure-bearing equipment in service based on organic mechanoluminescence disclosed and proposed by the present invention can be realized by those skilled in the art by referring to the content of this article and appropriately changing the conditions, routes and other links. Although the method and monitoring technology of the present invention have been described through preferred implementation examples, relevant technicians can obviously modify or recombine the methods and technical routes described in this article without departing from the content, spirit and scope of the present invention to realize the final monitoring technology. It is particularly important to point out that all similar substitutions and modifications are obvious to those skilled in the art, and they are all considered to be included in the spirit, scope and content of the present invention.

Claims

1. A method for quickly detecting defects of pressure-bearing equipment in service using organic mechanoluminescent materials, comprising the following steps: (1) Establishment of a database of pressure-bearing equipment defect-induced force-induced fluorescence images: (1.1) Prepare standard tensile specimens with internal or external defects of different sizes and depths; (1.2) selecting an organic mechanoluminescent material as a material and preparing an organic mechanoluminescent material solution; using a brush to evenly apply the organic mechanoluminescent material solution on the front and back surfaces of a standard tensile specimen containing defects, and heating the solution with a heating tool to form a film, thereby preparing a uniform organic mechanoluminescent coating; (1.3) A tensile test was conducted on a defective standard tensile specimen coated with organic mechanoluminescent material. An ultraviolet light source was used as the excitation light source. A CCD fluorescence acquisition system was used to record in real time the mechanoluminescent images of both sides of the specimen under different loads, defect lengths and depths to simulate the mechanoluminescent response caused by surface defects and internal defects of pressure-bearing equipment. (1.4) Processing of the mechanofluorescence image: The mechanofluorescence image is processed to obtain the characteristic value of the mechanofluorescence image. The mechanofluorescence image area is extracted through noise reduction and binarization processing, and the area and distance of the fluorescence in the binary image are calculated; (1.5) Establish a defect-induced force-induced fluorescence image database for pressure-bearing equipment: As the load and the length and depth of the defect increase, the area and distance of the force-induced fluorescence image on the sample surface gradually increase; Analyze the results obtained in the test (1.3) in detail, establish the relationship between the area and distance of the force-induced fluorescence image under different loads and defect sizes, and establish a defect-induced force-induced fluorescence image database; (2) Online monitoring method for defects of pressure-bearing equipment in service: (2.1) Selecting an organic mechanoluminescent material as a material and preparing an organic mechanoluminescent material solution; using a brush to evenly apply the organic mechanoluminescent material solution on the surface of the pressure-bearing equipment, and heating it with a heating tool to form a film, thereby preparing a uniform organic mechanoluminescent coating; (2.2) Use the CCD fluorescence acquisition system to scan and monitor the force-induced fluorescence signal on the surface of the pressure-bearing equipment in service in real time, and quickly locate the internal or external defects of the pressure-bearing equipment according to the location of the force-induced fluorescence; (2.3) Combining computer vision and artificial intelligence technology to build a residual attention network for the force-induced fluorescence images collected from the standard database, and extracting deep features of the fluorescence images with the deep network of the attention mechanism; (2.4) Extracting the characteristic values ​​of the fluorescence image based on image processing technology and enhancing the characteristic representation of the fluorescence image through feature fusion; Train and verify the constructed residual attention classification network to obtain the training model; (2.5) The force-induced fluorescence image results of the pressure-bearing equipment surface collected in real time online are input into the training model. The classification model will output the classification results of the fluorescence image and calculate the load and size of the force-induced fluorescence image induced by the defects of the pressure-bearing equipment according to the pre-defined category labels.

2. The method according to claim 1, characterized in that: The organic mechanoluminescent material is tetranitro-tetraphenylethylene material.

3. The method according to claim 1, characterized in that: The concentration of the organic fluorescent material solution is 0.01-0.05 g / mL; the organic solvent is chloroform.

4. The method according to claim 1, characterized in that: The heating tool uses a hot air gun, a heating furnace or a heating jacket; the heating temperature is 80°C to 300°C, and the heating time is 1 to 30 minutes.

5. The method according to claim 1, characterized in that: The CCD fluorescence acquisition system acquires in real time the force-induced fluorescence images of defective samples and the surfaces of pressure-bearing equipment in service after being subjected to stress.

6. The method according to claim 1, characterized in that: The CCD fluorescence collection system includes a CCD camera, a lens, an ultraviolet light source and a computer data analysis system. The fluorescence collection system is installed outside the standard sample and the service pressure container, and the collected force-induced fluorescence characteristic image is transmitted back to the computer data analysis system in real time.

7. The method according to claim 1, characterized in that: The method for processing the force-induced fluorescence image is to process the recorded force-induced fluorescence image of the defective sample and apply it to analyze fluorescence; The force-induced fluorescence image of the standard specimen surface during the deformation process was selected, and a 1-50 mm × 1-50 mm area was selected as the analysis object. It was converted into a grayscale image through image processing, and then the image was subjected to denoising, median filtering, smoothing and binarization to obtain the force-induced fluorescence area, and the area and distance occupied by the force-induced fluorescence area in the selected area were calculated.

8. The method according to claim 1, characterized in that: The defect-induced force-induced fluorescence image database is established by carrying out tensile tests on standard tensile specimens containing internal and external defects of different sizes and depths to obtain force-induced fluorescence images; The relationship between the area and distance of mechanofluorescence and the change of defect size, depth and load is established. Based on the test results of mechanofluorescence response of a series of defective specimens under different tensile loads, a defect-induced mechanofluorescence image database of pressure-bearing equipment is established.

9. The method according to claim 1, characterized in that: The rapid defect identification of the pressure-bearing equipment in service is to apply an organic mechanoluminescent coating on the surface of the pressure-bearing equipment, use a CCD fluorescence acquisition system to detect the response area of ​​mechanoluminescence online, and use the appearance position of mechanoluminescence on the surface of the component to quickly locate the location of the defect inside the component.

10. The method according to claim 1, characterized in that: The quantitative evaluation of defects of the in-service pressure-bearing equipment is to combine computer vision and artificial intelligence technology to perform characteristic value analysis on the images in the database and establish a training model; the characteristic values ​​of the force-induced fluorescence images of the surface of the pressure-bearing equipment collected in real time online are input into the training model to quantitatively obtain the depth and length of the defects of the in-service pressure-bearing equipment.

Images