Tensorflow-based neural network vermicularity prediction method for vermicular graphite cast iron

Abstract

The application discloses a kind of neural network vermicular graphite cast iron vermicularity prediction method based on Tensorflow, through the network structure design of neural network prediction model of vermicular graphite cast iron vermicularity and the collection of training data, in the case where the number of input neurons, output neurons, nerve activation function and implicit layer neurons of neural network are determined, preliminary established neural network prediction model is trained using preprocessed sample data, and then the neural network prediction model that can be used to detect vermicular graphite cast iron vermicularity is obtained.The application compared with traditional prediction model, the vermicularity is quickly predicted by neural network construction algorithm model, due to the characteristics of neural network itself, it has self-learning ability and self-adaptive ability, when production raw material or production condition changes, only need to provide enough data sample to retrain original model, network model can be updated automatically, and it has great guiding significance to vermicularity quality in production process.

Description

Technical Field

[0001] This invention belongs to the field of vermicular graphite cast iron creep rate prediction technology, specifically involving a neural network-based method for predicting the creep rate of vermicular graphite cast iron using Tensorflow. Background Technology

[0002] Due to the influence of production conditions and raw material quality, the multiple linear regression model constructed using traditional thermal analysis techniques to predict vermiformation rate and tissue morphology is difficult to adapt to the current stringent requirements of industrial production, and lacks high accuracy and adaptability. The formula fitted by the regression equation often introduces large errors. Therefore, it is necessary to seek a model that can be adaptive and self-trained.

[0003] The development of artificial neural networks has opened up a new avenue for predicting the microstructure and properties of vermicular graphite cast iron. It consists of many parallel computational elements connected by variable weights and adjusted during the learning process to achieve the desired outcome. It possesses strong nonlinear mapping capabilities, adaptability, and robustness, and can approximate any nonlinear system. Therefore, this project proposes a neural network-based model for predicting the vermicular graphite cast iron creep rate, and combines this with metallographic image processing data analysis to achieve accurate prediction of the creep rate. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing a neural network-based method for predicting the creep rate of vermicular graphite cast iron using Tensorflow. This method involves designing the network structure of a neural network prediction model for the creep rate of vermicular graphite cast iron and collecting training data. Given the determined number of input neurons, output neurons, activation functions, and hidden layer neurons, the pre-processed sample data is used to train the initially established neural network prediction model, thereby obtaining a neural network prediction model that can be used to detect the creep rate of vermicular graphite cast iron.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A neural network-based method for predicting the creep rate of vermicular graphite cast iron using Tensorflow includes the following steps:

[0007] Step S1. Record the temperature change of eutectic composition vermicular graphite cast iron molten iron over time using a computer-integrated temperature data acquisition card and plot the cooling curve. Select appropriate feature values ​​from the cooling curve to determine the input and output neurons of the neural network for detecting the vermicular graphite cast iron vermicular graphite cast iron vermicular graphite cast iron vermicular graphite cast iron vermicular graphite cast iron vermicular graphite cast iron vermicular graphite cast iron temperature change over time.

[0008] Step S2. Select a suitable activation function for the neural network neurons to ensure smooth network training and guarantee model quality;

[0009] Step S3. Based on the empirical formula Calculate the number of neurons in the hidden layer of the above neural network;

[0010] Step S4. Based on the neural network input neurons, output neurons, neural activation functions, and the number of neurons in the hidden layer determined in the above steps, establish a preliminary neural network prediction model;

[0011] Step S5. Preprocess the input and output sample data. The processed sample data is used to train the preliminary neural network prediction model established in step S4 to obtain a neural network prediction model that can be used to detect the vermicular graphite cast iron creep rate.

[0012] In step S1, the selection of appropriate feature values ​​from the cooling curve to determine the input and output neurons of the neural network for detecting the vermicular graphite cast iron creep rate is described. The input neurons include the minimum temperature TEU for eutectic growth of vermicular graphite cast iron, the eutectic end temperature TS, and the eutectic regeneration temperature TER. The output neuron is the vermicular graphite cast iron creep rate.

[0013] In step S2, a suitable activation function for a neural network neuron is selected, wherein the activation function is the ReLU function.

[0014] The step S3 described using the empirical formula Calculate the number of neurons in the hidden layer of the above neural network, where m is the number of neurons in the input layer, n is the number of neurons in the output layer, and c is a constant with a value ranging from 4 to 13.

[0015] In step S5, the input and output sample data are preprocessed by normalization, which transforms all sample data into data between 0 and 1.

[0016] The advantages of this invention over the prior art are:

[0017] Compared to traditional linear regression models, this invention uses a neural network to construct an algorithm model for rapid prediction of creep rate. Due to the inherent characteristics of neural networks, they possess self-learning and adaptive capabilities. When production raw materials or production conditions change, the network model can be automatically updated simply by providing sufficient data samples to retrain the original model. This provides significant guidance for creep quality during the production process. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a neural network-based method for predicting the creep rate of vermicular graphite cast iron based on Tensorflow, according to the present invention.

[0020] Figure 2 This is a schematic diagram of the neural network structure of the present invention;

[0021] Figure 3 This is a schematic diagram illustrating the variation of neural network error with step size in an embodiment of the present invention;

[0022] Figure 4 This is a schematic diagram of the creep rate prediction results in an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. 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.

[0024] Example: See Figure 1-4 .

[0025] like Figure 1 As shown, a neural network-based method for predicting the creep rate of vermicular graphite cast iron using Tensorflow includes the following steps:

[0026] Step S1. Record the temperature change of eutectic composition vermicular graphite cast iron molten iron over time using a computer-integrated temperature data acquisition card and plot the cooling curve. Select appropriate feature values ​​from the cooling curve to determine the input and output neurons of the neural network for detecting the vermicular graphite cast iron vermicular graphite cast iron vermicular graphite cast iron vermicular graphite cast iron vermicular graphite cast iron vermicular graphite cast iron vermicular graphite cast iron temperature change over time.

[0027] Step S2. In deep learning, commonly used activation functions include: sigmoid function, tanh function, and ReLU function. To ensure the smooth training of the network and guarantee the quality of the model, it is necessary to determine the appropriate activation function for the neurons of the neural network.

[0028] Step S3. The number of hidden layer neurons is an important parameter affecting the performance of a neural network. Generally speaking, too few neurons will reduce prediction performance and affect the stability of the model, while too many neurons will cause overfitting to the training data and increase the training time, affecting the practicality of the model. Therefore, according to the empirical formula... Calculate the number of neurons in the hidden layer of the above neural network;

[0029] Step S4. Based on the neural network input neurons, output neurons, neural activation functions, and the number of neurons in the hidden layer determined in the above steps, establish a preliminary neural network prediction model;

[0030] Step S5. Preprocess the input and output sample data. The processed sample data is used to train the preliminary neural network prediction model established in step S4 to obtain a neural network prediction model that can be used to detect the vermicular graphite cast iron creep rate.

[0031] Specifically, such as Figure 2 As shown, in step S1, selecting appropriate feature values ​​from the cooling curve to determine the input and output neurons of the neural network for detecting the vermicular graphite cast iron creep rate is described. The eutectic supercooling temperature (TEU) is the temperature at which the graphite nucleation process ends, after which graphite nuclei begin to grow. The higher the TEU, the better the spheroidization effect. The eutectic regeneration temperature (TER) is the temperature at which the cooling temperature rises again during eutectic solidification due to the latent heat of the eutectic. Therefore, the higher the TER, the more latent heat of solidification there is, and the faster the nucleation and growth rate will be. Consequently, the graphite area will increase, and the creep rate will improve. Therefore, the input layer selects TEU, TER, and TS as the input neurons for predicting the spheroidization rate, and the output neuron is the vermicular graphite cast iron creep rate.

[0032] In step S2, a suitable activation function for a neural network neuron is selected, wherein the activation function is the ReLU function.

[0033] The step S3 described using the empirical formula The number of neurons in the hidden layer of the above neural network is calculated, where m is the number of neurons in the input layer, n is the number of neurons in the output layer, and c is a constant ranging from 4 to 13. In this embodiment, the optimal number of neurons in the hidden layer is selected as 11 after calculation.

[0034] Step S5 involves preprocessing the input and output sample data. The preprocessing method is normalization, which transforms all sample data into data between 0 and 1.

[0035] Please see Figure 3-4 The following example further illustrates the method for predicting the creep rate of vermicular graphite cast iron based on Tensorflow neural networks according to the present invention.

[0036] Experimental Description: The main raw materials include vermicularizing agent, inoculant, scrap steel, and carbon raiser. The materials are prepared according to the on-site requirements for the vermicular brake drum, as detailed in Table 1 below.

[0037] Table 1. Composition of primary and final molten iron for vermicular graphite cast iron after control.

[0038]

[0039] Experimental procedure: Scrap steel was used as the basic furnace charge. Carbon and silicon contents were adjusted with carbon raisers and ferrosilicon. The furnace was smelted in a 300 kg medium-frequency induction furnace. Vermicularizer R2A (addition amount 0.05%~0.18%) and inoculant YFY-8 (addition amount 0.53%~0.62%) were used.

[0040] Both vermicularization and inoculation were performed using the bottom-pouring method; the tapping temperature of the molten iron was around 1510-1550℃, and the sample casting temperature was 1380-1420℃. Thermal analysis solidification curves of vermicular graphite cast iron that had just undergone vermicularization and inoculation were collected using a thermal analysis sample cup. Simultaneously, samples for microstructure analysis were cast, and the vermicularization rate of the solidified sample in the sample cup was measured. The collected stable thermal analysis solidification curves, corresponding characteristic values ​​(TEU, TER, TS), and vermicularization rate were then stored in a database. Experimental data are shown in Table 2.

[0041] Table 2: Partial characteristic temperature data collected in the experiment

[0042]

[0043] To fully train the neural network, 20 sets of data were randomly selected from 24 sets of collected cooling curve thermal analysis data as training samples, and the remaining 4 sets were used to test the prediction performance of the neural network. First, a preliminary neural network model was established according to the method of this invention. After determining the various parameters of the neural network, the Tensorflow Neural Network Toolbox was used to simulate and train the model. The weights of the connections between neurons were repeatedly adjusted until the error reached the desired value, such as... Figure 3 The figure shows the trend of network error during training. As can be seen from the figure, after training, the network error of the neural network model for predicting the vermicular graphite cast iron reached a minimum of 0.006, indicating that the BP neural network can accurately fit the relationship between the characteristic temperature and the vermicular rate.

[0044] Furthermore, to verify the accuracy of the above neural network prediction model, the feature temperature of the validation samples was used as the input to the trained neural network, and the corresponding prediction results of the creep rate are shown in Table 3 below.

[0045] Table 3 Comparison of Predicted Creep Rate Results

[0046]

[0047] Compare the actual values ​​with the predicted values ​​in the table above, such as... Figure 4 As shown in the figure, the predicted vermicularization rate is very close to the actual value, with a relative error of no more than 3.64%. This indicates that the application of the trained neural network algorithm to predict the vermicularization rate of vermicular graphite cast iron has high accuracy, and the prediction model can effectively characterize the relationship between the characteristic temperature points and the vermicularization rate of vermicular graphite cast iron.

[0048] In summary, compared to traditional linear regression models, the algorithm model based on neural networks in this invention is feasible for rapid prediction of creep rate. Furthermore, due to the inherent characteristics of neural networks, they possess self-learning and adaptive capabilities. Therefore, when production raw materials or conditions change, the network model can automatically update itself simply by providing sufficient new data samples to retrain the original model. Thus, the novel thermal analyzer using neural networks in this invention has broader feasibility and scalability, and is particularly significant for guiding creep quality in the production process.

[0049] The above description is merely a preferred embodiment of the present invention and does not constitute any limitation on the structure of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention shall fall within the scope of the technical solution of the present invention.

Claims

1. A Tensorflow-based neural network vermicularity prediction method for vermicular cast iron, characterized in that, The method comprises the following steps: Step S1. Record the change of the temperature of the eutectic composition vermicular cast iron with time by a computer integrated temperature data acquisition card, draw a cooling curve, and select appropriate characteristic values from the cooling curve to determine the input neurons and output neurons of the neural network for detecting the vermicular rate of the vermicular cast iron; The characteristic values selected from the cooling curve to determine the input neurons and output neurons of the neural network for detecting the vermicular rate of the vermicular cast iron include the lowest temperature TEU of the eutectic growth of the vermicular cast iron, the eutectic end temperature TS, and the eutectic regeneration temperature TER, and the output neurons are the vermicular rate of the vermicular cast iron; Step S2. Select an appropriate neural network neuron activation function to ensure the smooth training of the network and guarantee the quality of the model; Step S3. According to the empirical formula The number of neurons in the hidden layer of the neural network is calculated, wherein m is the number of neurons in the input layer, n is the number of neurons in the output layer, and c is a constant, and the value range is 4-13. Step S4. Establish a preliminary neural network prediction model according to the input neurons, output neurons, neural activation function, and the number of neurons in the hidden layer determined in the above steps; Step S5. Preprocess the input and output sample data, and use the processed sample data to train the preliminary neural network prediction model established in step S4 to obtain a neural network prediction model that can be used to detect the vermicular rate of the vermicular cast iron.

2. The Tensorflow-based neural network vermicularity prediction method for vermicular graphite cast iron according to claim 1, characterized in that, In step S2, the neural network neuron activation function is a Relu function. 3.The Tensorflow-based neural network method for predicting the vermicularity of vermicular graphite cast iron according to claim 1, characterized in that, In step S5, the preprocessing method of the input and output sample data is normalization processing, which converts all sample data to data between 0 and 1.

Images