A method and system for dynamic prediction of wall thickness of extruded pipes considering time delay

Abstract

The application discloses a kind of extruded pipe wall thickness dynamic prediction method and system considering time delay, its method includes: S1, obtains the extruded pipe wall thickness data in pipe extrusion process and the production data that influences extruded pipe wall thickness, as pipe extrusion basis data;S2, pipe extrusion basis data is preprocessed, and obtain preprocessed data set;S3, utilize genetic algorithm to carry out production data variable feature dimension reduction, and filter out the production related variable that is greater to extruded pipe wall thickness value Influence;S4, according to maximum information coefficient method, calculate the delay time of production related variable to extruded pipe wall thickness, obtain modeling data set;S5, based on DNN neural network, using step S4 obtains modeling data set and carries out training, to build extruded pipe wall thickness dynamic prediction model considering time delay;S6, using the trained extruded pipe wall thickness dynamic prediction model of step S5, the real-time extrusion wall thickness of target pipe is predicted.

Description

Technical Field

[0001] This invention belongs to the field of pipe forming process monitoring and intelligent control technology. Specifically, it relates to a method and system for dynamic prediction of extruded pipe wall thickness considering time delay. Background Technology

[0002] Pipe wall thickness is a core quality parameter for extruded pipes fitted with non-metallic idlers. The stability of the wall thickness directly determines the assembly accuracy of subsequent idlers, overall load-bearing strength, and finished product qualification rate, making it a key indicator for quality control in pipe extrusion production. During the pipe extrusion molding process, various process parameters, such as raw material properties, extruder screw speed, heating temperature of different sections of the barrel, vacuum pressure in the shaping section, and traction speed, are easily affected by factors such as equipment operating conditions, production environment, and raw material batches, causing dynamic fluctuations. This can easily lead to pipe wall thickness deviating from the preset standard tolerance range, resulting in product quality defects and production losses. Therefore, achieving high-precision, real-time online monitoring and prediction of wall thickness throughout the entire pipe extrusion process is a core requirement for ensuring the stability of pipe production quality and improving the level of intelligent management and control of the production line.

[0003] Currently, the mainstream pipe wall thickness monitoring and prediction solutions in the industry mainly rely on the extruder control system and ultrasonic thickness measuring equipment to collect production condition data and actual pipe wall thickness measurement data, and combine this with data-driven modeling algorithms to construct a wall thickness prediction model to achieve indirect prediction of pipe wall thickness. However, this technical solution has obvious defects in actual industrial applications: due to the limitations of production line equipment layout, there is a fixed physical distance between the ultrasonic thickness measuring equipment and the pipe extrusion molding point, and there is a fixed material transmission delay in the transmission of pipe material from the extrusion molding point to the thickness measurement detection point; at the same time, there are unavoidable signal transmission and processing delays in the entire data link of condition data acquisition, detection signal transmission, equipment data parsing and storage. Summary of the Invention

[0004] To address the issues of time delay and redundancy in the input data of existing pipe wall thickness dynamic prediction methods, and the asynchrony between wall thickness measurement and production condition data during extrusion, this invention provides a dynamic prediction method and system for extruded pipe wall thickness that considers time delay.

[0005] To achieve the above-mentioned technical objectives, the technical solution adopted by the present invention is as follows: A method for dynamically predicting the wall thickness of extruded pipes considering time delay includes the following steps: S1. Obtain the extruded pipe wall thickness data and production data affecting the extruded pipe wall thickness during the pipe extrusion process, as the basic data for pipe extrusion. ; S2, Extrude the pipe material from the base data Preprocessing is performed to obtain the preprocessed dataset. ; S3. Use a genetic algorithm to reduce the dimensionality of production data variables and screen out those that have a significant impact on the wall thickness of extruded pipes. One production-related variable; S4. Calculate based on the maximum information coefficient method. The modeling dataset is obtained by considering the time delay of production-related variables for the wall thickness of extruded pipes. ; S5. Based on the DNN neural network, the modeling dataset is obtained using step S4. The model is trained to construct a dynamic prediction model for the wall thickness of extruded pipes that takes into account time delay. S6. Using the trained dynamic prediction model for extruded pipe wall thickness obtained in step S5, the real-time extruded wall thickness of the target pipe is predicted.

[0006] Furthermore, production data that affect the wall thickness of extruded pipes include main machine speed, main machine current, feed speed, feed current, traction speed, traction current, vacuum pressure, heating chamber temperature, and die temperature.

[0007] Furthermore, the preprocessing method for the basic data of pipe extrusion in step S2 adopts normalization processing.

[0008] Furthermore, the detailed steps of step S3 include: S301. Construct a genetic algorithm fitness function based on a dynamic prediction model of extruded pipe wall thickness considering time delay; S302. Select a feature subset, normalize the feature subset, and divide it into a training set and a test set; S303. Feed the training set and test set into the learner for training and testing respectively, and calculate the fitness function; S304. Determine whether the fitness function value has reached the optimal value or the set number of iterations. If the fitness value has reached the optimal value or the set number of iterations, proceed to step S305. If the fitness value has not reached the optimal value or the set number of iterations, return to step S302 after the genetic operation to continue the optimization. S305. Output the optimal feature set, ultimately obtaining the feature set that has the greatest impact on the wall thickness of the extruded pipe. One production-related variable.

[0009] Furthermore, the objective function of the dynamic prediction model for extruded pipe wall thickness considering time delay during the training process is: in, The fitness function is MSE; A dynamic prediction model for the wall thickness of extruded pipes based on the DNN algorithm; The set is a combination of time-domain features The optimal feature set in.

[0010] Before dimensionality reduction, the chromosome encoding was set to binary; the learner used ELM; the fitness function was the mean squared error function; and the selection operator was selected using roulette wheel selection. The probability of each individual being selected is shown in the following formula: The crossover operator randomly sets the crossover point position using a single-point crossover method; the mutation operator mutates the original feature into other features by changing the binary code of the chromosome itself.

[0011] Furthermore, the detailed steps of step S4 include: S401. Select the extruded pipe wall thickness values ​​that have a significant impact from the optimal feature set. Each production-related variable is processed by time reconstruction to obtain the delay time calculation dataset corresponding to each related variable. The pipe extrusion wall thickness dataset is processed by time delay reconstruction to obtain the reconstructed dataset. S402. Calculate the information entropy between the column data of the delay time dataset and the pipe extrusion wall thickness dataset using the maximum information coefficient method; S403. Sort the obtained information entropy and find the column data corresponding to the column with the maximum information entropy value; S404. Delete all data except for the column corresponding to the maximum information entropy value. Repeat steps S401-S403 to obtain the delayed data. The sequence variables constitute the final modeling dataset.

[0012] Furthermore, step S5, which involves constructing a dynamic prediction model for the extruded pipe wall thickness based on a DNN neural network that considers time delay, includes: Define the hidden layer of the constructed DNN as layer l, and... As the data form of input variables for DNN, Defined as the output of a DNN; It is expressed as the output variable of the k-th hidden layer, where The number of neurons in the k-th hidden layer is defined as Therefore, the output of the i-th neuron in the k-th hidden layer The calculation formula is as follows: In the formula, The number of neurons in the (k-1)th hidden layer is defined as The output of the j-th neuron in the (k-1)-th hidden layer is , It represents the connection weight between the j-th neuron in the (k-1)-th hidden layer and the i-th neuron in the k-th hidden layer, while the symbol... This represents the bias of the i-th neuron in the k-th hidden layer; This represents the tanh activation function.

[0013] Furthermore, the specific formula for calculating the activation function is as follows: .

[0014] Furthermore, the formula for calculating the output expression of the DNN model is as follows: In the formula, variable b represents the bias of the output layer neuron, and variable... This represents the output of the i-th neuron in the l-th hidden layer, while the variable... This represents the weight between the i-th neuron in the l-th hidden layer and the neuron in the output layer.

[0015] Furthermore, in step S6, the real-time on-site operating data is input into the model trained in S5 to obtain the output of the extruded pipe wall thickness at the current moment.

[0016] A dynamic prediction system for extruded pipe wall thickness considering time delay includes a pipe extrusion basic data acquisition module, a pipe extrusion basic data preprocessing module, a production-related variable screening module, a modeling dataset calculation module, a dynamic prediction model construction module, and a pipe extrusion wall thickness prediction result output module. The pipe extrusion basic data acquisition module acquires the extruded pipe wall thickness data and the production data that affects the extruded pipe wall thickness during the pipe extrusion process, as the basic data for pipe extrusion. The pipe extrusion basic data preprocessing module preprocesses the pipe extrusion basic data to obtain a preprocessed dataset. The production-related variable screening module uses a genetic algorithm to reduce the dimensionality of production data variable features and screen out production-related variables that have a significant impact on the wall thickness of extruded pipes. The modeling dataset calculation module calculates the delay time of production-related variables with respect to the wall thickness of extruded pipes using the maximum information coefficient method, thus obtaining the modeling dataset. The dynamic prediction model construction module, based on the DNN neural network, uses the modeling dataset calculation module to obtain the modeling dataset for training, thereby constructing a dynamic prediction model of extruded pipe wall thickness that takes into account time delay. The pipe extrusion wall thickness prediction output module uses a trained dynamic prediction model for extruded pipe wall thickness to predict the real-time extrusion wall thickness of the target pipe.

[0017] Compared with the prior art, the present invention has the following advantages: By normalizing and preprocessing the basic data of pipe extrusion, and then using a genetic algorithm to reduce the dimensionality of production data variables, production-related variables with significant impact on the wall thickness of extruded pipes are selected. Based on the maximum information coefficient method, the time delay of these production-related variables with respect to the wall thickness of extruded pipes is calculated, resulting in a modeling dataset. This addresses the time lag and redundancy characteristics of the input data in dynamic prediction methods for pipe wall thickness. Furthermore, by constructing a dynamic prediction model for extruded pipe wall thickness that considers time delay, the problem of asynchronous wall thickness measurement and production condition data during pipe extrusion is resolved. Attached Figure Description

[0018] Figure 1 This is an overall flowchart of a dynamic prediction method for extruded pipe wall thickness that takes into account time delay, according to an embodiment of the present invention. Detailed Implementation

[0019] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to embodiments and accompanying drawings. The content mentioned in the embodiments is not intended to limit the present invention.

[0020] like Figure 1 As shown in the figure, this embodiment provides a method for dynamically predicting the wall thickness of extruded pipes considering time delay. The specific steps are as follows: S1: Obtain extruded pipe wall thickness data and production data affecting extruded pipe wall thickness during the pipe extrusion process, as basic data for pipe extrusion. ; S2: Extrude the pipe material from the base data Preprocessing is performed to obtain the preprocessed dataset. ; S3: Use a genetic algorithm to reduce the dimensionality of production data variables and screen out those that have a significant impact on the wall thickness of extruded pipes. One production-related variable; S4: Calculate based on the maximum information coefficient method. The modeling dataset is obtained by considering the time delay of production-related variables for the wall thickness of extruded pipes. ; S5: Based on the DNN neural network, obtain the modeling dataset using step S4. The model is trained to construct a dynamic prediction model for the wall thickness of extruded pipes that takes into account time delay. S6: Using the trained dynamic prediction model for extruded pipe wall thickness obtained in step S5, the real-time extruded wall thickness of the target pipe is predicted.

[0021] Step S1 involves obtaining the pipe extrusion process. Production dataset at time point ,in To affect the wall thickness of extruded pipes Several operating parameters (including main unit speed, main unit current, feeding speed, feeding current, traction speed, traction current, vacuum pressure, heating chamber temperature, and mold temperature). To generate a dataset for the wall thickness of extruded pipes. As shown in the following formula: Among them, dataset Contains There are 1 running variables, among which Indicates the first At that moment Variable data.

[0022] in, Indicates the first Extrusion wall thickness data at each moment.

[0023] Step S2 involves: targeting the dataset The preprocessed dataset is obtained by normalizing it using the following formula. : in, Based on the modeling dataset, The training dataset has been normalized. and The minimum and maximum values ​​in the dataset are used for basic modeling.

[0024] Step S3 specifically includes: S301: Construct the fitness function of the genetic algorithm. During the training process, the extruded pipe wall thickness dynamic prediction model considering time delay is based on the following objective function: in, The fitness function is MSE; A dynamic prediction model for the wall thickness of extruded pipes based on the DNN algorithm; The set is a combination of time-domain features The optimal feature set in; This represents the actual observed wall thickness of the extruded pipe for the i-th sample. For neural network weights; These are the weight coefficients of the hidden layer neurons; For neural network bias; To obtain the i-th set of input samples from the feature set D .

[0025] Before dimensionality reduction, the chromosome encoding was set to binary; the learner used ELM; the fitness function was the mean squared error function; and the selection operator was selected using roulette wheel selection. The probability of each individual being selected is shown in the following formula: The crossover operator randomly sets the crossover point position using a single-point crossover method; the mutation operator mutates the original feature into other features by changing the binary code of the chromosome itself.

[0026] S302: Select a subset of features, normalize the subset of features, and divide it into a training set and a test set; S303: Feed the training set and test set into the learner for training and testing respectively, and calculate the fitness function; S304: If the fitness value reaches the optimal level or the set number of iterations is reached, then the optimal feature set is output and the algorithm ends. Otherwise, after the genetic operation, return to step 2 to continue the optimization.

[0027] S305: The final result has a significant impact on the wall thickness of the extruded pipe. One production-related variable.

[0028] Step S4 specifically includes: S401: From the dataset After filtering Each variable is processed for time reconstruction, i.e., selecting... Using time as the starting variable, construct a delay time dataset to obtain a set of delay time calculation datasets corresponding to each relevant variable. : The pipe extrusion wall thickness dataset is reconstructed using time delay processing to obtain the reconstructed dataset. : in, ,express Real-time pipe extrusion wall thickness data, delayed data reconstruction and The length of each column of data is 1. .

[0029] S402: Calculate the delay time dataset using the maximum information coefficient method. of Column data and pipe extrusion wall thickness dataset The information entropy F between them is obtained. Information entropy, specifically expressed as ; S403: Obtain the information entropy Sort the data and find the maximum information entropy value. = The corresponding number Listed Column data; S404: Will Except Delete the data in that column, and repeat steps S41-S43 to obtain the delayed result. sequence variables ,in Represented as the first Under the variable, the first Delayed reconstruction data for columns, i.e., final modeling data .

[0030] Step S5 involves constructing a dynamic prediction model for the wall thickness of extruded pipes that considers time delay based on a DNN neural network. This specifically includes: Define the hidden layer of the constructed DNN as layer l, and... As the data form of input variables for DNN, Defined as the output of a DNN, where m represents the number of input variables. It is expressed as the output variable of the k-th hidden layer, where The number of neurons in the k-th hidden layer is defined as Therefore, the output of the i-th neuron in the k-th hidden layer The calculation formula is as follows:

[0031] In the formula, The number of neurons in the (k-1)th hidden layer is defined as The output of the j-th neuron in the (k-1)-th hidden layer is , It represents the connection weight between the j-th neuron in the (k-1)-th hidden layer and the i-th neuron in the k-th hidden layer, while the symbol... This represents the bias of the i-th neuron in the k-th hidden layer. This represents the tanh activation function, and its specific calculation formula is shown below.

[0032] The output expression of the DNN model is calculated as follows: In the formula, variable b represents the bias of the output layer neuron, and variable... This represents the output of the i-th neuron in the l-th hidden layer, while the variable... This represents the weight between the i-th neuron in the l-th hidden layer and the neuron in the output layer.

[0033] Step S6 includes: inputting the real-time on-site operating data into the model trained in S5 to obtain the output of the extruded pipe wall thickness at the current moment.

[0034] A dynamic prediction system for extruded pipe wall thickness considering time delay includes a pipe extrusion basic data acquisition module, a pipe extrusion basic data preprocessing module, a production-related variable screening module, a modeling dataset calculation module, a dynamic prediction model construction module, and a pipe extrusion wall thickness prediction result output module. The pipe extrusion basic data acquisition module acquires the extruded pipe wall thickness data and the production data that affects the extruded pipe wall thickness during the pipe extrusion process, as the basic data for pipe extrusion. The pipe extrusion basic data preprocessing module preprocesses the pipe extrusion basic data to obtain a preprocessed dataset. The production-related variable screening module uses a genetic algorithm to reduce the dimensionality of production data variable features and screen out production-related variables that have a significant impact on the wall thickness of extruded pipes. The modeling dataset calculation module calculates the delay time of production-related variables with respect to the wall thickness of extruded pipes using the maximum information coefficient method, thus obtaining the modeling dataset. The dynamic prediction model construction module, based on the DNN neural network, uses the modeling dataset calculation module to obtain the modeling dataset for training, thereby constructing a dynamic prediction model of extruded pipe wall thickness that takes into account time delay. The pipe extrusion wall thickness prediction output module uses a trained dynamic prediction model for extruded pipe wall thickness to predict the real-time extrusion wall thickness of the target pipe.

[0035] Compared with the prior art, the present invention has the following advantages: By normalizing and preprocessing the basic data of pipe extrusion, and then using a genetic algorithm to reduce the dimensionality of production data variables, production-related variables with significant impact on the wall thickness of extruded pipes are selected. Based on the maximum information coefficient method, the time delay of these production-related variables with respect to the wall thickness of extruded pipes is calculated, resulting in a modeling dataset. This addresses the time lag and redundancy characteristics of the input data in dynamic prediction methods for pipe wall thickness. Furthermore, by constructing a dynamic prediction model for extruded pipe wall thickness that considers time delay, the problem of asynchronous wall thickness measurement and production condition data during pipe extrusion is resolved.

[0036] The above provides a detailed description of a method and system for dynamic prediction of extruded pipe wall thickness considering time delay, as provided in this application. The specific embodiments are described only to aid in understanding the method and its core concepts. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the scope of protection of the claims.

Claims

1. A method for dynamically predicting the wall thickness of extruded pipes considering time delay, characterized in that, Including the following steps: S1. Obtain the wall thickness data of the extruded pipe and the production data that affect the wall thickness of the extruded pipe during the pipe extrusion process, as the basic data for pipe extrusion; S2. Preprocess the pipe extrusion base data to obtain a preprocessed dataset; S3. Use a genetic algorithm to reduce the dimensionality of production data variables and screen out production-related variables that have a significant impact on the wall thickness of extruded pipes. S4. Calculate the delay time of production-related variables with respect to the wall thickness of extruded pipes using the maximum information coefficient method to obtain the modeling dataset; S5. Based on the DNN neural network, the modeling dataset obtained in step S4 is used to train the modeling dataset to construct a dynamic prediction model for the wall thickness of extruded pipes that takes into account time delay. S6. Using the trained dynamic prediction model for extruded pipe wall thickness obtained in step S5, the real-time extruded wall thickness of the target pipe is predicted.

2. The method for dynamic prediction of extruded pipe wall thickness considering time delay according to claim 1, characterized in that, Production data that affect the wall thickness of extruded pipes include main machine speed, main machine current, feed speed, feed current, traction speed, traction current, vacuum pressure, heating chamber temperature, and die temperature.

3. The method for dynamic prediction of extruded pipe wall thickness considering time delay according to claim 2, characterized in that, In step S2, the preprocessing method for the basic data of pipe extrusion is normalization.

4. The method for dynamic prediction of extruded pipe wall thickness considering time delay according to claim 3, characterized in that, The detailed steps of step S3 include: S301. Construct a genetic algorithm fitness function based on a dynamic prediction model of extruded pipe wall thickness considering time delay; S302. Select a feature subset, normalize the feature subset, and divide it into a training set and a test set; S303. Feed the training set and test set into the learner for training and testing respectively, and calculate the fitness function; S304. Determine whether the fitness function value has reached the optimal value or the set number of iterations. If the fitness value has reached the optimal value or the set number of iterations, proceed to step S305. If the fitness value has not reached the optimal value or the set number of iterations, return to step S302 after the genetic operation to continue the optimization. S305. Output the optimal feature set to obtain the production-related variables that have a significant impact on the wall thickness of extruded pipes.

5. The method for dynamic prediction of extruded pipe wall thickness considering time delay according to claim 4, characterized in that, The objective function of the dynamic prediction model for extruded pipe wall thickness considering time delay during the training process is: in, The fitness function is MSE; A dynamic prediction model for the wall thickness of extruded pipes based on the DNN algorithm; The set is a combination of time-domain features The optimal feature set in; This represents the actual observed wall thickness of the extruded pipe for the i-th sample. For neural network weights; These are the weight coefficients of the hidden layer neurons; For neural network bias; To obtain the i-th set of input samples from the feature set D .

6. The method for dynamic prediction of extruded pipe wall thickness considering time delay according to claim 5, characterized in that, The detailed steps of step S4 include: S401. The production-related variables that have a significant impact on the wall thickness of the extruded pipe selected from the optimal feature set are subjected to time reconstruction processing to obtain the delay time calculation dataset corresponding to each related variable. The pipe extrusion wall thickness dataset is subjected to time delay reconstruction processing to obtain the reconstructed dataset. S402. Calculate the information entropy between the column data of the delay time dataset and the pipe extrusion wall thickness dataset using the maximum information coefficient method; S403. Sort the obtained information entropy and find the column data corresponding to the column with the maximum information entropy value; S404. Delete all data except for the column corresponding to the maximum information entropy value. Repeat steps S401-S403 to obtain the delayed data. The sequence variables constitute the final modeling dataset.

7. The method for dynamic prediction of extruded pipe wall thickness considering time delay according to claim 6, characterized in that, Step S5 involves constructing a dynamic prediction model for the wall thickness of extruded pipes that considers time delay based on a DNN neural network, including: Define the hidden layer of the constructed DNN as layer l, and... As the data form of input variables for DNN, Defined as the output of a DNN; It is expressed as the output variable of the k-th hidden layer, where The number of neurons in the k-th hidden layer is defined as Therefore, the output of the i-th neuron in the k-th hidden layer The calculation formula is as follows: In the formula, The number of neurons in the (k-1)th hidden layer is defined as The output of the j-th neuron in the (k-1)-th hidden layer is , It represents the connection weight between the j-th neuron in the (k-1)-th hidden layer and the i-th neuron in the k-th hidden layer, while the symbol... This represents the bias of the i-th neuron in the k-th hidden layer; This represents the tanh activation function.

8. The method for dynamic prediction of extruded pipe wall thickness considering time delay according to claim 7, characterized in that, The formula for calculating the output expression of a DNN model is: In the formula, variable b represents the bias of the output layer neuron, and variable... This represents the output of the i-th neuron in the l-th hidden layer, while the variable... This represents the weight between the i-th neuron in the l-th hidden layer and the neuron in the output layer.

9. The method for dynamic prediction of extruded pipe wall thickness considering time delay according to claim 1, characterized in that, In step S6, the real-time on-site operating data is input into the model trained in S5 to obtain the output of the extruded pipe wall thickness at the current moment.

10. A dynamic prediction system for extruded pipe wall thickness considering time delay, characterized in that, It includes a pipe extrusion basic data acquisition module, a pipe extrusion basic data preprocessing module, a production-related variable screening module, a modeling dataset calculation module, a dynamic prediction model construction module, and a pipe extrusion wall thickness prediction result output module; The pipe extrusion basic data acquisition module acquires the extruded pipe wall thickness data and the production data that affects the extruded pipe wall thickness during the pipe extrusion process, as the basic data for pipe extrusion. The pipe extrusion basic data preprocessing module preprocesses the pipe extrusion basic data to obtain a preprocessed dataset. The production-related variable screening module uses a genetic algorithm to reduce the dimensionality of production data variable features and screen out production-related variables that have a significant impact on the wall thickness of extruded pipes. The modeling dataset calculation module calculates the delay time of production-related variables with respect to the wall thickness of extruded pipes using the maximum information coefficient method, thus obtaining the modeling dataset. The dynamic prediction model construction module, based on the DNN neural network, uses the modeling dataset calculation module to obtain the modeling dataset for training, thereby constructing a dynamic prediction model of extruded pipe wall thickness that takes into account time delay. The pipe extrusion wall thickness prediction output module uses a trained dynamic prediction model for extruded pipe wall thickness to predict the real-time extrusion wall thickness of the target pipe.

Images