A data-driven method for controlling the sensitivity of ultrasonic testing before tempering of large rotor forgings

Abstract

This invention belongs to the field of ultrasonic testing technology for large rotor forgings, and relates to a data-driven method for controlling the ultrasonic testing sensitivity of large rotor forgings before heat treatment. The method includes the following steps: preprocessing the collected data to form a standardized dataset; performing correlation analysis on the standardized dataset to screen key parameters; conducting Pearson correlation analysis on the key parameters to determine the key influencing parameters of ultrasonic testing sensitivity; constructing a random forest regression benchmark fitting model based on the key influencing parameters and conducting partial dependency graph analysis; and using an improved adaptive weighted K-means clustering algorithm to perform clustering analysis and extract parameter combinations suitable for different detection regions. This invention deeply explores the coupling correlation between process parameters, forging characteristics, and ultrasonic testing sensitivity, optimizes the clustering algorithm to achieve precise parameter optimization, and realizes accurate and adaptive control of ultrasonic testing sensitivity, thereby improving the accuracy and efficiency of ultrasonic testing of forgings before heat treatment.

Description

Technical Field

[0001] This invention belongs to the field of ultrasonic testing technology for large rotor forgings, and particularly relates to a data-driven method for controlling the sensitivity of ultrasonic testing of large rotor forgings before heat treatment. Background Technology

[0002] Low-pressure rotors are core load-bearing components of energy equipment such as thermal power, nuclear power, and heavy-duty gas turbines. A single rotor can weigh up to hundreds of tons. The extreme service conditions of low-pressure rotor forgings, including high temperature, high pressure, high speed, and alternating load, result in strict requirements for the internal quality of the low-pressure rotor forgings. Ultrasonic testing before tempering is a key quality control point for eliminating defects in advance and avoiding waste in subsequent processing. It is also a core testing method that is mandatory by industry standards. However, the characteristics of low-pressure rotors after forging, such as coarse and uneven microstructure, severe sound attenuation, and poor surface coupling conditions, pose a great challenge to the precise control of ultrasonic testing sensitivity.

[0003] Currently, the adjustment of ultrasonic testing sensitivity relies on manual experience, lacks quantitative data support, and fails to integrate actual testing data to explore the correlation between process parameters and sensitivity. Parameter combinations exhibit strong randomness, and the industry has yet to develop a data-driven adaptive adjustment method for ultrasonic testing sensitivity specifically for large rotor forgings before heat treatment. With the development of intelligent manufacturing, data-driven and intelligent testing have become important upgrade directions in the field of nondestructive testing. Against this backdrop, there is an urgent need to establish a data-driven ultrasonic testing sensitivity adjustment method to address the pain points of traditional technologies, improve the reliability and accuracy of ultrasonic testing of large rotor forgings before heat treatment, and provide technical support for the high-quality manufacturing of large forgings in high-end equipment. Summary of the Invention

[0004] Based on the above analysis, the present invention aims to provide a data-driven method for controlling the sensitivity of ultrasonic testing before tempering of large rotor forgings, in order to solve the problems that the control of ultrasonic testing sensitivity before tempering of large rotor forgings relies on human experience, lacks quantitative data support, and has low reliability and accuracy.

[0005] This invention provides a data-driven method for controlling the sensitivity of ultrasonic testing before tempering of large rotor forgings, comprising the following steps: Step 1: Collect multi-data during the ultrasonic testing process of the rotor forging before heat treatment. The multi-data includes: rotor forging body feature data, rotor forging process parameters, and ultrasonic testing sensitivity. Preprocess the collected multi-data to form a standardized dataset. Step 2: Perform correlation analysis on the standardized dataset to screen out key parameters that are highly correlated with ultrasound detection sensitivity; Step 3: Conduct Pearson correlation analysis on the selected key parameters to determine the key parameters affecting the sensitivity of ultrasound detection; Step 4: Construct a random forest regression benchmark fitting model based on the key influencing parameters and conduct partial dependency graph analysis; Step 5: Based on the correlation ranking in Step 2, assign weights to each key influencing parameter, and use the improved adaptive weighted K-means clustering algorithm to perform cluster analysis and extract parameter combinations suitable for different detection regions.

[0006] Furthermore, step 1 specifically includes: 1.1 Select rotor forgings of specific materials as the research object and collect multivariate data; wherein, the rotor forging process parameters include: forging parameters and heat treatment parameters; 1.2 Perform missing value and outlier handling on the collected multi-data elements; 1.3 Initialization processing is used to perform dimensionless processing on the collected multi-data elements to obtain a standardized dataset.

[0007] Furthermore, step 2 specifically includes: 2.1 Using ultrasonic testing sensitivity as the reference sequence and rotor forging body feature data and rotor forging process parameters as the comparison sequence, an original data matrix is ​​formed; 2.2 Calculate the global absolute difference to obtain the global absolute difference matrix; 2.3 Determine the global minimum difference and the global maximum difference; 2.4 Introduce the resolution coefficient and calculate the correlation coefficient; 2.5 Calculate and sort the correlation.

[0008] Furthermore, step 2 specifically includes: 2.1 Using ultrasonic testing sensitivity as the reference sequence and rotor forging body characteristic data and rotor forging process parameters as the comparison sequence, the standardized reference sequence is as follows: The comparison sequence is ,in, To compare the number of sequences, The number of samples; 2.2 Calculate the absolute difference between the reference sequence and each comparison sequence for each sample: By iterating through all sample data, the global absolute difference matrix is ​​obtained; 2.3 Traverse the global absolute difference matrix to determine the global minimum difference. Global maximum difference ; 2.4 Introduction of resolution coefficient Calculate the correlation coefficient using the following formula:

[0009] 2.5 Grey Relationship between Various Parameters and Sensitivity ,according to Sort from largest to smallest, and select... The parameters are used as key parameters. The parameters are discarded as redundant parameters.

[0010] Furthermore, step 3 specifically includes: 3.1 Calculate the Pearson correlation coefficient between the key parameters and the ultrasonic detection sensitivity; 3.2 Based on the results of Pearson correlation analysis, the key parameters affecting the sensitivity of ultrasound detection were determined.

[0011] Furthermore, step 4 specifically includes: 4.1 Using the key influencing parameters selected in step 3 as input features and ultrasonic detection sensitivity as the output target, a random forest regression benchmark fitting model is established; 4.2 Based on the aforementioned random forest regression benchmark fitting model, calculate the partial dependence, plot the partial dependence curve with the values ​​of key influencing parameters on the horizontal axis and the corresponding partial dependence prediction values ​​of ultrasound detection sensitivity on the vertical axis.

[0012] Furthermore, in step 4.2, the partial dependency core calculation formula is as follows: for the parameter to be analyzed Iterate through all actual values ​​of this parameter in the dataset, keeping the other parameters unchanged, and calculate the mean predicted ultrasound detection sensitivity of the baseline model. The formula is: ,in, For parameters Values The partial dependency value at time, As the baseline fitting model, For the first In each sample, except Other parameter values, This represents the total sample size.

[0013] Furthermore, step 5 specifically includes: 5.1 The K-means++ algorithm is used to initialize the cluster centers; 5.2 Based on the correlation ranking obtained from the correlation analysis in step 2, weights are assigned to each key influencing parameter. The weights are positively correlated with the correlation between each key influencing parameter and sensitivity, as shown in the formula: ,in , The weight of the j-th key influencing parameter; 5.3 The optimal number of clusters k is determined by combining the silhouette coefficient and the Davidson-Bolding index; wherein the optimal number of clusters k is selected based on the k value with the largest silhouette coefficient and the smallest Davidson-Bolding index. 5.4 Run the improved adaptive weighted K-means clustering algorithm to cluster the standardized dataset, obtain k parameter combination clusters, and extract the optimal parameter combination.

[0014] Furthermore, in step 5.3, the contour coefficient takes values ​​in the range [-1, 1], and the calculation formula is as follows: ,in For the sample The average distance to other samples within the same cluster. For the sample The average distance to all samples in the nearest neighbor cluster.

[0015] Furthermore, in step 5.3, the Davidson-Bolding index ranges from [0, +∞), and the calculation formula is as follows: ,in, For the first The average distance of all samples within a cluster to the cluster center. For the first The average distance of all samples within a cluster to the cluster center. For the first Cluster and the first Distance from the cluster center.

[0016] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects: 1. This invention constructs a multi-source heterogeneous dataset at the data level. Relying on the combined analysis of grey relational degree, Pearson correlation and partial dependence graph, it realizes the full-dimensional quantitative correlation mining of intrinsic data, process parameters and ultrasonic detection sensitivity, accurately locks the optimal threshold of a single parameter, completely gets rid of the dependence on manual experience and provides scientific data support for parameter optimization.

[0017] 2. This invention proposes an improved adaptive weighted K-means clustering algorithm at the algorithm level. It integrates the partial dependency analysis threshold to optimize the initial center and feature weights. By using K-means++ initialization and automatic selection of k by dual indicators, it solves the problems of local optima, low accuracy and poor adaptability of traditional clustering. It can accurately adapt to the detection needs of different regions and working conditions.

[0018] 3. The present invention has a complete system that can be directly adapted to the testing of low-pressure rotors in thermal power and nuclear power plants. At the same time, it can be quickly extended to various large shaft forgings before heat treatment and ultrasonic testing scenarios. It has a wide range of applications and lays a solid foundation for the subsequent establishment of multi-parameter predictive ultrasonic testing sensitivity.

[0019] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained through the details specifically pointed out in the description and drawings. Attached Figure Description

[0020] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0021] Figure 1 This is a schematic flowchart of the method of the present invention; Figure 2 This is a partial dependence diagram of the six key influencing parameters selected in this embodiment of the invention on the ultrasonic detection sensitivity; Figure 3 This is a distribution diagram of the cluster analysis results of the five key influencing parameters selected in the embodiments of the present invention. Detailed Implementation

[0022] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of the embodiments, and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Furthermore, unless otherwise specified, the raw materials mentioned below are all commercially available products; and the process steps or preparation methods not mentioned in detail are all process steps or preparation methods known to those skilled in the art.

[0023] This invention discloses a data-driven method for controlling the sensitivity of ultrasonic testing before tempering of large rotor forgings, such as... Figure 1 As shown, it includes the following steps: Step 1: Collect multi-data during the ultrasonic testing process of the rotor forging before tempering. The multi-data includes: rotor forging body feature data, rotor forging process parameters and ultrasonic testing sensitivity; preprocess the collected multi-data to form a standardized dataset.

[0024] Step 1 specifically includes: 1.1 Select rotor forgings of specific materials as the research object and collect multivariate data; wherein, the rotor forging process parameters include: forging parameters and heat treatment parameters; 1.2 Perform missing value and outlier handling on the collected multi-data elements; 1.3 Initialization processing is used to perform dimensionless processing on the collected multi-data elements to obtain a standardized dataset.

[0025] It should be noted that tempering refers to a combined heat treatment process of quenching and high-temperature tempering, aimed at improving the strength, toughness, and overall mechanical properties of the rotor shaft. Ultrasonic testing sensitivity refers to the smallest defect size that the flaw detection equipment can identify; higher sensitivity means smaller defects can be detected. For low-pressure rotor shafts, the material uniformity, grain size, and internal stress affect the propagation and reflection of ultrasonic waves, thus influencing the testing sensitivity. The more non-uniform the material and the larger the grains, the stronger the ultrasonic wave scattering and the lower the testing sensitivity.

[0026] In the field of ultrasonic testing of large rotor forgings, the ultrasonic testing sensitivity is usually based on the equivalent size of the flat-bottomed hole (such as φ1.6 mm, φ2.0 mm flat-bottomed hole equivalent), that is, the testing system can reliably identify the smallest defect in the workpiece that is equivalent to the reflected echo amplitude of a flat-bottomed hole of a certain size.

[0027] For example, in step 1.1, a low-pressure rotor forging made of 30Cr2Ni4MoV material is selected. Based on existing actual production data records, the selected body characteristic data and historical process parameters mainly include steel ingot shape, delivery unit weight, forging blank shaft diameter, actual number of heat treatments for finished products, first heat holding temperature, first heat holding time, second heat holding temperature, second heat holding time, third heat holding temperature, third heat holding time, fourth heat holding temperature, fourth heat holding time, deformation amount of the last heat treatment, number of normalizing treatments after forging, and the proportion of ingots with a capacity of 242T or more in the same furnace, totaling 15 parameters.

[0028] Specifically, in step 1.2, for a small number of missing parameter data, the mean value filling method is used to fill in the missing data. For missing heat treatment process parameters, such as the third heat treatment time and the third heat treatment temperature when the finished product is only heated twice, NaN is used to fill in the missing data to avoid the data incompleteness affecting the analysis.

[0029] Specifically, in step 1.3, since the units and numerical magnitudes of the parameters are different, standardization is required. Initialization is performed using the following formula: ,in, For the original value of the k-th sample, The value is the standardized value. After standardization, all generated data is uniformly mapped to the interval [0,1].

[0030] Step 2: Perform correlation analysis on the standardized dataset to screen out key parameters that are highly correlated with ultrasound detection sensitivity.

[0031] Step 2 specifically includes: 2.1 Using ultrasonic testing sensitivity as the reference sequence and rotor forging body feature data and rotor forging process parameters as the comparison sequence, an original data matrix is ​​formed; 2.2 Calculate the global absolute difference to obtain the global absolute difference matrix; 2.3 Determine the global minimum difference and the global maximum difference; 2.4 Introduce the resolution coefficient and calculate the correlation coefficient; 2.5 Calculate and sort the correlation.

[0032] It should be noted that, using the ultrasonic testing sensitivity in the standardized dataset as the reference sequence, and the rotor forging body feature data and rotor forging process parameters as the comparison sequence, the degree of correlation between each parameter and the ultrasonic testing sensitivity is quantitatively calculated through grey relational analysis. Key parameters with satisfactory correlation and significant impact are selected, while redundant parameters are eliminated.

[0033] More specifically, step 2 includes: 2.1 Using ultrasonic testing sensitivity as the reference sequence and rotor forging body characteristic data and rotor forging process parameters as the comparison sequence, the standardized reference sequence is as follows: The comparison sequence is ,in, To compare the number of sequences, The number of samples; It should be noted that, Specifically, it refers to the sample data sequence of ultrasound detection sensitivity. This refers to the nth sample data in the ultrasound detection sensitivity data sequence; Specifically, it refers to the sample data sequence of a specific parameter (the i-th parameter) in the rotor forging process parameters and the characteristic data of the rotor forging body. This represents the nth sample data in the i-th parameter sequence.

[0034] 2.2 Calculate the absolute difference between the reference sequence and each comparison sequence for each sample: By iterating through all sample data, the global absolute difference matrix is ​​obtained; 2.3 Traverse the global absolute difference matrix to determine the global minimum difference. Global maximum difference ; 2.4 Introduction of resolution coefficient Calculate the correlation coefficient using the following formula:

[0035] 2.5 Grey Relationship between Various Parameters and Sensitivity ,according to Sort from largest to smallest, and select... The parameters are used as key parameters. The parameters are discarded as redundant parameters.

[0036] Step 3: Conduct Pearson correlation analysis on the selected key parameters to determine the key parameters affecting the sensitivity of ultrasound detection.

[0037] Step 3 specifically includes: 3.1 Calculate the Pearson correlation coefficient between the key parameters and the ultrasonic detection sensitivity; 3.2 Based on the results of Pearson correlation analysis, the key parameters affecting the sensitivity of ultrasound detection were determined.

[0038] It should be noted that Pearson correlation analysis was conducted on the selected key parameters to calculate the correlation coefficient between the key parameters and the ultrasound detection sensitivity, clarify the trend of the influence of each parameter change on the ultrasound detection sensitivity, and further deepen the exploration of the linear correlation between parameters and sensitivity.

[0039] Specifically, in step 3.1, the Pearson correlation coefficient ranges from [-1, 1]. According to the rule that the closer the positive correlation coefficient is to 1 and the closer the negative correlation coefficient is to -1, the stronger the correlation. Specifically, in step 3.2, parameters with an absolute value of Pearson correlation coefficient greater than 0.3 are selected as key influencing parameters.

[0040] It should be noted that the key influencing parameters are those that simultaneously satisfy the requirements of the correlation analysis. Parameters whose absolute value of the Pearson correlation coefficient is greater than 0.3.

[0041] Step 4: Construct a random forest regression benchmark fitting model based on the key influencing parameters and conduct partial dependency graph analysis.

[0042] Step 4 specifically includes: 4.1 Using the key influencing parameters selected in step 3 as input features and ultrasonic detection sensitivity as the output target, a random forest regression benchmark fitting model is established; 4.2 Based on the aforementioned random forest regression benchmark fitting model, calculate the partial dependence, plot the partial dependence curve with the values ​​of key influencing parameters on the horizontal axis and the corresponding partial dependence prediction values ​​of ultrasound detection sensitivity on the vertical axis.

[0043] It should be noted that a random forest regression benchmark fitting model was constructed based on key influencing parameters, and a partial dependency graph analysis was carried out. The remaining parameters were fixed as the sample mean to eliminate coupling interference. The partial dependency of the single parameter on the sensitivity of ultrasound detection was calculated and the dependency curve was plotted to analyze the marginal influence trend of the single parameter on the sensitivity.

[0044] Specifically, in step 4.1, the method for establishing the random forest regression benchmark fitting model is as follows: in the Python runtime environment, such as Python 3.10, import RandomForestRegressor from the sklearn.ensemble library and train the model under the default hyperparameter settings; Specifically, in step 4.2, the partial dependency core calculation formula is as follows: for the parameter to be analyzed Iterate through all actual values ​​of this parameter in the dataset, keeping the other parameters unchanged, and calculate the mean predicted ultrasound detection sensitivity of the baseline model. The formula is: ,in, For parameters Values The partial dependency value at time, As the baseline fitting model, For the first In each sample, except Other parameter values, This represents the total sample size.

[0045] Step 5: Based on the correlation ranking in Step 2, assign weights to each key influencing parameter, and use the improved adaptive weighted K-means clustering algorithm to perform cluster analysis and extract parameter combinations suitable for different detection regions.

[0046] Step 5 specifically includes: 5.1 The K-means++ algorithm is used to initialize the cluster centers; 5.2 Based on the correlation ranking obtained from the correlation analysis in step 2, weights are assigned to each key influencing parameter. The weights are positively correlated with the correlation between each key influencing parameter and sensitivity, as shown in the formula: ,in , The weight of the j-th key influencing parameter; 5.3 The optimal number of clusters k is determined by combining the silhouette coefficient and the Davidson-Bolding index; wherein the optimal number of clusters k is selected based on the k value with the largest silhouette coefficient and the smallest Davidson-Bolding index. 5.4 Run the improved adaptive weighted K-means clustering algorithm to cluster the standardized dataset, obtain k parameter combination clusters, and extract the optimal parameter combination.

[0047] It should be noted that, using the assigned key influencing parameters and sensitivity indicators as clustering features, combined with partial dependency analysis, an improved adaptive weighted K-means clustering algorithm is used to find the optimal combination of process parameters. The optimal number of clusters is automatically determined by adaptive weight assignment and dual indicators to complete the clustering, screen out high-quality parameter combination clusters that meet the standards, and extract the optimal combination of process parameters suitable for different detection areas.

[0048] Specifically, in step 5.3, the contour coefficient ranges from [-1, 1], and the calculation formula is as follows: ,in For the sample The average distance to other samples within the same cluster. For the sample The average distance to all samples in the nearest neighbor cluster.

[0049] Specifically, in step 5.3, the Davidson-Bolding index ranges from [0, +∞), and the calculation formula is as follows: ,in, For the first The average distance of all samples within a cluster to the cluster center. For the first The average distance of all samples within a cluster to the cluster center. For the first Cluster and the first Distance from the cluster center.

[0050] Understandably, by employing the aforementioned control methods, key parameters influencing the ultrasonic testing sensitivity of rotor forgings can be obtained. Further partial dependency graph analysis and cluster analysis can then yield specific values ​​for the key parameters corresponding to the highest ultrasonic testing sensitivity.

[0051] Compared with existing technologies, this invention achieves full-dimensional quantitative correlation mining of intrinsic data of forgings, process parameters and ultrasonic testing sensitivity, accurately locks the optimal threshold of single parameters, completely eliminates reliance on manual experience and provides scientific data support for parameter optimization.

[0052] The following specific embodiments are provided for further analysis.

[0053] Example 1 A data-driven method for controlling the sensitivity of ultrasonic testing before tempering of large rotor forgings includes the following steps: 1.1 Low-pressure rotor forgings made of 30Cr2Ni4MoV were selected, and valid sample data were collected. Based on existing actual production data records, the main body characteristic data and historical process parameters selected were: ingot shape, delivery unit weight, forging blank shaft diameter, actual number of heat treatments for finished products, first heat holding temperature, first heat holding time, second heat holding temperature, second heat holding time, third heat holding temperature, third heat holding time, fourth heat holding temperature, fourth heat holding time, deformation amount of the last heat treatment, number of normalizing treatments after forging, and the proportion of ingots with a capacity of 242 tons or more in the same furnace, totaling 15 parameters. A total of 136 valid data entries were collected.

[0054] For example, a low-pressure rotor forging made of 30Cr2Ni4MoV has an ingot size of 242 T, a delivery unit weight of 139.9 T, a forged blank shaft diameter of 2002 mm, an actual finished product heat treatment of 2 times, a first heat treatment holding temperature of 1250℃ and a first heat treatment holding time of 24 h, a second heat treatment holding temperature of 1200 ℃ and a second heat treatment holding time of 15.3 h, a final heat treatment deformation of 50%, a post-forging heat treatment normalizing number of times of 3 times, an ingot size of 242 T or above in the same furnace accounting for 50%, and an ultrasonic testing sensitivity (flat bottom hole equivalent) of φ3.0 mm before tempering.

[0055] 1.2 For a small number of missing parameter data, the mean value filling method is used to fill in the missing data. For missing heat treatment process parameters, such as the third heat treatment time and the third heat treatment temperature when the finished product is only heated twice, NaN is used to fill in the missing data to avoid the data incompleteness affecting the analysis.

[0056] 1.3 Since the units and numerical magnitudes of the parameters are different, standardization is required. Initialization is used, and the formula is as follows: ,in, For the original value of the k-th sample, These are the standardized values. After standardization, all generated data are uniformly mapped to the interval [0,1]. This yields the standardized dataset.

[0057] 2.1 Using ultrasonic testing sensitivity as the reference sequence and rotor forging body characteristic data and rotor forging process parameters as the comparison sequence, the standardized reference sequence is as follows: The comparison sequence is ,in, To compare the number of sequences, The number of samples; 2.2 Calculate the absolute difference between the reference sequence and each comparison sequence for each sample: By iterating through all sample data, the global absolute difference matrix is ​​obtained; 2.3 Traverse the global absolute difference matrix to determine the global minimum difference. Global maximum difference ; 2.4 Introduction of resolution coefficient Calculate the correlation coefficient using the following formula:

[0058] 2.5 Grey Relationship between Various Parameters and Sensitivity ,according to Sort from largest to smallest, and select... The parameters are used as key parameters. The parameters are discarded as redundant parameters.

[0059] In this embodiment, the key parameters selected through grey relational analysis are: forging blank shaft diameter, second heat holding time, steel ingot shape, delivery unit weight, first heat holding time, proportion of ingots of 242T or more in the same furnace, second heat holding temperature, fourth heat holding temperature, fourth heat holding time, and third heat holding temperature.

[0060] 3.1 Calculate the Pearson correlation coefficients between the above key parameters and the ultrasonic detection sensitivity; 3.2 Selection of correlation analysis Furthermore, parameters with an absolute value of Pearson correlation coefficient greater than 0.3 are considered key parameters affecting the sensitivity of ultrasound detection.

[0061] In this embodiment, the key influencing parameters that were finally determined were the diameter of the forged blank shaft, the second heat holding time, the ingot shape, the third heat holding temperature, the delivery unit weight, and the first heat holding time.

[0062] 4.1 Using the key influencing parameters selected in step 3 as input features and ultrasound detection sensitivity as the output target, in a Python runtime environment, such as Python 3.10, import RandomForestRegressor from the sklearn.ensemble library, train the model under the default hyperparameter settings, and establish a random forest regression benchmark fitting model.

[0063] 4.2 Based on the aforementioned random forest regression benchmark fitting model, partial dependence is calculated. A partial dependence curve is plotted with the values ​​of key influencing parameters on the horizontal axis and the corresponding partial dependence prediction values ​​of ultrasound detection sensitivity on the vertical axis. The core formula for calculating partial dependence is: For the parameter to be analyzed... Iterate through all actual values ​​of this parameter in the dataset, keeping the other parameters unchanged, and calculate the mean predicted ultrasound detection sensitivity of the baseline model. The formula is: ,in, For parameters Values The partial dependency value at time, As the baseline fitting model, For the first In each sample, except Other parameter values, This represents the total sample size.

[0064] In this embodiment, the partial dependence graph of six key influencing parameters on ultrasonic testing sensitivity—forged blank shaft diameter, second-heat holding time, ingot shape, third-heat holding temperature, delivery unit weight, and first-heat holding time—is shown below. Figure 2 As shown.

[0065] Depend on Figure 2 It can be seen that as the diameter of the forged blank shaft and the shape of the steel ingot increase, the equivalent of the flat-bottom hole increases slowly, but the overall increase is low. This indicates that the larger the size of the forging, the more likely it is to have uneven internal structure and coarse grains, resulting in lower ultrasonic detection sensitivity. When the single delivery weight exceeds 160 T, the equivalent of the flat-bottom hole increases significantly, and the ultrasonic detection sensitivity decreases. The holding time of the first and second heats is positively correlated with the equivalent of the flat-bottom hole, with the increase of the holding time of the second heat having a more significant impact on the ultrasonic detection sensitivity. The holding temperature of the third heat is negatively correlated with the equivalent of the flat-bottom hole, indicating that the increase of the holding temperature of the third heat can refine the grains, thereby reducing internal defects in the forging, decreasing the equivalent of the flat-bottom hole, and increasing the ultrasonic detection sensitivity.

[0066] 5.1 The K-means++ algorithm is used to initialize the cluster centers.

[0067] 5.2 Based on the correlation ranking obtained from the correlation analysis in step 2, weights are assigned to each key influencing parameter. The weights are positively correlated with the correlation between each key influencing parameter and sensitivity, as shown in the formula: ,in , Let be the weight of the j-th key influencing parameter.

[0068] 5.3 Based on the maximum profile coefficient and the minimum Davidson-Bolding index, the final value of k is selected as 3.

[0069] 5.4 Run the improved adaptive weighted K-means clustering algorithm to cluster the standardized dataset, obtain three parameter combination clusters, and obtain the optimal parameter combination.

[0070] In this embodiment, since the insulation temperature of the three-heater is mostly filled with NaN, this type of data cannot be clustered. Therefore, the final clustering analysis results of the five key influencing parameters are as follows: Figure 3 As shown, the horizontal axis (Cluster) represents 3 combined clusters.

[0071] According to the cluster analysis results, Cluster2 has the smallest flat-bottomed hole equivalent φ3.09 mm, meaning it has the highest ultrasonic detection sensitivity. Figure 3The cluster analysis results output the mean values ​​of the parameters corresponding to cluster2, and obtain the values ​​of the five key influencing parameters: the mean diameter of the forged blank shaft is 1998.3 mm, the mean second heat holding time is 24.5 h, the mean steel ingot shape is 254.2 T, the mean delivery unit weight is 107.5 T, and the mean first heat holding time is 31 h.

[0072] Understandably, by employing the aforementioned control methods, six key influencing parameters affecting the ultrasonic testing sensitivity of rotor forgings can be obtained: the diameter of the forged blank shaft, the second heating holding time, the ingot shape, the third heating holding temperature, the delivery unit weight, and the first heating holding time. Further partial dependency graph analysis and cluster analysis can then yield the specific values ​​of the key influencing parameters corresponding to the highest ultrasonic testing sensitivity.

[0073] Referring to the values ​​of the above five key influencing parameters and the third-heat holding temperature (one of the key influencing parameters), the specific parameters of the preparation process were adjusted (where the diameter of the forged blank shaft is 1998.3 mm, the second-heat holding time is 24.5 h, the ingot size is 254.2 T, the delivery unit weight is 107.5 T, and the first-heat holding time is 31 h). When producing the 30Cr2Ni4MoV low-pressure rotor, the ingot size is 242 T, the diameter of the forged blank shaft is 2005 mm, the delivery unit weight is 102.9 T, the first-heat holding time is 30 h, the second-heat holding time is 24 h, and the third-heat holding temperature is 1200℃. After forging, ultrasonic equipment was used to perform flaw detection on the rotor before tempering, and the equivalent flat-bottom hole (ultrasonic detection sensitivity) was measured to be φ3.0 mm.

[0074] In actual production, due to the influence of parameters and conditions of equipment such as smelting, forging, and heat treatment furnaces, as well as process parameters of preceding and following processes (such as cold working allowance requirements), it is impossible to completely and accurately match the recommended values ​​for parameters such as steel ingot shape and delivery unit weight. Therefore, adjustments should be made accordingly based on the situation.

[0075] According to the test results, the sensitivity of ultrasonic testing can be precisely controlled by adjusting six key process parameters: the diameter of the forging blank shaft, the second heat holding time, the ingot shape, the third heat holding temperature, the delivery unit weight, and the first heat holding time.

[0076] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0077] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A data-driven method for controlling the sensitivity of ultrasonic testing before tempering of large rotor forgings, characterized in that, Includes the following steps: Step 1: Collect multi-data during the ultrasonic testing process of the rotor forging before heat treatment. The multi-data includes: rotor forging body feature data, rotor forging process parameters, and ultrasonic testing sensitivity. Preprocess the collected multi-data to form a standardized dataset. Step 2: Perform correlation analysis on the standardized dataset to screen out key parameters that are highly correlated with ultrasound detection sensitivity; Step 3: Conduct Pearson correlation analysis on the selected key parameters to determine the key parameters affecting the sensitivity of ultrasound detection; Step 4: Construct a random forest regression benchmark fitting model based on the key influencing parameters and conduct partial dependency graph analysis; Step 5: Based on the correlation ranking in Step 2, assign weights to each key influencing parameter, and use the improved adaptive weighted K-means clustering algorithm to perform cluster analysis and extract parameter combinations suitable for different detection regions.

2. The data-driven method for controlling the sensitivity of ultrasonic testing before tempering of large rotor forgings according to claim 1, characterized in that, Step 1 specifically includes: 1.1 Select rotor forgings of specific materials as the research object and collect multivariate data; wherein, the rotor forging process parameters include: forging parameters and heat treatment parameters; 1.2 Perform missing value and outlier handling on the collected multi-data elements; 1.3 Initialization processing is used to perform dimensionless processing on the collected multi-data elements to obtain a standardized dataset.

3. The data-driven method for controlling the sensitivity of ultrasonic testing before tempering of large rotor forgings according to claim 1, characterized in that, Step 2 specifically includes: 2.1 Using ultrasonic testing sensitivity as the reference sequence and rotor forging body feature data and rotor forging process parameters as the comparison sequence, an original data matrix is ​​formed; 2.2 Calculate the global absolute difference to obtain the global absolute difference matrix; 2.3 Determine the global minimum difference and the global maximum difference; 2.4 Introduce the resolution coefficient and calculate the correlation coefficient; 2.5 Calculate and sort the correlation.

4. The data-driven method for controlling the sensitivity of ultrasonic testing before tempering of large rotor forgings according to claim 1, characterized in that, Step 2 specifically includes: 2.1 Using ultrasonic testing sensitivity as the reference sequence and rotor forging body characteristic data and rotor forging process parameters as the comparison sequence, the standardized reference sequence is as follows: The comparison sequence is ,in, To compare the number of sequences, The number of samples; 2.2 Calculate the absolute difference between the reference sequence and each comparison sequence for each sample: By iterating through all sample data, the global absolute difference matrix is ​​obtained; 2.3 Traverse the global absolute difference matrix to determine the global minimum difference. Global maximum difference ; 2.4 Introduction of resolution coefficient Calculate the correlation coefficient using the following formula: 2.5 Grey Relationship between Various Parameters and Sensitivity ,according to Sort by largest to smallest, then select The parameters are used as key parameters. The parameters are discarded as redundant parameters.

5. The data-driven method for controlling the sensitivity of ultrasonic testing before tempering of large rotor forgings according to claim 1, characterized in that, Step 3 specifically includes: 3.1 Calculate the Pearson correlation coefficient between the key parameters and the ultrasonic detection sensitivity; 3.2 Based on the results of Pearson correlation analysis, the key parameters affecting the sensitivity of ultrasound detection were determined.

6. The data-driven method for controlling the sensitivity of ultrasonic testing before tempering of large rotor forgings according to claim 1, characterized in that, Step 4 specifically includes: 4.1 Using the key influencing parameters selected in step 3 as input features and ultrasonic detection sensitivity as the output target, a random forest regression benchmark fitting model is established; 4.2 Based on the aforementioned random forest regression benchmark fitting model, calculate the partial dependence, plot the partial dependence curve with the values ​​of key influencing parameters on the horizontal axis and the corresponding partial dependence prediction values ​​of ultrasound detection sensitivity on the vertical axis.

7. The data-driven method for controlling the sensitivity of ultrasonic testing before tempering of large rotor forgings according to claim 6, characterized in that, In step 4.2, the partial dependency core calculation formula is: for the parameter to be analyzed Iterate through all actual values ​​of this parameter in the dataset, keeping the other parameters unchanged, and calculate the mean predicted ultrasound detection sensitivity of the baseline model. The formula is: ,in, For parameters Values The partial dependency value at time, As the baseline fitting model, For the first In each sample, except Other parameter values, This represents the total sample size.

8. The data-driven method for controlling the sensitivity of ultrasonic testing before tempering of large rotor forgings according to claim 1, characterized in that, Step 5 specifically includes: 5.1 The K-means++ algorithm is used to initialize the cluster centers; 5.2 Based on the correlation ranking obtained from the correlation analysis in step 2, weights are assigned to each key influencing parameter. The weights are positively correlated with the correlation between each key influencing parameter and sensitivity, as shown in the formula: ,in , The weight of the j-th key influencing parameter; 5.3 The optimal number of clusters k is determined by combining the silhouette coefficient and the Davidson-Bolding index; wherein the optimal number of clusters k is selected based on the k value with the largest silhouette coefficient and the smallest Davidson-Bolding index. 5.4 Run the improved adaptive weighted K-means clustering algorithm to cluster the standardized dataset, obtain k parameter combination clusters, and extract the optimal parameter combination.

9. The data-driven method for controlling the sensitivity of ultrasonic testing before tempering of large rotor forgings according to claim 8, characterized in that, In step 5.3, the contour coefficient ranges from [-1, 1], and the calculation formula is as follows: ,in For the sample The average distance to other samples within the same cluster. For the sample The average distance to all samples in the nearest neighbor cluster.

10. The data-driven method for controlling the sensitivity of ultrasonic testing before tempering of large rotor forgings according to claim 8, characterized in that, In step 5.3, the Davidson-Bolding index ranges from [0, +∞), and the calculation formula is as follows: ,in, For the first The average distance of all samples within a cluster to the cluster center. For the first The average distance of all samples within a cluster to the cluster center. For the first Cluster and the first Distance from the cluster center.

Images