Reliability analysis and optimization method for edge recognition accuracy of inner diameter of heat exchanger tube hole

Abstract

This invention provides a method for analyzing and optimizing the accuracy and reliability of tube hole edge recognition in heat exchangers. The method includes: selecting light intensity and Gaussian kernel size as random variables; conducting morphological and positional recognition accuracy experiments based on these random variables to obtain morphological and positional data samples; training the morphological and positional data samples to obtain a functional relationship between recognition accuracy and random variables; performing reliability and reliability sensitivity analyses on the recognition accuracy to obtain the reliability and sensitivity of tube hole recognition position and morphology; optimizing the tube hole recognition accuracy and reliability based on the reliability and sensitivity; and verifying the recognition accuracy of the tube hole based on the optimized parameters and determining the required light intensity and corresponding Gaussian kernel size. This invention significantly improves the accuracy and reliability of tube hole recognition during welding, optimizes welding quality, and enhances the stability and performance of the system.

Description

Technical Field

[0001] The present invention relates to the field of machine vision technology, and in particular to a method for analyzing and optimizing the accuracy of edge recognition of inner diameter of a heat exchanger tube hole. Background Art

[0002] At present, most edge recognition algorithms mainly evaluate their performance by accuracy, efficiency and robustness. For example, the circle detection algorithm proposed by Lan et al. has been verified by a large number of experiments and is superior to other algorithms in all three aspects. Similarly, a facial contour line generation algorithm proposed by Yang has shown faster detection speed and significantly improved correctness in medical image verification. Although these algorithms perform well on various images, they still fail to fully consider the impact of the uncertainty of algorithm input parameters and the random distribution of lighting conditions on recognition performance. At the same time, reliability theory provides an effective method to deeply understand and optimize performance by evaluating the sensitivity and reliability of the system to various random factors. Reliability theory has achieved remarkable results in the fields of structural motion accuracy analysis, system fault diagnosis and structural failure prediction. However, this theory has not yet been applied to the performance evaluation of edge recognition algorithms. Summary of the invention

[0003] In view of this, the present invention is committed to providing a reliability analysis and optimization method for the recognition accuracy of the inner diameter edge of the heat exchanger tube hole, which is used to solve the problem that the uncertainty of the algorithm input parameters and the random distribution of lighting conditions on the recognition performance are not fully considered in the prior art.

[0004] In a first aspect, the present invention provides a method for reliability analysis and optimization of the edge recognition accuracy of the inner diameter of a heat exchanger tube hole, the method comprising:

[0005] Select light intensity and Gaussian kernel size as random variables;

[0006] Performing a morphological recognition accuracy experiment and a position recognition accuracy experiment based on the random variables to obtain a morphological data sample and a position data sample;

[0007] Training the morphological data samples and the position data samples to obtain a functional relationship between recognition accuracy and random variables;

[0008] Performing reliability and reliability sensitivity analysis on the recognition accuracy to obtain the reliability and reliability sensitivity of the pore recognition position and shape;

[0009] Based on the reliability and the reliability sensitivity, the reliability optimization design of the pore recognition accuracy is performed; based on the parameters obtained after the optimization, the recognition accuracy of the pore recognition is verified and the required illumination intensity and the corresponding Gaussian kernel size are determined.

[0010] Optionally, the performing a morphology recognition accuracy experiment and a position recognition accuracy experiment based on the random variable to obtain a morphology data sample and a position data sample includes:

[0011] Under multiple sets of random variables with different light intensities and Gaussian kernel sizes, different pores are identified and the pore radius is obtained.

[0012] Under multiple sets of random variables with different light intensities and Gaussian kernel sizes, the same pore is identified and the coordinates of the pore center are obtained.

[0013] The difference between the tube hole radius and the real tube hole radius is calculated to obtain the morphological data sample, and the distance between the tube hole circle center coordinates and the real tube hole circle center coordinates is calculated to obtain the position data sample.

[0014] Optionally, the method further comprises verifying the functional relationship between the recognition accuracy and the random variable on a test set.

[0015] Optionally, the method for performing reliability and reliability sensitivity analysis on the recognition accuracy is a method for performing reliability and reliability sensitivity analysis on the recognition accuracy using a first-order second moment method, specifically comprising:

[0016] Create a function;

[0017] Get the initial coordinate value of the design point;

[0018] The reliability index is solved by partial derivatives, coefficients and the performance function at the design point.

[0019] Optionally, the method further comprises:

[0020] Recalculate the design point coordinates;

[0021] Determine whether the relative error of the reliability index of two iterations is less than a given value.

[0022] Optionally, the method further comprises, when the relative error is greater than a given value, using the recalculated design point coordinates as new design points.

[0023] Optionally, the performing the reliability optimization design of the pipe hole identification accuracy based on the reliability and the reliability sensitivity includes:

[0024] Based on the functional relationship between the recognition accuracy and the random variable, an optimization objective function is established;

[0025] Determine the value range of the random variable and the initial value of the algorithm iteration;

[0026] The accuracy and reliability of pipe hole identification are optimized through the FMINCON function.

[0027] Optionally, the training of the morphological data samples and the position data samples utilizes a BP neural network, and the BP neural network consists of an input layer, an output layer and a hidden layer.

[0028] Optionally, the functional relationship between the recognition accuracy and the random variable is:

[0029] ;

[0030] in, Indicates The output value of the neuron; The input and output values ​​of each neuron are expressed as ; Represents the network connection weight from the input layer to the hidden layer; Represents the threshold from the input layer to the hidden layer; represents the hidden layer transfer function; , They represent the network connection weights and thresholds from the hidden layer to the output layer respectively.

[0031] Optionally, the recognition accuracy experiment and the position recognition accuracy experiment use the same neural network structure and parameters.

[0032] According to the solution of the present invention, by applying reliability theory to the performance evaluation of the edge recognition algorithm, a comprehensive reliability analysis of the algorithm recognition accuracy is achieved. This technical means significantly improves the recognition accuracy and reliability of the tube holes during welding, which not only optimizes the welding quality, but also enhances the overall stability and performance of the system. At the same time, the present invention can optimize the reliability design of the tube hole recognition accuracy, ensuring that the recognition results have high accuracy and stability under different light intensities and Gaussian kernel sizes, thereby effectively improving the accuracy and consistency of tube hole recognition.

[0033] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention and implement it according to the contents of the specification, the preferred embodiments of the present invention are described in detail as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A schematic flow chart of a method for analyzing and optimizing the accuracy of heat exchanger tube hole inner diameter edge recognition according to an embodiment of the present invention is shown;

[0035] Figure 2 Shows Figure 1 A schematic flow chart of a method for performing a morphological recognition accuracy experiment and a position recognition accuracy experiment based on random variables in step S200 to obtain a morphological data sample and a position data sample;

[0036] Figure 3 Shows Figure 1 A schematic flow chart of a method for performing reliability and reliability sensitivity analysis on recognition accuracy in step S400 to obtain the reliability and reliability sensitivity of pore recognition position and morphology;

[0037] Figure 4 Shows Figure 1 A schematic flow chart of a method for optimizing the reliability of pore identification accuracy based on reliability and reliability sensitivity in step S500;

[0038] Figure 5 A schematic diagram showing a camera field of view for pore morphology recognition according to an embodiment of the present invention is shown;

[0039] Figure 6 A schematic diagram showing the relative relationship of the radius in the ROI in the pore morphology recognition according to an embodiment of the present invention;

[0040] Figure 7 A schematic diagram showing a camera field of view for pore position recognition according to an embodiment of the present invention;

[0041] Figure 8 A schematic diagram showing the relative relationship of various points in the ROI for pore position recognition according to an embodiment of the present invention;

[0042] Fig. 9 A schematic diagram of the BP neural network structure according to an embodiment of the present invention is shown;

[0043] Fig.10 A schematic diagram of the morphological recognition accuracy network training error result according to an embodiment of the present invention is shown;

[0044] Fig.11 A schematic diagram of relative error of morphological recognition accuracy test according to an embodiment of the present invention is shown;

[0045] Fig.12 A schematic diagram showing the position recognition accuracy network training error result according to an embodiment of the present invention is shown;

[0046] Fig.13 A schematic diagram of relative error of position recognition accuracy test according to an embodiment of the present invention is shown;

[0047] Fig.14 A schematic diagram showing the mean sensitivity of morphological recognition accuracy according to an embodiment of the present invention is shown;

[0048] Fig.15 A schematic diagram showing the mean sensitivity of position recognition accuracy according to an embodiment of the present invention is shown;

[0049] Fig.16 A schematic diagram showing the results of the mean sensitivity of morphological recognition accuracy before and after optimization according to an embodiment of the present invention;

[0050] Fig.17 A schematic diagram showing the results of the mean sensitivity of the position recognition accuracy before and after optimization according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0051] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are described in detail below in conjunction with the accompanying drawings. It is to be understood that the specific embodiments described herein are only used to explain the present application, rather than to limit the present application. It should also be noted that, for ease of description, only some structures related to the present application are shown in the accompanying drawings, rather than all structures. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0052] The terms "including" and "having" and any variations thereof in this application are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device comprising a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products or devices.

[0053] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0054] Figure 1 FIG. 1 is a schematic flow chart of a method for analyzing and optimizing the accuracy of heat exchanger tube hole inner diameter edge recognition according to an embodiment of the present invention. Figure 1 As shown, the method comprises the following steps:

[0055] Step S100, selecting light intensity and Gaussian kernel size as random variables.

[0056] When identifying the inner diameter edge of the heat exchanger tube hole, the two parameters of light intensity and Gaussian kernel size have the greatest impact on the recognition accuracy. Therefore, light intensity and Gaussian kernel size are used as random variables for the recognition accuracy of the tube hole edge.

[0057] Step S200, performing a morphological recognition accuracy experiment and a position recognition accuracy experiment based on random variables to obtain a morphological data sample and a position data sample.

[0058] The reliability of the pore recognition accuracy is determined by the reliability of the morphological recognition accuracy and the reliability of the position recognition accuracy. When analyzing the morphological recognition accuracy reliability, the difference between the recognized pore radius and the true pore radius is used as the morphological recognition accuracy reliability parameter. When analyzing the position recognition accuracy reliability, the difference between the recognized circle center coordinates and the true circle center coordinates is used as the position recognition accuracy reliability parameter.

[0059] Step S300, training the morphological data samples and the position data samples to obtain a functional relationship between the recognition accuracy and the random variable.

[0060] In some embodiments, the morphological data samples and the position data samples are trained and fitted by a BP neural network, and the BP neural network is composed of an input layer, an output layer and a hidden layer. In other embodiments, other appropriate algorithms or network models may be used to train the morphological data samples and the position data samples, which are not limited in detail here.

[0061] Step S400 , performing reliability and reliability sensitivity analysis on the recognition accuracy to obtain the reliability and reliability sensitivity of the pore recognition position and shape.

[0062] In some embodiments, the reliability and reliability sensitivity of pore identification accuracy are analyzed by using an improved first-order second-order moment method to obtain the reliability and reliability sensitivity of pore identification position accuracy and the reliability and reliability sensitivity of pore identification morphology accuracy.

[0063] Step S500: Based on the reliability and reliability sensitivity, the reliability optimization design of the pipe hole identification accuracy is performed.

[0064] In some embodiments, an approximate function method is selected to optimize the reliability of pore recognition accuracy. By optimizing the reliability of pore recognition accuracy, a constrained nonlinear target optimization problem can be solved. By optimizing the reliability of pore recognition accuracy, the optimal parameter combination can be found under different random variable conditions such as light intensity and Gaussian kernel size to maximize recognition accuracy.

[0065] Step S600: verifying the recognition accuracy of pore recognition based on the optimized parameters.

[0066] The technical solution of the present invention realizes a comprehensive reliability analysis of the algorithm recognition accuracy by applying reliability theory to the performance evaluation of the edge recognition algorithm. This technical means significantly improves the recognition accuracy and reliability of the tube holes during welding, which not only optimizes the welding quality, but also enhances the overall stability and performance of the system. At the same time, the present invention can optimize the reliability design of the tube hole recognition accuracy, ensuring that the recognition results have high accuracy and stability under different light intensities and Gaussian kernel sizes, thereby effectively improving the accuracy and consistency of tube hole recognition.

[0067] Figure 2 Shows Figure 1 A schematic flow chart of a method for performing a morphological recognition accuracy experiment and a position recognition accuracy experiment based on random variables in step S200 to obtain a morphological data sample and a position data sample. Figure 2 As shown, step S200 includes:

[0068] S210, identifying different pores under multiple groups of random variables with different light intensities and Gaussian kernel sizes to obtain pore radii.

[0069] The morphological recognition accuracy experiment can cover a variety of pore morphologies by identifying different pores at multiple different parameter levels, so that the recognition algorithm can be tested when dealing with a variety of practical situations. At the same time, the diverse parameters and pore morphologies can provide richer information, improve the statistical reliability of the experimental results, reduce the impact of accidental factors on the results, and help to more comprehensively reflect the performance of the algorithm.

[0070] S220, identifying the same pore under multiple groups of random variables with different light intensities and Gaussian kernel sizes, and obtaining the coordinates of the center of the pore.

[0071] The position recognition accuracy experiment selected multiple sets of different parameters to identify the same pore. By keeping the pore shape unchanged, the interference of morphological changes on the experimental results was reduced, making the experiment more able to reflect the direct impact of parameter levels on position recognition. This setting can more finely analyze the impact of parameter changes on position recognition accuracy, ensure the comparability of experimental results, and make it more reliable to compare the impact of different parameter levels on position recognition accuracy.

[0072] S230, calculating the difference between the tube hole radius and the real tube hole radius to obtain a morphological data sample, calculating the distance between the tube hole circle center coordinates and the real tube hole circle center coordinates to obtain a position data sample.

[0073] The difference calculation is performed on each set of data to obtain the corresponding morphological data samples and position data samples.

[0074] In some embodiments, after obtaining the functional relationship between the recognition accuracy and the random variable, the functional relationship between the recognition accuracy and the random variable is verified on the data of the test set, which can further confirm the reliability of the functional relationship between the recognition accuracy and the random variable, ensure that the functional relationship can work stably under different conditions, and provide reliable technical support for practical applications.

[0075] In some embodiments, the above-mentioned recognition accuracy experiment and position recognition accuracy experiment use the same neural network structure and parameters.

[0076] Figure 3 Shows Figure 1 The reliability and reliability sensitivity analysis of the recognition accuracy in step S400 is a schematic flow chart of the method for obtaining the reliability and reliability sensitivity of the pore recognition position and morphology. In some embodiments, the reliability and reliability sensitivity analysis of the recognition accuracy can be performed using the first order second moment method. Figure 3 As shown, step S400 includes:

[0077] S410, establishing a functional function.

[0078] A functional function is established based on the illumination intensity, Gaussian kernel size parameters and the functional relationship between recognition accuracy and random variables. In the optimization of pore recognition accuracy, the functional function represents the relationship between recognition accuracy and random variables.

[0079] S420, obtaining the initial coordinate value of the design point.

[0080] In reliability and reliability sensitivity analysis of recognition accuracy, design points refer to specific values ​​of random variable combinations, such as light intensity and Gaussian kernel size.

[0081] S430, solving the reliability index through the partial derivatives, coefficients and performance functions at the design points.

[0082] Partial derivatives are used to measure the rate at which the performance function changes around the design point and can help determine how to adjust the design point to improve performance. The partial derivative calculation coefficient at is as follows:

[0083] .

[0084] After calculating the coefficients through the above formula, substitute the design point coordinates into the performance function to solve the reliability index .

[0085] The design point coordinate calculation formula is:

[0086] .

[0087] S440, recalculate the design point coordinates.

[0088] According to the above calculation results, the coordinates of the calculated design points are readjusted to improve the value of the performance function, and the formula in the above step S430 is still used.

[0089] S450, determining whether the relative error of the reliability index of two iterations is less than a given value.

[0090] By comparing the difference between the reliability indicators of two iterative calculations with the set error tolerance, it is determined whether the analysis and calculation process has converged.

[0091] S460, when the relative error is greater than a given value, the recalculated design point coordinates are used as new design points; when the relative error is less than a given value, the reliability sensitivity is calculated using the failure probability.

[0092] When the relative error in step S450 is greater than a given value, the recalculated design point coordinates are used as new design points, and the above steps S430 to S450 are repeated until the relative error is less than a given value. When the relative error is less than a given value, the reliability sensitivity is calculated based on the reliability index and the failure probability, and the calculation formula is as follows:

[0093] ; .

[0094] Figure 4 Shows Figure 1 A schematic flow chart of a method for optimizing the reliability design of pore identification accuracy based on reliability and reliability sensitivity in step S500. Figure 4 As shown, step S500 includes:

[0095] S510, establishing an optimization objective function based on a functional relationship between recognition accuracy and random variables.

[0096] S520, determine the value range of the random variable and the initial value of the algorithm iteration.

[0097] Define the range of random variables for the optimization algorithm. For example, the range of light intensity can be 0 to 1000 Lux, and the Gaussian kernel size can be 1 to 10 pixels. These ranges are usually based on actual conditions or experimental data. At the same time, set the initial value of the optimization algorithm, that is, the initial guess value of the design variable when starting the optimization. These initial values ​​can be empirical values ​​or values ​​obtained from previous experiments. By defining the range of random variables, it can be ensured that the optimization algorithm searches within a reasonable range and avoids unrealistic design points. The appropriate initial value can speed up the convergence of the optimization algorithm, reduce calculation time, and improve optimization efficiency.

[0098] S530, optimize the accuracy and reliability of pipe hole identification through FMINCON function.

[0099] Use the FMINCON function in MATLAB to solve the optimization problem, that is, minimize the objective function and satisfy the preset constraints, such as the range of the design variables. Through iterative optimization, FMINCON can find the optimal solution that minimizes the objective function and improve the accuracy and reliability of pipe hole identification.

[0100] The above scheme establishes a functional relationship between recognition accuracy and random variables, determines a reasonable parameter range and initial value, and uses the FMINCON function for optimization. The whole process systematically improves the accuracy and reliability of pore recognition and ensures the scientificity and effectiveness of the optimization process.

[0101] Optionally, the functional relationship between the above recognition accuracy and the random variable is:

[0102] ;

[0103] in, Indicates The output value of the neuron; The input and output values ​​of each neuron are expressed as ; Represents the network connection weight from the input layer to the hidden layer; Represents the threshold from the input layer to the hidden layer; represents the hidden layer transfer function; , They represent the network connection weights and thresholds from the hidden layer to the output layer respectively.

[0104] The reliability analysis and optimization method for the inner diameter edge recognition accuracy of a heat exchanger tube hole of the present invention will be further explained below with a specific implementation process.

[0105] Figure 5 A schematic diagram of a camera field of view for pore morphology recognition according to an embodiment of the present invention is shown. Figure 6 FIG. 4 is a schematic diagram showing the relative relationship of the radius in the ROI in the pore morphology recognition according to an embodiment of the present invention. Figure 6 As shown, represents the center coordinate of the real pore, Indicates the coordinates of the center of the circle identified by the algorithm; represents the actual pore radius, Represents the radius of the pore identified by the algorithm. The actual hole radius The absolute value of the difference is calculated and the reliability parameter of the pore shape recognition accuracy is obtained. .

[0106] Figure 7 A schematic diagram showing a camera field of view for pore position recognition according to an embodiment of the present invention. Figure 8 FIG. 4 is a schematic diagram showing the relative relationship of various points in the ROI for pore position identification according to an embodiment of the present invention. Figure 8 As shown, represents the image center coordinates, represents the center coordinate of the real pore, Indicates the coordinates of the center of the circle identified by the algorithm. The green circle in the figure represents the shape and size of the pore, and the blue circle represents the floating range of the coordinates of the center of the circle identified by the algorithm. The actual pore center position Calculate the distance difference, and the difference is the reliability parameter of the hole position recognition accuracy. .

[0107] Taking light intensity and Gaussian kernel size as random variables, two groups of recognition accuracy experiments were carried out under multiple sets of parameters with different light intensities and Gaussian kernel sizes, namely, the morphological recognition accuracy experiment and the position recognition accuracy experiment.

[0108] Different pores were identified under 500 different illumination intensities and Gaussian kernel size parameter levels, and the difference between the pore radius obtained each time and the true pore radius was calculated. Based on the different differences obtained under different parameter conditions, a data sample for the reliability analysis of pore morphology recognition accuracy was established, as shown in Table 1. Indicates the light intensity, represents the Gaussian kernel size, Indicates the reliability parameter of pore morphology recognition accuracy. At the same time, the same pore was identified under 500 different illumination intensities and Gaussian kernel size parameter levels, and the distance between the center coordinates of each identified circle and the true pore center was obtained. Based on the different distance values ​​obtained under different parameter conditions, a data sample for pore position recognition accuracy reliability analysis was established, as shown in Table 2 below. In Table 2, Indicates the light intensity, represents the Gaussian kernel size, It represents the reliability parameter of the hole position identification accuracy.

[0109]

[0110] The BP neural network is used to train the random variable samples of the morphological recognition accuracy reliability analysis and the random variable samples of the position recognition accuracy reliability analysis obtained above. The BP neural network consists of 1 input layer, 1 output layer and 1 hidden layer. The Levenberg-Marquardt algorithm is selected for training, and the above two groups of experiments use the same network structure and parameters. Fig. 9 FIG. 2 shows a schematic diagram of a BP neural network structure according to an embodiment of the present invention. Fig. 9 As shown in the figure, the maximum allowed number of trainings for the BP neural network is set to 2000 times, the learning rate of the network is 0.01, the momentum factor is set to 0.85, and the minimum performance gradient is set to 10e-6, thereby obtaining an approximate functional relationship between the recognition accuracy and the random variable.

[0111] Among them, the formula for the function approximate fitting process between the fitting result and the random variable in the BP neural network is:

[0112] ;

[0113] ;

[0114] ;

[0115] .

[0116] In the above formula, The input and output values ​​of each neuron are expressed as . , They represent the network connection weights and thresholds from the input layer to the hidden layer respectively. , Respectively represent The input and output values ​​of a neuron. represents the hidden layer transfer function. , They represent the network connection weights and thresholds from the hidden layer to the output layer respectively. , Respectively represent The input and output values ​​of a neuron. represents the output layer transfer function.

[0117] Thus, the functional relationship expression between the output layer and the random variable is obtained, that is, the functional relationship expression between the recognition accuracy and the random variable is:

[0118] .

[0119] Fig.10A schematic diagram of the morphological recognition accuracy network training error results according to an embodiment of the present invention is shown. Fig.11 A schematic diagram of relative error of morphological recognition accuracy test according to an embodiment of the present invention is shown. Fig.12 A schematic diagram of the position recognition accuracy network training error results according to an embodiment of the present invention is shown. Fig.13 FIG. 2 shows a schematic diagram of relative error of position recognition accuracy test according to an embodiment of the present invention. Fig.10 and Fig.12 As shown in the figure, the neural networks of the two figures converged during the training process. The morphology recognition training network converged after 27 iterations, with a mean square error of 0.0099. The location recognition training network converged at the 1791th iteration, with a mean square error of 0.037. Fig.11 , Fig.13 From the data distribution, we can see that the relative error between the true value and the predicted value is small.

[0120] Fig.14 A schematic diagram showing the mean sensitivity of morphological recognition accuracy according to an embodiment of the present invention is shown. Fig.15 FIG. 4 is a schematic diagram showing the mean sensitivity of the position recognition accuracy according to an embodiment of the present invention. Fig.14 and Fig.15 As shown, based on the functional relationship between the recognition accuracy and the random variable obtained by the above training, the limit state function is established as the response quantity, and the reliability of the pore shape and position recognition accuracy is obtained by the improved second-order moment method, which is 0.92485 and 0.97136 respectively.

[0121] Fig.16 A schematic diagram showing the results of the mean sensitivity of morphological recognition accuracy before and after optimization according to an embodiment of the present invention is shown. Fig.17 FIG. 4 is a schematic diagram showing the result of the mean sensitivity of the position recognition accuracy before and after optimization according to an embodiment of the present invention. Fig.16 As shown in the figure, the yellow column represents the mean sensitivity of the morphological recognition accuracy corresponding to the random variable before optimization, and the green column represents the mean sensitivity of the morphological recognition accuracy corresponding to the random variable after optimization. Fig.17 As shown in the figure, the yellow column represents the mean sensitivity of the position identification accuracy corresponding to the random variable before optimization, and the green column represents the mean sensitivity of the position identification accuracy corresponding to the random variable after optimization. After obtaining the reliability and reliability sensitivity, the reliability optimization design of the pipe hole identification accuracy is carried out. The optimization objective function is established based on the approximate function fitted by the above BP neural network. Determine the value range of the light intensity and the Gaussian kernel size and the initial value of the algorithm iteration. Reasonable initial values ​​can be set artificially according to the characteristics of the problem to be solved and the constraints. Solve the optimization through the FMINCON function in MATLAB.

[0122] The recognition accuracy of the optimized data parameters in the tube hole recognition is verified. The following is a comparison experiment of the tube hole recognition accuracy before and after optimization, taking the morphological recognition accuracy as an example. During the comparison experiment, the light intensity that the tube hole end face can receive is set to 220 lx, 570 lx, and 710 lx, respectively, and the Gaussian kernel size in the recognition algorithm is set to 9. Among them, the light intensity of 570 lx is the light intensity parameter calculated after optimization. Based on this, 5 types of tube holes with different characteristics are identified respectively. The feature includes all the situations that occur in the recognition process, namely, a single tube hole, two tube holes, three tube holes, two tube holes, rust and welding spots, and three tube holes, rust and welding spots. Table 3 below shows the comparison of the tube hole morphological recognition accuracy before and after optimization under different tube hole characteristics, in mm.

[0123]

[0124] As shown in Table 3, when the illumination intensity is 220 lx and 710 lx, the recognition accuracy of the pores with different characteristics will fail, regardless of the shape or position recognition accuracy, and cannot meet the actual production requirements. Only when the illumination intensity is 570 lx and the Gaussian kernel size is 9, the recognition accuracy of the pores with different characteristics is very small, within the foreseeable and acceptable range, so it can be well overlapped with the edge of the real pore.

[0125] In summary, according to the above embodiments, the technical solution of the present invention fully considers the influence of the uncertainty of ambient lighting conditions and algorithm parameters on the recognition performance by selecting light intensity and Gaussian kernel size as random variables. At the same time, through the morphological recognition accuracy experiment and the position recognition accuracy experiment, the morphological data samples and the position data samples are respectively obtained, and these samples are used for training to obtain the functional relationship between the recognition accuracy and the random variables, thereby ensuring the accuracy and reliability of the model. In addition, by performing reliability and reliability sensitivity analysis on the recognition accuracy, the reliability and reliability sensitivity of the tube hole recognition position and morphology can be accurately evaluated, thereby providing a scientific basis for the optimization design. Based on these analysis results, the reliability optimization design of the tube hole recognition accuracy is performed to further improve the accuracy and stability of the recognition algorithm. Finally, the optimized parameters are verified by experiments to ensure the high efficiency and accuracy in practical applications, and effectively improve the accuracy and reliability of the edge recognition of the inner diameter of the heat exchanger tube hole.

[0126] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0127] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A reliability analysis and optimization method for the edge recognition accuracy of the inner diameter of a heat exchanger tube hole, characterized in that: The method comprises: Select light intensity and Gaussian kernel size as random variables; Performing a morphological recognition accuracy experiment and a position recognition accuracy experiment based on the random variables to obtain a morphological data sample and a position data sample; Training the morphological data samples and the position data samples to obtain a functional relationship between recognition accuracy and random variables; Performing reliability and reliability sensitivity analysis on the recognition accuracy to obtain the reliability and reliability sensitivity of the pore recognition position and shape; Based on the reliability and the reliability sensitivity, a reliability optimization design of the pipe hole identification accuracy is performed; Based on the optimized parameters, the recognition accuracy of pore recognition is verified and the required illumination intensity and the corresponding Gaussian kernel size are determined; The morphological recognition accuracy experiment and the position recognition accuracy experiment are performed based on the random variables to obtain the morphological data samples and the position data samples, including: Under multiple sets of random variables with different light intensities and Gaussian kernel sizes, different pores are identified and the pore radius is obtained. Under multiple sets of random variables with different light intensities and Gaussian kernel sizes, the same pore is identified and the coordinates of the pore center are obtained. Calculating the difference between the tube hole radius and the real tube hole radius to obtain the morphological data sample, and calculating the distance between the tube hole circle center coordinates and the real tube hole circle center coordinates to obtain the position data sample; The reliability optimization design of the pipe hole identification accuracy based on the reliability and the reliability sensitivity includes: Based on the functional relationship between the recognition accuracy and the random variable, an optimization objective function is established; Determine the value range of the random variable and the initial value of the algorithm iteration; The accuracy and reliability of pipe hole identification are optimized through the FMINCON function.

2. The method according to claim 1, characterized in that The method also includes verifying the functional relationship between the recognition accuracy and the random variable on a test set.

3. The method according to claim 1, characterized in that The method for performing reliability and reliability sensitivity analysis on the recognition accuracy is a method for performing reliability and reliability sensitivity analysis on the recognition accuracy using a first-order second-order moment method, specifically comprising: Create a function; Get the initial coordinate value of the design point; The reliability index is solved by partial derivatives, coefficients and the performance function at the design point.

4. The method according to claim 3, characterized in that The method further comprises: Recalculate the design point coordinates; Determine whether the relative error of the reliability index of two iterations is less than a given value.

5. The method according to claim 4, characterized in that The method further includes, when the relative error is greater than a given value, using the recalculated design point coordinates as new design points; and when the relative error is less than a given value, calculating the reliability sensitivity through the failure probability.

6. The method according to claim 1, characterized in that The BP neural network is used to train the morphological data samples and the position data samples. The BP neural network consists of an input layer, an output layer and a hidden layer.

7. The method according to claim 6, characterized in that The functional relationship between the recognition accuracy and the random variable is: Among them, y oθ represents the output value of the θth neuron; the input and output values ​​of the kth neuron are both represented by x k ;w kj represents the network connection weight from the input layer to the hidden layer; b 1j Represents the threshold from the input layer to the hidden layer; represents the hidden layer transfer function; v jθ 、b 2θ They represent the network connection weights and thresholds from the hidden layer to the output layer respectively.

8. The method according to claim 6, characterized in that The recognition accuracy experiment and the position recognition accuracy experiment use the same neural network structure and parameters.

Images