Neural network-based austenitic stainless steel mechanical property calculation method and system

Abstract

The invention provides an austenitic stainless steel mechanical property calculation method and system based on a neural network, and relates to the technical field of mechanical property analysis, the physical property of austenitic stainless steel at a standard temperature is measured, the influence of temperature, humidity and corrosion medium concentration on the change of the physical property is analyzed, and a sensitivity parameter is generated. And determining the influence weight of each environmental factor by using an orthogonal table, and calculating the contribution value and the dominance degree of the environmental parameters in combination with the influence weight and the sensitivity. And constructing a deep learning model, training a physical property prediction model and a strength prediction model by taking the environmental parameters, the standard physical properties and the dominating degree labels as input, and finally outputting the yield strength of the austenitic stainless steel to be measured. According to the method, by quantifying the dynamic influence of environmental factors, the accuracy and applicability of mechanical property prediction are remarkably improved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of mechanical property analysis, in particular to an austenitic stainless steel mechanical property calculation method and system based on a neural network. BACKGROUND

[0002] Austenitic stainless steel is widely used in chemical industry, energy industry, medical industry and other fields due to its excellent corrosion resistance and good mechanical properties. However, the mechanical properties of austenitic stainless steel are affected by various factors, including the microstructure of the material itself and external environmental conditions. Traditional methods mainly determine these properties through experiments, but the process is tedious and it is difficult to fully consider the dynamic influence of environmental factors. In addition, the existing technology lacks a method that can comprehensively consider material parameters and environmental factors and predict mechanical properties through efficient calculation, making it difficult to quickly and accurately evaluate material properties in practical applications, especially under complex and variable environmental conditions. Therefore, developing an austenitic stainless steel mechanical property calculation method based on a neural network has become the key to solving this technical problem.

[0003] In the prior art, the method for calculating the mechanical properties of austenitic stainless steel using a neural network disclosed in CN112597604A uses a Ludwik model to establish a neural network, analyzes the depth indentation curve through the neural network, and calculates the mechanical properties of the austenitic stainless steel. The specific steps of the method are as follows: 1) establishing a Ludwik model; 2) establishing a finite element simulation analysis database: in order to establish a neural network database, Ansys finite element software is used for simulation calculation; 3) creating a neural network; 4) neural network training: a pair of randomly selected input and output vectors is used to train the neural network; through the training process, the neural network can learn the approximate relationship between the input and output data; 5) identifying material parameters: through an optimization program, the material parameters are found to minimize the error between the experimental observation value and the neural network simulation prediction value.

[0004] The main problem of the above-mentioned scheme is that it needs to use finite element software to establish a simulation database, and when designing the nonlinear material behavior such as plastic deformation of austenitic stainless steel, the simulation accuracy and calculation efficiency are difficult to balance; only relying on the Ludwik model may not accurately reflect the dynamic influence of environmental factors on material properties; the training data of the neural network of this scheme comes entirely from finite element simulation, and the simulation model itself has simplifying assumptions, resulting in deviations between the training data and real experimental data; this scheme needs to adjust the material parameters through an optimization program to minimize the error between the experimental observation value and the neural network prediction value, however, the optimization process may fall into local optimization due to the nonlinearity of the objective function, resulting in inaccurate parameter identification.

[0005] The above information disclosed in the Background section is only for enhancing the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art that is already known to those of ordinary skill in the art. SUMMARY

[0006] The purpose of the present application is to provide an austenitic stainless steel mechanical property calculation method and system based on neural network, to solve the problems raised in the background.

[0007] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0008] An austenitic stainless steel mechanical property calculation method based on neural network, the specific steps include:

[0009] Step 1: Measure the standard physical properties of the austenitic stainless steel sample at the standard temperature, including weight, average grain size, nitrogen content and cold working deformation, and record the humidity and corrosion medium concentration at the time of measurement, set as standard humidity and standard corrosion medium concentration;

[0010] Step 2: Change the temperature, humidity and corrosion medium concentration one by one by the control variable method to obtain the physical properties of the austenitic stainless steel sample under different temperature, humidity and corrosion medium concentration combinations;

[0011] Step 3: Analyze the physical properties of the austenitic stainless steel sample at different temperatures, generate the first, second and third temperature change sensitivities based on the changes of average grain size, nitrogen content and cold working deformation at different temperatures, generate the second and third humidity change sensitivities based on the changes of nitrogen content and cold working deformation at different humidities, and generate the corrosion concentration sensitivity based on the deformation of weight at different corrosion medium concentrations;

[0012] Step 4: Construct an L9(3 4 ) orthogonal table, determine the influence weight of temperature, humidity and corrosion medium concentration on average grain size, nitrogen content and cold working deformation based on the orthogonal table;

[0013] Step 5: Based on the sensitivity and influence weight of each physical property under different environmental factors, generate the contribution value of each environmental parameter to the physical property, normalize the contribution value to generate the dominance degree of each environmental parameter, construct a deep learning model, collect the austenitic stainless steel data of known physical properties and yield strength and its environmental parameters, take temperature, humidity, corrosion medium concentration and standard physical properties as input, add dominance degree label to each input, take actual physical properties as label, train the physical property prediction model, take the predicted actual physical properties as input, take yield strength as label, and train the strength prediction model;

[0014] Step 6: Collecting the environmental parameters of the austenitic stainless steel to be predicted, inputting the environmental parameters and standard physical properties into the physical property prediction model to generate the actual physical properties, and inputting the actual physical properties into the strength prediction model to output the yield strength of the austenitic stainless steel to be predicted.

[0015] Further, the specific scheme of the control variable method is: controlling the humidity and the concentration of the corrosion medium to keep the standard humidity and the standard concentration of the corrosion medium unchanged, setting the sampling temperature at equal temperature intervals in the range of 100 to 1000 DEG C, and detecting the physical properties of the austenitic stainless steel sample at different sampling temperatures one by one; controlling the temperature and the concentration of the corrosion medium to keep the standard temperature and the standard concentration of the corrosion medium unchanged, setting the sampling humidity at equal humidity intervals in the range of 10 to 90%, and detecting the physical properties of the austenitic stainless steel sample at different sampling humidities one by one; controlling the temperature and the humidity to keep the standard temperature and the standard temperature unchanged, setting the sampling concentration at equal intervals in the range of 1 to 20% of the concentration of the corrosion medium, and detecting the physical properties of the austenitic stainless steel at different sampling concentrations one by one.

[0016] Further, the principles for generating the first, second and third temperature change sensitivities are:

[0017] The formula for generating the first temperature change sensitivity is:

[0018]

[0019] wherein S P,i represents the change sensitivity of the average grain size at the ith sampling temperature, i represents the index of the sampling temperature, and i∈[1, I], I represents the total number of sampling temperatures, P i represents the average grain size at the ith sampling temperature, P0 represents the average grain size at the standard temperature, T i represents the ith sampling temperature, T0 represents the standard temperature, S P,T represents the first temperature change sensitivity;

[0020] The formula for generating the second temperature change sensitivity is:

[0021]

[0022] wherein S N,i represents the change sensitivity of the nitrogen content at the ith sampling temperature, N i represents the nitrogen content at the ith sampling temperature, N0 represents the nitrogen content at the standard temperature, S N,T represents the second temperature change sensitivity;

[0023] The principle for generating the third temperature change sensitivity is:

[0024] All sampling temperatures with a decrease of cold working deformation relative to the standard cold working deformation exceeding 30% are screened out, and the minimum value among them is taken as the critical temperature. Sampling temperatures not higher than the critical temperature are defined as low sampling temperatures, and sampling temperatures higher than the critical temperature are defined as high sampling temperatures. A low-temperature sensitivity is generated for each low sampling temperature, and the formula is:

[0025]

[0026] wherein, S low,Q,l represents the low-temperature sensitivity of the lth low sampling temperature, l represents the index of the low sampling temperature, Q l represents the cold working deformation of the lth low sampling temperature, Q0 represents the cold working deformation at the standard temperature, T l represents the lth low sampling temperature;

[0027] A high-temperature sensitivity is generated for each high sampling temperature, and the formula is:

[0028]

[0029] wherein, S high,Q,h represents the high-temperature sensitivity of the hth high sampling temperature, h represents the index of the high sampling temperature, Q h represents the cold working deformation of the hth high sampling temperature, Q c represents the cold working deformation at the critical temperature, T h represents the hth high sampling temperature, T c represents the critical temperature, and a represents a high-temperature sensitivity fitting parameter;

[0030] The formula for generating the third temperature change sensitivity is:

[0031]

[0032] wherein, S Q,T represents the third temperature change sensitivity, L represents the number of low sampling temperatures, and H represents the number of high sampling temperatures.

[0033] Further, the principles for generating the second and third humidity change sensitivities are:

[0034] The formula for generating the second humidity change sensitivity is:

[0035]

[0036] wherein, S N,j represents the change sensitivity of nitrogen content at the jth sampling humidity, j represents the index of the sampling humidity, and j∈[1,J], J represents the number of sampling humidities, N jNj represents the nitrogen content at the jth sampling humidity, N0 represents the nitrogen content at the standard humidity, W0 represents the standard humidity, and W represents the jth sampling humidity j Nj represents the nitrogen content at the jth sampling humidity, N0 represents the nitrogen content at the standard humidity, W0 represents the standard humidity, and W represents the jth sampling humidity N,W S2 represents the second humidity change sensitivity.

[0037] The formula for generating the third humidity change sensitivity is:

[0038]

[0039] where S Q,j Qj represents the change sensitivity of the cold working deformation at the jth sampling humidity, Q0 represents the cold working deformation at the standard temperature, and S represents the jth sampling humidity j Qj represents the cold working deformation at the jth sampling humidity, Q0 represents the cold working deformation at the standard temperature, and S represents the jth sampling humidity Q,W S3 represents the third humidity change sensitivity.

[0040] Further, the principle for generating the corrosion concentration sensitivity is:

[0041] The formula for calculating the corrosion rate is:

[0042]

[0043] where V r Vr represents the corrosion rate at the rth sampling concentration, r represents the index of the sampling concentration, and r ∈ [1, R], R represents the number of sampling concentrations, M0 represents the initial weight of the sample, and M r Mr represents the dry weight of the sample after being treated by the rth sampling concentration, A represents the surface area of the sample exposed to the corrosion medium, and t represents the corrosion time.

[0044] The different corrosion medium concentrations and the corresponding corrosion rates are fitted as a functional relationship:

[0045] V r = k × C r b

[0046] where C r represents the rth sampling concentration, k and b represent the pre-coefficient and exponential term parameters of the fitted curve, respectively.

[0047] Take the natural logarithm of both sides of the functional relationship: lnV r = lnk + b × lnC r Solve lnk and b by the least square method, and the corrosion concentration sensitivity is:

[0048]

[0049] D(V r ) = k × b × Cr b-1

[0050] wherein F represents corrosion concentration sensitivity, D(V r ) represents derivation of V r .

[0051] Further, the principle for constructing the L9(3 4 ) orthogonal table is as follows:

[0052] Three temperatures 100℃, 200℃ and 300℃ are selected from the sampling temperature, three humidities 30%, 40% and 50% are selected from the sampling humidity, and three concentrations 1%, 3% and 5% are selected from the sampling concentration.

[0053] The L9(3 4 ) orthogonal table has 9 rows and 4 columns, 9 rows correspond to the number of experiments, 4 columns correspond to the maximum of 4 influencing factors, each column corresponds to one influencing factor, and each influencing factor corresponds to 3 experiment times.

[0054] Further, the principle for determining the influence weight of temperature, humidity and corrosion medium concentration on the average grain size, nitrogen content and cold working deformation based on the orthogonal table is as follows:

[0055] Each row in the orthogonal table corresponds to an experimental condition, i.e. temperature, humidity and corrosion medium concentration. Under each experimental condition, the average grain size, nitrogen content and cold working deformation are measured and filled in the same row as the experimental condition in the orthogonal table. A total of 9 groups of average grain size, nitrogen content and cold working deformation are measured. For the average grain size, an average grain size mean value is calculated at each temperature. Similarly, an average grain size mean value is calculated at each humidity and concentration. The maximum value minus the minimum value of the three average grain size mean values at the same temperature gives the temperature influence range D T , the maximum value minus the minimum value of the three average grain size mean values at the same humidity gives the humidity influence range D W , and the maximum value minus the minimum value of the three average grain size mean values at the same concentration gives the concentration influence range D C . The temperature influence weight, humidity influence weight and concentration influence weight are generated based on the temperature influence range, humidity influence range and concentration influence range. The formula is as follows:

[0056]

[0057] wherein D T represents the temperature influence range, D W represents the humidity influence range, D C represents the concentration influence range, and w T,1Indicates the weight of the effect of temperature on grain size, w W,1 Indicates the weight of the effect of humidity on grain size, w C,1 Indicates the weight of the effect of corrosive medium concentration on grain size;

[0058] According to the above steps, the influence weights of temperature, humidity and corrosive medium concentration on average grain size were calculated. Similarly, the influence weights of temperature, humidity and corrosive medium concentration on nitrogen content and cold working deformation were calculated.

[0059] Furthermore, the principle for generating the dominance of each environmental parameter is as follows:

[0060] For each physical property, the contribution of each environmental parameter is calculated based on the following formula:

[0061] G T =w T,1 ×S P,T +w T,2 ×S N,T +w T,3 ×S Q,T

[0062] G W =w W,2 ×S N,W +w W,3 ×S Q,W

[0063] G C =w C,1 ×F+w C,2 ×F+w C,3 ×F

[0064] Among them, G T Indicates the temperature contribution value, G W Represents the humidity contribution value, G C Indicates the contribution value of the corrosive medium concentration, S P,T Indicates the first temperature change sensitivity, S N,T Indicates the second temperature change sensitivity, S Q,T Indicates the third temperature change sensitivity, w W,2 Indicates the weight of the effect of humidity on nitrogen content, S N,W Indicates the second humidity change sensitivity, w W,3 Represents the weight of the influence of humidity on cold working deformation, S Q,W represents the third humidity change sensitivity, F represents the corrosion concentration sensitivity, w C,2 Indicates the weight of the influence of corrosive medium concentration on nitrogen content, w C,3 Indicates the influence weight of the corrosive medium concentration on the cold working deformation;

[0065] Dominance of each environmental parameter is:

[0066]

[0067] Wherein, Y T represents the temperature dominance, Y W represents the humidity dominance, Y C represents the corrosion medium concentration dominance;

[0068] The dominance label is input into the deep learning model together with the original input, and the mean square error function introduced with the dominance is taken as the loss function:

[0069]

[0070] Y = max(Y T ,Y W ,Y C )

[0071] Wherein, tau represents the loss function, K represents the total number of samples input into the model for training, k represents the sample index of the model training, represents the actual physical property prediction value of the kth sample of the model training, represents the actual physical property true value of the kth sample of the model training, Y represents the dominance label, beta belongs to [0.1, 1], and represents a hyperparameter for controlling the influence of the dominance on the loss.

[0072] The application also provides an austenitic stainless steel mechanical property calculation system based on a neural network, which is used to realize the austenitic stainless steel mechanical property calculation method based on the neural network, and specifically comprises:

[0073] A data acquisition and initialization module is configured to measure standard physical properties of an austenitic stainless steel sample at a standard temperature, wherein the physical properties include weight, average grain size, nitrogen content and cold working deformation, and the humidity and corrosion medium concentration during measurement are recorded as standard humidity and standard corrosion medium concentration;

[0074] A data processing module is configured to change the temperature, humidity and corrosion medium concentration one by one through the control variable method to obtain the physical properties of the austenitic stainless steel sample under different combinations of temperature, humidity and corrosion medium concentration.

[0075] An environmental change calculation module is configured to analyze the physical properties of the austenitic stainless steel sample at different temperatures, generate first, second and third temperature change sensitivities based on the changes of the average grain size, nitrogen content and cold working deformation at different temperatures, generate second and third humidity change sensitivities based on the changes of the nitrogen content and cold working deformation at different humidities, and generate a corrosion concentration sensitivity based on the deformation of the weight at different corrosion medium concentrations.

[0076] a weight analysis module, configured to construct an L9(3 4 ) orthogonal table, and determine the influence weights of temperature, humidity and corrosion medium concentration on the average grain size, nitrogen content and cold working deformation based on the orthogonal table;

[0077] a model training module, configured to generate a contribution value of each environmental parameter to the physical property based on the sensitivity and influence weight of each physical property under different environmental factors, perform normalization processing on the contribution value to generate a dominance degree of each environmental parameter, construct a deep learning model, collect austenitic stainless steel data with known physical properties and yield strength and environmental parameters thereof, take temperature, humidity, corrosion medium concentration and standard physical properties as inputs, add a dominance degree label to each input, take actual physical properties as labels, train a physical property prediction model, take the predicted actual physical properties as inputs, take yield strength as labels, and train a strength prediction model;

[0078] a data output module, configured to collect environmental parameters of an austenitic stainless steel to be predicted, input the environmental parameters and standard physical properties into the physical property prediction model to generate actual physical properties, input the actual physical properties into the strength prediction model, and output the yield strength of the austenitic stainless steel to be predicted.

[0079] Compared with the prior art, the present application has the following advantages:

[0080] The present application can comprehensively understand the influence of environmental factors such as temperature, humidity and corrosion medium concentration on the performance of austenitic stainless steel through repeated experiments under different environmental conditions, and can obtain more comprehensive feature expressions.

[0081] The present application further constructs an L9(3 4 ) orthogonal table, identifies the influence weight of each environmental factor on the physical property, helps to optimize the input features of the model subsequently, improves the prediction accuracy of the model on the physical property, quantifies the dominance degree of the environmental parameters, can intuitively understand the relative contribution of each factor to the physical property, and the introduction of the dominance degree label enables the deep learning model not only to make predictions but also to provide interpretable results, helps users to understand why certain environmental conditions have a significant impact on the performance of the material, and improves the prediction accuracy of the model. BRIEF DESCRIPTION OF DRAWINGS

[0082] Figure 1 is a flowchart of the method of the embodiment of the present application;

[0083] Figure 2A high-temperature sensitivity change with temperature diagram for an embodiment of the present application;

[0084] Figure 3 A high-temperature sensitivity fitting curve for an embodiment of the present application;

[0085] Figure 4 A corrosion concentration sensitivity change broken line graph for an embodiment of the present application;

[0086] Figure 5 A system module schematic diagram for an embodiment of the present application. DETAILED DESCRIPTION

[0087] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with specific embodiments.

[0088] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present application should be understood as the general meaning understood by those skilled in the art to which the present application belongs. The "first", "second" and similar words used in the present application do not represent any order, quantity or importance, but are only used to distinguish different components. "Include" or "contain" and similar words mean that the elements or objects before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connected" or "connected" and similar words are not limited to physical or mechanical connection, but can include electrical connection, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to represent the relative positional relationship, when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0089] Embodiment:

[0090] Please refer to Figures 1 to 4 The present application provides a technical solution:

[0091] An austenitic stainless steel mechanical property calculation method based on a neural network, the specific steps comprising:

[0092] Step 1: Measure the standard physical properties of the austenitic stainless steel sample at the standard temperature, the physical properties including weight, average grain size, nitrogen content and cold working deformation, while recording the humidity and corrosion medium concentration at the time of measurement, set as standard humidity and standard corrosion medium concentration;

[0093] In this embodiment, the mass of the austenitic stainless steel sample is measured by an electronic balance, the standard temperature is set to 25℃, the sample is cleaned on the surface after using anhydrous ethanol at the standard temperature and dried and weighed, the weighing result is recorded, the weighing result is repeated three times and the average value is taken as the weight;

[0094] Put the sample under metallographic microscope, randomly select 8 fields of view, take 500 times metallographic photos, calculate the grain size based on the intercept method, draw a straight line on the photo, count the number of intersections between the straight line and the grain boundary, and the calculation formula is: Take the average value of all field detection results as the average grain size;

[0095] Cut 1g of cubic small pieces from the austenitic stainless steel sample, polish to remove the surface oxide layer, and use anhydrous ethanol. Weigh the cut sample, put it into a high-purity graphite crucible and heat it to complete melting. The nitrogen content is detected by a thermal conductivity detector.

[0096] Uniaxial tension is carried out on the sample by a universal material testing machine until the plastic deformation stage, and the gage length before and after tension is recorded,

[0097] The measured temperature, humidity and corrosion medium concentration represent the environmental temperature, humidity and corrosion medium concentration of the austenitic stainless steel sample.

[0098] Step 2: Change the temperature, humidity and corrosion medium concentration one by one to obtain the physical properties of the austenitic stainless steel sample under different temperature, humidity and corrosion medium concentration combinations;

[0099] In this embodiment, the humidity and corrosion medium concentration are controlled to keep the standard humidity and standard corrosion medium concentration unchanged, and the sampling temperature is set at equal temperature intervals in the range of 100℃ to 1000℃. The physical properties of the austenitic stainless steel sample at different sampling temperatures are detected one by one. The temperature and corrosion medium concentration are controlled to keep the standard temperature and standard corrosion medium concentration unchanged, and the sampling humidity is set at equal humidity intervals in the range of 10% to 90%. The physical properties of the austenitic stainless steel sample at different sampling humidities are detected one by one. The temperature and humidity are controlled to keep the standard temperature and standard temperature unchanged. The corrosion medium concentration here is selected as NaCl concentration for calculation. Because austenitic stainless steel is often in contact with chlorine-containing corrosion medium in marine environment, food processing and chemical equipment application, NaCl is the main representative of chlorine-containing corrosion medium. The sampling concentration is set at equal intervals in the range of 1% to 20% NaCl concentration, and the physical properties of the austenitic stainless steel at different sampling concentrations are detected one by one.

[0100] Step 3: Analyze the physical properties of the austenitic stainless steel sample at different temperatures, generate first, second and third temperature change sensitivities based on the changes of average grain size, nitrogen content and cold working deformation at different temperatures, generate second and third humidity change sensitivities based on the changes of nitrogen content and cold working deformation at different humidities, and generate corrosion concentration sensitivity based on the deformation of weight at different corrosion medium concentrations;

[0101] In this embodiment, the principles for generating the first, second, and third temperature change sensitivities are as follows:

[0102] The formula for generating the first temperature change sensitivity is:

[0103]

[0104] Among them, S P,i represents the sensitivity of the average grain size change at the i-th sampling temperature, i represents the index of the sampling temperature, and i∈[1,I], I represents the total number of sampling temperatures, P i represents the average grain size at the i-th sampling temperature, P0 represents the average grain size at the standard temperature, and T i represents the i-th sampling temperature, T0 represents the standard temperature, S P,T Indicates the first temperature change sensitivity;

[0105] The first temperature change sensitivity reflects the sensitivity of the single variable of average grain size to temperature change, S P,i Indicates the grain size change rate at a single temperature, (P i -P0) / P0 represents the temperature T i Under the condition of grain size P i The relative change rate of the grain size P0 at the standard temperature can eliminate the absolute value difference of the grain size and standardize the sample results with different initial values. (T i -T0) / T0 represents the real-time temperature T i The relative rate of change relative to the standard temperature T0 eliminates the interference of the absolute value of temperature; The physical meaning of is the relative change in grain size caused by the relative change in unit temperature. The result is dimensionless and can be directly used to compare the effects of different environmental parameters on the same physical property. For each sampling temperature, a change sensitivity can be obtained. The average of the change sensitivities of all sampling temperatures is the first temperature change sensitivity that reflects the change sensitivities of all sampling temperatures. P,T A positive value indicates that the temperature increases and the grain size increases, and S P,T The higher the value, the more obvious the change of grain size is as the temperature increases. P,T The sensitivity S of each sample temperature change P,i Proportional to S P,i It is proportional to the average grain size at the sampling temperature and inversely proportional to the difference between the sampling temperature and the standard temperature.

[0106] The formula for generating the second temperature change sensitivity is:

[0107]

[0108] wherein S N,i represents the change sensitivity of nitrogen content at the ith sampling temperature, N i represents the nitrogen content at the ith sampling temperature, N0 represents the nitrogen content at the standard temperature, S N,T represents the second temperature change sensitivity;

[0109] The second temperature change sensitivity reflects the sensitivity of nitrogen content to temperature change, and its formula principle is the same as that of the first temperature change sensitivity.

[0110] The principle for generating the third temperature change sensitivity is:

[0111] All sampling temperatures with a decrease of cold working deformation relative to the standard cold working deformation exceeding 30% are screened out, and the minimum value among them is taken as the critical temperature. Sampling temperatures not higher than the critical temperature are defined as low sampling temperatures, and sampling temperatures higher than the critical temperature are defined as high sampling temperatures. A low-temperature sensitivity is generated for each low sampling temperature, and the formula is:

[0112]

[0113] wherein S low,Q,l represents the low-temperature sensitivity of the lth low sampling temperature, l represents the index of the low sampling temperature, Q l represents the cold working deformation at the lth low sampling temperature, Q0 represents the cold working deformation at the standard temperature, T l represents the lth low sampling temperature;

[0114] A high-temperature sensitivity is generated for each high sampling temperature, and the formula is:

[0115]

[0116] wherein S high,Q,h represents the high-temperature sensitivity of the hth high sampling temperature, h represents the index of the high sampling temperature, Q h represents the cold working deformation at the hth high sampling temperature, Q c represents the cold working deformation at the critical temperature, T h represents the hth high sampling temperature, T c represents the critical temperature, a represents the high-temperature sensitivity fitting parameter;

[0117] The cold working deformation of austenitic stainless steel is significantly affected by temperature, and the dominant mechanism is different in different temperature ranges. The temperature point at which the cold working deformation is reduced by more than 30% compared with the standard temperature is set as the critical temperature. The critical temperature is the temperature at which the cold working deformation behavior of austenitic stainless steel changes significantly. The reduction ratio of deformation at each temperature relative to the standard temperature is calculated, and the first temperature that satisfies the reduction ratio greater than or equal to 30% is the critical temperature T c When the sampling temperature is less than the critical temperature, it is in the low temperature range, and the dominant mechanism in the low temperature range is dislocation slip and accumulation. The cold working deformation changes slowly with temperature, and the overall relationship is linear. When the sampling temperature is higher than the critical temperature, it is in the high temperature range, and the dominant mechanism in the high temperature range is dynamic recovery, recrystallization or phase change. The cold working deformation changes sharply, and the temperature and cold working deformation are nonlinear. An exponential term is introduced to reflect the nonlinear response. The high temperature behavior of the material is exponentially related to the temperature, which can be regarded as a normalized thermal activation term. T h -T c represents the difference between the high temperature range and the critical temperature, reflecting the thermal driving force of the high temperature range. The fitting parameter a is used to control the rate of nonlinear growth. The smaller a is, the steeper the exponential term grows, indicating that the sample is more sensitive to temperature changes. The larger a is, the more gentle the exponential term grows, indicating that the temperature has a weaker effect. The low temperature sensitivity is proportional to the deformation increase value and inversely proportional to the temperature difference. The high temperature sensitivity is proportional to the deformation and inversely proportional to a. Taking the natural logarithm of the high temperature range formula to linearize it, we get the linearized expression: with T h -T c as the independent variable x, and the dependent variable y, the expression is: The a is fitted by least squares method. The critical temperature is 400℃, and a is 100. The change of high temperature sensitivity with temperature is shown in Table 1. Table 1 reflects the change of high temperature sensitivity of austenitic stainless steel with temperature at high sampling temperature. From 420℃ to 1000℃, the high temperature sensitivity shows an accelerating upward trend, and the higher the temperature, the more obvious the accelerating upward trend.

[0118] Table 1. High temperature sensitivity change table

[0119]

[0120]

[0121] The formula for generating the third temperature change sensitivity is:

[0122]

[0123] where SQ,t represents the third temperature change sensitivity, L represents the number of low sampling temperatures, and H represents the number of high sampling temperatures.

[0124] The principle for generating the second and third humidity change sensitivities is as follows:

[0125] The formula for generating the second humidity change sensitivity is as follows:

[0126]

[0127] wherein, S N,j represents the change sensitivity of the nitrogen content at the jth sampling humidity, j represents the index of the sampling humidity, and j∈[1,J], J represents the number of sampling humidities, N j represents the nitrogen content at the jth sampling temperature, N0 represents the nitrogen content at the standard humidity, W0 represents the standard humidity, W j represents the jth sampling humidity, S N,W represents the second humidity change sensitivity.

[0128] The second humidity change rate reflects the sensitivity of the nitrogen content to the humidity change, and the logarithmic form is used to capture the nonlinear relationship between the humidity and the nitrogen content, and respectively represent the values of the nitrogen content and the humidity relative to the standard state, the influence of the humidity on the diffusion or chemical reaction of the nitrogen often presents an exponential or power relationship, the nonlinear relationship is converted into a linear relationship by taking the logarithm, and the change range of the humidity is relatively large, the response of the nitrogen content is sensitive in a small humidity interval and gentle in a large humidity interval, the elasticity coefficient of the nitrogen content to the unit humidity change is obtained by taking the logarithm and making the quotient, to reflect this non-uniform response, and the average value of the sensitivities at all sampling humidities is calculated to eliminate random errors and reflect the overall change trend, S N,j is proportional to the nitrogen content and inversely proportional to the sampling temperature.

[0129] The formula for generating the third humidity change sensitivity is as follows:

[0130]

[0131] wherein, S Q,j represents the change sensitivity of the cold working deformation at the jth sampling humidity, Q j represents the cold working deformation at the jth sampling humidity, Q0 represents the cold working deformation at the standard temperature, S Q,W represents the third humidity change sensitivity.

[0132] The third humidity change sensitivity is used to quantify the sensitivity of the cold working deformation to the humidity change, and reflects the influence of the humidity on the plastic deformation capacity of the material, (Q jQ0) / Q0 represents the relative change of cold working deformation, (W j W0) / W0 represents the relative change of humidity, and the ratio of the two represents the change rate of cold working deformation caused by unit humidity change; for each sampling humidity, a change sensitivity can be obtained, and the average of the change sensitivities of all sampling humidities is taken to obtain a third humidity change sensitivity reflecting the change sensitivities of all sampling humidities; S Q,W is positive, indicating that the increase in humidity leads to an increase in cold working deformation, and S Q,W The higher the value, the more obvious the change in cold working deformation with the increase in humidity, S Q,W is proportional to the change sensitivity S Q,j of each sampling humidity, and S Q,j is proportional to the cold working deformation at the sampling humidity and inversely proportional to the difference between the sampling humidity and the standard humidity.

[0133] The principle for generating the corrosion concentration sensitivity is as follows:

[0134] The formula for calculating the corrosion rate is as follows:

[0135]

[0136] wherein, V r represents the corrosion rate at the rth sampling concentration, r represents the index of the sampling concentration, and r ∈ [1, R], R represents the number of sampling concentrations, M0 represents the initial weight of the sample, M r represents the dry weight of the sample after being treated by the rth sampling concentration, A represents the surface area of the sample exposed to the corrosion medium, and t represents the corrosion time;

[0137] The different corrosion medium concentrations and the corresponding corrosion rates are fitted as a functional relationship:

[0138] V r = k × C r b

[0139] wherein, C r represents the rth sampling concentration, and k and b represent the pre-coefficient and exponential term parameters of the fitted curve, respectively;

[0140] Take the natural logarithm of both sides of the functional relationship: lnV r = lnk + b × lnC r Solve lnk and b by the least square method, and the corrosion concentration sensitivity is:

[0141]

[0142] D(V r ) = k × b × C rb-1

[0143] wherein F represents corrosion concentration sensitivity, D(V r ) represents derivation of V r .

[0144] Corrosion concentration sensitivity reflects the rate change of weight loss of austenitic stainless steel under different corrosion medium concentrations, the relationship between corrosion rate and concentration is nonlinear, the increase of concentration may cause corrosion acceleration or passivation, first, the weight loss of the sample under different corrosion medium concentrations is measured by experiment, the corrosion rate is calculated, the corrosion rate reflects the weight loss of the unit surface area of the sample per unit time, the higher the corrosion rate, the more serious the corrosion, the physical meaning of corrosion concentration sensitivity is the degree of influence of the slight change of corrosion medium concentration on the corrosion rate, therefore, the corrosion concentration sensitivity is obtained by derivation of the corrosion rate; the corrosion concentration sensitivity is proportional to the corrosion rate, taking the initial weight as 10g, the corrosion time as 24 hours, the surface area as 5cm 2 , the corrosion concentration sensitivity changes with the corrosion rate as shown in Table 2, and 30 groups of data are selected from 1% to 20% in Table 2, other sampling concentrations can be selected for experiment outside the 30 groups of data, Table 2 reflects that the weight after corrosion decreases with the increase of concentration, and the corrosion rate increases with the increase of concentration, the corrosion concentration sensitivity and the sampling concentration show a nonlinear relationship, but the corrosion concentration sensitivity shows an overall upward trend with the increase of sampling concentration.

[0145] Table 2. Corrosion concentration sensitivity change table

[0146]

[0147]

[0148] Step 4: construct L9(3 4 ) orthogonal table, determine the influence weight of temperature, humidity and corrosion medium concentration on average grain size, nitrogen content and cold working deformation based on the orthogonal table;

[0149] In this embodiment, the principle for constructing L9(3 4 ) orthogonal table is as follows:

[0150] Three equidistant temperatures 100℃, 200℃ and 300℃ are selected from the sampling temperature, three equidistant humidities 30%, 40% and 50% are selected from the sampling humidity, and three equidistant concentrations 1%, 3% and 5% are selected from the sampling concentration;

[0151] L9(3 4) The orthogonal table has 9 rows and 4 columns, 9 rows correspond to the number of experiments, and 4 columns correspond to the maximum 4 influencing factors. Each column corresponds to one influencing factor, and each influencing factor corresponds to 3 experiment times. The constructed orthogonal table is shown in Table 3.

[0152] Table 3. L9 (3 4 ) Orthogonal table

[0153]

[0154]

[0155] The principle for determining the influence weight of temperature, humidity and corrosion medium concentration on the average grain size, nitrogen content and cold working deformation amount based on the orthogonal table is as follows:

[0156] Each row in the orthogonal table corresponds to an experimental condition, i.e. temperature, humidity and corrosion medium concentration. Under each experimental condition, the average grain size, nitrogen content and cold working deformation amount are measured and filled in the same row as the experimental condition in the orthogonal table. A total of 9 groups of average grain size, nitrogen content and cold working deformation amount are measured. For the average grain size, an average grain size mean value is calculated at each temperature. Similarly, an average grain size mean value is calculated at each humidity and concentration. The maximum value minus the minimum value of the three average grain size mean values at the same temperature gives the temperature influence range D T , the maximum value minus the minimum value of the three average grain size mean values at the same humidity gives the humidity influence range D w , and the maximum value minus the minimum value of the three average grain size mean values at the same concentration gives the concentration influence range D C . The temperature influence weight, humidity influence weight and concentration influence weight are generated based on the temperature influence range, humidity influence range and concentration influence range. The formula is as follows:

[0157]

[0158] wherein, D T represents the temperature influence range, D W represents the humidity influence range, D C represents the concentration influence range, w T,1 represents the influence weight of temperature on grain size, w W,1 represents the influence weight of humidity on grain size, and w C,1 represents the influence weight of corrosion medium concentration on grain size.

[0159] According to the above steps, the influence weight of temperature, humidity and corrosion medium concentration on the average grain size is obtained. Similarly, the influence weight of temperature, humidity and corrosion medium concentration on the nitrogen content and cold working deformation amount is obtained.

[0160] Step 5: Based on the sensitivity and influence weight of each physical property under different environmental factors, the contribution value of each environmental parameter to the physical property is generated, the contribution value is normalized to generate the dominance degree of each environmental parameter, a deep learning model is constructed, the data of the known physical property and yield strength of the austenitic stainless steel and the environmental parameters thereof are collected, the temperature, humidity, corrosion medium concentration and standard physical property are taken as inputs, the dominance degree label is added to each input, and the actual physical property is taken as a label, a physical property prediction model is trained, the predicted actual physical property is taken as an input, and the yield strength is taken as a label to train a strength prediction model;

[0161] In this embodiment, the principle for generating the dominance degree of each environmental parameter is:

[0162] For each physical property, the contribution value of each environmental parameter is calculated, and the formula is:

[0163] G T = w T,1 × S P,T + w T,2 × S N,T + w T,3 × S Q,T

[0164] G W = w W,2 × S N,W + w w,3 × S Q,W

[0165] G C = w C,1 × F + w C,2 × F + w C,3 × F

[0166] Wherein, G T represents the temperature contribution value, G W represents the humidity contribution value, G C represents the corrosion medium concentration contribution value, S P,T represents the first temperature change sensitivity, S N,T represents the second temperature change sensitivity, S Q,T represents the third temperature change sensitivity, w W,2 represents the influence weight of humidity on nitrogen content, S N,W represents the second humidity change sensitivity, w W,3 represents the influence weight of humidity on cold working deformation, S Q,W represents the third humidity change sensitivity, F represents the corrosion concentration sensitivity, w C,2 represents the influence weight of corrosion medium concentration on nitrogen content, w C,3represents the weight of the influence of the corrosion medium concentration on the cold working deformation amount;

[0167] The contribution value of the environmental parameter integrates the influence of each environmental parameter on different physical properties by weighted summation, and the influence of each environmental parameter on the physical property is embodied by two parts: weight and sensitivity; the independent influence weight of temperature, humidity and corrosion medium concentration on each physical property is quantified by orthogonal experiment, and the sensitivity reflects the sensitive degree of the physical property with the change of the environmental parameter; the product of the two represents the local contribution of a certain environmental parameter to a certain physical property, and the total contribution value G T , G W , G C is the sum of these local contributions, which embodies the global influence of a single environmental parameter on all physical properties; the contribution value of the environmental parameter is proportional to the sensitivity of each physical property affected by the environmental parameter; the temperature contribution value G T is the weighted sum of the influence of temperature on each physical property, which integrates the global influence of temperature on the microstructure and mechanical properties of the material, and temperature may simultaneously affect the grain size, nitrogen content and cold working deformation amount, G T integrates these influences and reflects the comprehensive effect of temperature on the three physical properties, and the higher the temperature sensitivity, the higher the temperature contribution value; the humidity contribution value G W only considers the influence of humidity on nitrogen content and cold working deformation amount, because the influence of humidity on the average grain size is small and can be ignored, and the higher the humidity sensitivity, the higher the humidity contribution value; the corrosion medium concentration contribution value G C is the weighted sum of the influence of corrosion medium on all physical properties, since corrosion mainly affects the material through weight loss, and other will indirectly change the grain structure and nitrogen content, therefore each physical property is evaluated by weight, and the corrosion medium concentration contribution value is proportional to the corrosion medium sensitivity.

[0168] The dominance of each environmental parameter is:

[0169]

[0170] Among them, Y T represents the temperature dominance, Y W represents the humidity dominance, Y c represents the corrosion medium concentration dominance;

[0171] The dominance degree is obtained by normalizing the contribution value, which quantifies the relative influence degree of temperature, humidity and corrosion medium concentration on physical properties. During the training process, the model will preferentially learn the correlation between the environment parameters with high dominance degree and mechanical properties, so as to more accurately capture the action mechanism of the main influencing factors. The contribution value reflects the influence of a single environmental parameter on the three physical properties respectively. The dominance degree is calculated by the range method, the nonlinear benefits of temperature, humidity and corrosion medium concentration are converted into quantifiable weights, the numerical weight of each environmental parameter is determined, and the prediction accuracy of the model is improved.

[0172] The dominance degree label is input into the deep learning model together with the original input, and the mean square error function with the introduced dominance degree is used as the loss function:

[0173]

[0174] Y = max(Y T ,Y W ,Y C )

[0175] Where τ represents the loss function, K represents the total number of samples input into the model for training, k represents the sample index of the model training, represents the actual physical property prediction value of the kth sample of the model training, represents the actual physical property true value of the kth sample of the model training, Y represents the dominance degree label, β ∈ [0.1, 1], and represents the hyperparameter controlling the influence of the dominance degree on the loss.

[0176] When constructing the deep learning network, the mean square error function is usually used as the loss function, and the mean square error function is In order to reflect the importance of different environmental parameters on the result in the loss calculation, the dominance degree weight part (1+β×Y) is introduced, which is a global constant representing the overall influence of environmental factors on the mechanical properties of the material. Even if the dominance degree label Y is 0, the loss function can maintain the original mean square error function. The higher Y is, the greater the influence of environmental factors is, and the lower Y is, the smaller the influence of environmental factors is. For the value of β, different β values are tried in the range of [0.1, 1] with a fixed step of 0.1, the loss function corresponding to each β is recorded, and the β value corresponding to the minimum loss function is selected.

[0177] The deep learning network structure for constructing the physical property prediction model is:

[0178] Input layer: contains 5 neurons, used to input temperature, humidity, corrosion medium concentration, standard physical property and dominance degree label;

[0179] First hidden layer: contains 128 neurons, activated by ReLU function;

[0180] Second hidden layer: contains 64 neurons, activated by ReLU function;

[0181] Output layer: contains 1 neuron, used for outputting actual physical properties;

[0182] The deep learning network structure for building the strength prediction model is as follows:

[0183] Input layer: contains 1 neuron, used for inputting actual physical properties;

[0184] First hidden layer: contains 128 neurons, activated by ReLU function;

[0185] Second hidden layer: contains 64 neurons, activated by ReLU function;

[0186] Output layer: contains 1 neuron, used for outputting yield strength.

[0187] Step 6: Collecting the environmental parameters of the austenitic stainless steel to be predicted, inputting the environmental parameters and standard physical properties into the physical property prediction model to generate actual physical properties, and inputting the actual physical properties into the strength prediction model to output the yield strength of the austenitic stainless steel to be predicted.

[0188] Please refer to Figure 5 The application also provides an austenitic stainless steel mechanical property calculation system based on neural network, which is used to realize the above-mentioned austenitic stainless steel mechanical property calculation method based on neural network, and specifically comprises:

[0189] A data acquisition and initialization module is used to measure the standard physical properties of the austenitic stainless steel sample at a standard temperature, the physical properties include weight, average grain size, nitrogen content and cold working deformation, and the humidity and corrosion medium concentration during measurement are recorded as standard humidity and standard corrosion medium concentration;

[0190] A data processing module is used to change the temperature, humidity and corrosion medium concentration one by one through the control variable method to obtain the physical properties of the austenitic stainless steel sample under different combinations of temperature, humidity and corrosion medium concentration;

[0191] An environmental change calculation module is used to analyze the physical properties of the austenitic stainless steel sample at different temperatures, generate first, second and third temperature change sensitivities based on the changes of the average grain size, nitrogen content and cold working deformation at different temperatures, generate second and third humidity change sensitivities based on the changes of the nitrogen content and cold working deformation at different humidities, and generate corrosion concentration sensitivity based on the deformation of the weight at different corrosion medium concentrations;

[0192] A weight analysis module is used to build L9(34 )orthogonal table, based on the orthogonal table respectively determine the influence weight of temperature, humidity and corrosion medium concentration on average grain size, nitrogen content and cold working deformation;

[0193] The model training module is configured to generate a contribution value of each environmental parameter to the physical property based on the sensitivity and the influence weight of each physical property under different environmental factors, normalize the contribution value to generate a dominance degree of each environmental parameter, construct a deep learning model, collect data of the austenitic stainless steel with known physical properties and yield strength and environmental parameters thereof, take temperature, humidity, corrosion medium concentration and standard physical properties as inputs, add a dominance degree label to each input, take actual physical properties as labels, train a physical property prediction model, take the predicted actual physical properties as inputs, take yield strength as labels, and train a strength prediction model.

[0194] The data output module is configured to collect environmental parameters of the austenitic stainless steel to be predicted, input the environmental parameters and standard physical properties into the physical property prediction model to generate actual physical properties, input the actual physical properties into the strength prediction model, and output the yield strength of the austenitic stainless steel to be predicted.

[0195] The above formulas are all dimensionless numerical calculations, and the formulas are obtained by software simulation of a large amount of data to obtain a formula of the most recent real situation, and the preset parameters in the formula are set by a person skilled in the art according to the actual situation.

[0196] The above embodiments can be realized wholly or partially by software, hardware, firmware or any combination thereof. When realized by software, the above embodiments can be realized in the form of a computer program product wholly or partially. Those skilled in the art can realize that the units and algorithm steps of the examples described in connection with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized by hardware or software methods depends on the specific application and design constraints of the technical solutions.

[0197] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, which can be located in one place or distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiments.

[0198] The above is merely specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for calculating the mechanical properties of austenitic stainless steel based on neural network, characterized in that: The specific steps include: Step 1: Measure the standard physical properties of the austenitic stainless steel sample at standard temperature, including weight, average grain size, nitrogen content, and cold working deformation. At the same time, record the humidity and corrosive medium concentration during measurement and set them as standard humidity and standard corrosive medium concentration. Step 2: Change the temperature, humidity and corrosive medium concentration one by one by controlling the variable method to obtain the physical properties of the austenitic stainless steel sample under different combinations of temperature, humidity and corrosive medium concentration; Step 3: Analyze the physical properties of the austenitic stainless steel sample at different temperatures, generate the first, second, and third temperature change sensitivities based on the changes in average grain size, nitrogen content, and cold working deformation at different temperatures, generate the second and third humidity change sensitivities based on the changes in nitrogen content and cold working deformation at different humidity levels, and generate the corrosion concentration sensitivity based on the weight deformation at different corrosive medium concentrations; Step 4: Build L9(3 4 ) orthogonal table, based on which the influence weights of temperature, humidity and corrosive medium concentration on average grain size, nitrogen content and cold working deformation are determined respectively; Step 5: Based on the sensitivity and influence weight of each physical property under different environmental factors, the contribution value of each environmental parameter to the physical property is generated. The contribution value is normalized to generate the dominance of each environmental parameter. A deep learning model is constructed. Austenitic stainless steel data with known physical properties and yield strength and its environmental parameters are collected. Temperature, humidity, corrosive medium concentration, and standard physical properties are used as inputs. After adding a dominance label to each input, the physical property prediction model is trained using the actual physical properties as labels. The strength prediction model is trained using the predicted actual physical properties as input and the yield strength as a label. Step 6: Collect the environmental parameters of the austenitic stainless steel to be predicted, input the environmental parameters and standard physical properties into the physical property prediction model to generate actual physical properties, input the actual physical properties into the strength prediction model, and output the yield strength of the austenitic stainless steel to be predicted.

2. The method for calculating mechanical properties of austenitic stainless steel based on a neural network according to claim 1, characterized in that: The specific scheme of the control variable method in step 2 is: controlling the humidity and the concentration of the corrosive medium to keep the standard humidity and the standard corrosive medium concentration unchanged, setting the sampling temperature at equal temperature intervals in the range of 100°C to 1000°C, and detecting the physical properties of the austenitic stainless steel samples at different sampling temperatures one by one; controlling the temperature and the concentration of the corrosive medium to keep the standard temperature and the standard corrosive medium concentration unchanged, setting the sampling humidity at equal humidity intervals in the range of 10% to 90%, and detecting the physical properties of the austenitic stainless steel samples at different sampling humidity one by one; controlling the temperature and the humidity to keep the standard temperature and the standard temperature unchanged, setting the sampling concentration at equal intervals in the range of 1% to 20% of the corrosive medium concentration, and detecting the physical properties of the austenitic stainless steel at different sampling concentrations one by one.

3. The method for calculating mechanical properties of austenitic stainless steel based on a neural network according to claim 1, characterized in that: The principle for generating the first, second and third temperature change sensitivities in step 3 is: The formula for generating the first temperature change sensitivity is: Among them, S P,i represents the sensitivity of the average grain size change at the i-th sampling temperature, i represents the index of the sampling temperature, and i∈[1,I], I represents the total number of sampling temperatures, P i represents the average grain size at the i-th sampling temperature, P0 represents the average grain size at the standard temperature, and T i represents the i-th sampling temperature, T0 represents the standard temperature, S P,T Indicates the first temperature change sensitivity; The formula for generating the second temperature change sensitivity is: Among them, S N,i Indicates the sensitivity of nitrogen content at the i-th sampling temperature, N i represents the nitrogen content at the i-th sampling temperature, N0 represents the nitrogen content at the standard temperature, S N,T Indicates the second temperature change sensitivity; The principle for generating the third temperature change sensitivity is: All sampling temperatures where the cold working deformation decreases by more than 30% relative to the standard cold working deformation are screened out, and the minimum value among them is taken as the critical temperature. The sampling temperature not higher than the critical temperature is defined as the low sampling temperature, and the sampling temperature higher than the critical temperature is defined as the high sampling temperature. The low temperature sensitivity is generated for each low sampling temperature, based on the following formula: Among them, S low,Q,l Indicates the low temperature sensitivity of the lth low sampling temperature, l represents the index of the low sampling temperature, Q l represents the cold working deformation at the first low sampling temperature, Q0 represents the cold working deformation at the standard temperature, T l Indicates the lth low sampling temperature; The high temperature sensitivity is generated for each high sample temperature according to the formula: Among them, S high,Q,h represents the high temperature sensitivity of the hth high sampling temperature, h represents the index of the high sampling temperature, Q h represents the cold working deformation at the hth high sampling temperature, Q c Indicates the cold working deformation at the critical temperature, T h Indicates the hth high sampling temperature, T c represents the critical temperature, a represents the high temperature sensitivity fitting parameter; The formula for generating the third temperature change sensitivity is: Among them, S Q,T represents the third temperature change sensitivity, L represents the number of low sampling temperatures, and H represents the number of high sampling temperatures.

4. The method for calculating mechanical properties of austenitic stainless steel based on a neural network according to claim 1, characterized in that: The principle for generating the second and third humidity change sensitivities in step 3 is as follows: The formula for generating the second humidity change sensitivity is: Among them, S N,j represents the sensitivity of nitrogen content to the jth sampling humidity, j represents the index of the sampling humidity, and j∈[1,J], J represents the number of sampling humidity, N j represents the nitrogen content at the Jth sampling humidity, N0 represents the nitrogen content at the standard humidity, W0 represents the standard humidity, and W j represents the jth sample humidity, S N,W Indicates the second humidity change sensitivity; The formula for generating the third humidity change sensitivity is: Among them, S Q,j represents the sensitivity of cold working deformation under the jth sampling humidity, Q j represents the cold working deformation at the jth sampling humidity, Q0 represents the cold working deformation at the standard temperature, S Q,W Indicates the third humidity change sensitivity.

5. The method for calculating mechanical properties of austenitic stainless steel based on a neural network according to claim 1, characterized in that: The principle for generating the corrosion concentration sensitivity in step 3 is: The corrosion rate is calculated based on the formula: Among them, V r represents the corrosion rate at the rth sampling concentration, r represents the index of the sampling concentration, and r∈[1,R], R represents the number of sampling concentrations, M0 represents the initial weight of the sample, M r represents the dry weight of the sample after being treated with the rth sampling concentration, A represents the surface area of ​​the sample exposed to the corrosive medium, and t represents the corrosion time; Fit the functional relationship between different corrosive medium concentrations and corresponding corrosion rates: In r =k×C r b Among them, C r represents the rth sampling concentration, k and b represent the pre-coefficient and exponential term parameters of the fitting curve, respectively; Take the natural logarithm of both sides of the function relationship: lnV r =lnk+b×lnC r , lnk, b are solved by the least squares method, and the corrosion concentration sensitivity is: D(V r )=k×b×C r b-1 Where F represents the corrosion concentration sensitivity, D(V r ) indicates the value of V r Find the derivative.

6. The method for calculating mechanical properties of austenitic stainless steel based on a neural network according to claim 2, characterized in that: In step 4, L9(3 4 ) The principle of orthogonal array is: Select three equally spaced temperatures of 100°C, 200°C, and 300°C from the sampled temperatures; select three equally spaced humidity values ​​of 30%, 40%, and 50% from the sampled humidity; and select three equally spaced concentration values ​​of 1%, 3%, and 5% from the sampled concentrations. L9(3 4 ) The orthogonal table has 9 rows and 4 columns. The 9 rows correspond to the number of experiments, and the 4 columns correspond to a maximum of 4 influencing factors. Each column corresponds to 1 influencing factor, and each influencing factor corresponds to 3 experimental times.

7. The method for calculating mechanical properties of austenitic stainless steel based on a neural network according to claim 6, characterized in that: The principle for determining the influence weights of temperature, humidity and corrosive medium concentration on average grain size, nitrogen content and cold working deformation based on the orthogonal table is as follows: Each row in the orthogonal table corresponds to an experimental condition, namely temperature, humidity and corrosive medium concentration. Under each experimental condition, the average grain size, nitrogen content and cold working deformation are measured and filled in the row of the orthogonal table with the same experimental conditions. A total of 9 groups of average grain size, nitrogen content and cold working deformation are measured. For the average grain size, an average grain size mean is calculated at each temperature. Similarly, an average grain size mean is also calculated at each humidity and concentration. The temperature effect range D is obtained by subtracting the maximum value from the minimum value of the three average grain size means at the same temperature. T The maximum value minus the minimum value among the three average grain sizes under the same humidity is the humidity effect range D. W The maximum value minus the minimum value among the three average grain sizes at the same concentration is the concentration effect range D C , based on the temperature influence range, humidity influence range and concentration influence range, the temperature influence weight, humidity influence weight and concentration influence weight are generated according to the following formula: Among them, D T Indicates that the temperature effect is extremely poor, D W Indicates that humidity has a very poor effect, D C Indicates that the concentration has a very poor effect, w T,1 Indicates the weight of the effect of temperature on grain size, w W,1 Indicates the weight of the effect of humidity on grain size, w C,1 Indicates the weight of the effect of corrosive medium concentration on grain size; According to the above steps, the influence weights of temperature, humidity and corrosive medium concentration on average grain size were calculated. Similarly, the influence weights of temperature, humidity and corrosive medium concentration on nitrogen content and cold working deformation were calculated.

8. The method for calculating mechanical properties of austenitic stainless steel based on a neural network according to claim 7, characterized in that: The principle for generating the dominance of each environmental parameter in step 6 is: For each physical property, the contribution of each environmental parameter is calculated based on the following formula: G T =w T,1 ×S P,T +w T,2 ×S N,T +w T,3 ×S Q,T G W =w W,2 ×S N,W +w W,3 ×S Q,W G C =w C,1 ×F+w C,2 ×F+w C,3 ×F Among them, G T Indicates the temperature contribution value, G W Represents the humidity contribution value, G C Indicates the contribution value of the corrosive medium concentration, S P,T Indicates the first temperature change sensitivity, S N,T Indicates the second temperature change sensitivity, S Q,T Indicates the third temperature change sensitivity, w W,2 Indicates the weight of the effect of humidity on nitrogen content, S N,W Indicates the second humidity change sensitivity, w W,3 Represents the weight of the influence of humidity on cold working deformation, S Q,W represents the third humidity change sensitivity, F represents the corrosion concentration sensitivity, w C,2 Indicates the weight of the influence of corrosive medium concentration on nitrogen content, w C,3 Indicates the influence weight of the corrosive medium concentration on the cold working deformation; The dominance of each environmental parameter is: Among them, Y T Indicates the temperature dominance, Y W Indicates humidity control degree, Y C Indicates the dominance of corrosive medium concentration; The dominance labels are fed into the deep learning model along with the original input, and the mean squared error function with dominance introduced is used as the loss function: Y=max(Y T ,AND W ,AND C ) Among them, τ represents the loss function, K represents the total number of samples input into the model for training, and k represents the sample index of the model training. represents the actual physical property prediction value of the kth sample trained by the model, represents the actual physical property value of the kth sample trained by the model, Y represents the dominance label, and β∈[0.1,1] represents the hyperparameter that controls the impact of dominance on the loss.

9. A neural network-based austenitic stainless steel mechanical properties calculation system, characterized by: The system is used to implement the neural network-based austenitic stainless steel mechanical property calculation method according to any one of claims 1 to 8, specifically comprising: The data acquisition and initialization module is used to measure the standard physical properties of austenitic stainless steel samples at standard temperature, including weight, average grain size, nitrogen content, and cold working deformation. The humidity and corrosive medium concentration during measurement are also recorded and set as standard humidity and standard corrosive medium concentration. A data processing module is used to change the temperature, humidity and corrosive medium concentration one by one through the control variable method to obtain the physical properties of the austenitic stainless steel sample under different combinations of temperature, humidity and corrosive medium concentration; Environmental change calculation module, used to analyze the physical properties of austenitic stainless steel samples at different temperatures, generate the first, second, and third temperature change sensitivities based on the changes in average grain size, nitrogen content, and cold working deformation at different temperatures, generate the second and third humidity change sensitivities based on the changes in nitrogen content and cold working deformation at different humidity levels, and generate the corrosion concentration sensitivity based on the weight deformation at different corrosive medium concentrations; Weight analysis module, used to construct L9(3 4 ) orthogonal table, based on which the influence weights of temperature, humidity and corrosive medium concentration on average grain size, nitrogen content and cold working deformation are determined respectively; The model training module is used to generate the contribution value of each environmental parameter to the physical property based on the sensitivity and influence weight of each physical property under different environmental factors, normalize the contribution value to generate the dominance of each environmental parameter, build a deep learning model, collect austenitic stainless steel data with known physical properties and yield strength and its environmental parameters, use temperature, humidity, corrosive medium concentration and standard physical properties as input, and add a dominance label to each input. The physical property prediction model is trained using the actual physical properties as the label, and the strength prediction model is trained using the predicted actual physical properties as input and the yield strength as the label; The data output module is used to collect the environmental parameters of the austenitic stainless steel to be predicted, input the environmental parameters and standard physical properties into the physical property prediction model to generate actual physical properties, input the actual physical properties into the strength prediction model, and output the yield strength of the austenitic stainless steel to be predicted.

Images