Infrared thermal imaging composite material defect detection method based on slow feature analysis

Abstract

The invention discloses an infrared thermal imaging composite material defect detection method based on slow feature analysis. The method comprises the following steps: 1) data acquisition: acquiring thermal image data of internal defects of a polymer composite material by using an infrared thermal imager; 2) data preprocessing: preprocessing the original thermal image data acquired by the sensor; 3) constructing an SFT model: establishing a slow feature thermal image data analysis model; 4) analyzing an SFT model result: visualizing a corresponding vector in the slow feature; and 5) quantitatively evaluating the effect of the SFT model: calculating a signal-to-noise ratio index. According to the method, thermal imaging data is understood as a slow feature and fast feature mixed time-varying signal, and the slow feature is extracted to distinguish defect and background information, so that accurate detection of the internal defect of the polymer composite material is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of infrared thermal imaging nondestructive testing of composite material defects, and in particular to an infrared thermal imaging composite material defect detection method based on slow feature analysis. Background Art

[0002] Polymer composites have become an important material for manufacturing products due to their good properties. Whether in the defense industry such as aerospace and shipbuilding, or in civilian life such as electronic energy, sports and leisure, composite materials have indispensable application value. However, in the manufacturing process of composite materials, various types of invisible defects such as fiber breakage, voids, delamination, and interface peeling will appear. These defects will have a negative impact on the quality of composite products and even cause product failures and endanger life safety. Therefore, accurate evaluation of composite quality has become an important and very necessary task.

[0003] Non-destructive Testing (NDT) technology has become the preferred choice for composite defect detection due to its non-destructive and comprehensive detection advantages. Commonly used NDT methods include infrared thermography (IRT), ultrasonic testing, resistance testing and ray backscattering testing. Among them, IRT is popular because of its low cost, easy operation and wide scanning range. IRT is a technology that uses an infrared camera to obtain the surface temperature of the target sample, and then constructs the corresponding temperature curve into a thermal image for further analysis. IRT can be divided into active or passive. Passive infrared thermography (PIRT) monitors the temperature generated by internal heat sources. However, there is no internal heat source in the specimens manufactured in industry. Therefore, active infrared thermography (AIRT) must be used for defect detection of composite materials.

[0004] In recent years, IRT has attracted much attention due to its low cost, easy operation and fast scanning speed. To apply IRT, a photographic flash is usually used to generate a short energy pulse to heat the surface of the target object in a short time, and then the heat flux diffuses into the material, and an infrared thermal imager is used to record the surface temperature response as a time series of thermal images. However, it is usually difficult to identify defects in thermal images by visual inspection alone, mainly because thermal images usually contain measurement noise and uneven background. In addition, high-frequency infrared cameras often capture a large number of thermal images during the test process, and visual inspection one by one is laborious and time-consuming. Therefore, it is very necessary to adopt effective thermal imaging data processing methods to improve the results of thermal imaging defect analysis.

[0005] At present, many common thermal imaging data processing technologies have been developed, such as principal component analysis (PCT) and independent component analysis (ICT). Among them, PCT uses linear projection to expect the data variance to be the largest in the projection dimension to achieve dimensionality reduction and maximize the retention of the characteristics of the original data structure, thereby realizing defect recognition in thermal images. ICT uses the ICA algorithm to treat thermal imaging data as mixed signals from multiple sources, such as materials, defects, noise, etc., and each source is independent of each other. ICA is used to separate signals from various sources. After decomposition, defect information may be highlighted in a small number of images. Previous studies have shown that these methods analyze thermal imaging data from different angles based on their own algorithmic principles to achieve the purpose of removing uneven background, reducing noise and extracting features. However, these defect assessment methods are still very limited in the extraction of defect information, and problems such as incomplete and unfocused defect detection have always existed, resulting in the recognition of defects that are not accurate, fast, and efficient, and the level of defect detection still has a lot of room for improvement.

[0006] Slow feature analysis (SFA) is a linear factor model that uses the invariance of time signals to discover information from time signals. This invariance of time-varying signals helps to analyze and classify data. The goal of SFA is to identify the slowest time-varying signal, that is, the slow feature. When using AIRT for defect detection, due to the presence of defects in the sample being tested, the temperature distribution of the defective part at different times is different from that of the intact part. Therefore, the rate at which the temperature of the defective area changes over time will be "faster" (for heat-absorbing defects) or "slower" (for insulation defects) than that of the intact area, while the rate at which the temperature of the intact area changes over time is relatively slow. Here, the slow characteristics of the defective area and the intact area are different. Usually, the defective area is "faster" and the intact area is "slower". In this way, SFA can distinguish the background from the defect while identifying the slow features, thereby improving the visibility of the defect. It is understood that no one has used SFA for thermal imaging data analysis, so this paper proposes a thermal imaging data analysis method based on slow feature analysis (Slow featurethermography, SFT). Summary of the invention

[0007] Aiming at the difficulty in extracting features from high-dimensional thermal imaging data during infrared thermal imaging detection of composite material defects, and the fact that thermal imaging sequences have time attributes, the present invention proposes an infrared thermal imaging composite material defect detection method based on slow feature analysis; thermal imaging data is understood as a time-varying signal mixed with "slow" features and "fast" features, and slow features are extracted to distinguish defects from background information, so as to achieve accurate detection of internal defects in polymer composite materials.

[0008] The technical solution adopted by the present invention to solve its technical problem is:

[0009] A method for detecting composite material defects by infrared thermal imaging based on slow feature analysis comprises the following steps:

[0010] 1) Data acquisition: Use infrared thermal imager to obtain thermal imaging data of internal defects of polymer composite materials;

[0011] The thermal imaging data acquisition process includes two processes: pulse heating of the sample and capturing of the thermal image of the sample surface;

[0012] First, the sample is heated using a heat pulse provided by a flash lamp that can deliver high energy; then, during the sample cooling phase, an infrared camera is used to capture a thermal image of the object surface using a certain resolution and sampling rate.

[0013] 2) Data preprocessing: preprocess the collected raw thermal image data;

[0014] The quality of thermal image data plays a decisive role in the defect detection performance of the slow feature analysis method. Data preprocessing includes two parts: first, the 3-D thermal image data matrix is ​​expanded into a 2-D matrix that can be used for SFT analysis; second, thermal imaging data contains a lot of noise information, and the SFT method is very sensitive to noise, so the uniform distribution technology is used to preprocess the thermal image data to make the thermal image data distribution similar and reduce the influence of outliers.

[0015] 3) Construct a slow feature thermal imaging model: calculate the weight matrix.

[0016] The thermal image data after data preprocessing is represented by X, x i Represents each thermal image. Based on the SFA algorithm, first, the input data X is normalized, and the normalized matrix has a mean of 0 and a variance of 1; secondly, the normalized data is nonlinearly expanded, and the expanded matrix is ​​z; thirdly, the whitening matrix S of the nonlinear expansion matrix z is calculated. Finally, the J eigenvectors with the smallest derivative variance of z are found, and the eigenvectors are arranged from small to large. The optimal weight vector that meets this condition returns the weight matrix W; at this time, J slow features are obtained.

[0017] 4) Result analysis and visualization: Result visualization of the slow feature thermal imaging model.

[0018] To visualize the defects, resize the J slow features to n x ×n y Then, J thermal images that can identify defects are obtained. In the process of analyzing thermal imaging data by this method, in order to achieve data compression, the dimension of J is usually selected to be smaller than X.

[0019] By observing the reshaped thermal image, judging the shape, position, size and contrast of the defect, observing the effect of noise removal and background separation, and comprehensively analyzing the results, we can preliminarily determine the effect of the SFT model on the internal defect detection of polymer composite materials. If the visual effect is too poor, it is necessary to re-select the parameters to establish the SFT model.

[0020] 5) Quantitative evaluation of model effect: calculation of signal-to-noise ratio index.

[0021] In order to evaluate the performance of the model more objectively, the signal-to-noise ratio (SNR) index, which is widely used in the comparison of thermal imaging data processing methods, is calculated. The formula of SNR is as follows:

[0022]

[0023] Where M def and M in are the average pixel values ​​of defective and intact areas, σ in is the standard deviation of the pixel values ​​of the intact area. The SNR index is dimensionless and reflects the thermal contrast between defective and non-defective areas. Therefore, a large SNR value indicates better defect recognition capability.

[0024] Further, the process of step (2) is:

[0025] Step 2.1: Data expansion

[0026] The thermal imaging dataset recorded by the data acquisition can be viewed as a three-dimensional (3-D) matrix containing a series of thermal images collected at different sampling time instances. During the cooling phase of the test, n t Frame thermal image, each frame consists of n x ×n y The color of each pixel represents the surface temperature at the corresponding position. Since SFT cannot be directly applied to the three-dimensional thermal image data matrix, the three-dimensional (3-D) matrix is ​​converted into a two-dimensional (2-D) expansion matrix. In the process, each frame of thermal image is expanded into a length of n x n y Column vector x i (i=1,...,nt ). i Each element of corresponds to a pixel of the original image.

[0027] Step 2.2: Even distribution

[0028] Considering that thermal image data is affected by factors such as uneven heating during the acquisition process, the collected original thermal images contain a large amount of uneven background and noise, which covers up the effective defect features in the image, resulting in the thermal image analysis method not being ideal for defect detection. The SFA algorithm is very sensitive to noise and requires the existence of static points in the thermal image sequence. Therefore, before applying the SFT algorithm (the basic algorithm of SFT is SFA), the thermal image data expansion matrix should be preprocessed for uniform distribution, and the cumulative distribution function of the feature should be found through nonlinear transformation, and the quantile transformation should be used to place all features in the same uniform distribution, so that the thermal image data distribution is similar and the influence of outliers is reduced.

[0029] Furthermore, the process of step (3) is as follows:

[0030] Step 3.1: Normalize the data

[0031] The SFT model uses pixels as samples, and the number of samples is equal to the number of pixels n. x ×n y , the sample dimension is the number of thermal image pictures n t ; The input data is a two-dimensional (2-D) expansion matrix after uniform distribution preprocessing. The expansion matrix is ​​normalized. The normalized matrix has a mean of 0 and a variance of 1. express.

[0032] Step 3.2: Nonlinear expansion

[0033] Find the nonlinear expansion function h(x) for Perform nonlinear expansion, and the nonlinear expansion matrix satisfies z=h(X)=(x1,...,x I ,x1x1,x1x2,...,x I x I ) T , calculate the whitening matrix S of z.

[0034] Step 3.3: Derivative of the nonlinear expansion matrix

[0035] Derivative of the nonlinear expansion matrix Find out The J vectors whose derivative variance is the smallest.

[0036] Step 3.4: Return weights

[0037] Arrange the J feature vectors from small to large, and return the optimal weight vector that meets this condition to the weight matrix W; SFA calculates J slow features, which meet the constraints of SFA, namely zero mean, mean square error, and independence. Resize the J slow features to a size of n x ×n y Then J thermal images that can identify defects are obtained.

[0038] The beneficial effects of the present invention are mainly manifested in: the present invention proposes a thermal imaging data analysis method based on slow feature analysis, which understands thermal imaging data as a time-varying signal mixed with "slow" features and "fast" features, extracts slow features to distinguish defects from background information, and realizes accurate detection of internal defects of polymer composite materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is a flow chart of the method of the present invention;

[0040] Figure 2 It is a schematic diagram of 3D thermal imaging data collected by the present invention;

[0041] Figure 3 It is a schematic diagram of the expansion of 3D thermal imaging data into a 2D data matrix of the present invention;

[0042] Figure 4 is an illustration of the slow feature thermal image visualization proposed by the present invention;

[0043] Figure 5 is a schematic diagram of sample defects;

[0044] Figure 6 is the original thermal image;

[0045] Figure 7 It is the SFT result graph;

[0046] Figure 8 It is the PCT result graph;

[0047] Fig. 9 It is the ICT result graph;

[0048] Fig.10 This is a comparison chart of SNR of three methods. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments of the specification. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0050] On the contrary, the present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention as defined by the claims. Further, in order to make the public have a better understanding of the present invention, some specific details are described in detail in the detailed description of the present invention below. Those skilled in the art can fully understand the present invention without the description of these details.

[0051] See attached Figures 1 to 10 A method for detecting composite material defects using infrared thermal imaging based on slow feature analysis is presented. Taking the detection of internal defects of a carbon fiber reinforced polymer (CFRP) sample manufactured in a laboratory as an example, the method comprises the following steps:

[0052] Step 1) Data acquisition: using an infrared thermal imager to obtain thermal imaging data of internal defects of polymer composite materials.

[0053] In order to effectively demonstrate the effect of the SFT model proposed in the present invention, a laboratory-made CFRP sample was used as a defect detection object. During the manufacturing process of the CFRP specimen, three flat-arranged Teflon strips with different shapes were inserted into the carbon fiber plate during the lamination step to purposefully generate defects of different positions, shapes, and depths. The shallowest defect is located at the lower right of the specimen and is generated by inserting a trapezoidal Teflon tape under a single layer of carbon fiber plate with a depth of approximately 0.05 cm. In the middle part of the specimen, there is a circular defect under two layers of carbon fiber plates, and its depth is approximately 0.10 cm. The deepest defect is located in the upper left corner, which is a diamond-shaped defect covered by three layers of carbon fiber with a depth of 0.15 cm. The size of each defect area is approximately 3 square centimeters. After the epoxy resin is injected, these defects are invisible. The defect information is shown as follows Figure 5 As shown, Figure 5 The horizontal and vertical coordinates are both pixel values.

[0054] Pulse heating: The CFRP specimens were heated using a heat pulse provided by a flash lamp with a transfer energy of 3200J within 3ms. Capture of thermal images of the specimen surface: During the cooling phase of the specimen, the thermal image of the object surface was captured by an infrared camera (TAS-G100EXD, NEC model), where the sensor installed in the camera was an uncooled microbolometer focal plane array. The experimental layout was set up in reflection mode. The resolution and sampling rate used in the experiment were 320×240 pixels and 30 frames / second, respectively. A total of 90 thermal images were captured, some of which are shown in Figure 2. Figure 6 shown.

[0055] Step 2) Data preprocessing: Preprocess the raw thermal image data collected by the sensor. The quality of thermal image data plays a decisive role in the defect detection performance of the slow feature model. Data preprocessing mainly includes the following three parts:

[0056] a) Image valid area selection:

[0057] The invalid edges of the original thermal image collected initially are removed, and the remaining area is used as the region of interest (ROI), and the ROI contains 308×212 pixels.

[0058] b) Data expansion:

[0059] Each frame of thermal imaging image in the 3-D matrix is ​​expanded into a column vector x of length 65296 in the 2-D matrix i (i=1,...,n t ), each x i An element of corresponds to a pixel of the original image.

[0060] c) Homogenization treatment:

[0061] The homogenization method is to find the cumulative distribution function of the feature through nonlinear transformation and use quantile transformation to put all features into the same uniform distribution.

[0062] Step 3) Construct SFT model: Construct the SFT model of slow feature thermal image. The specific steps are as follows:

[0063] Step 3.1: Data transformation and dimension definition

[0064] The two-dimensional matrix X after data preprocessing is as follows: Figure 3 As shown, the pixels are taken as samples. At this time, the number of samples is equal to the number of pixels 65296, and the sample dimension is equal to the number of thermal images taken 90.

[0065] Step 3.2: Normalize the data

[0066] The SFT model uses pixels as samples, and the number of samples is equal to the number of pixels n. x ×n y , the sample dimension is the number of thermal image pictures n t ; The input data is a two-dimensional (2-D) expansion matrix after uniform distribution preprocessing. The expansion matrix is ​​normalized. The normalized matrix has a mean of 0 and a variance of 1. express.

[0067] Step 3.3: Nonlinear expansion

[0068] Find the nonlinear expansion function h(x) for Perform nonlinear expansion, and the nonlinear expansion matrix satisfies z=h(X)=(x1,...,x I ,x1x1,x1x2,...,x I x I ) T , calculate the whitening matrix S of z.

[0069] Step 3.4: Derivative of the nonlinear expansion matrix

[0070] Derivative of the nonlinear expansion matrix Find out The J vectors whose derivative variance is the smallest.

[0071] Step 3.5: Return weights

[0072] Arrange the J feature vectors from small to large, and return the optimal weight vector that meets this condition to the weight matrix W. SFT calculates J slow features, which meet the constraints of SFA, namely zero mean, mean square error, and independence. Resize the J slow features to a size of n x ×n y Then J thermal images that can identify defects are obtained.

[0073] The slow features are reduced to 6 dimensions (ie J = 6).

[0074] (4) SFT model result analysis: slow feature visualization

[0075] The slow feature is the representation of the original data matrix X in the feature space. In order to visualize the defects, each column vector in the slow feature (slow feature, abbreviated as SF) is resized into a two-dimensional matrix of size 308×212, and then 6 thermal images SF1-SF6 that can identify defects are obtained, as shown in Figure 7 The results of PCT and ICT are shown in Figure 8 , Fig. 9 shown.

[0076] Analysis shows that Figure 7 SF4 highlights the information of shallowly buried trapezoidal and circular defects in the test, with clear outlines, sharp edges, and good defect display effects. It removes noise to a certain extent and separates the uneven background caused by uneven heating from the defect features, making the defect information easy to identify. SFT has better detection effects on circular and trapezoidal defects, which are stronger than PCT and ICT; the detection effect of diamond defects is poor. Figure 7 The location and shape information of defects in artificial defective samples can be clearly detected, which shows that the SFT model proposed in this paper has a certain effect on defect detection of polymer composite materials and is a good thermal imaging data processing model.

[0077] (5) Quantitative evaluation of the SFT model effect: Establishing a slow feature thermal image analysis model

[0078] In order to more objectively evaluate the performance of the SFT model, this method is compared with the existing polymer composite thermal imaging analysis methods PCT and ICT, using SNR as the evaluation index. Fig.10 shown.

[0079] From the results, it can be seen that the SFT method has certain advantages over the other two methods in the defect assessment of polymer composite materials, which is reflected in the detection ability of circular defects and trapezoidal defects, and the detection ability of diamond defects is weaker than the other two methods. In general, the proposed SFT method has a higher overall SNR index and has better comprehensive defect detection performance.

[0080] The present invention proposes an infrared thermal imaging composite material defect detection method based on slow feature analysis, which can effectively identify internal defects of polymer composite materials. At the same time, it provides a demonstration for thermal imaging data analysis, which can be applied to analyze different types of IRT heating programs and thermal imaging data recorded by different types of test materials.

[0081] Abbreviations, terms and explanations:

[0082]

[0083]

[0084] The contents described in the embodiments of this specification are merely an enumeration of the implementation forms of the inventive concept. The protection scope of the present invention should not be regarded as limited to the specific forms described in the embodiments. The protection scope of the present invention also extends to equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A method for detecting composite material defects by infrared thermal imaging based on slow feature analysis, characterized in that: The following steps are involved: Step 1) Data acquisition: using an infrared thermal imager to obtain thermal imaging data of internal defects of polymer composite materials; Step 2) Data preprocessing: preprocessing the raw thermal image data collected by the sensor; Step 3) Constructing a slow feature thermal image SFT model: constructing a slow feature thermal image SFT model based on the SFA algorithm; Step 4) SFT model result analysis: observe the effect of the SFT model on the internal defect detection of polymer composites; Step 5) Quantitative evaluation of the effect of the SFT model.

2. The method for composite material defect detection based on infrared thermal imaging and slow feature analysis according to claim 1, characterized in that: The process of step 1) is: The data acquisition process of thermal imaging data includes two processes: pulse heating of the sample and capturing the thermal image of the sample surface. First, the sample is heated by the heat pulse provided by the flash lamp. Then, during the cooling stage of the sample, the thermal image of the object surface is captured by the infrared camera using the set resolution and sampling rate.

3. The method for composite material defect detection based on infrared thermal imaging and slow feature analysis according to claim 1, characterized in that: The process of step 2) is: Step 2.1: Data expansion: The thermal image data set recorded by data acquisition is regarded as a three-dimensional 3-D matrix, which is expanded into a 2-D matrix that can be used for SFT analysis. That is, X belongs to the real number space Each frame of thermal image is expanded into a length of n x n y Column vector x i (i=1,...,n t ), x i Each element of corresponds to a pixel of the original image; Step 2.2: Uniform distribution processing: The 2-D matrix is ​​preprocessed for uniform distribution. The cumulative distribution function of the feature is found through nonlinear transformation, and the quantile transformation is used to put all features in the same uniform distribution, so that the distribution of thermal imaging data is similar and the influence of outliers is reduced.

4. The method for composite material defect detection based on infrared thermal imaging and slow feature analysis according to claim 1, characterized in that: The process of step 3) is: Step 3.1: Normalize the data: The SFT model uses pixels as samples, and the number of samples is equal to the number of pixels n. x ×n y , the sample dimension is the number of thermal image pictures n t ; The input data is a two-dimensional 2-D expansion matrix after uniform distribution preprocessing. The 2-D expansion matrix is ​​normalized. The normalized matrix has a mean of 0 and a variance of 1. express; Step 3.2: Nonlinear expansion: Find the nonlinear expansion function h(x) for Perform nonlinear expansion, and the nonlinear expansion matrix satisfies z=h(X)=(x1,...,x I ,x1x1,x1x2,...,x I x I ) T , calculate the whitening matrix S of z; Step 3.3: Derivative of the nonlinear expansion matrix: Derivative of the nonlinear expansion matrix Find out The direction of J where the variance of the derivative is the smallest; Step 3.4: Return weights: Arrange the J eigenvectors from small to large, and return the optimal weight vector that meets this condition to the weight matrix W.

5. The method for detecting composite material defects by infrared thermal imaging based on slow feature analysis according to claim 4, characterized in that: The process of step 4) is: Resize J slow features to size n x ×n y Then, J reshaped thermal images capable of identifying defects are obtained; By observing the reconstructed thermal images and judging the shape, location, size and contrast of the defects, as well as the effects of noise removal and background separation, the effect of the SFT model on the internal defect detection of polymer composites was preliminarily determined.

6. The method for composite material defect detection based on infrared thermal imaging and slow feature analysis according to claim 1, characterized in that: The process of step 5) is: Calculate the signal-to-noise ratio index: Use the signal-to-noise ratio (SNR) index to objectively evaluate the performance of the model. The formula for SNR is as follows: Where M def and M in are the average pixel values ​​of defective and intact areas, σ in It is the standard deviation of the pixel values ​​in the complete area; the larger the SNR value, the stronger the defect recognition ability.

Images