Mechanical property prediction system and prediction method for hot-rolled steel bar after heat treatment

Abstract

The application discloses a kind of mechanical properties prediction system after hot-rolled steel bar heat treatment, the system includes database communication module, process control module, machine learning algorithm module, physical metallurgy model module.The system real-time monitoring collection field production data, and according to the production data of different steel rolling process, it is not under the condition of heat treatment to predict its mechanical properties after heat treatment according to national standard heat treatment process, including tensile yield strength, tensile tensile strength, tensile elongation after fracture, tensile yield ratio test, surface shrinkage.The application realizes the on-line prediction of the mechanical properties of 42CrMo steel bar after heat treatment by using deep convolutional neural network, and the relevant prediction results can be used to replace the actual heat treatment inspection process to reduce production cost.

Description

Technical Field

[0001] This invention belongs to the field of steel rolling technology, and particularly relates to a mechanical property prediction system and method for 42CrMo bars. Background Technology

[0002] In today's industrial society, hot-rolled bars are widely used in various applications, such as construction, shipbuilding, vehicle manufacturing, and furniture. These industries require bars to possess good mechanical properties to ensure project quality and avoid safety hazards. The mechanical properties of bars are closely related to their composition and processing parameters, such as heating, rolling, and cooling processes, and the sequence of these processes. Variations in the mechanical properties of bars due to the randomness of elemental composition and process parameters can cause several problems. First, bar production requires heat treatment before mechanical testing to determine if it meets the mechanical properties required for its intended use. This process typically takes several days, and if a portion of the steel produced in a batch fails to meet performance requirements, the entire batch may be discarded, resulting in significant waste. Second, sampling and testing processes require substantial manual labor, which in turn increases production costs and time. Given these drawbacks, there is an urgent need to design a new method for online prediction of the mechanical properties of bars.

[0003] Because the production process of bar stock is an extremely complex multi-factor coupled process, predicting the microstructure and properties of wire rods is quite difficult, making it one of the research directions of great interest to metallurgists both domestically and internationally. A large amount of basic research related to this topic has also been launched. Existing data shows that patent application 201010232151.1, a mechanical property prediction system for 40Cr wire rods, uses a BP neural network to predict the mechanical properties of 40Cr wire rods. However, the relevant model does not consider the influence of heat treatment data fluctuations, and the BP neural network structure is relatively simple, making it difficult to achieve high prediction accuracy. In addition, some studies have investigated the temperature, deformation, or simulated microstructure evolution of high-carbon steel wire rods from initial rolled slabs; however, there are no reports on the ability to predict the mechanical properties of 42CrMo bars after heat treatment. Summary of the Invention

[0004] Purpose of the invention: The purpose of this invention is to provide a system and method for predicting the mechanical properties of hot-rolled 42CrMo steel bars after heat treatment. This system can predict the mechanical properties of 42CrMo steel after heat treatment following the completion of the rolling process, replacing the heat treatment performance testing step and reducing production costs.

[0005] Technical Solution: The present invention provides a system for predicting the mechanical properties of hot-rolled steel bars after heat treatment. This system monitors and collects on-site production data in real time and, based on the rolling process data of different steels, predicts the mechanical properties (including tensile yield strength, tensile strength, elongation at break, tensile yield ratio, and reduction of area) after heat treatment according to national standard processes, without heat treatment. The system mainly consists of a database communication module, a process control module, a machine learning algorithm module, and a physical metallurgical model module. During the prediction process, the on-site production data of each batch of steel is input into the machine learning algorithm module, which calculates the preliminary mechanical properties and stores them in the on-site database.

[0006] This invention discloses a prediction method for a system for predicting the mechanical properties of hot-rolled steel bars after heat treatment, comprising the following steps:

[0007] (1) Process the experimental data of heat treatment mechanical properties stored in the database;

[0008] (2) A DNN neural network is used for reverse derivation. The input parameters are the experimental data of mechanical properties after heat treatment and the heat treatment process. The output parameters are the typical mechanical property parameters after eliminating the influence of fluctuations in heat treatment process parameters.

[0009] (3) After completing the data processing, utilize the processed heat treatment performance data and the production-related process data in the database;

[0010] (4) The deep convolutional neural network is trained based on the process data. During the training process, the production-related process data is first processed into a two-dimensional array in the form of 4 rows and 5 columns and input into the deep convolutional neural network for prediction. The deep convolutional neural network structure is 4*5 input parameters and 3*4+2*3+5 output parameters, where the output parameters are the processed heat treatment performance data.

[0011] Furthermore, in step (2), the DNN neural network structure is 20*100*5.

[0012] Furthermore, in step (2), the mechanical property test data include room temperature tensile yield strength, room temperature tensile tensile strength, room temperature tensile elongation after fracture, room temperature tensile yield strength ratio, and area shrinkage.

[0013] Further, in step (2), the heat treatment process includes the diameter of the heat-treated part, room temperature, minimum furnace temperature, maximum furnace temperature, intermediate furnace temperature, furnace temperature rise rate, billet temperature rise rate, furnace time, air cooling time, oil temperature, oil quenching time, water temperature, and water cooling time.

[0014] Furthermore, in step (2), the typical mechanical property parameters include average tensile yield strength, average tensile tensile strength, average tensile elongation after fracture, average tensile yield ratio, and average shrinkage.

[0015] Further, in step (3), the production-related process data includes the finished steel grade, finished diameter, thickness, width, length, carbon element, silicon element, manganese element, phosphorus element, aluminum element, vanadium element, titanium element, chromium element, furnace heating temperature, furnace heating time, initial rolling temperature, sizing inlet temperature, and heat treatment process number.

[0016] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0017] (1) Using production data such as steel alloy elements and production process information stored in the system database, a deep convolutional neural network was developed using Python software. The mechanical properties after heat treatment were inversely derived using the DNN neural network to eliminate the influence of fluctuations in heat treatment process parameters.

[0018] (2) This invention utilizes a deep convolutional neural network to achieve online prediction of the mechanical properties of 42CrMo steel bars after heat treatment. The prediction results can be used to replace the actual heat treatment inspection process and reduce production costs. Attached Figure Description

[0019] Figure 1 This is a diagram of a deep convolutional neural network structure.

[0020] Figure 2 This is a diagram of a deep fully connected neural network structure. Detailed Implementation

[0021] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0022] This invention provides a system for predicting the mechanical properties of hot-rolled 42CrMo steel bars after heat treatment. This system connects to the on-site MES (Manufacturing Execution System) database, continuously running and monitoring the database as a backend service. When new rolling production information is added, the system writes the relevant heat-treated mechanical properties of the product into the MES database after the rolling process is completed, allowing on-site personnel to query them on the MES terminal. The model is retrained every ten days based on the new production data and heat treatment data.

[0023] In actual operation, the experimental data of heat-treated mechanical properties stored in the database are first processed, and a DNN neural network is used for reverse derivation. The input parameters are the experimental data of mechanical properties after heat treatment (room temperature tensile yield strength, room temperature tensile tensile strength, room temperature tensile elongation after fracture, room temperature tensile yield strength ratio, and area reduction) and the heat treatment process (diameter of heat-treated part, room temperature, minimum furnace temperature, maximum furnace temperature, intermediate furnace temperature, heating rate, heating rate 2, furnace time, air cooling time, oil temperature, oil quenching time, water temperature, and water cooling time). The output parameters are typical mechanical property parameters (average tensile yield strength, average tensile strength, average tensile elongation after fracture, average tensile yield strength ratio, and average area reduction) after eliminating the influence of fluctuations in heat treatment process parameters. This processing method can converge the prediction data range and improve the accuracy of the prediction results. The DNN neural network structure is 20*100*5.

[0024] After data processing, a deep convolutional neural network was trained based on the processed heat treatment performance data and production-related process data from the database (20 process parameters including finished steel grade, finished diameter, thickness, width, length, carbon, silicon, manganese, phosphorus, aluminum, vanadium, titanium, chromium, furnace heating temperature, furnace heating time, initial rolling temperature, sizing inlet temperature, and heat treatment process number). During training, the 20 production-related process parameters were first processed into a two-dimensional array (4 rows, 5 columns) and input into the deep convolutional neural network for prediction. The deep convolutional neural network structure was 4*5 (input parameters) + 3*4 + 2*3 + 5 (output parameters), where the output parameters were the processed heat treatment performance data (average tensile yield strength, average tensile tensile strength, average tensile elongation after fracture, average tensile yield-to-tensile ratio, and average reduction of area).

Claims

1. A prediction method for a system for predicting the mechanical properties of hot-rolled steel bars after heat treatment, characterized in that, The mechanical property prediction system for hot-rolled steel bars after heat treatment includes a database communication module, a process control module, a machine learning algorithm module, and a physical metallurgical model module. During the prediction process, the on-site production data of each batch of steel is input into the machine learning algorithm module, which calculates the preliminary mechanical properties and stores them in the on-site database. The prediction method of the mechanical property prediction system for hot-rolled steel bars after heat treatment includes the following steps: (1) Process the experimental data of heat treatment mechanical properties stored in the database; (2) A DNN neural network is used for reverse derivation. The input parameters are the experimental data of mechanical properties after heat treatment and the heat treatment process. The output parameters are typical mechanical property parameters after eliminating the influence of fluctuations in heat treatment process parameters. The heat treatment process includes the diameter of the heat-treated part, room temperature, minimum furnace temperature, maximum furnace temperature, intermediate furnace temperature, furnace temperature rise rate, billet temperature rise rate, furnace time, air cooling time, oil temperature, oil quenching time, water temperature, and water cooling time. The typical mechanical property parameters include average tensile yield strength, average tensile tensile strength, average tensile elongation after fracture, average tensile yield ratio, and average area reduction. (3) After data processing is completed, the processed heat treatment performance data and production-related process data in the database are used; the production-related process data includes finished steel grade, finished diameter, thickness, width, length, carbon element, silicon element, manganese element, phosphorus element, aluminum element, vanadium element, titanium element, chromium element, furnace heating temperature, furnace heating time, initial rolling temperature, sizing inlet temperature, and heat treatment process number. (4) The deep convolutional neural network is trained based on the process data. During the training process, the production-related process data is first processed into a two-dimensional array in the form of 4 rows and 5 columns and input into the deep convolutional neural network for prediction. The deep convolutional neural network structure is 4*5 input parameters and 3*4+2*3+5 output parameters, where the output parameters are the processed heat treatment performance data.

2. The prediction method of the mechanical property prediction system for hot-rolled steel bars after heat treatment according to claim 1, characterized in that, In step (2), the DNN neural network structure is 20*100*5.

3. The prediction method of the mechanical property prediction system for hot-rolled steel bars after heat treatment according to claim 1, characterized in that, In step (2), the mechanical property test data include room temperature tensile yield strength, room temperature tensile tensile strength, room temperature tensile elongation after fracture, room temperature tensile yield strength ratio, and area shrinkage.

Images