Method and system for predicting casting process parameters of brake drum

Abstract

The invention relates to the technical field of brake drum casting, and discloses a prediction method and system for brake drum casting process parameters, and the prediction method comprises the steps: obtaining a training data set, the training data set comprises characteristic variables and a shrinkage defect volume y, and the characteristic variables comprise brake drum size variables and process parameter variables; training the constructed feed-forward neural network by using a training data set to obtain parameter values of the feed-forward neural network, and obtaining a prediction model of the brake drum casting process parameters; measuring each numerical value of the brake drum size variable of the brake drum workpiece, selecting a plurality of groups of process parameter variables, substituting the brake drum size variable and each group of process parameter variables into the prediction model, and obtaining the shrinkage porosity defect volume y corresponding to each group of process parameter variables, and the technological parameter variable corresponding to the minimum shrinkage defect volume y is selected as the optimal technological parameter of the brake drum workpiece. And more suitable casting process parameters of brake drums with different sizes can be quickly obtained, and the time and economic cost is lower.

Description

Technical Field

[0001] The invention relates to the technical field of brake drum casting, and in particular to a method and system for predicting brake drum casting process parameters. Background Art

[0002] Gray cast iron brake drums are key components of drum brake systems in large trucks and engineering vehicles. Due to their excellent thermal conductivity, wear resistance and casting properties, they meet the braking needs of vehicles at a low cost. However, casting defects such as shrinkage in brake drum workpieces will seriously reduce their service life and even affect vehicle driving safety. The sizes of brake drums used by vehicles with different loads are also different. Therefore, in production, the optimal pouring temperature, pouring time, runner size, riser size and other process parameters for brake drums of different sizes are often different. Traditionally, an initial process parameter is set based on experience, and then the optimal result is gradually obtained through trial and error. This process not only consumes manpower to modify the mold and adjust the process equipment, which increases costs, but also affects the operation process of the automated production line and delays delivery time. In the prior art, technicians use computing tools such as finite element simulation to simulate the casting process of the workpiece, predict the filling and solidification conditions of the workpiece under different process parameters in advance, and assist in adjusting the process parameters in actual production. However, the more suitable process parameters for brake drums of different sizes are different. Although using computing tools such as finite element simulation is more cost-effective than testing on the production line, there is also the problem of setting the initial process parameters based on experience and then gradually obtaining better results through trial and error, and it will consume a lot of computing resources and computing time costs.

[0003] Therefore, there is an urgent need for a method and system for predicting brake drum casting process parameters to solve the above problems. Summary of the invention

[0004] Based on the above, the purpose of the present invention is to provide a method and system for predicting brake drum casting process parameters, which can quickly obtain more suitable casting process parameters for brake drums of different sizes with low time and economic costs.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for predicting brake drum casting process parameters, comprising:

[0007] Obtain a training data set, the training data set includes characteristic variables and shrinkage defect volume y, the characteristic variables include brake drum size variables and process parameter variables, the brake drum size variables include brake drum cup inner diameter D g , the length of the vertical part of the drum wall L 1 , Length of transition from drum wall to flange L 2, flange width f, drum wall thickness t, the process parameter variables include the upper diameter D of the riser m , the runner width W of the pouring system h , pouring temperature T and pouring rate v;

[0008] Constructing a feedforward neural network, using the training data set to train the constructed feedforward neural network, obtaining parameter values ​​of the feedforward neural network, and obtaining a prediction model for brake drum casting process parameters;

[0009] Measure each value of the brake drum size variable of the brake drum workpiece, select multiple groups of the process parameter variables, bring the brake drum size variables and each group of process parameter variables into the prediction model, obtain the shrinkage defect volume y corresponding to each group of the process parameter variables, and select the process parameter variable corresponding to the minimum shrinkage defect volume y as the optimal process parameter of the brake drum workpiece.

[0010] As a preferred solution of a method for predicting brake drum casting process parameters, the feedforward neural network includes an input layer, an output layer, and at least one hidden layer connected between the input layer and the output layer in a fully connected manner, the input layer includes nine input neurons, the nine input neurons receive the brake drum size variables and the process parameter variables respectively, the output layer includes one output neuron, and receives the shrinkage defect volume y during training;

[0011] Setting or determining through training the number of hidden layers, the number of neuron nodes in each hidden layer, and the activation function of each neuron node;

[0012] The feedforward neural network is trained by using a training data set to optimize weight parameters between the input layer and the hidden layer, between the hidden layers, and between the hidden layer and the output layer, as well as the bias parameters of each neuron node and the output neuron.

[0013] As a preferred solution of a method for predicting brake drum casting process parameters, the hidden layer is set to two layers, the number of neuron nodes in each hidden layer is 10, the activation function between the input layer and the first hidden layer adopts ReLU, and the output of the first hidden layer is:

[0014] h 1 =ReLU(W 1 ·x+b 1 )=max(0,W 1 ·x+b 1 ) (1)

[0015] Among them, the input vector x is the feature variable vector, x={D g , L 1 , L2 ,f,t,D m , W h ,T,v};

[0016] The activation function between the two hidden layers uses ReLU, which is located at the output of the second hidden layer:

[0017] h 2 =ReLU(W 2 ·h 1 +b 2 )=max(0,W 2 ·h 1 +b 2 ) (2)

[0018] Substituting formula (1) into formula (2), we get:

[0019] h 2 =max(0,W 2 max(0,W 1 ·x+b 1 )+b 2 ) (3)

[0020] The activation function between the second hidden layer and the output layer adopts Linear, and the output of the output layer is obtained:

[0021] y=Linear(W 3 ·h 2 +b 3 )=W 3 ·h 2 +b 3 (4)

[0022] Substituting formula (3) into formula (4), we get:

[0023] y=W 3 max(0,W 2 max(0,W 1 ·x+b 1 )+b 2 )+b 3 (5)

[0024] Among them, W 1 is the weight parameter between the input layer and the hidden layer, W 1 is a 10×9 matrix, b 1 is the bias parameter of the neuron node in the first hidden layer, b 1 is a 10×1 matrix, W 2 is the weight parameter between two hidden layers, W 2 is a 10×10 matrix, b 2is the bias parameter of the neuron node in the second hidden layer, b 2 is a 10×1 matrix, W 3 is the weight parameter between the second hidden layer and the output layer, W 3 is a 1×10 matrix, b 3 is the bias parameter of the output neuron, b 3 is a scalar.

[0025] As a preferred solution of a method for predicting process parameters of a brake drum casting, obtaining a training data set includes using a finite element model to obtain a mapping relationship between characteristic variables and shrinkage defect volume y:

[0026] The dimensions of the brake drum workpiece and the gating system structure are simplified, and different characteristic variables are used to describe them parametrically. Based on the different characteristic variables, multiple different parametric 3D models of the brake drum workpiece and the gating system structure are established.

[0027] Set simulation boundary conditions;

[0028] Different parameterized three-dimensional models are calculated and post-processed to obtain the shrinkage defect volume corresponding to each parameterized three-dimensional model.

[0029] As a preferred solution of a method for predicting brake drum casting process parameters, before obtaining a training data set, the following steps are also included:

[0030] Selecting a number of manufactured brake drum workpieces, each of which has a corresponding characteristic variable, performing an X-ray flaw detection test on each of the brake drum workpieces to obtain a first actual shrinkage defect volume corresponding to each of the brake drum workpieces;

[0031] Using a finite element model to obtain a first predicted shrinkage defect volume corresponding to each brake drum workpiece;

[0032] The error between the first actual shrinkage defect volume and the first predicted shrinkage defect volume is compared, and if the error is within a preset range, the training data set is obtained by calculating using the finite element model.

[0033] As a preferred solution of a method for predicting the process parameters of a brake drum casting, after obtaining the prediction model of the process parameters of the brake drum casting, the accuracy of the prediction model is verified, including the following steps:

[0034] Selecting a number of manufactured brake drum workpieces, each of which has a corresponding characteristic variable, performing an X-ray flaw detection test on each of the brake drum workpieces to obtain a second actual shrinkage defect volume corresponding to each of the brake drum workpieces;

[0035] Bringing the characteristic variable corresponding to each of the brake drum workpieces into the prediction model to obtain a second predicted shrinkage defect volume of each of the brake drum workpieces;

[0036] A prediction error of the prediction model is calculated according to the second actual shrinkage defect volume and the second predicted shrinkage defect volume.

[0037] As a preferred method for predicting the process parameters of brake drum casting, D g The value range is 320-500mm, L 1 The value range is 140-250mm, L 2 The value range of is 55-117mm, the value range of f is 40-65mm, the value range of t is 10-30mm, and D m The value range is 40-60mm, W h The value range of is 20-30mm, the value range of T is 1350-1450℃, and the value range of v is 2-4kg / s.

[0038] A prediction system for brake drum casting process parameters, comprising:

[0039] An input module, used for inputting brake drum dimension variables and multiple process parameter variables;

[0040] A calculation module, used for receiving the brake drum size variable and a plurality of process parameter variables input by the input module, and calculating the shrinkage defect volume y corresponding to each process parameter variable;

[0041] The comparison module is used to compare the shrinkage defect volumes y corresponding to different process parameter variables, and output the process parameter variable corresponding to the minimum shrinkage defect volume y.

[0042] The beneficial effects of the present invention are:

[0043] The present invention provides a method for predicting brake drum casting process parameters. A prediction model is obtained by training a feedforward neural network with a data training set. When brake drum workpieces of different sizes need to be cast, better process parameters can be obtained according to the prediction model, thereby facilitating the direct design of the casting system structure according to the process parameters without the need for repeated numerical simulation or trial and error optimization of experiments, thereby saving design and production time and cost and improving design and production efficiency.

[0044] The present invention also provides a prediction system for brake drum casting process parameters, which includes an input module, a calculation module and a comparison module. When a new casting system structure is designed for a brake drum workpiece that lacks an existing casting system structure, it is only necessary to input the brake drum size variables of the brake drum workpiece into the prediction system to obtain better process parameters. The new casting system structure can be directly designed based on the process parameters without repeated numerical simulation or trial and error optimization, thereby saving design and production time and cost and improving design and production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the contents of the embodiments of the present invention and these drawings without paying any creative work.

[0046] Figure 1 It is a flow chart of a method for predicting brake drum casting process parameters provided by an embodiment of the present invention;

[0047] Figure 2 It is a structural schematic diagram of a brake drum workpiece and a pouring system provided by an embodiment of the present invention;

[0048] Figure 3 is a cross-sectional view of a brake drum workpiece and a riser provided by an embodiment of the present invention;

[0049] Figure 4 is a cross-sectional view of a brake drum workpiece provided by an embodiment of the present invention;

[0050] Figure 5 is a cross-sectional view of a riser provided by an embodiment of the present invention;

[0051] Figure 6 is a top view of a pouring system provided by an embodiment of the present invention;

[0052] Figure 7 is a schematic diagram of the structure of a feedforward neural network provided by an embodiment of the present invention;

[0053] Figure 8 The present invention provides a flowchart of a system for predicting brake drum casting process parameters according to an embodiment of the present invention.

[0054] In the figure:

[0055] 1. Brake drum workpiece; 2. Riser; 3. Casting system. DETAILED DESCRIPTION

[0056] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0057] At present, the production process of gray cast iron brake drums is generally sand casting on a static pressure line. The raw materials include pig iron, scrap steel, recycled materials, carburizer, ferrochrome inoculant, etc. The content of each main element in the composition is controlled to be C: 3.2-3.4%, Si: 1.7-1.9%, Mn: 0.6-0.8%, S: 0.01-0.1%, P: 0.01-0.1%, Cu: 0.1-0.3%, Cr: 0.2-0.4%. The raw materials are smelted in a medium-frequency induction furnace with an iron-out temperature of about 1500°C. An automated static pressure production line is used for casting. Note: The length and width of the sand box are 1200mm and 900mm respectively, the height of the upper and lower sand boxes are 250mm and 350mm respectively, the mold is a one-mold two-cavity design, the template used is a metal template, and the selected molding sand is a clay sand mixed with raw sand, clay, sawdust, and coal. The sand mold is made by a multi-contact high-pressure sand machine. The preheating temperature of the sand box is 300℃. The pouring temperature is generally 1350-1450℃ depending on the size of the workpiece. The pouring time is generally 15-30s. After 100 minutes after pouring, the casting is taken out and the subsequent cleaning process is carried out. In traditional casting, most of them set an initial process parameter based on experience, and then gradually obtain better results through trial and error. This process not only consumes manpower to modify the mold and adjust the process equipment, which increases the cost, but also affects the operation process of the automated production line and delays delivery time. In current technology, technicians will use computing tools such as finite element simulation to simulate the casting process of the workpiece, predict the filling and solidification conditions of the workpiece under different process parameters in advance, and assist in adjusting the process parameters in actual production. However, the more suitable process parameters for brake drums of different sizes are different. Although using computing tools such as finite element simulation is more cost-effective than testing on the production line, there is also the problem of setting the initial process parameters based on experience and then gradually obtaining better results through trial and error, and it will consume a lot of computing resources and computing time costs.

[0058] Based on the above technical issues, such as Figures 1 to 7 As shown, this embodiment provides a method for predicting brake drum casting process parameters to solve the problem of economic and time waste caused by relying on simulation and production test trial and error in the prior art. The prediction method specifically includes the following steps:

[0059] S100: Obtaining a training data set, the training data set includes characteristic variables and shrinkage defect volume y, the characteristic variables include brake drum size variables and process parameter variables, the brake drum size variables include the brake drum cup inner diameter D g , the length of the vertical part of the drum wall L 1 , Length of transition from drum wall to flange L 2 , flange width f, drum wall thickness t, the process parameter variables include the upper diameter D of the riser m , the width of the pouring system runner W h , pouring temperature T and pouring rate v; among which, the inner diameter D of the brake drum cup in the brake drum size variable g It can also be replaced by the outer diameter of the brake drum cup.

[0060] Specifically, obtaining the training data set includes using a finite element model to obtain a mapping relationship between characteristic variables and shrinkage defect volume y:

[0061] S101: simplifying the dimensions of the brake drum workpiece and the gating system structure, using different characteristic variables to parametrically describe them, and establishing a plurality of different parametric three-dimensional models of the brake drum workpiece and the gating system structure according to the different characteristic variables;

[0062] Simplification refers to ignoring or setting the structural features that have little impact on the forming quality to a constant value, such as ignoring the slight wall thickness change of the bowl wall of the brake drum workpiece, ignoring the smaller cast hole at the flange, ignoring the reinforcing ribs at the bottom of the bowl mouth and the bowl wall, and setting the transition angle from the drum wall to the flange to 135°.

[0063] Among them, the selection of characteristic variables is to parametrically describe the various dimensions of the brake drum workpiece and the casting system structure, and then select the main dimensional features that have a significant impact on the forming quality in combination with the casting process conditions. Different characteristic variables represent brake drum workpieces and casting system structures of different sizes.

[0064] In this embodiment, the value range of each parameter of the characteristic variable is determined according to casting experience, D g The value range is 320-500mm, for example, D g The value of L is 320mm, 400mm, 450mm or 500mm, etc. 1 The value range is 140-250mm, for example, L 1 The value of L is 140mm, 200mm, 150mm, etc. 2 The value range is 55-117mm, for example, L 2The value of is 55mm, 100mm, 117mm, etc.; the value range of f is 40-65mm, for example, the value of f is 40mm, 50mm, 65mm, etc.; the value range of t is 10-30mm, for example, the value of t is 10mm, 20mm, 30mm, etc.; D m The value range is 40-60, for example, D m The values ​​of are 40mm, 50mm, 60mm; W h The value range is 20-30mm, for example, W h The value of is 20mm, 15mm or 30mm, etc.; the value range of T is 1350-1450℃, for example, the value of T is 1350℃, 1400℃, 1450℃, etc.; the value range of v is 2-4kg / s, for example, the value of v is 2kg / s, 3kg / s, 4kg / s, etc. The brake drum cast under the parameters beyond the range has large defects. Determining the value range of each parameter based on experience can effectively avoid wasting subsequent calculation resources.

[0065] S102: Setting simulation boundary conditions; for example, setting ambient temperature, type of heat transfer medium, material properties, workpiece displacement limit, etc.

[0066] S103: Calculate and post-process different parameterized three-dimensional models to obtain the shrinkage defect volume corresponding to each parameterized three-dimensional model. Record and organize multiple groups of characteristic variables and corresponding shrinkage defect volumes to obtain a training data set.

[0067] In order to improve the final prediction accuracy, the following steps are also included before obtaining the training data set:

[0068] S1021: Select several manufactured brake drum workpieces, each brake drum workpiece has a corresponding characteristic variable, perform an X-ray flaw detection test on each brake drum workpiece, and obtain a first actual shrinkage defect volume corresponding to each brake drum workpiece; wherein, it is optional but not limited to using ultrasonic non-destructive testing equipment to perform X-ray flaw detection test on the brake drum workpiece.

[0069] S1022: using a finite element model to obtain the first predicted shrinkage defect volume corresponding to each brake drum workpiece; wherein, the numerical simulation method refers to steps S101 and S102, and illustratively, the meshing adopts tetrahedral meshes with a size of 1-10 mm and a number of about 3 million; the initial time step is 1e-03s, the maximum time step for filling is 1e-01s, and the maximum time step for other stages is 1s, and the PETSC converger is used with a convergence value of 1e-04.

[0070] S1023: comparing the error between the first actual shrinkage defect volume and the first predicted shrinkage defect volume, and if the error is within a preset range, using a finite element model to calculate and obtain the training data set.

[0071] If the error between the first actual shrinkage defect volume and the first predicted shrinkage defect volume is within the preset range, it means that the established finite element model is reasonable, and the data of the training data set obtained through numerical simulation is close to the actual data, ensuring the prediction accuracy of the prediction model obtained after subsequent training, and then the finite element model is used to simulate and calculate to obtain the training data set; if the error between the first actual shrinkage defect volume and the first predicted shrinkage defect volume exceeds the preset range, adjust the finite element model, such as adjusting the simplified three-dimensional model and adjusting the boundary conditions, until the error between the first actual shrinkage defect volume and the first predicted shrinkage defect volume is within the preset range, and use the adjusted finite element model to calculate the characteristic variables within the range to obtain the training data set.

[0072] Exemplarily, the error between the first actual shrinkage defect volume and the first predicted shrinkage defect volume does not exceed ±10%. For example, 10 groups of manufactured brake drum workpieces are selected, and a finite element model is established to perform numerical simulation on the 10 groups of brake drum workpieces, and the shrinkage defect volumes corresponding to the characteristic variables of the 10 groups of brake drum workpieces are calculated. If the errors between the first actual shrinkage defect volumes and the first predicted shrinkage defect volumes corresponding to the 10 groups of brake drum workpieces do not exceed ±10%, subsequent training data acquisition is performed. For example, referring to Table 1, the finite element model is verified by experiment. In Table 1, the experimental shrinkage defect volume is the first actual shrinkage defect volume, and the simulated shrinkage defect volume is the first predicted shrinkage defect volume. It can be seen from Table 1 that the error between the first actual shrinkage defect volume and the first predicted shrinkage defect volume does not exceed ±10%.

[0073] S200: constructing a feedforward neural network, training the constructed feedforward neural network using a training data set, obtaining parameter values ​​of the feedforward neural network, and obtaining a prediction model for brake drum casting process parameters;

[0074] Specifically, step S200 includes:

[0075] S201: Use the pytorch framework to build a feedforward neural network, such as Figure 7 As shown, the feedforward neural network includes an input layer (input layer in the figure), an output layer (output layer in the figure), and at least one hidden layer (hidden layer in the figure) connected between the input layer and the output layer in a fully connected manner. The input layer includes nine input neurons, and the nine input neurons receive brake drum size variables and process parameter variables respectively. The output layer includes one output neuron, which receives shrinkage defect volume y during training;

[0076] S202: Setting or determining through training the number of hidden layers, the number of neuron nodes in each hidden layer, and the activation function of each neuron node;

[0077] In this embodiment, two hidden layers are set, the number of neuron nodes in each hidden layer is 10, the activation function between the input layer and the first hidden layer is ReLU, the activation function between the two hidden layers is ReLU, and the activation function between the second hidden layer and the output layer is linear activation, i.e. Linear. Of course, in other embodiments, the training data set can also be used to train the feedforward neural network using the stochastic gradient descent method, and the thinking evolution algorithm can be used to optimize the hyperparameters such as the learning rate, the number of hidden layers, and the number of neurons in each hidden layer. For example, the feature variable x in the training data set is set to {D g , L 1 , L 2 ,f,t,D m , W h , T, v} are input into the input layer of the feedforward neural network, and the shrinkage defect volume y in the training data set is input into the output layer of the feedforward neural network. The initial weight parameters and bias parameters are given for training to determine more appropriate hyperparameters.

[0078] S203: Train the feedforward neural network through the training data set, optimize the weight parameters between the input layer and the hidden layer, between each hidden layer, and between the hidden layer and the output layer, and optimize the bias parameters of each neuron node and the output neuron:

[0079] Extract the weight parameter W between the output layer and the first hidden layer 1 and the bias parameter b 1 , where W 1 and b 1 are 10x9 matrices and 10x1 vectors respectively. According to formula (1), the output h of the first hidden layer is obtained 1 :

[0080] h 1 =ReLU(W 1 ·x+b 1 )=max(0,W 1 ·x+b 1 ) (1)

[0081] Among them, the input vector x is the feature variable vector, x={D g , L 1 , L 2 ,f,t,D m , W h ,T,v};

[0082] Among them, ReLU (W 1 ·x+b 1 ) indicates the 1 ·x+b 1The elements in the vector are compared with 0 one by one, and the larger value is taken to replace the original W 1 ·x+b 1 The value of the corresponding element in the vector, that is, max(0,W 1 ·x+b 1 ).

[0083] Preferably, before training, the formula Normalize x and preprocess it, where x max and x mi n is the maximum and minimum value of each parameter in the feature variable vector. Normalization facilitates subsequent data processing and accelerates training convergence.

[0084] Then extract the weight parameter W between the first hidden layer and the second hidden layer 2 and the bias parameter b 2 , where W 2 and b 2 are 10x10 matrices and 10x1 vectors respectively. Then, the output h of the second hidden layer is obtained according to the following formula (2): 2 :

[0085] h 2 =ReLU(W 2 ·h 1 +b 2 )=max(0,W 2 ·h 1 +b 2 )(2)

[0086] Substituting formula (1) into formula (2), we get:

[0087] h 2 =max(0,W 2 max(0,W 1 ·x+b 1 )+b 2 )(3)

[0088] Then extract the weight parameter W between the second hidden layer and the output layer 3 and the bias parameter b 3 , where W 3 and b 3 are 1x10 matrices and scalars respectively, and the output y of the output layer is obtained according to the following formula (4):

[0089] y=Linear(W 3 ·h 2 +b 3 )=W 3 ·h 2 +b 3 (4)

[0090] Substituting formula (3) into formula (4), we get:

[0091] y=W 3 max(0,W 2 max(0,W 1 ·x+b 1 )+b 2 )+b 3 (5)

[0092] Among them, W 1 is the weight parameter between the input layer and the hidden layer, W 1 is a 10×9 matrix, b 1 is the bias parameter of the neuron node in the first hidden layer, b 1 is a 10×1 matrix, W 2 is the weight parameter between two hidden layers, W 2 is a 10×10 matrix, b 2 is the bias parameter of the neuron node in the second hidden layer, b 2 is a 10×1 matrix, W 3 is the weight parameter between the second hidden layer and the output layer, W 3 is a 1×10 matrix, b 3 is the bias parameter of the output neuron, b 3 is a scalar. That is, each input neuron receives a parameter in the feature variable, and all feature variables are weighted and input to each neuron node in the first layer. After adding the bias, if the threshold of the activation function is reached, the output is used as the weighted input of the neuron node of the next hidden layer, and is passed in sequence until the output value of the output neuron is obtained.

[0093] In the present embodiment, after obtaining the training data set, the training data set is divided into a training set, a validation set and a test set according to a certain ratio, for example, the training data set is divided into a training set, a validation set and a test set according to a ratio of 70%, 10% and 20%. Of course, in other embodiments, other ratios may also be used, for example, the training data set is divided into a training set, a validation set and a test set according to a ratio of 75%, 10% and 15%, which is specifically set according to actual needs. In step S203, the training set in the training data set is used to train the feedforward neural network. In the summary of the training process, the weight parameters between the input layer and the hidden layer, between each hidden layer, between the hidden layer and the output layer in the neural network, and the bias parameters of each neuron node and the output neuron will be continuously iterated and updated until the sum of the mean square error of the output value of the current neural network and the output value in the training set (i.e., the shrinkage defect volume value in the training set) is minimized. The validation set is used to detect the hyperparameters in the current neural network, such as the number of hidden layers, the number of neuron nodes in the hidden layer, the weight parameters, the bias parameters, etc., so as to combine the thinking evolution algorithm to obtain more appropriate values ​​of the hyperparameters, so that the neural network model can converge and have better accuracy during training. The test set is used to test the trained prediction model. The data in the test set does not overlap with the data in the training set, which can evaluate the generalization ability of the prediction model. The smaller the error of the test set test result, the higher the prediction accuracy of the prediction model.

[0094] After the prediction model is obtained by training the training data set, the expression in the above formula (5) is finally obtained. For example, 100 sets of training data sets are obtained by numerical simulation method, and the neural network is trained by the training data sets. After optimization, the following is obtained:

[0095]

[0096] W 3 =[0.3333 0.3098 0.2867 0.5432 0.4684 -0.1411 -0.0183 0.5277 -0.0883 0.3958],

[0097] b 3 =[0.6695];

[0098] W 1 , W 2 , W 3 , b 1 , b 2 , b 3 Substituting it into equation (5) and simplifying it, we can get the relationship between the brake drum shrinkage defect and the casting process parameters.

[0099] Furthermore, after obtaining the prediction model of the brake drum casting process parameters, the accuracy of the prediction model is verified, including the following steps:

[0100] S204: selecting a number of manufactured brake drum workpieces, each brake drum workpiece having a corresponding characteristic variable, performing an X-ray flaw detection test on each brake drum workpiece, and obtaining a second actual shrinkage defect volume corresponding to each brake drum workpiece;

[0101] S205: bringing the characteristic variables corresponding to each brake drum workpiece into the prediction model to obtain a second predicted shrinkage defect volume of each brake drum workpiece;

[0102] S206: Calculate a prediction error of the prediction model according to the second actual shrinkage defect volume and the second predicted shrinkage defect volume.

[0103] By calculating the error of the prediction model through experiments, it is convenient to adjust the prediction model according to the prediction error of the prediction model, that is, to further train and optimize the neural network so that the prediction model has higher prediction accuracy and smaller error, etc., which is conducive to improving the practicality of the prediction model.

[0104] For example, referring to Table 1, the prediction model is verified by experiments. In Table 1, the experimental shrinkage defect volume is the second actual shrinkage defect volume, and the shrinkage defect volume predicted by the prediction model is the second predicted shrinkage defect volume. It can be seen from Table 1 that the error between the second actual shrinkage defect volume and the second predicted shrinkage defect volume does not exceed ±15%

[0105] Table 1 Test verification results of logarithmic finite element model and prediction model

[0106]

[0107]

[0108] Furthermore, after obtaining a prediction model with a prediction accuracy within a reasonable range, the prediction model can be used to predict the brake drum casting process parameters:

[0109] S300: Measure each value of the brake drum size variable of the brake drum workpiece, select multiple groups of process parameter variables, bring the brake drum size variable and each group of process parameter variables into the prediction model, obtain the shrinkage defect volume y corresponding to each group of process parameter variables, and select the process parameter variable corresponding to the minimum shrinkage defect volume y as the optimal process parameter of the brake drum workpiece.

[0110] When it is necessary to cast brake drum workpieces of different sizes, the optimal process parameters can be obtained according to the prediction model, which makes it easy to design the casting system structure directly according to the process parameters, without the need for repeated numerical simulation or trial and error optimization, saving design and production time and cost and improving design and production efficiency.

[0111] like Figure 7 As shown, this embodiment also provides a prediction system for brake drum casting process parameters, which includes an input module, a calculation module and a comparison module. The input module is used to input brake drum size variables and multiple process parameter variables; the calculation module is used to receive the brake drum size variables and multiple process parameter variables input by the input module, and calculate the shrinkage defect volume y corresponding to each process parameter variable; the comparison module is used to compare the shrinkage defect volumes y corresponding to different process parameter variables, and output the process parameter variables corresponding to the minimum shrinkage defect volume y. Through the prediction system for brake drum casting process parameters, when a new gating system structure is designed for a brake drum workpiece lacking an existing gating system structure, it is only necessary to input the brake drum size variables of the brake drum workpiece into the prediction system to obtain better process parameters, and a new gating system structure can be directly designed according to the process parameters, without the need for repeated numerical simulation or trial and error optimization, saving design and production time and cost, and improving design and production efficiency.

[0112] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A method for predicting brake drum casting process parameters, characterized in that: include: Obtain a training data set, the training data set includes characteristic variables and shrinkage defect volume y, the characteristic variables include brake drum size variables and process parameter variables, the brake drum size variables include brake drum cup inner diameter D g , the length of the vertical part of the drum wall L1, the length of the transition part from the drum wall to the flange L2, the flange width f, the drum wall thickness t, and the process parameter variables include the upper diameter D of the riser m , the width of the pouring system runner W h , pouring temperature T and pouring rate v; Constructing a feedforward neural network, using the training data set to train the constructed feedforward neural network, obtaining parameter values ​​of the feedforward neural network, and obtaining a prediction model for brake drum casting process parameters; Measure each value of the brake drum size variable of the brake drum workpiece, select multiple groups of the process parameter variables, bring the brake drum size variables and each group of process parameter variables into the prediction model, obtain the shrinkage defect volume y corresponding to each group of the process parameter variables, and select the process parameter variable corresponding to the minimum shrinkage defect volume y as the optimal process parameter of the brake drum workpiece.

2. The method for predicting brake drum casting process parameters according to claim 1, characterized in that: The feedforward neural network includes an input layer, an output layer, and at least one hidden layer connected between the input layer and the output layer in a fully connected manner, the input layer includes nine input neurons, the nine input neurons receive the brake drum size variable and the process parameter variable respectively, and the output layer includes one output neuron, which receives the shrinkage defect volume y during training; Setting or determining through training the number of hidden layers, the number of neuron nodes in each hidden layer, and the activation function of each of the neuron nodes; The feedforward neural network is trained by using a training data set to optimize weight parameters between the input layer and the hidden layer, between the hidden layers, and between the hidden layer and the output layer, as well as the bias parameters of each neuron node and the output neuron.

3. The method for predicting brake drum casting process parameters according to claim 2, characterized in that: The number of hidden layers is set to two, namely the first hidden layer and the second hidden layer. The number of neuron nodes in each hidden layer is 10. The activation function between the input layer and the first hidden layer adopts ReLU, and the output of the first hidden layer is: h1=ReLU(W1·x+b1)=max(0,W1·x+b1) (1) Among them, the input vector x is the feature variable vector, x={D g , L1, L2, f, t, D m , W h ,T,v}; The activation function between the two hidden layers uses ReLU, which is located at the output of the second hidden layer: h2=ReLU(W2·h1+b2)=max(0,W2·h1+b2) (2) Substituting formula (1) into formula (2), we get: h2=max(0, W2·max(0,W1·x+b1)+b2) (3) The activation function between the second hidden layer and the output layer adopts Linear, and the output of the output layer is obtained: y=Linear(W3·h2+b3)=W3·h2+b3 (4) Substituting formula (3) into formula (4), we get: y=W3·max(0,W2·max(0,W1·x+b1)+b2)+b3 (5) Among them, W1 is the weight parameter between the input layer and the hidden layer, W1 is a 10×9 matrix, b1 is the bias parameter of the neuron node of the first hidden layer, b1 is a 10×1 matrix, W2 is the weight parameter between the two hidden layers, W2 is a 10×10 matrix, b2 is the bias parameter of the neuron node of the second hidden layer, b2 is a 10×1 matrix, W3 is the weight parameter between the second hidden layer and the output layer, W3 is a 1×10 matrix, b3 is the bias parameter of the output neuron, and b3 is a scalar.

4. The method for predicting brake drum casting process parameters according to claim 1, characterized in that: Acquiring a training data set includes using a finite element model to obtain a mapping relationship between characteristic variables and shrinkage defect volume y: The dimensions of the brake drum workpiece and the gating system structure are simplified, and different characteristic variables are used to describe them parametrically. Based on the different characteristic variables, multiple different parametric 3D models of the brake drum workpiece and the gating system structure are established. Set simulation boundary conditions; Different parameterized three-dimensional models are calculated and post-processed to obtain the shrinkage defect volume corresponding to each parameterized three-dimensional model.

5. The method for predicting brake drum casting process parameters according to claim 4, characterized in that: Before obtaining the training data set, the following steps are also included: Selecting a number of manufactured brake drum workpieces, each of which has a corresponding characteristic variable, performing an X-ray flaw detection test on each of the brake drum workpieces to obtain a first actual shrinkage defect volume corresponding to each of the brake drum workpieces; Using a finite element model to obtain a first predicted shrinkage defect volume corresponding to each brake drum workpiece; The error between the first actual shrinkage defect volume and the first predicted shrinkage defect volume is compared, and if the error is within a preset range, the training data set is obtained by calculating using the finite element model.

6. The method for predicting brake drum casting process parameters according to claim 1, characterized in that: After obtaining the prediction model of the brake drum casting process parameters, the accuracy of the prediction model is verified, including the following steps: Selecting a number of manufactured brake drum workpieces, each of which has a corresponding characteristic variable, performing an X-ray flaw detection test on each of the brake drum workpieces to obtain a second actual shrinkage defect volume corresponding to each of the brake drum workpieces; Bringing the characteristic variable corresponding to each of the brake drum workpieces into the prediction model to obtain a second predicted shrinkage defect volume of each of the brake drum workpieces; A prediction error of the prediction model is calculated according to the second actual shrinkage defect volume and the second predicted shrinkage defect volume.

7. The method for predicting brake drum casting process parameters according to claim 1, characterized in that: D g The value range of is 320-500mm, the value range of L1 is 140-250mm, the value range of L2 is 55-117mm, the value range of f is 40-65mm, the value range of t is 10-30mm, and D m The value range is 40-60mm, W h The value range of is 20-30mm, the value range of T is 1350-1450℃, and the value range of v is 2-4kg / s.

8. A prediction system for brake drum casting process parameters, characterized in that: include: An input module, used for inputting brake drum dimension variables and multiple process parameter variables; A calculation module, used for receiving the brake drum size variable and a plurality of process parameter variables input by the input module, and calculating the shrinkage defect volume y corresponding to each process parameter variable; The comparison module is used to compare the shrinkage defect volumes y corresponding to different process parameter variables, and output the process parameter variable corresponding to the minimum shrinkage defect volume y.

Images