An artificial intelligence-based optimization method and system for automotive hot-dip galvanizing process parameters

Abstract

The present invention relates to the technical field of hot-dip galvanizing. The present invention discloses an artificial intelligence-based method and system for optimizing process parameters of automobile hot-dip galvanizing. The method comprises: performing window division processing on initial cooling parameters, constructing a cooling rate matrix, and finely characterizing the heat exchange state of different cooling sections, thereby solving the problem of insufficient cooling control accuracy under conditions of different alternating cooling rates. The main axis direction of the grain is extracted in combination with a microscopic image, the mean and range of the angle are calculated, a consistency index set is generated, the problem of dispersion of grain growth direction is quantitatively described, the cooling rate and the consistency index are jointly matched, a cooling rate structure offset matrix is ​​constructed, spatial mutation areas are identified, and abnormally coarse grains or abnormally arranged positions are accurately located. Finally, the target cooling parameters are screened according to the cooling index values ​​corresponding to the abnormal areas, and local adjustment of the cooling process is achieved, thereby reducing the abnormal grain growth rate and improving the grain direction consistency.

Description

Technical Field

[0001] The present invention relates to the technical field of hot-dip galvanizing, and more particularly to an artificial intelligence-based method and system for optimizing process parameters of automobile hot-dip galvanizing. Background Art

[0002] Hot-dip galvanizing is widely used in industries such as construction, transportation, energy, and automobile manufacturing due to its advantages such as good corrosion resistance, fast film formation speed, and wide application range. In particular, in the automobile manufacturing process, the hot-dip galvanizing layer not only plays a protective role, but also directly affects the coating effect and service life of the car body. During the hot-dip galvanizing process, the steel surface reacts with the molten zinc liquid to form a coating with a crystalline structure. The formation of grains is affected by many factors, including the composition of the zinc liquid, the activity of the substrate, the immersion time, and the cooling rate. Spangles, as a typical grain morphology of the hot-dip galvanized surface, are formed by the crystallization growth of the liquid zinc layer during the cooling process. Their size and orientation have a significant impact on the appearance uniformity, adhesion, and corrosion resistance of the coating.

[0003] However, in the existing hot-dip galvanizing production process, the grain growth process is still obviously uncontrollable, especially under conditions of rapid cooling or alternating cooling rates. It is easy for zinc flower grains to become abnormally coarse and the growth direction to be dispersed, resulting in uneven surface structure of the coating, affecting the consistency and overall aesthetics of subsequent coating. In addition, the cooling gradient changes caused by different materials or equipment will also aggravate the volatility of grain growth behavior and cause inconsistency in the microstructure of the zinc layer. How to reduce the abnormal growth rate of zinc flower grains while enhancing the consistency of grain orientation under different cooling rates, thereby ensuring the overall uniformity of the coating surface structure? As the requirements for coating performance continue to increase, this problem needs to be solved in actual production.

[0004] In view of this, the present invention proposes an artificial intelligence-based automobile hot-dip galvanizing process parameter optimization method and system to solve the above problems. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides an artificial intelligence-based method and system for optimizing automobile hot-dip galvanizing process parameters.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] In the first aspect, a method for optimizing process parameters of automotive hot-dip galvanizing based on artificial intelligence is provided, comprising:

[0008] The initial cooling parameters and microscopic images of the hot-dip galvanizing process are obtained. Window partitioning is performed on the initial cooling parameters to obtain a cooling rate matrix. The microscopic images are used to characterize the arrangement of grains in the spatial region.

[0009] Performing principal axis direction extraction processing on the microscopic image according to the cooling rate matrix to obtain a grain direction set;

[0010] According to the angle between any two main axis direction vectors in the grain direction set, consistency calculation processing is performed to obtain the angle mean and angle range, and a consistency index set is generated based on the angle mean and angle range;

[0011] Perform matrix joint matching processing based on the cooling rate matrix and the consistency index set to construct a cooling rate structure offset matrix; extract a mutation index set based on the spatial distribution mutation of the consistency index in the cooling rate structure offset matrix, and map the abnormal grain area based on the mutation index set to obtain an abnormal grain area set;

[0012] According to the cooling index values ​​in the abnormal grain region set, target cooling parameters corresponding to the cooling index values ​​are screened from the initial cooling parameters.

[0013] In some embodiments, the initial cooling parameters include a furnace zone temperature change sequence, a wind speed adjustment sequence, and a cooling segment. A method for performing window partitioning on the initial cooling parameters to obtain a cooling rate matrix includes:

[0014] Perform sliding window division processing on the furnace zone temperature change sequence according to the time axis, extract the average temperature gradient within each sliding window, and aggregate the average temperature gradient by time period to generate a furnace temperature interval sequence;

[0015] The wind speed adjustment sequence is mapped to the cooling section position, and the wind speed change rate sequence corresponding to each cooling section is extracted. The wind speed change rates are arranged in section order to form a wind speed interval matrix.

[0016] According to the cooling section, the furnace temperature interval sequence and the wind speed interval matrix are synchronously aggregated, and the average temperature gradient and wind speed change rate in each cooling section are combined into a cooling feature vector set;

[0017] Normalizing the cooling feature vector set to obtain a normalized vector, combining all normalized vectors in the order of cooling segments to generate a cooling feature matrix;

[0018] According to the characteristic row corresponding to each cooling section in the cooling characteristic matrix, its average cooling rate value is calculated, and all average cooling rate values ​​are combined into a cooling rate matrix.

[0019] In some embodiments, a method for performing principal axis direction extraction processing on a microscopic image according to a cooling rate matrix to obtain a set of grain directions includes:

[0020] Extracting grayscale edge contour data corresponding to each grain region in the microscopic image, and converting the grayscale edge contour data into a two-dimensional boundary coordinate point set;

[0021] According to the average cooling rate value of each cooling section in the cooling rate matrix, a composite function transformation is performed to obtain a density threshold interval, and the coordinate point density of the local area of ​​each boundary point in the two-dimensional boundary coordinate point set is calculated. The local area of ​​the boundary point whose coordinate point density is greater than the upper limit of the density threshold interval is extracted as the grain core contour area;

[0022] Performing a least squares fitting process on each grain core contour area to generate an ellipse model parameter set, the ellipse model parameter set including a major axis length, a minor axis length, and an inclination angle value;

[0023] Extract the tilt angle value from the ellipse model parameter set, construct the main axis direction vector, organize all the main axis direction vectors by grain number index, and generate a main axis direction list;

[0024] The main axis direction list is mapped back to the microscopic image space coordinates according to the grain number, forming a grain direction set with the pixel position as the index and the main axis direction as the value.

[0025] In some embodiments, a method for extracting a tilt angle value from an ellipse model parameter set and constructing a principal axis direction vector includes:

[0026] ;

[0027] Where, Indicates the The principal axis direction vector of each grain, represents the major axis tilt angle extracted from the ellipse model, represents the disturbance adjustment coefficient, Indicates the The cooling rate gradient of the cooling rate at the location of each grain in the lateral position.

[0028] In some embodiments, a method for performing consistency calculation based on the angle between any two principal axis direction vectors in the grain direction set to obtain the angle mean and angle range includes:

[0029] Based on each principal axis direction vector in the grain direction set, all principal axis direction vectors are organized in order of grain numbers to generate a principal axis direction vector sequence;

[0030] Extract all adjacent direction vector pairs from the main axis direction vector sequence, perform inner product calculation on each adjacent direction vector pair, and convert the inner product value into an angle value through the arc cosine function to form an angle value sequence;

[0031] According to the grain number corresponding to each main axis direction vector in the main axis direction vector sequence, a grain number sequence is constructed, adjacent difference calculation processing is performed on the grain number sequence to generate a number difference sequence, the number position in the number difference sequence that is greater than the jump threshold is used as an abnormal mark index, the angle value of the corresponding index position is extracted, and an angle abnormal value set is constructed, and the angle values ​​in the abnormal value set are removed from the angle value sequence to generate a denoised angle value set;

[0032] Perform mean calculation processing on the denoised angle value set to obtain the arithmetic mean of all angle values ​​in the set to obtain the angle mean; perform range calculation processing on the denoised angle value set to extract the difference between the maximum and minimum values ​​to obtain the angle range.

[0033] In some embodiments, a method for constructing a set of angle outlier values ​​includes: using the numbered positions in the numbered difference sequence that are greater than a jump threshold as abnormality marker indexes, extracting the angle values ​​corresponding to the indexed positions, and constructing the set of angle outlier values.

[0034] Performing a difference calculation process on each pair of adjacent numbers in the grain number sequence to obtain a number difference sequence, wherein the number difference represents the position span between the current main axis direction vector and the previous main axis direction vector on the original grain number;

[0035] Compare each number difference in the number difference sequence with a preset jump threshold, mark the number difference position greater than the jump threshold as an abnormal flag bit, and form an abnormal number index list;

[0036] According to the index position in the abnormal number index list, the angle value at the corresponding position in the angle value sequence is extracted to generate an angle abnormal value set.

[0037] In some embodiments, the jump threshold setting logic is as follows:

[0038] ;

[0039] in, represents the jump threshold, Indicates the current position index of the disturbance enhancement number difference sequence sorted by index, Indicates the number index position participating in the cumulative calculation, 、 、 Respectively represent The weights, means and variances of the Gaussian distribution components, represents the Gaussian probability density function, represents the first The difference in grain numbers, Indicates index at all positions In the equation, find the function value with the largest absolute value of the second-order derivative value, is the number of Gaussian components used to fit the distribution of number differences in the Gaussian mixture model.

[0040] In some embodiments, a matrix joint matching process is performed based on the cooling rate matrix and the consistency index set to construct a cooling rate structure offset matrix, including:

[0041] According to the matrix row index order of each cooling segment in the cooling rate matrix, the corresponding matrix row is extracted for each cooling segment, the average value of all elements in the cooling rate matrix is ​​calculated, and the average cooling rate value of the cooling segment is obtained. The average cooling rate values ​​of all cooling segments are arranged in sequence according to the original numbering order of the matrix row index, and a linear vector structure is constructed to generate a cooling rate sequence;

[0042] Construct a consistency indicator sequence according to the order of extracting the consistency indicator values ​​of each segment in the consistency indicator set;

[0043] The cooling rate sequence and the consistency index sequence are paired within the cooling segment according to the cooling segment number to construct a joint feature vector set. The vectors corresponding to two adjacent segments in the joint feature vector set are subtracted to calculate the cooling rate change difference and the consistency index change difference respectively.

[0044] The cooling rate change difference and the consistency index change difference are spliced ​​by dimension to form a two-dimensional offset vector. After summarizing all vectors, a two-dimensional offset set is formed. According to the index position of each two-dimensional vector in the two-dimensional offset set, a cooling rate structure offset matrix is ​​constructed.

[0045] In some embodiments, a method for performing intra-segment pairing operations on the cooling rate sequence and the consistency index sequence according to the cooling segment numbers to construct a joint feature vector set includes:

[0046] According to the average cooling rate value of each cooling segment in the cooling rate sequence, the corresponding segment number index is extracted to generate a cooling number sequence;

[0047] According to each consistency score value in the consistency index sequence, the corresponding segment number index is extracted to generate a consistency number sequence;

[0048] Perform index alignment processing on the cooling number sequence and the consistency number sequence, extract the cooling rate value and consistency score value with consistent index, and combine the consistent cooling rate value and consistency score value into a two-dimensional feature vector pair;

[0049] Arrange all two-dimensional feature vector pairs in ascending order of segment numbers and construct a joint feature vector set.

[0050] In a second aspect, an artificial intelligence-based automobile hot-dip galvanizing process parameter optimization system is provided, which is used to implement the above-mentioned artificial intelligence-based automobile hot-dip galvanizing process parameter optimization method, including:

[0051] Data acquisition module: used to obtain the initial cooling parameters and microscopic images in the hot-dip galvanizing process, perform window partitioning on the initial cooling parameters to obtain the cooling rate matrix, and the microscopic images are used to characterize the arrangement of grains in the spatial region;

[0052] The first processing module is used to perform principal axis direction extraction processing on the microscopic image according to the cooling rate matrix to obtain a grain direction set;

[0053] The second processing module is used to perform consistency calculation processing based on the angle between any two main axis direction vectors in the grain direction set, obtain the angle mean and angle range, and generate a consistency index set based on the angle mean and angle range;

[0054] Region detection module: used to perform matrix joint matching processing based on the cooling rate matrix and the consistency index set to construct a cooling rate structure offset matrix; extract the mutation index set based on the spatial distribution mutation of the consistency index in the cooling rate structure offset matrix, and map the abnormal grain area based on the mutation index set to obtain the abnormal grain area set;

[0055] Parameter optimization module: used to select target cooling parameters corresponding to the cooling index values ​​from the initial cooling parameters according to the cooling index values ​​in the abnormal grain area set.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] The present invention first constructs a cooling rate matrix by performing window partitioning processing on the initial cooling parameters, thereby refining the heat exchange state of different cooling sections, and specifically addresses the constraint condition of insufficient control accuracy under "different cooling rate alternating conditions"; further, combined with the microscopic image, the main axis direction of the grain in the spatial region is extracted, and the angle mean and angle range are calculated based on the angle between the main axis directions, thereby generating a consistency index set, and realizing a structured quantitative expression of the "growth direction dispersion" problem of the grain; subsequently, the cooling rate matrix and the consistency index set are jointly matched to construct a cooling rate structure offset matrix, and a mutation index set is extracted based on the spatial distribution mutation of the consistency index, thereby locating the abnormal grain area and effectively identifying the spatial position of "abnormally coarse grains" or "abnormal arrangement"; finally, according to the cooling index value in the abnormal grain area set, the corresponding target cooling parameters are screened from the initial cooling parameters to realize local optimization adjustment of the cooling process, and complete the dual control effects of "reducing abnormal growth rate" and "enhancing grain direction consistency". BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 Schematic diagram of a process flow of an artificial intelligence-based automotive hot-dip galvanizing process parameter optimization method of the present invention;

[0059] Figure 2 The figure is a schematic structural diagram of an artificial intelligence-based automobile hot-dip galvanizing process parameter optimization system in the present invention. DETAILED DESCRIPTION

[0060] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings. In the following detailed description, many specific details are set forth to provide a thorough understanding of the described exemplary embodiments. However, it is obvious to those skilled in the art that the described embodiments can be practiced without some or all of these specific details. In other exemplary embodiments, well-known structures are not described in detail to avoid unnecessarily obscuring the concepts of the present disclosure. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. At the same time, the various aspects described in the embodiments can be arbitrarily combined without conflict.

[0061] Example 1

[0062] See also Figure 1 As shown, this embodiment discloses a method for optimizing automotive hot-dip galvanizing process parameters based on artificial intelligence, including:

[0063] S10: obtaining initial cooling parameters and microscopic images in the hot-dip galvanizing process, performing window partitioning processing on the initial cooling parameters to obtain a cooling rate matrix, and the microscopic image is used to characterize the arrangement state of grains in the spatial region;

[0064] In this embodiment, the hot-dip galvanizing process refers to a surface treatment process in which a zinc layer covering structure is formed by immersing a metal substrate in molten zinc liquid, which is mainly used to enhance the corrosion resistance, adhesion and coating performance of the steel surface. The initial cooling parameters refer to the temperature control-related process parameters collected and recorded by the equipment during the hot-dip galvanizing process. The initial cooling parameters include the furnace zone temperature change sequence, the wind speed adjustment sequence and the cooling section, etc., which reflect the heat exchange rate and control rhythm of the galvanized layer in the cooling stage, and are the basic data source for constructing the cooling rate matrix.

[0065] Specifically, the furnace zone temperature change sequence refers to the temperature time series data obtained by continuously monitoring each temperature-controlled furnace section heated to the zinc pot before discharging in the hot-dip galvanizing process. This data reflects the heating process and temperature gradient change of the steel strip in each heating zone. It is usually composed of multiple sub-segments such as the front furnace, the middle furnace, and the rear furnace, and has a direct impact on the cooling rate judgment. The wind speed adjustment sequence refers to the set of wind speed control parameters set or adjusted in real time for the cold air blowing device in the forced cooling area during the cooling process of the galvanized layer. This parameter can be continuously changed according to time or position segments, and is used to regulate the heat exchange intensity per unit time. It is an important external control variable that affects the rapid solidification and microstructure formation of grains. The cooling section refers to the parameter set that divides the continuous cooling path into several physical or logical sub-segments, which is used to identify the start and end range of each cooling process. It usually includes areas such as rapid cooling section, slow cooling section and constant temperature section, which constitute the boundary constraint conditions of the window division processing.

[0066] Taking the initial cooling parameters including the furnace zone temperature change sequence, wind speed adjustment sequence, and cooling section as an example, the method of performing window partitioning on the initial cooling parameters to obtain the cooling rate matrix includes:

[0067] Perform sliding window division processing on the furnace zone temperature change sequence according to the time axis, extract the average temperature gradient within each sliding window, and aggregate the average temperature gradient by time period to generate a furnace temperature interval sequence;

[0068] The wind speed adjustment sequence is mapped to the cooling section position, and the wind speed change rate sequence corresponding to each cooling section is extracted. The wind speed change rates are arranged in section order to form a wind speed interval matrix.

[0069] According to the cooling section, the furnace temperature interval sequence and the wind speed interval matrix are synchronously aggregated, and the average temperature gradient and wind speed change rate in each cooling section are combined into a cooling feature vector set;

[0070] Normalizing the cooling feature vector set to obtain a normalized vector, combining all normalized vectors in the order of cooling segments to generate a cooling feature matrix;

[0071] According to the characteristic row corresponding to each cooling section in the cooling characteristic matrix, its average cooling rate value is calculated, and all average cooling rate values ​​are combined into a cooling rate matrix.

[0072] In this embodiment, sliding window division processing is performed on the furnace zone temperature change sequence in order to decompose the continuous temperature change process into multiple windows with local time period characteristics. The average temperature gradient is calculated in each window to reflect the heating trend of the time period; the average temperature gradients of all windows are aggregated in chronological order to construct a furnace temperature interval sequence, which provides a continuous temperature basis for subsequent matching with the cooling section. The wind speed adjustment sequence is subjected to segment mapping processing according to the cooling section position, which is to bind each physical cooling area with the corresponding wind speed change rate, so that the wind speed interval matrix can accurately express the heat exchange intensity change of each section. This processing is one of the prerequisite data steps for constructing the cooling rate. Synchronous aggregation processing is performed on the furnace temperature interval sequence and the wind speed interval matrix according to the cooling section in order to combine the temperature gradient and the wind speed change into a cooling feature vector within each cooling section, so that the same cooling section has a uniformly characterized thermal condition.

[0073] It should be noted that the normalization process is performed on the cooling feature vector set in order to unify the dimensions and scales of the features in each dimension, so as to facilitate the consistency comparison of subsequent processing; all normalized vectors are arranged in segment order to form a cooling feature matrix, which serves as the basic structure for the subsequent calculation of the cooling rate. The average cooling rate value is calculated according to the feature rows corresponding to each segment in the cooling feature matrix in order to compress the combined features into a single heat exchange index, and finally generate a cooling rate matrix by combining the cooling rate values ​​of all segments to reflect the actual cooling efficiency of different segments during the galvanizing process.

[0074] S20: performing principal axis direction extraction processing on the microscopic image according to the cooling rate matrix to obtain a grain direction set;

[0075] In this embodiment, a method for performing principal axis direction extraction processing on a microscopic image according to a cooling rate matrix to obtain a set of grain directions includes:

[0076] Extracting grayscale edge contour data corresponding to each grain region in the microscopic image, and converting the grayscale edge contour data into a two-dimensional boundary coordinate point set;

[0077] According to the average cooling rate value of each cooling section in the cooling rate matrix, a composite function transformation is performed to obtain a density threshold interval, and the coordinate point density of the local area of ​​each boundary point in the two-dimensional boundary coordinate point set is calculated. The local area of ​​the boundary point whose coordinate point density is greater than the upper limit of the density threshold interval is extracted as the grain core contour area;

[0078] Performing a least squares fitting process on each grain core contour area to generate an ellipse model parameter set, the ellipse model parameter set including a major axis length, a minor axis length, and an inclination angle value;

[0079] Extract the tilt angle value from the ellipse model parameter set, construct the main axis direction vector, organize all the main axis direction vectors by grain number index, and generate a main axis direction list;

[0080] The main axis direction list is mapped back to the microscopic image space coordinates according to the grain number, forming a grain direction set with the pixel position as the index and the main axis direction as the value.

[0081] In this embodiment, the grayscale edge contour data corresponding to each grain area is extracted from the microscopic image in order to capture the contour structure of the grain shape. The grayscale edge contour data can be converted into a two-dimensional boundary coordinate point set to facilitate the subsequent mathematical modeling of the grain morphology. The average cooling rate value of each cooling section in the cooling rate matrix will affect the grain boundary density distribution during the heat conduction process. Therefore, this value is used as input and the density threshold interval is obtained through composite function transformation in order to dynamically set the judgment standard for identifying higher density areas from the boundary coordinate point set. The area where the coordinate point density is greater than the upper limit of the density threshold interval is extracted as the grain core contour area, which can effectively eliminate boundary noise and retain only the boundary shape that can reflect the direction of the grain body, thereby improving the subsequent modeling accuracy.

[0082] It should be added that, in this embodiment, the row and column indices of the cooling rate structure offset matrix maintain a linear correspondence with the grid coordinate system of the microscopic image. Each matrix unit is mapped to a pixel area with an area size of N×N (for example, N=50 pixels). This mapping relationship provides a coordinate index basis for subsequent positioning of abnormal areas.

[0083] It should be noted that the least squares fitting process is performed on the core contour area of ​​the grain in order to obtain a stable and quantifiable set of ellipse model parameters, which includes the major axis length, minor axis length and tilt angle value, which helps to unify the directional expression of different grains. The tilt angle value is extracted from the ellipse model parameter set, the main axis direction vector is constructed, and it is organized into a main axis direction list according to the grain number index. This is to ensure a one-to-one correspondence between the grain direction information and the spatial position of the microscopic image. Finally, it can be mapped back to the pixel coordinates of the microscopic image to form a grain direction set, which serves as the input for subsequent consistency calculation and distribution recognition.

[0084] In this embodiment, to enhance the ability of the grain principal axis direction vector to perceive changes in the local cooling state, a disturbance correction mechanism based on the cooling rate variation trend is introduced. To achieve the computability and engineering feasibility of this disturbance term, it is necessary to clarify the lateral variation in the cooling rate at the location of the grain. Specifically, in the microscopic image, the cooling rate values ​​within a range of several adjacent pixels to the left and right of each grain can be selected to form a lateral distribution sequence. Based on this sequence, the rate variation trend in the lateral direction of the grain is estimated by calculating the difference in cooling rate between the center position and the positions to the left and right. This trend value is superimposed on the original principal axis direction angle as a disturbance factor, thereby forming a dynamically adjusted principal axis direction vector. This allows the grain direction characterization results to more realistically reflect the uneven effects of the actual cooling process.

[0085] The composite function is as follows:

[0086] ;

[0087] Where, Indicates the coordinate point density threshold, Indicates the The average cooling rate of the cooling section, represents the weight coefficient of the exponential decay term, Indicates the degree of attenuation of the cooling rate to the threshold, represents the adjustment coefficient of the disturbance term, represents the disturbance frequency control parameter, is a natural constant.

[0088] It is understood that, for example, , , , can be obtained by fitting and optimizing the historical data of hot-dip galvanizing process. The composite function proposed in this embodiment is used to The average cooling rate of the cooling section is used to calculate the density threshold of the coordinate points. The function consists of two parts: the first part is the exponential decay term, and the second part is the sinusoidal perturbation term. It aims to dynamically adjust the screening criteria for grain boundary density under different cooling conditions to improve the recognition accuracy of the core contour area. It is particularly suitable for complex scenarios with uneven cooling rates in hot-dip galvanizing processes. The function structure embodies the innovative design of the present invention in modeling heat treatment behavior. The exponential decay term reflects the physical phenomenon that the higher the cooling rate, the lower the boundary density threshold. This is because under conditions of high cooling rates, the boundary area formed by rapid solidification of grains is usually sparse. Therefore, if a fixed threshold is used, the core boundary points may be mistakenly eliminated. The exponential term sets the weight coefficient and the cooling rate attenuation degree parameter so that the boundary judgment criterion can be automatically adjusted with the cooling intensity, thereby improving processing flexibility and physical rationality.

[0089] Furthermore, the disturbance term adopts a sinusoidal structure to introduce a periodic fine-tuning mechanism to simulate the grain boundary density fluctuations caused by heat transfer disturbances or local cold air instability. This item sets two adjustment coefficients, the disturbance amplitude and the disturbance frequency, so that the threshold can be dynamically increased or decreased in the medium cooling rate segment, thereby improving the response ability to boundary density fluctuations, so that the core area can still be accurately identified when the spatial distribution is non-uniform. Unlike the fixed threshold or single linear model used in the existing technology, this composite structure enhances the fitting effect of the parameters to the real physical state, and the technical effect is better.

[0090] Methods for extracting the tilt angle value from the ellipse model parameter set and constructing the main axis direction vector include:

[0091] ;

[0092] Where, Indicates the The principal axis direction vector of each grain, represents the major axis tilt angle extracted from the ellipse model, represents the disturbance adjustment coefficient, Indicates the The cooling rate gradient of the cooling rate at the location of each grain in the lateral position.

[0093] It can be understood that the method of constructing the main axis direction vector in this embodiment is to introduce a disturbance adjustment term based on the main axis inclination angle extracted from the elliptical model parameter set. The disturbance term uses the gradient value of the cooling rate at the lateral position of the grain as an adjustment factor to correct the original main axis direction, thereby improving the responsiveness of the main axis direction vector to changes in the local cooling environment. Compared with the traditional method that relies only on a fixed main axis angle, this method can better reflect the temperature gradient driven behavior in the actual grain formation process.

[0094] It should be noted that the traditional method only constructs the main axis direction vector based on the main axis inclination angle, ignoring the influence of the cooling rate change at the position of the grain during the generation process on the fine-tuning of its final orientation. However, in the rapid cooling stage of hot-dip galvanizing, due to the unevenness of the cooling air velocity or heat flow channel, the growth direction of the grain boundary is easily affected by the local temperature gradient disturbance. Therefore, a disturbance coefficient is introduced in this method, multiplied by the lateral gradient value of the cooling rate, and superimposed on the original angle to achieve dynamic correction of the main axis direction, thereby constructing a main axis direction vector that is closer to the actual growth path of the grain, thereby improving the accuracy and engineering applicability of the model.

[0095] S30: performing consistency calculation processing based on the angle between any two principal axis direction vectors in the grain direction set to obtain an angle mean and an angle range, and generating a consistency index set based on the angle mean and the angle range;

[0096] Methods for performing consistency calculation based on the angle between any two principal axis direction vectors in the grain direction set to obtain the angle mean and angle range include:

[0097] Based on each principal axis direction vector in the grain direction set, all principal axis direction vectors are organized in order of grain numbers to generate a principal axis direction vector sequence;

[0098] Extract all adjacent direction vector pairs from the main axis direction vector sequence, perform inner product calculation on each adjacent direction vector pair, and convert the inner product value into an angle value through the arc cosine function to form an angle value sequence;

[0099] According to the grain number corresponding to each main axis direction vector in the main axis direction vector sequence, a grain number sequence is constructed, adjacent difference calculation processing is performed on the grain number sequence to generate a number difference sequence, the number position in the number difference sequence that is greater than the jump threshold is used as an abnormal mark index, the angle value of the corresponding index position is extracted, and an angle abnormal value set is constructed, and the angle values ​​in the abnormal value set are removed from the angle value sequence to generate a denoised angle value set;

[0100] Perform mean calculation processing on the denoised angle value set to obtain the arithmetic mean of all angle values ​​in the set to obtain the angle mean; perform range calculation processing on the denoised angle value set to extract the difference between the maximum and minimum values ​​to obtain the angle range.

[0101] The method of using the number position in the number difference sequence that is greater than the jump threshold as the abnormal mark index, extracting the angle value of the corresponding index position, and constructing the angle abnormal value set includes:

[0102] Performing a difference calculation process on each pair of adjacent numbers in the grain number sequence to obtain a number difference sequence, wherein the number difference represents the position span between the current main axis direction vector and the previous main axis direction vector on the original grain number;

[0103] Compare each number difference in the number difference sequence with a preset jump threshold, mark the number difference position greater than the jump threshold as an abnormal flag bit, and form an abnormal number index list;

[0104] According to the index position in the abnormal number index list, the angle value at the corresponding position in the angle value sequence is extracted to generate an angle abnormal value set.

[0105] It is understandable that in order to accurately evaluate the consistency of grain arrangement directions in microscope images, it is necessary to quantitatively analyze the angles between the main axis direction vectors, thereby calculating the angle mean and angle range, and constructing a core indicator reflecting the structural stability of the grain arrangement direction. Specifically, first, based on each main axis direction vector in the grain direction set, a main axis direction vector sequence is generated in the order of grain numbering. This vector sequence ensures the consistency of spatial layout and structural evolution; then, all adjacent crystal faces are extracted from the sequence, and the direction similarity is calculated by performing a vector inner product operation, and then the inverse cosine function is used to convert it into an angle value, finally forming an angle value sequence. This sequence truly reflects the degree of change in the main axis direction between adjacent grains and is the basis for constructing a direction consistency assessment. For example, in a certain cooling stage area, if the angles of the vectors are 8°, 10°, and 12°, respectively, it indicates that the grain arrangement directions are highly consistent. The subsequent obtained angle mean and range can represent the arrangement trend and direction disturbance range, respectively, providing a basis for process optimization.

[0106] Furthermore, in order to eliminate the erroneous angle interference caused by grain number jumps, this embodiment analyzes the grain number differences, constructs a number difference sequence and combines the jump threshold to perform anomaly elimination processing, thereby improving the reliability of statistical indicators. Specifically, a difference calculation is performed on each pair of adjacent numbers to obtain a number difference sequence, and the bit quantity greater than the jump threshold is used as the abnormal mark bit; then, the angle of the corresponding position is extracted from the angle value sequence, and a set of angle abnormal values ​​is generated, and it is deleted from the sequence to form a denoised angle value set. This processing logic is intended to eliminate non-continuous numbering problems caused by factors such as image occlusion and edge truncation. For example, if a certain section of grain number jumps more than 20 units, its corresponding main axis direction vector may not have spatial correlation, and its angle value is identified as abnormal and eliminated to prevent it from affecting the overall consistency measurement.

[0107] It should be noted that the final denoised angle value set is used to calculate the angle mean and angle range. The purpose is to simultaneously measure the "central tendency" and "fluctuation intensity" of the main axis direction arrangement with a dual-index structure, so as to improve the judgment ability of the stability of the grain structure in the cooling stage. The angle mean reflects the average orientation trend of the main axis direction, and the angle range quantitatively represents the maximum fluctuation range of the main axis direction in the spatial distribution. Taking a certain cooling stage as an example, if the removed angle set is [9°, 11°, 10°], the mean is 10° and the range is 2°, indicating that the grain orientation in this section is highly consistent; if the range is as high as 25°, it indicates that there is significant arrangement instability, and it is necessary to add cooling rate or wind speed adjustment to the parameter control. This combined dimensional design of mean + range can capture both the overall trend and local fluctuations, which helps to accurately evaluate the consistency of the grain structure and guide parameter callback.

[0108] In this embodiment, the jump threshold setting logic is as follows:

[0109] ;

[0110] in, represents the jump threshold, Indicates the current position index of the disturbance enhancement number difference sequence sorted by index, Indicates the number index position participating in the cumulative calculation, 、 、 Respectively represent The weights, means and variances of the Gaussian distribution components, represents the Gaussian probability density function, represents the first The difference in grain numbers, Indicates index at all positions In the equation, find the function value with the largest absolute value of the second-order derivative value, is the number of Gaussian components used to fit the distribution of number differences in the Gaussian mixture model.

[0111] It is understood that, in this embodiment, the Gaussian mixture model is fitted to the number difference sequence using the expectation maximization (EM) algorithm to calculate each difference. The probability density of the cumulative probability density is approximated by the difference method. In order to accurately identify the possible jump positions in the grain numbering sequence and thus improve the statistical accuracy of the subsequent consistency indicators, a jump threshold setting logic based on a probability model is introduced. This logic is based on the perturbation enhancement sequence of the numbering difference, and the Gaussian mixture model is used to fit the difference distribution. The point with the largest absolute value of the second-order derivative of the fitted distribution is found in all position indexes as the jump threshold to ensure that the identified jump positions are statistically significant and the distribution structure is reasonable.

[0112] It should be noted that the core idea of ​​this method is that there may be multiple local extreme points in the number difference sequence after disturbance enhancement, and these extreme points often reflect the area of ​​grain number jump or abnormal fluctuation. The traditional method may only be set through empirical threshold setting, which is difficult to adapt to the diversity of difference distribution under different cooling conditions. Therefore, this embodiment models all number differences based on the Gaussian mixture distribution probability density function, extracts the second-order derivative of its change rate through the change trend of the cumulative probability density function, and finds the point with the largest absolute value of the derivative among all possible positions as the threshold, so as to ensure that the threshold is exactly in the area with the most drastic transition of the probability density. This point often corresponds to the dividing position between the distribution peaks, that is, the jump boundary, which helps to eliminate interference values ​​and optimize the calculation boundary of the consistency index.

[0113] A method for generating a consistency index set based on the angle mean and angle range can be to take the angle mean as the center value, and set upper and lower limits in combination with the angle range, where the limit range can be determined by the angle mean minus half of the angle range and the angle mean plus half of the angle range; traverse all the angle means in the grain direction set, and mark the angle means that fall within the limit range as consistent samples; count the number of consistent samples for all principal axis direction vectors in segmented order according to the numbering sequence, and calculate the proportion of consistent samples in each segment as the direction consistency score value of the segment, thereby generating a consistency index set.

[0114] S40: performing matrix joint matching processing according to the cooling rate matrix and the consistency index set to construct a cooling rate structure offset matrix; extracting a mutation index set according to the spatial distribution mutation of the consistency index in the cooling rate structure offset matrix, mapping the abnormal grain region according to the mutation index set to obtain an abnormal grain region set;

[0115] A method for constructing a cooling rate structure offset matrix by performing matrix joint matching processing based on a cooling rate matrix and a consistency index set includes:

[0116] According to the matrix row index order of each cooling segment in the cooling rate matrix, the corresponding matrix row is extracted for each cooling segment, the average value of all elements in the cooling rate matrix is ​​calculated, and the average cooling rate value of the cooling segment is obtained. The average cooling rate values ​​of all cooling segments are arranged in sequence according to the original numbering order of the matrix row index, and a linear vector structure is constructed to generate a cooling rate sequence;

[0117] Construct a consistency indicator sequence according to the order of extracting the consistency indicator values ​​of each segment in the consistency indicator set;

[0118] The cooling rate sequence and the consistency index sequence are paired within the cooling segment according to the cooling segment number to construct a joint feature vector set. The vectors corresponding to two adjacent segments in the joint feature vector set are subtracted to calculate the cooling rate change difference and the consistency index change difference respectively.

[0119] The cooling rate change difference and the consistency index change difference are spliced ​​by dimension to form a two-dimensional offset vector. After summarizing all vectors, a two-dimensional offset set is formed. According to the index position of each two-dimensional vector in the two-dimensional offset set, a cooling rate structure offset matrix is ​​constructed.

[0120] It can be understood that in order to achieve the correlation modeling between the cooling rate parameters and the grain direction consistency index, this embodiment introduces matrix joint matching processing in step S40 to construct a cooling rate structure offset matrix that reflects the dynamic change relationship between the two. This step first extracts the row data corresponding to each cooling segment according to the matrix row index based on the structural characteristics of the cooling rate matrix, and calculates the average value of all elements in the row data to obtain the average cooling rate value of the cooling segment. For example, if the value of the cooling segment in the 5th row is [2.1, 2.3, 2.0], then its average value is (2.1+2.3+2.0) / 3=2.13, which is a typical representative of the overall heat exchange intensity of the cooling segment. The average cooling rate values ​​of all cooling segments are arranged in the order of the original row index, that is, a cooling rate sequence is constructed, which can be regarded as a linear expression of the temperature control process.

[0121] Furthermore, a consistency index sequence is generated based on the order of extracting the consistency scores of each segment in the consistency index set. For example, the score value of the 5th segment is 0.78 and that of the 6th segment is 0.55, indicating that the distribution of the main axis direction of the grains in the 5th segment is relatively concentrated, and the direction fluctuation of the 6th segment is relatively large. This consistency index sequence provides a measurement reference for the impact of the cooling stage on the stability of the grain structure. Subsequently, a number alignment operation is performed on the above-mentioned cooling rate sequence and the consistency index sequence, that is, by indexing the same segment number, the average cooling rate value and the consistency score value of the corresponding segment in the two are combined into a two-dimensional feature pair, for example: (2.13, 0.78), indicating that the grain direction consistency of the 5th segment is 0.78 under the condition of a cooling rate of 2.13. Multiple such feature pairs can be constructed as a joint feature vector set.

[0122] It should be noted that to further obtain the state change trend between adjacent cooling segments, this embodiment performs a subtraction process between any two adjacent segments in the joint feature vector set, and calculates their cooling rate difference and consistency score difference, respectively. For example, segment 5 is (2.13, 0.78), and segment 6 is (1.98, 0.55), with a difference of (-0.15, -0.23). This difference reflects the dynamic impact of changes in cooling conditions on grain orientation consistency. After completing the difference calculation for all adjacent segments, this embodiment concatenates each set of differences into a two-dimensional offset vector to form a two-dimensional state change expression. For example, (-0.15, -0.23) indicates that the cooling rate decreases by 0.15 and the consistency decreases by 0.23, indicating that changes in heat exchange significantly affect structural stability. All two-dimensional offset vectors are aggregated and organized according to their source segment index position to generate a cooling rate-structure offset matrix. This matrix not only quantitatively describes the linkage effect of cooling rate changes on grain consistency distribution but also provides a structured basis for subsequent spatial identification of abnormal areas.

[0123] The method for performing intra-segment pairing operation on the cooling rate sequence and the consistency index sequence according to the cooling segment number to construct a joint feature vector set includes:

[0124] According to the average cooling rate value of each cooling segment in the cooling rate sequence, the corresponding segment number index is extracted to generate a cooling number sequence;

[0125] According to each consistency score value in the consistency index sequence, the corresponding segment number index is extracted to generate a consistency number sequence;

[0126] Perform index alignment processing on the cooling number sequence and the consistency number sequence, extract the cooling rate value and consistency score value with consistent index, and combine the consistent cooling rate value and consistency score value into a two-dimensional feature vector pair;

[0127] Arrange all two-dimensional feature vector pairs in ascending order of segment numbers and construct a joint feature vector set.

[0128] It can be understood that this step aims to match the feature data from two different sources, namely the cooling rate sequence and the consistency index sequence, one-to-one through a unified numbering index to construct a two-dimensional joint feature structure required for subsequent modeling. Specifically, first, according to the average cooling rate value of each cooling segment in the cooling rate sequence, the segment number index of its position is extracted to generate a cooling number sequence. This number sequence is essentially the row index sequence of the cooling rate matrix, reflecting the original order and unique identification of the cooling segment. Similarly, according to each consistency score value in the consistency index sequence, its segment number in the original sequence is extracted to generate a consistency number sequence, which is used to identify the segment source of each consistency score value to ensure that its number has a one-to-one correspondence with the number in the cooling rate sequence.

[0129] Furthermore, index alignment processing is performed on the cooling number sequence and the consistency number sequence, and all cooling rate values ​​and consistency score values ​​with the same number are extracted, and they are combined to construct a two-dimensional feature vector pair. Each vector pair consists of a set of cooling rate values ​​and consistency score values, forming a joint feature unit for expressing the degree of change of grain direction in this cooling process. Taking a practical example, if the cooling number sequence is [1,2,3], the corresponding cooling rate value is [12.3,11.7,10.9], the consistency number sequence is [1,2,3], and the consistency score value is [0.86,0.79,0.74], then it can be constructed. The three sets of joint vectors are [(12.3, 0.86), (11.7, 0.79), (10.9, 0.74)]. This structure can simultaneously reflect the coupling between the cooling efficiency and directional consistency of a certain section, and is the necessary input data format for the subsequent calculation of the cooling rate structure offset. It should be noted that the set of joint feature vectors established by the numbering index alignment mechanism has structural unity and numbering closed-loop properties, which can effectively avoid data mismatch problems caused by segment order disorder or index loss, and ensure that in the subsequent offset calculation and spatial mutation extraction, the pairing results of the cooling state and grain consistency information are highly consistent and traceable.

[0130] The method of performing subtraction processing on vectors corresponding to two adjacent segments in the joint feature vector set and respectively calculating the cooling rate change difference and the consistency index change difference includes:

[0131] ;

[0132] Where, Indicates the Paragraph and Section two-dimensional offset vector between segments, Indicates the cooling rate sequence The average cooling rate value of the segment, Indicates the consistency index sequence The directional consistency score of the segment, Indicates the difference in cooling rate between adjacent segments, Indicates the difference in consistency score changes between adjacent segments.

[0133] Methods for obtaining spatial distribution mutations of consistency indicators in the cooling rate structure offset matrix include:

[0134] According to the position index of each row in the cooling rate structure offset matrix, the consistency index change difference in each row is extracted, and the consistency index change difference is placed in the two-dimensional coordinate position corresponding to its row index to form a consistency distribution matrix;

[0135] For each coordinate position in the consistency distribution matrix, calculate the coordinate difference between the coordinate position and the four adjacent upper, lower, left and right positions to form a local difference matrix;

[0136] Perform mean and variance calculations on all coordinate differences in the local difference matrix to obtain the local change mean and local change variance;

[0137] When the local change mean is greater than the preset local change threshold, the coordinate position is directly marked as a spatial distribution mutation. When the local change mean is less than or equal to the preset local change threshold, it is further determined whether the local change variance is greater than the preset local variance threshold. If so, the coordinate position is marked as a spatial distribution mutation.

[0138] It can be understood that in order to extract the location area with drastic changes in directional consistency from the cold rate structure offset matrix, it is necessary to construct a consistency distribution matrix to reveal the overall spatial distribution trend. Specifically, according to the position index of each row in the cold rate structure offset matrix, the consistency index change difference in the row is extracted, and the difference is filled in the corresponding two-dimensional coordinate position. For example, if the 5th row represents the offset change from the 10th segment to the 11th segment, and the row index is mapped to the (2,3)th coordinate in the matrix, the change difference of 0.18 is filled in the (2,3) position, thereby constructing a spatial position matrix. The consistency distribution matrix is ​​indexed and takes the indicator change as the value. This matrix converts the offset information into an image-like distribution structure, which is convenient for subsequent mutation detection. Furthermore, to determine whether there is a drastic change in the consistency indicator in the local area, this embodiment calculates the difference between each coordinate position in the consistency distribution matrix and its four adjacent positions above, below, left and right to form a local difference matrix. This processing logic is equivalent to performing a four-neighborhood difference estimation on a two-dimensional matrix, reflecting the change gradient between the target position and the neighborhood. For example, the value of the (3,4) position is 0.85, and the values ​​of the four adjacent positions are:

[0139] ;

[0140] The corresponding difference set is , which reflects the difference level of the point relative to the neighborhood and is the prerequisite data for detecting boundary disturbances or discontinuous areas.

[0141] It should be noted that in order to quantify the aggregation trend and fluctuation intensity of local differences, this embodiment calculates the mean and variance of the above difference set, where the mean represents the overall directional change level of the point, and the variance represents the discrete degree of the change value. For example, when the difference set is [0.05, 0.03, 0.02, 0.02], the corresponding mean is 0.03 and the variance is 0.0002. This statistical value reflects that the consistency change of the position is small and the fluctuation is stable. This calculation process provides a quantitative basis for whether to determine whether it is a mutation point in the future, avoiding misjudgment caused by a single point extreme value. It can be understood that in order to determine whether a certain coordinate position is spatially distributed, To determine the spatial mutation point, this embodiment introduces a dual-threshold judgment mechanism: first, it is judged whether the local change mean is greater than the preset local change threshold. For example, if the threshold is set to 0.04 and the mean of the point is 0.05, the point can be directly marked as a spatial mutation point; if the mean does not exceed the threshold, such as 0.03, then the second judgment stage is entered to compare whether its variance exceeds the preset local variance threshold. For example, if the variance threshold is set to 0.005 and the variance of the point is 0.008, it is also regarded as a spatial mutation point. This dual judgment mechanism can simultaneously identify the situation of overall directional consistency deviation and local disturbance enhancement. For example, a set of local difference values ​​is:

[0142] [0.01, 0.01, 0.02, 0.08], with a mean of 0.03 but a variance of 0.007, can be identified as a point of severe boundary disturbance, laying an accurate foundation for the identification of abnormal grain areas.

[0143] In this embodiment, the logic for extracting a mutation index set based on the spatial distribution mutation of the consistency index in the cooling rate structure offset matrix is ​​as follows: first, using the consistency distribution matrix generated in the aforementioned step, the judgment result of whether each coordinate position constitutes a spatial distribution mutation is filled in the corresponding two-dimensional position in the form of a Boolean value to construct a mutation Boolean matrix. The Boolean value of each coordinate position in the mutation Boolean matrix indicates whether the position meets the mutation condition. Then, for all coordinate positions with true values ​​in the mutation Boolean matrix, the row index and column index are extracted according to their two-dimensional coordinate format, and the coordinate position is used as the position record of a spatial mutation event. Then, all position indexes that meet the mutation condition are organized in row-first or column-first order to generate a mutation index set. The mutation index set is essentially a spatial position set of abnormal change areas analyzed from the consistency distribution matrix, which is used for subsequent mapping of abnormal grain areas in microscopic images. It has a clear two-dimensional coordinate structure, a limited number, and strong directionality. This completes the conversion operation from spatial mutation results to locatable and traceable structural units to form a mutation index set.

[0144] It should be noted that in order to ensure that each data unit in the cooling rate structure offset matrix can accurately correspond to the spatial area of ​​the microscopic image, this embodiment establishes a mapping rule from the matrix row index to the two-dimensional coordinates of the image. The rule is based on the grid-slicing method of the microscopic image, dividing the entire image into multiple cooling area units of equal size, and numbering them in order from top to bottom and from left to right. On this basis, the cooling segment number represented by each matrix row index can be converted into its grid position in the image. For example, when the image is divided into several columns horizontally, each row index can be determined by integer division and remainder operations to determine its longitudinal and lateral positions, respectively, to achieve a one-to-one spatial positioning relationship. Through this mapping method, the cooling state changes in the matrix structure can be accurately associated with the grain structure in the microscopic image, ensuring that the subsequent spatial mutation detection results can be accurately located in the image area, thereby supporting the accurate identification of abnormal grain areas and subsequent parameter tracking.

[0145] According to the logic of mapping abnormal grain areas according to the mutation index set, each coordinate position in the mutation index set is used as the positioning entry of the grain state distribution structure in the cooling rate structure offset matrix, and the corresponding local grain image area in the microscopic image is extracted according to the index position. The known grain number information is extracted in each local grain image area, and all grain numbers are deduplicated and summarized to construct an abnormal number set. The grain numbers in the abnormal number set are mapped to the main axis direction list and the consistency index set. For example, by recording the corresponding index positions of the grain number and the cooling section number in the main axis direction list, number consistency and rapid mapping are achieved, thereby completing the labeling and tracking of the abnormal grain area, and obtaining the abnormal grain area set as a reference basis for subsequent hot-dip galvanizing process parameter adjustment.

[0146] S50: According to the cooling index value in the abnormal grain region set, a target cooling parameter corresponding to the cooling index value is selected from the initial cooling parameters.

[0147] In this embodiment, the logic of screening the target cooling parameters corresponding to the cooling index values ​​from the initial cooling parameters according to the cooling index values ​​in the abnormal grain area set is: first, based on the abnormal grain area set obtained in the previous step, number each abnormal grain, find the corresponding cooling segment number in the spindle direction list or consistency index set, and deduplicate and standardize all corresponding cooling segment numbers to construct a cooling index set.

[0148] Next, each cooling section number in the cooling index set is used as a retrieval condition to screen and process the source data constituting the cooling rate matrix in the initial cooling parameters, including: extracting the temperature time period corresponding to the cooling section number in the furnace zone temperature change sequence, extracting the wind speed control value at the corresponding position in the wind speed adjustment sequence, and extracting the physical segmentation parameters consistent with the number in the cooling section. Then, the furnace temperature sequence fragment, wind speed adjustment sequence fragment and cooling section boundary corresponding to each cooling section number are combined into a target cooling parameter triplet. Finally, all target cooling parameter triplets are organized in order of cooling section numbers to form a target cooling parameter set, which provides targeted support for the reconstruction of local cooling strategies and parameter optimization in the subsequent hot-dip galvanizing process.

[0149] Example 2

[0150] See also Figure 2 As shown, based on the same inventive concept, this embodiment discloses an artificial intelligence-based automobile hot-dip galvanizing process parameter optimization system. For details not provided in this embodiment, please refer to the description of the relevant parts in Example 1. The system includes:

[0151] Data acquisition module: used to obtain the initial cooling parameters and microscopic images in the hot-dip galvanizing process, perform window partitioning on the initial cooling parameters to obtain the cooling rate matrix, and the microscopic images are used to characterize the arrangement of grains in the spatial region;

[0152] The first processing module is used to perform principal axis direction extraction processing on the microscopic image according to the cooling rate matrix to obtain a grain direction set;

[0153] The second processing module is used to perform consistency calculation processing based on the angle between any two main axis direction vectors in the grain direction set, obtain the angle mean and angle range, and generate a consistency index set based on the angle mean and angle range;

[0154] Region detection module: used to perform matrix joint matching processing based on the cooling rate matrix and the consistency index set to construct a cooling rate structure offset matrix; extract the mutation index set based on the spatial distribution mutation of the consistency index in the cooling rate structure offset matrix, and map the abnormal grain area based on the mutation index set to obtain the abnormal grain area set;

[0155] Parameter optimization module: used to select target cooling parameters corresponding to the cooling index values ​​from the initial cooling parameters according to the cooling index values ​​in the abnormal grain area set.

[0156] The detailed description set forth above in conjunction with the accompanying drawings describes examples and does not represent all examples that can be implemented or fall within the scope of the claims. The terms "example" and "exemplary" when used in this specification mean "used as an example, instance or illustration" and do not mean "better than or better than other examples."

[0157] Reference throughout this specification to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present invention. Therefore, use of these phrases may refer to more than just one embodiment, and further, the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0158] It should also be noted that these embodiments may be described as a process depicted as a flowchart, structure diagram, or block diagram, and that although the flowchart may describe the operations as a sequential process, many of the operations can be performed in parallel or concurrently, and the order of the operations may be rearranged.

Claims

1. An artificial intelligence-based method for optimizing process parameters of automotive hot-dip galvanizing, characterized in that: include: The initial cooling parameters and microscopic images of the hot-dip galvanizing process are obtained. Window partitioning is performed on the initial cooling parameters to obtain a cooling rate matrix. The microscopic images are used to characterize the arrangement of grains in the spatial region. Performing principal axis direction extraction processing on the microscopic image according to the cooling rate matrix to obtain a grain direction set; According to the angle between any two main axis direction vectors in the grain direction set, consistency calculation processing is performed to obtain the angle mean and angle range, and a consistency index set is generated based on the angle mean and angle range; Perform matrix joint matching processing based on the cooling rate matrix and the consistency index set to construct a cooling rate structure offset matrix; extract a mutation index set based on the spatial distribution mutation of the consistency index in the cooling rate structure offset matrix, and map the abnormal grain area based on the mutation index set to obtain an abnormal grain area set; According to the cooling index values ​​in the abnormal grain region set, target cooling parameters corresponding to the cooling index values ​​are screened from the initial cooling parameters.

2. The method for optimizing automobile hot-dip galvanizing process parameters based on artificial intelligence according to claim 1, characterized in that: The initial cooling parameters include a furnace zone temperature change sequence, a wind speed adjustment sequence, and a cooling segment. Window partitioning is performed on the initial cooling parameters to obtain a cooling rate matrix. The method includes: Perform sliding window division processing on the furnace zone temperature change sequence according to the time axis, extract the average temperature gradient within each sliding window, and aggregate the average temperature gradient by time period to generate a furnace temperature interval sequence; The wind speed adjustment sequence is mapped to the cooling section position, and the wind speed change rate sequence corresponding to each cooling section is extracted. The wind speed change rates are arranged in section order to form a wind speed interval matrix. According to the cooling section, the furnace temperature interval sequence and the wind speed interval matrix are synchronously aggregated, and the average temperature gradient and wind speed change rate in each cooling section are combined into a cooling feature vector set; Normalizing the cooling feature vector set to obtain a normalized vector, combining all normalized vectors in the order of cooling segments to generate a cooling feature matrix; According to the characteristic row corresponding to each cooling section in the cooling characteristic matrix, its average cooling rate value is calculated, and all average cooling rate values ​​are combined into a cooling rate matrix.

3. The method for optimizing automobile hot-dip galvanizing process parameters based on artificial intelligence according to claim 2, characterized in that: Methods for extracting the principal axis direction from a microscopic image according to a cooling rate matrix to obtain a set of grain directions include: Extracting grayscale edge contour data corresponding to each grain region in the microscopic image, and converting the grayscale edge contour data into a two-dimensional boundary coordinate point set; According to the average cooling rate value of each cooling section in the cooling rate matrix, a composite function transformation is performed to obtain a density threshold interval, and the coordinate point density of the local area of ​​each boundary point in the two-dimensional boundary coordinate point set is calculated. The local area of ​​the boundary point whose coordinate point density is greater than the upper limit of the density threshold interval is extracted as the grain core contour area; Performing a least squares fitting process on each grain core contour area to generate an ellipse model parameter set, the ellipse model parameter set including a major axis length, a minor axis length, and an inclination angle value; Extract the tilt angle value from the ellipse model parameter set, construct the main axis direction vector, organize all the main axis direction vectors by grain number index, and generate a main axis direction list; The main axis direction list is mapped back to the microscopic image space coordinates according to the grain number, forming a grain direction set with the pixel position as the index and the main axis direction as the value.

4. The method for optimizing automobile hot-dip galvanizing process parameters based on artificial intelligence according to claim 3, characterized in that: Methods for extracting the tilt angle value from the ellipse model parameter set and constructing the main axis direction vector include: ； Where, Indicates the The principal axis direction vector of each grain, represents the major axis tilt angle extracted from the ellipse model, represents the disturbance adjustment coefficient, Indicates the The cooling rate gradient of the cooling rate at the location of each grain in the lateral position.

5. The method for optimizing automobile hot-dip galvanizing process parameters based on artificial intelligence according to claim 3, characterized in that: Methods for performing consistency calculation based on the angle between any two principal axis direction vectors in the grain direction set to obtain the angle mean and angle range include: Based on each principal axis direction vector in the grain direction set, all principal axis direction vectors are organized in order of grain numbers to generate a principal axis direction vector sequence; Extract all adjacent direction vector pairs from the main axis direction vector sequence, perform inner product calculation on each adjacent direction vector pair, and convert the inner product value into an angle value through the arc cosine function to form an angle value sequence; According to the grain number corresponding to each main axis direction vector in the main axis direction vector sequence, a grain number sequence is constructed, adjacent difference calculation processing is performed on the grain number sequence to generate a number difference sequence, the number position in the number difference sequence that is greater than the jump threshold is used as an abnormal mark index, the angle value of the corresponding index position is extracted, and an angle abnormal value set is constructed, and the angle values ​​in the abnormal value set are removed from the angle value sequence to generate a denoised angle value set; Perform mean calculation processing on the denoised angle value set to obtain the arithmetic mean of all angle values ​​in the set to obtain the angle mean; perform range calculation processing on the denoised angle value set to extract the difference between the maximum and minimum values ​​to obtain the angle range.

6. The method for optimizing automobile hot-dip galvanizing process parameters based on artificial intelligence according to claim 5, characterized in that: The method of using the number position in the number difference sequence that is greater than the jump threshold as the abnormal mark index, extracting the angle value of the corresponding index position, and constructing the angle abnormal value set includes: Performing a difference calculation process on each pair of adjacent numbers in the grain number sequence to obtain a number difference sequence, wherein the number difference represents the position span between the current main axis direction vector and the previous main axis direction vector on the original grain number; Compare each number difference in the number difference sequence with a preset jump threshold, mark the number difference position greater than the jump threshold as an abnormal flag bit, and form an abnormal number index list; According to the index position in the abnormal number index list, the angle value at the corresponding position in the angle value sequence is extracted to generate an angle abnormal value set.

7. The method for optimizing automobile hot-dip galvanizing process parameters based on artificial intelligence according to claim 6, characterized in that: The logic for setting the jump threshold is as follows: ; in, represents the jump threshold, Indicates the current position index of the disturbance enhancement number difference sequence sorted by index, Indicates the number index position participating in the cumulative calculation, 、 、 Respectively represent The weights, means and variances of the Gaussian distribution components, represents the Gaussian probability density function, represents the first The difference in grain numbers, Indicates index at all positions In the equation, find the function value with the largest absolute value of the second-order derivative value, is the number of Gaussian components used to fit the distribution of number differences in the Gaussian mixture model.

8. The method for optimizing automobile hot-dip galvanizing process parameters based on artificial intelligence according to claim 6, characterized in that: A method for constructing a cooling rate structure offset matrix by performing matrix joint matching processing based on a cooling rate matrix and a consistency index set includes: According to the matrix row index order of each cooling segment in the cooling rate matrix, the corresponding matrix row is extracted for each cooling segment, the average value of all elements in the cooling rate matrix is ​​calculated, and the average cooling rate value of the cooling segment is obtained. The average cooling rate values ​​of all cooling segments are arranged in sequence according to the original numbering order of the matrix row index, and a linear vector structure is constructed to generate a cooling rate sequence; Construct a consistency indicator sequence according to the order of extracting the consistency indicator values ​​of each segment in the consistency indicator set; The cooling rate sequence and the consistency index sequence are paired within the cooling segment according to the cooling segment number to construct a joint feature vector set. The vectors corresponding to two adjacent segments in the joint feature vector set are subtracted to calculate the cooling rate change difference and the consistency index change difference respectively. The cooling rate change difference and the consistency index change difference are spliced ​​by dimension to form a two-dimensional offset vector. After summarizing all vectors, a two-dimensional offset set is formed. According to the index position of each two-dimensional vector in the two-dimensional offset set, a cooling rate structure offset matrix is ​​constructed.

9. The method for optimizing automobile hot-dip galvanizing process parameters based on artificial intelligence according to claim 8, characterized in that: The method for performing intra-segment pairing operation on the cooling rate sequence and the consistency index sequence according to the cooling segment number to construct a joint feature vector set includes: According to the average cooling rate value of each cooling segment in the cooling rate sequence, the corresponding segment number index is extracted to generate a cooling number sequence; According to each consistency score value in the consistency index sequence, the corresponding segment number index is extracted to generate a consistency number sequence; Perform index alignment processing on the cooling number sequence and the consistency number sequence, extract the cooling rate value and consistency score value with consistent index, and combine the consistent cooling rate value and consistency score value into a two-dimensional feature vector pair; Arrange all two-dimensional feature vector pairs in ascending order of segment numbers and construct a joint feature vector set.

10. An artificial intelligence-based automobile hot-dip galvanizing process parameter optimization system, which is used to implement an artificial intelligence-based automobile hot-dip galvanizing process parameter optimization method according to any one of claims 1 to 9, characterized in that: include: Data acquisition module: used to obtain the initial cooling parameters and microscopic images in the hot-dip galvanizing process, perform window partitioning on the initial cooling parameters to obtain the cooling rate matrix, and the microscopic images are used to characterize the arrangement of grains in the spatial region; The first processing module is used to perform principal axis direction extraction processing on the microscopic image according to the cooling rate matrix to obtain a grain direction set; The second processing module is used to perform consistency calculation processing based on the angle between any two main axis direction vectors in the grain direction set, obtain the angle mean and angle range, and generate a consistency index set based on the angle mean and angle range; Region detection module: used to perform matrix joint matching processing based on the cooling rate matrix and the consistency index set to construct a cooling rate structure offset matrix; extract the mutation index set based on the spatial distribution mutation of the consistency index in the cooling rate structure offset matrix, and map the abnormal grain area based on the mutation index set to obtain the abnormal grain area set; Parameter optimization module: used to select target cooling parameters corresponding to the cooling index values ​​from the initial cooling parameters according to the cooling index values ​​in the abnormal grain area set.

Images