Cold welding defect detection method and system based on multi-source data fusion

Abstract

The present application relates to the technical field of cold welding defect recognition, and discloses a cold welding defect detection method and system based on multi-source data fusion, comprising: synchronously collecting acoustic emission signals, infrared thermal imaging data, laser scanning geometric data and embedded mechanical sensing data, performing feature extraction, and defining a cold welding feature set; preprocessing the cold welding feature set to generate a normalized cold welding sensitive feature set; calculating a dynamically weighted cold welding risk index based on the cold welding sensitive feature set; classifying and outputting a cold welding defect type according to the cold welding sensitive feature set and a preset defect type feature library; mapping the cold welding risk index to a three-level risk grade based on a mapping relationship calibrated by an orthogonal test; and outputting a defect type and risk grade report with three-dimensional space coordinates. The present application can realize dual precise recognition of defect types and risk grades, and provide auxiliary reference for maintenance strategies.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cold welding defect recognition, and particularly relates to a cold welding defect detection method and system based on multi-source data fusion. BACKGROUND

[0002] Cold welding is a defect in the process of metal connection caused by interface pollution, oxidation or surface adsorption, etc., which leads to a significant decrease in the strength of the welded joint. It is difficult to detect. Cold welding defect detection is a key link to ensure the safety of welded structures, especially in high-reliability fields such as nuclear power and aerospace. Traditional methods rely on manual experience or single-source sensors (such as ultrasonic waves and visual detection), which are difficult to capture hidden defects of cold welding and lack a quantitative correlation mechanism for mechanical performance degradation, resulting in a lack of scientific basis for maintenance decisions.

[0003] However, existing cold welding defect recognition mainly relies on a single sensing method or static threshold determination, which has significant limitations. In traditional methods, optical detection (such as AOI) can determine cold welding by observing the orange peel-like appearance or insufficient collapse height of the surface of the welding point, but it cannot capture the internal metallurgical reaction state, resulting in missed detection of slight cold welding. Although infrared monitoring can obtain temperature distribution, it is difficult to quantify the correlation between thermal diffusion coefficient and mechanical performance. In wave-soldering or reflow-soldering processes, cold welding can be improved by adjusting the welding temperature and time parameters, but the parameter setting relies on experience and lacks a quantitative grading mechanism for defect severity. In addition, existing technologies do not establish a mapping model between cold welding characteristics and mechanical performance degradation, and can only output a binary defect determination (presence / absence), which cannot evaluate the potential risk level (such as the rate of decrease in tensile force), resulting in a lack of targeted maintenance strategies.

[0004] Therefore, there is an urgent need for a cold welding defect detection method and system based on multi-source data fusion, which can achieve dual-precision identification of defect type and risk level and provide auxiliary reference for maintenance strategies. SUMMARY

[0005] To solve the above technical problems, the present application provides a cold welding defect detection method and system based on multi-source data fusion, which can achieve dual-precision identification of defect type and risk level and provide auxiliary reference for maintenance strategies.

[0006] The present application provides a cold welding defect detection method based on multi-source data fusion, comprising the following steps:

[0007] S1, synchronously collecting acoustic emission signals, infrared thermal imaging data, laser scanning geometric data and embedded mechanical sensing data, performing feature extraction, and defining a cold welding feature set;

[0008] S2, preprocessing the cold welding feature set to generate a normalized cold welding sensitive feature set;

[0009] S3, calculating a dynamically weighted cold welding risk index based on the cold welding sensitive feature set;

[0010] S4, classifying and outputting a cold welding defect type according to the cold welding sensitive feature set and a preset defect type feature library;

[0011] S5, mapping the cold welding risk index to a three-level risk grade based on a mapping relationship calibrated by an orthogonal test;

[0012] S6, outputting a defect type and risk grade report with three-dimensional spatial coordinates.

[0013] Further, the cold welding feature set includes acoustic emission energy entropy, impact count rate, thermal diffusivity, low temperature difference gradient, surface geometric distortion rate and pressure fluctuation standard deviation.

[0014] Further, in S2, the preprocessing specifically includes data cleaning, time-space alignment, feature dimension reduction and normalization processing.

[0015] Further, in S3, calculating a dynamically weighted cold welding risk index based on the cold welding sensitive feature set includes:

[0016] S31, calculating a confidence degree of each cold welding sensitive feature in the cold welding sensitive feature set;

[0017] S32, calculating a dynamic weight of each cold welding sensitive feature according to a signal-to-noise ratio of each sensor and a feature sensitivity coefficient of each cold welding sensitive feature;

[0018] S33, calculating a cold welding risk index according to a feature value, a confidence degree and a dynamic weight of each cold welding sensitive feature.

[0019] Further, in S33, calculating a cold welding risk index according to each cold welding sensitive feature, a confidence degree and a dynamic weight, the calculation formula is as follows:

[0020] ;

[0021] wherein, CWRI represents the cold welding risk index, i represents the i th cold welding sensitive feature, n represents the total number of cold welding sensitive features, λ represents a decay factor, w i represents the dynamic weight of the i th cold welding sensitive feature, S i represents the feature value of the i th cold welding sensitive feature, Confidence i represents the confidence degree of the i th cold welding sensitive feature.

[0022] Further, in S4, classifying and outputting a cold welding defect type according to the cold welding sensitive feature set and a preset defect type feature library includes:

[0023] S41, constructing a defect type feature library according to the normalized sensitive feature set of the historical cold welding sample and the cold welding defect type;

[0024] S42, calculating a feature mean vector of each type of cold welding defect type as a defect center of the cold welding defect type according to the defect type feature library;

[0025] S43, constructing a feature vector according to the cold welding sensitive feature set, calculating a matching probability of the feature vector and each type of defect center using a Gaussian kernel function, and outputting a final classification result.

[0026] Further, in S43, a feature vector is constructed according to the cold welding sensitive feature set, a matching probability of the feature vector and each type of defect center is calculated using a Gaussian kernel function, and a final classification result is outputted, and the calculation formula is as follows:

[0027] ;

[0028] ;

[0029] Wherein, P(k|S) represents a conditional probability density, S represents a feature vector of the cold welding sensitive feature set, k represents the kth type of cold welding defect, σ k represents the feature space variance of the kth type of cold welding defect, μ k represents the defect center of the kth type of cold welding defect, DefectType represents the final classification result, and K represents a set of cold welding defect categories.

[0030] Further, in S5, the cold welding risk index is mapped to a three-level risk level based on the mapping relationship calibrated by the orthogonal test, including:

[0031] S51, constructing an experimental result database of the cold welding risk index and the tension drop rate through an orthogonal test;

[0032] S52, constructing a cold welding risk index-tension drop rate mapping function based on a nonlinear regression model;

[0033] S53, defining the tension drop rate of each risk level boundary;

[0034] S54, according to the tension drop rate of each risk level boundary, inversely solving the corresponding cold welding risk index through the cold welding risk index-tension drop rate mapping function, as a theoretical boundary cold welding risk index;

[0035] S55, correcting the theoretical boundary cold welding risk index through a confidence interval to obtain a boundary cold welding risk index;

[0036] S56, obtaining a mapping relationship of the cold welding risk index and each risk level according to the boundary cold welding risk index.

[0037] Further, in S55, the theoretical boundary cold welding risk index is corrected by a confidence interval to obtain a boundary cold welding risk index, including:

[0038] S551, input the theoretical boundary cold welding risk index into a cold welding risk index-tension drop rate mapping function to obtain a regression prediction value of the corresponding tension drop rate;

[0039] S552, according to the regression prediction value of the tension drop rate, the difference between the regression prediction value of the tension drop rate and the true value of the tension drop rate in the test result database, the theoretical boundary cold welding risk index and the average value of the cold welding risk index in the test result database, calculate the 95% confidence lower limit of the tension drop rate;

[0040] S553, according to the 95% confidence lower limit of the tension drop rate, the corresponding cold welding risk index is inversely solved by the cold welding risk index-tension drop rate mapping function as the boundary cold welding risk index.

[0041] The application also provides a cold welding defect detection system based on multi-source data fusion, which is used to execute the cold welding defect detection method based on multi-source data fusion, and the system comprises:

[0042] A data acquisition module is configured to synchronously acquire acoustic emission signals, infrared thermal imaging data, laser scanning geometric data and embedded mechanical sensing data, perform feature extraction, and define a cold welding feature set.

[0043] A preprocessing module is configured to preprocess the cold welding feature set to generate a normalized cold welding sensitive feature set.

[0044] A risk index calculation module is configured to calculate a dynamically weighted cold welding risk index based on the cold welding sensitive feature set.

[0045] A category prediction module is configured to classify and output a cold welding defect type according to the cold welding sensitive feature set and a preset defect type feature library.

[0046] A risk level prediction module is configured to map the cold welding risk index to a three-level risk level based on a mapping relationship calibrated by orthogonal tests.

[0047] An output module is configured to output a defect type and a risk level report with three-dimensional spatial coordinates.

[0048] The embodiments of the application have the following technical effects:

[0049] The present application constructs a multi-dimensional synchronous acquisition system of acoustic emission, infrared, laser scanning and mechanical sensing, extracts energy entropy, thermal diffusivity and other characteristics, and designs a dynamic weighted cold welding risk index, which integrates the signal-to-noise ratio weight of the sensor and the feature confidence, suppresses the interference of low-quality data through an exponential decay term, uses a probabilistic neural network for defect classification, calculates the matching probability of the feature vector and the preset defect type library by using a Gaussian kernel function, outputs the specific defect category, and grades the risk based on the nonlinear mapping function calibrated by orthogonal test, relates the cold welding risk index to the tension drop rate, and introduces the confidence interval correction theory boundary value, finally generates a three-level risk grade, breaks through the nonlinear correlation limitation of morphological characteristics and mechanical properties, realizes the dual-precision output of defect type and risk grade, drives differentiated maintenance decisions with grading strategy, and avoids over-maintenance. BRIEF DESCRIPTION OF DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the specific embodiments or prior art of the present application, the drawings needed in the specific embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0051] Figure 1 is a flowchart of the cold welding defect detection method based on multi-source data fusion provided by the embodiment of the present application;

[0052] Figure 2 is a cold welding risk index-mechanical degradation mapping curve diagram provided by the embodiment of the present application;

[0053] Figure 3 is a structural schematic diagram of the cold welding defect detection system based on multi-source data fusion provided by the embodiment of the present application. DETAILED DESCRIPTION

[0054] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions of the present application will be described clearly and completely below. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of the present application.

[0055] Figure 1 is a flowchart of the cold welding defect detection method based on multi-source data fusion provided by the embodiment of the present application. Referring to Figure 1 , specifically includes:

[0056] S1, synchronously collect acoustic emission signals, infrared thermal imaging data, laser scanning geometric data and embedded mechanical sensing data, perform feature extraction, and define a cold welding feature set.

[0057] The cold welding feature set includes acoustic emission energy entropy, impact count rate, thermal diffusion coefficient, low temperature difference gradient, surface geometric distortion rate and pressure fluctuation standard deviation. These data respectively reflect the key features of material internal energy release, heat conduction behavior, surface morphology change and mechanical response in the welding process, which can describe the formation mechanism and form of cold welding defects from different angles.

[0058] S2, pre-process the cold welding feature set to generate a normalized cold welding sensitive feature set.

[0059] The pre-processing specifically includes data cleaning, time-space alignment, feature dimension reduction and normalization processing.

[0060] Specifically, data cleaning is used to solve noise interference and denoise the acoustic emission signal: wavelet packet transform is used to decompose high-frequency transient waves, and effective components with energy greater than a threshold are retained; time-space alignment processing includes injecting a unified timestamp for each sensor in the edge computing network, and converting other sensor coordinates to the BIM model coordinate system with the laser scanner coordinate system as the origin; feature dimension reduction can be achieved through linear discriminant analysis (LDA), reducing redundant information, and finally solving the dimensional difference between features through normalization processing.

[0061] S3, calculate a dynamically weighted cold welding risk index based on the cold welding sensitive feature set.

[0062] In some embodiments, S3 specifically includes the following sub-steps:

[0063] S31, calculate the confidence of each cold welding sensitive feature in the cold welding sensitive feature set.

[0064] The calculation formula is as follows:

[0065] ;

[0066] wherein Confidence i represents the confidence of the i-th cold welding sensitive feature, represents the variance of the i-th cold welding sensitive feature in the sliding time window, and the sliding time window can be set to 10 seconds, for example, represents a smoothing coefficient, which is 10 by default −6 to prevent the denominator from being zero.

[0067] The confidence reflects the reliability and stability of the feature in the current detection environment. For example, when the acoustic emission sensor is in a strong electromagnetic interference environment, the acoustic emission energy entropy collected by the acoustic emission sensor may be disturbed, causing the data to fluctuate violently, and at this time the confidence of the feature should be correspondingly reduced.

[0068] S32, calculating a dynamic weight of each cold welding sensitive feature according to the signal-to-noise ratio of each sensor and the feature sensitive coefficient of each cold welding sensitive feature.

[0069] The calculation formula is as follows:

[0070] ;

[0071] wherein w i represents the dynamic weight of the i th cold welding sensitive feature, i represents the i th cold welding sensitive feature, j represents the j th cold welding sensitive feature, n represents the total number of cold welding sensitive features, α i represents the feature sensitive coefficient of the i th cold welding sensitive feature, α j represents the feature sensitive coefficient of the j th cold welding sensitive feature, the feature sensitive coefficient can be calibrated by orthogonal test, SNR i represents the signal-to-noise ratio of the feature generation channel corresponding to the i th cold welding sensitive feature, SNR j represents the signal-to-noise ratio of the feature generation channel corresponding to the j th cold welding sensitive feature.

[0072] The weight distribution not only considers the sensitivity of the feature itself to cold welding, but also combines the signal-to-noise ratio of the sensor on which the feature depends and the discrimination ability of the feature in the current working condition. The higher the signal-to-noise ratio of the sensor, the stronger the reliability of the data collected by the sensor, and the higher the weight of the corresponding feature. At the same time, the feature sensitive coefficient is introduced, which is obtained by calibration and reflects the discrimination ability of a certain feature in distinguishing normal welding from cold welding state. For example, in some welding processes, the surface geometric distortion rate may be more sensitive to cold welding defects, and its sensitive coefficient is higher, so it occupies a larger proportion in weight calculation. The calculation of dynamic weight adopts a weighted combination method, which takes the signal-to-noise ratio and the feature sensitive coefficient as the influencing factors together, to ensure that the weight distribution reflects not only the hardware performance, but also the discrimination value of the feature itself.

[0073] S33, calculating a cold welding risk index according to the feature value, the confidence and the dynamic weight of each cold welding sensitive feature.

[0074] The calculation formula of the cold welding risk index is as follows:

[0075] ;

[0076] wherein CWRI represents the cold welding risk index, i represents the i th cold welding sensitive feature, n represents the total number of cold welding sensitive features, λ represents the attenuation factor, wi S represents the dynamic weight of the ith cold welding sensitive feature i Confidence represents the feature value of the ith cold welding sensitive feature i Confidence represents the confidence of the ith cold welding sensitive feature.

[0077] By introducing the attenuation factor, the contribution of high-risk features is adjusted to avoid excessive influence of individual abnormal feature values on the overall evaluation result. When a feature value deviates significantly from the normal range, if its confidence is low or its weight is small, its influence on the final risk index will be suppressed, thereby improving the robustness of the evaluation result. The whole calculation process realizes the adaptive fusion of multi-source features, which can dynamically adjust the contribution of each feature according to the real-time data state, making the cold welding risk index more close to the actual welding quality condition. Confidence is used to measure the reliability of the feature data in the current detection environment, and dynamic weight reflects the discrimination ability of the feature under the current welding process condition. The higher the feature value, the stronger the confidence, and the greater the weight, the more significant the contribution of the feature to the cold welding risk index. The role of the attenuation factor is to nonlinearly compress the high feature value, so that its contribution does not increase linearly, thereby improving the robustness of the whole risk evaluation model. The cold welding risk index calculated by the formula provides a quantitative basis for subsequent risk level division and defect classification, which helps to realize accurate identification and grading evaluation of cold welding defects.

[0078] S4, according to the cold welding sensitive feature set and the preset defect type feature library, the cold welding defect type is classified and output.

[0079] In some embodiments, S4 specifically includes the following sub-steps:

[0080] S41, according to the normalized sensitive feature set of historical cold welding samples and the cold welding defect type, a defect type feature library is constructed.

[0081] Among them, the cold welding defect type includes pore defect, crack defect and incomplete fusion defect. The sensitive feature combination corresponding to the pore defect is high surface geometric distortion rate and low thermal diffusion coefficient; the sensitive feature combination corresponding to the crack defect is high acoustic emission energy entropy and high pressure fluctuation standard deviation; the sensitive feature combination corresponding to the incomplete fusion defect is high impact count rate and low low temperature difference gradient; the threshold values of the sensitive features corresponding to each defect can be determined by cluster analysis.

[0082] When constructing the defect type feature library, the historical cold welding samples need to be normalized to make them comparable with the cold welding sensitive features extracted in the current detection.

[0083] S42, according to the defect type feature library, the feature mean vector of each type of cold welding defect is calculated as the defect center of the cold welding defect type.

[0084] The characteristic data of each type of cold welding defect is statistically analyzed, and the characteristic mean vector is calculated as the "defect center" of the defect type. These defect centers represent the typical distribution positions of various types of cold welding defects in the feature space, and are key reference points in the classification process.

[0085] S43, according to the cold welding sensitive feature set, a feature vector is constructed, and a Gaussian kernel function is used to calculate the matching probability of the feature vector and each type of defect center, and the final classification result is output.

[0086] The calculation formula is as follows:

[0087] ;

[0088] ;

[0089] Where P(k|S) represents the conditional probability density, S represents the feature vector of the cold welding sensitive feature set, k represents the kth type of cold welding defect, represents the feature space variance of the kth type of cold welding defect (which can be dynamically optimized and determined by an artificial bee colony algorithm), μ k represents the defect center of the kth type of cold welding defect, DefectType represents the final classification result, and K represents the set of cold welding defect categories.

[0090] Based on the cold welding sensitive feature set extracted from the current welding area, a feature vector is constructed, and similarity analysis is performed with each type of defect center. Due to the certain fuzziness and overlap of the distribution of cold welding defects in the feature space, a Gaussian kernel function is used as a similarity measurement tool, which can effectively capture the nonlinear relationship between the feature vector and each type of defect center. The Gaussian kernel function calculates the Euclidean distance between the feature vector and the defect center, and combines the variance information of the feature space to obtain a probability density value, reflecting the possibility of the feature vector belonging to a certain type of cold welding defect. For example, when the feature vector is close to multiple defect centers in a certain detection, the Gaussian kernel function can reasonably allocate the matching probability of each type of defect according to the distribution density of the feature space, thereby avoiding misclassification. Finally, the defect type with the highest matching probability is selected as the classification result of this detection.

[0091] S5, based on the mapping relationship calibrated based on the orthogonal test, the cold welding risk index is mapped to a three-level risk level.

[0092] In some embodiments, Figure 2 is a cold welding risk index-mechanical degradation mapping curve provided by the embodiments of the present application, as shown in Figure 2 S5 specifically includes the following sub-steps:

[0093] S51, an experimental result database of cold welding risk index and tensile force reduction rate is constructed by orthogonal test.

[0094] Specifically, the test factors and levels in the test are defined, the test factor is a key process parameter actively controlled in the experiment, a core variable affecting the welding quality, and the level is a discrete value of each factor set in advance, representing different test conditions of the parameter.

[0095] In this embodiment, the test factors include the heating plate temperature, the heat absorption time and the welding pressure. For example, the test factors and levels are set as shown in Table 1:

[0096] Table 1 Test factor and level parameters

[0097] Test factor Level 1 Level 2 Level 3 Hot plate temperature 190℃ 210℃ 225℃ Heat-up time 30s 60s 95s Welding pressure 0.8 MPa 1.2 MPa 1.6 MPa

[0098] In Table 1, the engineering significance of the heating plate temperature level is that too low temperature → insufficient melting; and too high temperature → material degradation. The engineering significance of the heat absorption time level is that insufficient time → insufficient molecular entanglement; and too long time → thermal damage. The engineering significance of the welding pressure level is that insufficient pressure → insufficient bonding; and too large pressure → material deformation. Through the three-factor and three-level orthogonal test, the correlation between the process parameters and the cold welding defects is analyzed.

[0099] The calculation formula of the drop rate is as follows:

[0100] Drop=(F normal -F weld ) / F normal ×100%;

[0101] Wherein, Drop represents the drop rate, which is used to quantify the mechanical property degradation caused by cold welding, F normal represents the standard tensile strength of the base material (defect-free material), and F weld represents the actual tensile strength of the test piece containing the weld.

[0102] S52, constructing a cold welding risk index-drop rate mapping function based on a nonlinear regression model.

[0103] Since the influence of cold welding defects on the welding strength is not a linear relationship, but shows certain nonlinear characteristics, a nonlinear model such as polynomial regression, support vector regression or neural network is selected for fitting, to ensure that the mapping function can accurately reflect the complex relationship between the two.

[0104] In this embodiment, the cold welding risk index-drop rate mapping function expression is as follows:

[0105] Drop=a×exp(b×CWRI)+c;

[0106] wherein a represents a scale factor, b represents an exponential growth coefficient, and c represents an intercept offset term, and exemplarily, a = 1.85, b = 2.63, and c = -1.72 are obtained by least square fitting.

[0107] S53, defining a tension drop rate of each risk level boundary.

[0108] According to the engineering requirements, the tension drop rate boundary corresponding to each risk level is defined, and the boundary value can be set based on the strength requirement of the welded joint, the process standard or the industry specification, to ensure that the risk level division has practical engineering significance.

[0109] Exemplarily, in the embodiment, the risk level is divided into three levels, the tension drop rate Drop = 5% is defined as a two-level boundary, and the tension drop rate Drop = 15% is defined as a two-level boundary, that is, Drop < 5% is a first risk level, the structural strength attenuation can be ignored, 5% ≤ Drop < 15% is a second risk level, and the planned maintenance is required, and Drop ≥ 15% is a third risk level, and there is an immediate failure risk.

[0110] S54, according to the tension drop rate of each risk level boundary, the corresponding cold welding risk index is inversely solved by a cold welding risk index-tension drop rate mapping function, as a theoretical boundary cold welding risk index.

[0111] In the embodiment, the corresponding cold welding risk index obtained by inverse solution is: Drop = 5% → CWRI bound = 0.32, Drop = 15% → CWRI bound = 0.68, and CWRI bound is a theoretical boundary cold welding risk index. In the process of cold welding risk level division, the determination of the theoretical boundary cold welding risk index depends on the mapping function between the cold welding risk index and the tension drop rate. However, due to the volatility of the test data and the error of the model prediction, directly using the theoretical boundary may lead to an unstable risk level division. Therefore, the confidence interval is introduced to modify the theoretical boundary to improve the statistical reliability of the boundary value and ensure that the risk level division has sufficient safety margin.

[0112] S55, the theoretical boundary cold welding risk index is modified by the confidence interval to obtain a boundary cold welding risk index.

[0113] In order to improve the reliability of the boundary value, the theoretical boundary cold welding risk index is modified by the confidence interval, which considers the model error and the volatility of the test data, so that the risk level division is more stable and reliable, which specifically includes:

[0114] S551, inputting the theoretical boundary cold welding risk index into the cold welding risk index-tension drop rate mapping function to obtain a regression prediction value of the corresponding tension drop rate.

[0115] The predicted value represents the average tensile strength reduction level estimated by the model under the given cold welding risk index. Since the mechanical properties of the actual welded joint are affected by various random factors, there is a deviation between the regression predicted value and the true tensile strength reduction rate measured in the test. To quantify this uncertainty, the residual distribution characteristics of the model need to be analyzed, i.e. the difference between the true tensile strength reduction rate and the predicted value in all test samples.

[0116] S552, according to the regression predicted value of the tensile strength reduction rate, the difference between the regression predicted value of the tensile strength reduction rate and the true value of the tensile strength reduction rate in the test result database, the theoretical boundary cold welding risk index and the average of the cold welding risk index in the test result database, calculate the 95% confidence lower limit of the tensile strength reduction rate.

[0117] The calculation formula is as follows:

[0118]

[0119] where CWRI low represents the 95% confidence lower limit of the tensile strength reduction rate corresponding to the theoretical boundary cold welding risk index, represents the regression predicted value of the tensile strength reduction rate, t 0.975 represents the t-distribution critical value, which is obtained from the statistical t-distribution table, s represents the residual standard deviation, i.e. the difference between the regression predicted value of the tensile strength reduction rate and the true value of the tensile strength reduction rate in the test result database, N represents the number of test samples, i.e. the number of orthogonal test groups in the test result database, CWRI bound represents the theoretical boundary cold welding risk index (for example, CWRI bound = 0.32 and CWRI bound = 0.68) in this embodiment, CWRI p represents the cold welding risk index of the pth test in the test result database, represents the average of the cold welding risk index in the test result database.

[0120] This confidence lower limit reflects that under the current cold welding risk index level, the tensile strength reduction rate has a 95% probability of not being lower than this value. This statistical boundary takes into account the degree of dispersion of the data and the prediction uncertainty of the model, and has higher engineering safety. Especially in the high risk area, using the confidence lower limit can avoid misjudgment due to the model underestimating the severity of the defect.

[0121] S553, according to the 95% confidence lower limit of the tensile strength reduction rate, the corresponding cold welding risk index is inversely solved by the cold welding risk index-tensile strength reduction rate mapping function, as the boundary cold welding risk index.

[0122] ​The calculated 95% confidence lower limit of the tension drop is input again into the cold welding risk index-tension drop mapping function, and the corresponding cold welding risk index is obtained by back-solving. Since this process is based on a more conservative performance boundary for back-solving, the obtained cold welding risk index is more stringent than the original theoretical boundary, and can be used as the boundary cold welding risk index in actual application. This value serves as the dividing point between adjacent risk levels, effectively reducing the risk of misclassification due to measurement errors or model bias. The discrimination mechanism that prefers misclassification to missed judgment for high-risk defects is realized, ensuring zero missed judgment of real high-risk defects and avoiding safety hazards.

[0123] In some embodiments, the 95% confidence lower limit of the tension drop can also be used to obtain the boundary cold welding risk index by weighted summation of the corresponding cold welding risk index and the theoretical boundary cold welding risk index, which balances the risks of misclassification and missed judgment.

[0124] S56, obtain the mapping relationship between the cold welding risk index and each risk level according to the boundary cold welding risk index.

[0125] In this embodiment, the boundary cold welding risk index obtained by back-solving is: Drop=5%→CWRI low =0.29, Drop=15%→CWRI low =0.66. Accordingly, the mapping relationship between the cold welding risk index and each risk level is: CWRI<0.29 is level one risk, 0.29≤CWRI<0.66 is level two risk, and CWRI≥0.66 is level three risk.

[0126] By introducing the confidence interval correction mechanism, the boundary cold welding risk index not only reflects the average trend between cold welding defects and mechanical properties, but also takes into account the influence of data uncertainty and model error, making the final risk level division more robust and engineering practical.

[0127] S6, output the defect type and risk level report with three-dimensional spatial coordinates.

[0128] The application extracts energy entropy, thermal diffusivity and the like features by constructing a multi-dimensional synchronous acquisition system of acoustic emission, infrared, laser scanning and mechanical sensing, and designs a dynamic weighted cold welding risk index, the index fuses sensor signal-to-noise ratio weight and feature confidence, suppresses low-quality data interference through an index attenuation term, defect classification adopts a probabilistic neural network, uses a Gaussian kernel function to calculate the matching probability of a feature vector and a preset defect type library, outputs a specific defect category, risk grading is based on a nonlinear mapping function calibrated based on an orthogonal test, relates the cold welding risk index to the tension drop rate, and introduces a confidence interval correction theory boundary value, finally generates a three-level risk grade, breaks through the nonlinear correlation limitation of morphological features and mechanical properties, realizes dual-precision output of defect type and risk grade, differential maintenance decision is driven by grading strategy, and over-maintenance is avoided.

[0129] The application embodiment further provides a cold welding defect detection system based on multi-source data fusion, Figure 3 is a structural schematic diagram of the cold welding defect detection system based on multi-source data fusion provided by the application embodiment, referring to Figure 3 The system comprises the following modules:

[0130] A data acquisition module is configured to synchronously acquire acoustic emission signals, infrared thermal imaging data, laser scanning geometric data and embedded mechanical sensing data, perform feature extraction, and define a cold welding feature set.

[0131] A preprocessing module is configured to preprocess the cold welding feature set to generate a normalized cold welding sensitive feature set.

[0132] A risk index calculation module is configured to calculate a dynamic weighted cold welding risk index based on the cold welding sensitive feature set.

[0133] A category prediction module is configured to classify and output a cold welding defect type according to the cold welding sensitive feature set and a preset defect type feature library.

[0134] A risk level prediction module is configured to map the cold welding risk index to a three-level risk grade based on a mapping relationship calibrated based on an orthogonal test.

[0135] An output module is configured to output a defect type and a risk level report with three-dimensional space coordinates.

[0136] The system corresponds to the method provided by the above-mentioned embodiments, and will not be described here one by one.

[0137] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions recorded in the above embodiments can still be modified, or some or all of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the technical solutions of the embodiments of the present application.

Claims

1. A cold welding defect detection method based on multi-source data fusion, characterized in that, The method comprises the following steps: S1, synchronously collecting acoustic emission signals, infrared thermal imaging data, laser scanning geometric data and embedded mechanical sensing data, performing feature extraction, and defining a cold welding feature set; the cold welding feature set comprises acoustic emission energy entropy, impact count rate, thermal diffusivity, low temperature difference gradient, surface geometric distortion rate and pressure fluctuation standard deviation; S2, preprocessing the cold welding feature set to generate a normalized cold welding sensitive feature set; S3, calculating a dynamically weighted cold welding risk index based on the cold welding sensitive feature set; Specifically comprising: S31, calculating the confidence of each cold welding sensitive feature in the cold welding sensitive feature set; S32, calculating the dynamic weight of each cold welding sensitive feature according to the signal-to-noise ratio of each sensor and the feature sensitivity coefficient of each cold welding sensitive feature; S33, calculating the cold welding risk index according to the feature value of each cold welding sensitive feature, the confidence and the dynamic weight; S4, classifying and outputting the cold welding defect type according to the cold welding sensitive feature set and a preset defect type feature library; S5, mapping the cold welding risk index to a three-level risk level based on the mapping relationship calibrated by orthogonal test; Specifically comprising: S51, constructing an experimental result database of cold welding risk index and tensile force reduction rate through orthogonal test; S52, constructing a cold welding risk index-tensile force reduction rate mapping function based on a nonlinear regression model; S53, defining the tensile force reduction rate of each risk level boundary; S54, inversely solving the corresponding cold welding risk index through the cold welding risk index-tensile force reduction rate mapping function according to the tensile force reduction rate of each risk level boundary, as a theoretical boundary cold welding risk index; S55, correcting the theoretical boundary cold welding risk index through a confidence interval to obtain a boundary cold welding risk index; S56, obtaining the mapping relationship between the cold welding risk index and each risk level according to the boundary cold welding risk index; S6, outputting a defect type and risk level report with three-dimensional spatial coordinates.

2. The cold weld defect detection method based on multi-source data fusion according to claim 1, characterized in that, In the S2, the preprocessing specifically comprises data cleaning, time-space alignment, feature dimension reduction and normalization processing.

3. The cold weld defect detection method based on multi-source data fusion according to claim 1, characterized in that, In the S33, the cold welding risk index is calculated according to each cold welding sensitive feature, the confidence and the dynamic weight, and the calculation formula is as follows: ； wherein CWRI represents a cold weld risk index, i represents an i-th cold weld sensitive feature, n represents a total number of cold weld sensitive features, λ represents a decay factor, w i represents a dynamic weight of the i-th cold weld sensitive feature, S i represents a feature value of the i-th cold weld sensitive feature, Confidence i represents a confidence of the i-th cold weld sensitive feature.

4. The cold weld defect detection method based on multi-source data fusion according to claim 3, characterized in that, In the S4, the cold welding defect type is classified and outputted according to the cold welding sensitive feature set and a preset defect type feature library, comprising: S41, constructing a defect type feature library according to the normalized sensitive feature set and the cold welding defect type of historical cold welding samples; S42, calculating the feature mean vector of each type of cold welding defect type as the defect center of the cold welding defect type according to the defect type feature library; S43, constructing a feature vector according to the cold welding sensitive feature set, calculating the matching probability of the feature vector and each defect center using a Gaussian kernel function, and outputting the final classification result.

5. The cold weld defect detection method based on multi-source data fusion according to claim 4, characterized in that, In the S43, the feature vector is constructed according to the cold welding sensitive feature set, the matching probability of the feature vector and each defect center is calculated using a Gaussian kernel function, and the final classification result is outputted, and the calculation formula is as follows: ； ； wherein P(k|S) represents a conditional probability density, S represents a feature vector of the cold welding sensitive feature set, and k represents a kth cold welding defect, represents a feature space variance of the kth cold welding defect, μ k represents a defect center of the kth cold welding defect, DefectType represents a final classification result, and K represents a cold welding defect category set.

6. The cold weld defect detection method based on multi-source data fusion according to claim 1, characterized in that, The boundary cold welding risk index is obtained by correcting the theoretical boundary cold welding risk index by using a confidence interval, and the method comprises the following steps: S551, inputting the theoretical boundary cold welding risk index into the cold welding risk index-tension drop rate mapping function to obtain a regression prediction value of the corresponding tension drop rate; S552, calculating a 95% confidence lower limit of the tension drop rate according to the regression prediction value of the tension drop rate, a difference between the regression prediction value of the tension drop rate and a true value of the tension drop rate in the test result database, the theoretical boundary cold welding risk index, and a mean value of the cold welding risk index in the test result database; S553, inversely solving a corresponding cold welding risk index as the boundary cold welding risk index by using the cold welding risk index-tension drop rate mapping function according to the 95% confidence lower limit of the tension drop rate.

7. A cold weld defect detection system based on multi-source data fusion, characterized in that, The system is used for executing the cold welding defect detection method based on multi-source data fusion according to any one of claims 1-6, and the system comprises: a data acquisition module, which is used for synchronously collecting acoustic emission signals, infrared thermal imaging data, laser scanning geometric data and embedded mechanical sensing data, performing feature extraction, and defining a cold welding feature set; a preprocessing module, which is used for preprocessing the cold welding feature set to generate a normalized cold welding sensitive feature set; a risk index calculation module, which is used for calculating a dynamically weighted cold welding risk index based on the cold welding sensitive feature set; a category prediction module, which is used for classifying and outputting a cold welding defect type according to the cold welding sensitive feature set and a preset defect type feature library; a risk level prediction module, which is used for mapping the cold welding risk index to a three-level risk level based on a mapping relationship calibrated by orthogonal tests; an output module, which is used for outputting a defect type and a risk level report with three-dimensional space coordinates.

Images