CN116000297A

CN116000297A - Preparation device and method for high-strength tungsten lanthanum wire

Info

Publication number: CN116000297A
Application number: CN202310004419.3A
Authority: CN
Inventors: 何学文; 徐俊良; 刘星星; 张洪杰
Original assignee: Ganzhou Sunny Non Ferrous Metals Co ltd
Current assignee: Ganzhou Sunny Non Ferrous Metals Co ltd
Priority date: 2023-01-03
Filing date: 2023-01-03
Publication date: 2023-04-25

Abstract

The application relates to the field of intelligent control, and particularly discloses a preparation device and a preparation method of a high-strength tungsten lanthanum wire, which are used for improving the sintering quality of a vertical smelting sintering strip by adaptively adjusting the temperature control of vertical smelting sintering based on the structural thermodynamic diagram of the vertical smelting sintering strip.

Description

Preparation device and method for high-strength tungsten lanthanum wire

Technical Field

The application relates to the field of intelligent control, and more particularly relates to a preparation device and a preparation method of a high-strength tungsten lanthanum wire.

Background

At present, tungsten wires used by bulbs are doped tungsten wires added with potassium, aluminum and silicon elements. However, the tungsten wire has the defects of small cold resistance, small hot resistance and poor photoelectric property, and the tungsten wire can be hardened during rotary forging and drawing in the process of processing and generating the tungsten wire, so that the internal stress is increased, the tungsten wire is easy to break during drawing, and the production difficulty is increased. The hardening of the alloy and the existence of internal stress lead to the production of tungsten filaments with shorter service life and poorer shock resistance, and the filaments are easy to sag after the bulb is used.

Therefore, a more optimized tungsten filament structure and its preparation are desired.

Disclosure of Invention

The present application has been made in order to solve the above technical problems. Embodiments of the present application provide a manufacturing apparatus of high strength tungsten lanthanum wire and a method thereof, which improve sintering quality of a sintered vertical melting bar by adaptively adjusting temperature control of the sintered vertical melting based on a structural thermodynamic diagram of the sintered vertical melting bar.

According to one aspect of the present application, there is provided a preparation device of a high-strength tungsten lanthanum wire, comprising:

the data monitoring and collecting unit is used for obtaining power values of the heater at a plurality of preset time points in a preset time period and thermodynamic diagrams of the vertical fusion sintering strips at the plurality of preset time points in the preset time period;

a sintered bar thermodynamic diagram coding unit, configured to obtain a thermodynamic diagram of a sintered bar at each predetermined time point in the thermodynamic diagrams of sintered bars at the predetermined time points by using a first convolutional neural network model of spatial attention;

the sintering bar thermodynamic diagram dynamic feature extraction unit is used for obtaining a thermodynamic change feature diagram of the space structure through a second convolution neural network model using a time attention mechanism;

The power time sequence feature coding unit is used for arranging power values of the heaters at a plurality of preset time points into power input vectors according to a time dimension and then obtaining power feature vectors by using a third convolution neural network model of a one-dimensional convolution kernel;

the dimension reduction unit is used for carrying out global mean value pooling on each feature matrix of the spatial structure thermodynamic change feature diagram along the channel dimension so as to obtain a spatial structure thermodynamic change feature vector;

the responsiveness estimation unit is used for calculating responsiveness estimation of the power characteristic vector relative to the space structure thermodynamic change characteristic vector so as to obtain a classification characteristic matrix; and

and the control result generating unit is used for passing the classification characteristic matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the power value of the heater at the current time point should be increased or decreased.

In the above-mentioned preparation facilities of high strength tungsten lanthanum silk, sintering strip thermodynamic diagram coding unit includes: a convolution coding subunit, configured to input a thermodynamic diagram of the vertical fusion sintering strips at each predetermined time point in the convolution coding subunit into a multi-layer convolution layer of the first convolution neural network model to obtain a thermodynamic distribution feature map; a spatial attention subunit, configured to pass the thermodynamic distribution feature map through a spatial attention module of the first convolutional neural network model to obtain a spatial attention feature matrix; and a notice applying subunit, configured to calculate a multiplication of each feature matrix of the thermodynamic distribution feature map along the channel dimension and the position-wise point of the spatial attention feature matrix to obtain the thermodynamic feature map of the spatial structure.

In the above-mentioned preparation facilities of high strength tungsten lanthanum silk, the space attention subunit is further used for: respectively carrying out global average pooling and global maximum pooling along the channel dimension on the thermodynamic distribution feature map by using the spatial attention module so as to obtain a first feature matrix and a second feature matrix; the first feature matrix and the second feature matrix are aggregated to obtain a space feature diagram; convolving the spatial feature map with the spatial attention module to obtain a spatial attention matrix; and performing nonlinear activation on the spatial attention moment matrix by using a Sigmoid function to obtain the spatial attention characteristic matrix.

In the above-mentioned preparation facilities of high strength tungsten lanthanum silk, sintering strip thermodynamic diagram dynamic characteristics draws unit includes: an extraction subunit, configured to extract a first spatial structure thermodynamic characteristic map and a second spatial structure thermodynamic characteristic map of two adjacent predetermined time points of the plurality of spatial structure thermodynamic characteristic maps; the time accumulation subunit is used for calculating the position-wise point multiplication of the first space structure thermodynamic characteristic diagram and the second space structure thermodynamic characteristic diagram to obtain a time accumulation space structure thermodynamic characteristic diagram; an activation subunit, configured to activate the time-cumulative spatial structure thermodynamic feature map with a Softmax activation function to obtain a time attention feature map; and a time attention applying subunit, configured to multiply the time attention feature map and the second space structure thermodynamic feature map by location points to obtain a time enhanced second space structure thermodynamic feature map corresponding to the second space structure thermodynamic feature map.

In the above-mentioned preparation device of high-strength tungsten lanthanum wire, the responsiveness estimation unit is further configured to: calculating the response estimation of the power feature vector relative to the space structure thermodynamic change feature vector by the following formula to obtain a classification feature matrix;

wherein, the formula is:

V ₂ ＝M*V

wherein V is ₂ And representing the power characteristic vector, V represents the thermal variation characteristic vector of the space structure, and M represents the classification characteristic matrix.

In the above-mentioned preparation device of high-strength tungsten lanthanum wire, the control result generating unit is further configured to: processing the classification feature matrix by using the classifier in the following formula to obtain a classification result;

wherein, the formula is: o=softmax { (W) _n ,B _n ):…:(W ₁ ,B ₁ )|Project(F)}，

Wherein Project (F) represents projecting the classification feature matrix as a vector, W ₁ To W _n For each layerWeight matrix of full connection layer, B ₁ To B _n Representing the bias vector for each fully connected layer.

In the preparation device of the high-strength tungsten lanthanum wire, the preparation device further comprises a training module for training the first convolution neural network model using spatial attention, the second convolution neural network model using a temporal attention mechanism and the third convolution neural network model using a one-dimensional convolution kernel.

In the above-mentioned preparation facilities of high strength tungsten lanthanum silk, training module includes: the training data acquisition unit is used for acquiring training data, wherein the training data comprises training power values of the heater at a plurality of preset time points in a preset time period, training thermodynamic diagrams of the vertical fusion sintering strips at a plurality of preset time points in the preset time period, and real label values of which the power values of the heater at the current time point are increased or reduced; the training sintering bar thermodynamic diagram coding unit is used for enabling training thermodynamic diagrams of the vertical fusion sintering bars at each preset time point in the training thermodynamic diagrams of the vertical fusion sintering bars at the preset time points to pass through the first convolution neural network model using the space attention so as to obtain a plurality of training space structure thermodynamic characteristic diagrams; the training sintering bar thermodynamic diagram dynamic characteristic extraction unit is used for enabling the training space structure thermodynamic characteristic diagrams to pass through the second convolution neural network model using the time attention mechanism so as to obtain training space structure thermodynamic variation characteristic diagrams; the training power time sequence feature coding unit is used for arranging training power values of the heaters at a plurality of preset time points into training power input vectors according to time dimension and obtaining training power feature vectors through the third convolution neural network model using the one-dimensional convolution kernel; the training dimension reduction unit is used for carrying out global averaging and pooling on each feature matrix of the training space structure thermodynamic change feature diagram along the channel dimension so as to obtain a training space structure thermodynamic change feature vector; the training response estimation unit is used for calculating the response estimation of the training power feature vector relative to the training space structure thermodynamic change feature vector so as to obtain a training classification feature matrix; the classification loss unit is used for passing the training classification characteristic matrix through a classifier to obtain a classification loss function value; a context loss unit, configured to calculate a context-counted local scene metric loss function value of the training spatial structure thermal variation feature vector, where the context-counted local scene metric loss function value is related to statistical features of feature value sets of all positions of the training spatial structure thermal variation feature vector; and a training unit for training the first convolutional neural network model using spatial attention, the second convolutional neural network model using temporal attention mechanism, and the third convolutional neural network model using one-dimensional convolution kernel with a weighted sum of the classification loss function value and the context-statistical local scene metric loss function value as a loss function value.

In the above-mentioned preparation device of high-strength tungsten lanthanum wire, the context loss unit is further configured to: calculating a local scene measurement loss function value of the context statistics of the training space structure thermodynamic change feature vector according to the following formula;

wherein, the formula is:

wherein v is _i Is the eigenvalue of each position of the thermodynamic change eigenvector of the space structure, and mu and sigma are the v _i L is the length of the spatially structured thermodynamic change feature vector relative to the mean and variance of the feature set of the spatially structured thermodynamic change feature vector.

According to another aspect of the present application, there is provided a method for preparing a high strength tungsten lanthanum wire, comprising:

acquiring power values of heaters at a plurality of preset time points in a preset time period and a thermodynamic diagram of a vertical fusion sintering strip at the plurality of preset time points in the preset time period;

obtaining a plurality of space structure thermodynamic characteristic diagrams through a first convolution neural network model of space attention according to thermodynamic diagrams of the vertical fusion sintering strips at each preset time point in the thermodynamic diagrams of the vertical fusion sintering strips at the preset time points;

the plurality of spatial structure thermodynamic characteristic diagrams are subjected to a second convolution neural network model using a time attention mechanism to obtain spatial structure thermodynamic change characteristic diagrams;

The power values of the heaters at a plurality of preset time points are arranged into power input vectors according to the time dimension, and then a power characteristic vector is obtained through a third convolution neural network model using a one-dimensional convolution kernel;

carrying out global averaging pooling on each feature matrix of the spatial structure thermodynamic change feature diagram along the channel dimension to obtain a spatial structure thermodynamic change feature vector;

calculating the response estimation of the power feature vector relative to the space structure thermodynamic change feature vector to obtain a classification feature matrix; and

the classification feature matrix is passed through a classifier to obtain a classification result, which indicates whether the power value of the heater at the current time point should be increased or decreased.

Compared with the prior art, the preparation device and the preparation method for the high-strength tungsten lanthanum wire are capable of improving the sintering quality of the vertical smelting sintering strip by adaptively adjusting the temperature control of vertical smelting sintering based on the structural thermodynamic diagram of the vertical smelting sintering strip.

Drawings

The foregoing and other objects, features and advantages of the present application will become more apparent from the following more particular description of embodiments of the present application, as illustrated in the accompanying drawings. The accompanying drawings are included to provide a further understanding of embodiments of the application and are incorporated in and constitute a part of this specification, illustrate the application and not constitute a limitation to the application. In the drawings, like reference numerals generally refer to like parts or steps.

FIG. 1 illustrates an application scenario diagram of a preparation apparatus for high strength tungsten lanthanum filaments according to an embodiment of the present application;

FIG. 2 illustrates a block diagram of a preparation apparatus for a high strength tungsten lanthanum wire in accordance with an embodiment of the present application;

FIG. 3 illustrates a block diagram of a training module in a high strength tungsten lanthanum wire preparation apparatus according to an embodiment of the present application;

FIG. 4 illustrates a system architecture diagram of an inference module in a high strength tungsten lanthanum wire fabrication apparatus according to an embodiment of the present application;

FIG. 5 illustrates a block diagram of a sintered bar thermodynamic diagram coding unit in a high strength tungsten lanthanum wire preparation apparatus according to an embodiment of the present application;

FIG. 6 illustrates a flowchart of a first convolutional neural network encoding process in a high strength tungsten lanthanum wire preparation apparatus in accordance with an embodiment of the present application;

FIG. 7 illustrates a block diagram of a sintered bar thermodynamic diagram dynamic feature extraction unit in a high strength tungsten lanthanum wire fabrication apparatus according to an embodiment of the present application;

FIG. 8 illustrates a system architecture diagram of a training module in a high strength tungsten lanthanum wire preparation apparatus according to an embodiment of the present application;

fig. 9 illustrates a flow chart of a method of preparing a high strength tungsten lanthanum wire according to an embodiment of the present application.

Detailed Description

Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be apparent that the described embodiments are only some of the embodiments of the present application and not all of the embodiments of the present application, and it should be understood that the present application is not limited by the example embodiments described herein.

Scene overview

As described above, the hardening of the alloy and the existence of the internal stress thereof lead to a short service life of the tungsten filament produced, poor shock resistance and easy sagging of the filament after use. Therefore, a more optimized tungsten filament structure and its preparation are desired.

Aiming at the technical problems, the applicant selects to add lanthanum element into tungsten wire, and controls the content of lanthanum element in tungsten alloy by changing the adding mode of lanthanum element, thereby changing the alloy composition and the size of internal crystal grains thereof to improve the performance of tungsten lanthanum wire. Those of ordinary skill in the art will appreciate that adding lanthanum to tungsten filaments to alter and optimize the properties of the tungsten filaments is a relatively common technique, the key being the process of their preparation.

Specifically, the applicant of the present application adopts the following preparation process, firstly, a tungsten blank strip doped with lanthanum element is prepared, and presintered under the protection of hydrogen to obtain a presintered strip; then, the presintered strip is subjected to vertical fusion sintering under the protection of hydrogen to obtain a vertical fusion sintered strip; and then, forging the vertical smelting sintering strip by using rotary forging equipment, so that the section of the vertical smelting sintering strip is gradually reduced, the length is increased, and the tungsten structure, the crystal grain state and the surface state in the blank strip are changed, so that the tungsten rod is prepared. Finally, the tungsten rod is made into the required tungsten filament through a drawing process.

Through researches, the applicant of the application finds that the structure of the vertical fusion sintering strip is particularly important for the forming quality of a final tungsten filament, in the existing vertical fusion sintering process, the vertical fusion sintering temperature is 2650-2950 ℃, the sintering time is 60-70 minutes, and the density of the obtained vertical fusion sintering strip is 17.2-17.6g/cm ³ However, due to fluctuation of temperature and time of the vertical fusion sintering, uniformity and uniformity of internal structure of the finally produced vertical fusion sintered bar are poor and it is difficult to maintain uniformity. Accordingly, an optimized high strength tungsten lanthanum wire manufacturing apparatus is desired that is capable of adaptively adjusting the temperature control of the green sintering based on the structural thermodynamic diagram of the green sintered strip to improve the sintering quality of the green sintered strip.

Specifically, in the technical scheme of the application, firstly, power values of heaters at a plurality of preset time points in a preset time period and thermodynamic diagrams of vertical fusion sintering strips at a plurality of preset time points in the preset time period are obtained. Here, considering that the sintering quality of the vertical fusion sintering is directly related to the thermodynamic distribution of the sintered bar during the vertical fusion sintering, a thermodynamic diagram of the vertical fusion sintered bar is selected as the monitoring data.

Next, the thermodynamic diagrams of the vertical fusion sintering strips at each preset time point in the thermodynamic diagrams of the vertical fusion sintering strips at the preset time points are obtained through a first convolution neural network model using spatial attention so as to obtain a plurality of spatial structure thermodynamic characteristic diagrams. That is, a deep convolutional neural network model is used as a feature extractor to capture the thermodynamic distribution characteristics of the vertical fusion sintering bar at each predetermined point in time, considering that the vertical fusion sintering bar is an organic whole, and capturing the thermodynamic distribution salient features of the vertical fusion sintering bar in its spatial domain is expected when extracting the thermodynamic distribution features. Therefore, in the technical solution of the present application, a spatial attention mechanism is integrated in the first convolutional neural network model.

Specifically, the encoding process of the first convolutional neural network includes: firstly, performing deep convolution coding on thermodynamic diagrams of the vertical fusion sintering strips at all preset time points by using a plurality of convolution layers of the first convolution neural network model to obtain thermodynamic distribution characteristic diagrams; the thermodynamic distribution profile is then subjected to an average pooling and maximum pooling operation along the channel, followed by a separation of the aggregated channel attention profile into two-dimensional maps: f (F) _MAX And F _Avg The average pooling feature and the maximum pooling feature in the channel are respectively represented, and are activated through a Sigmoid function after standard convolution, so that a two-dimensional space attention feature map is generated. And then multiplying the two-dimensional space attention feature map and each feature matrix of the thermodynamic distribution feature map along the channel dimension by position points to obtain the space structure thermodynamic feature map.

The plurality of spatially structured thermal profiles are then passed through a second convolutional neural network model using a temporal attention mechanism to obtain a spatially structured thermal variation profile. That is, the change in the thermal profile of the penumbra in the time dimension is captured by using a second convolutional neural network model of the time-attentive mechanism.

Specifically, the encoding process of the second convolutional neural network model includes: firstly, extracting two space structure thermodynamic characteristic diagrams of two adjacent preset time points in the plurality of space structure thermodynamic characteristic diagrams, and recording the two space structure thermodynamic characteristic diagrams as a first space structure thermodynamic characteristic diagram and a second space structure thermodynamic characteristic diagram; then, calculating the position-wise product of the first space structure thermodynamic characteristic diagram and the second space structure thermodynamic characteristic diagram to obtain a time accumulation space structure thermodynamic characteristic diagram, and activating the time accumulation space structure thermodynamic characteristic diagram by using a Softmax activation function to obtain a time attention characteristic diagram; and then, multiplying the time attention characteristic diagram and the second space structure thermodynamic characteristic diagram according to position points to obtain a time enhancement second space structure thermodynamic characteristic diagram corresponding to the second space structure thermodynamic characteristic diagram. And then, carrying out cyclic iteration in the same mode, wherein a time-enhanced spatial structure thermodynamic characteristic diagram corresponding to the spatial structure thermodynamic characteristic diagram at the last preset time point in the plurality of spatial structure thermodynamic characteristic diagrams is the spatial structure thermodynamic change characteristic diagram.

In the technical scheme of the application, the power values of the heaters at a plurality of preset time points in the preset time period are subjected to one-dimensional convolution coding by using a third convolution neural network model of a one-dimensional convolution kernel to obtain power feature vectors. Specifically, the power values of the heaters at the plurality of preset time points are firstly arranged into power input vectors according to a time dimension, then the power input vectors are used as the input of the third convolutional neural network model, and the pattern features of the power distribution in the local time span in the power input vectors are extracted by the third convolutional neural network model, so that the power feature vectors are obtained.

In particular, in the technical solution of the present application, the power value of the heater is the cause of the change in the thermal distribution of the vertical melting sintering bar, that is, the power value of the heater is the condition data, and the thermal distribution of the vertical melting sintering bar is the ending index. The influence of the power of the expression heater on the thermal distribution of the final vertical fusion sintering strip can be optimized by fully utilizing the high-dimensional implicit response rule of the condition data on the ending indexes. In particular, a responsiveness estimate of the power feature vector relative to the spatially structured thermodynamic change feature map is calculated to represent a high-dimensional implicit response pattern between the condition data to the ending index.

However, the power eigenvector is a one-dimensional eigenvector, and the spatial structure thermodynamic change eigenvector is a three-dimensional eigenvector, and there is a difference in eigenvector dimensions between the two, so that it is difficult to directly perform a responsiveness estimation operation at the data level. In order to facilitate operation, in the technical solution of the present application, dimensions of the spatial structure thermal variation feature map and the power feature vector are unified first, and in a specific example, global average pooling is performed on each feature matrix of the spatial structure thermal variation feature map along a channel dimension to obtain a spatial structure thermal variation feature vector. Then, a responsiveness estimate of the power feature vector relative to the spatially structured thermodynamic change feature vector is calculated to obtain a classification feature matrix. In a specific example, the implicit association of responsiveness between the power feature vector and the spatially structured thermodynamic change feature vector is represented by a transfer matrix of the two feature vectors.

Then, the transfer matrix is used as a classification feature matrix to pass through a classifier to obtain a classification result, wherein the classification result is used for representing that the power value of the heater at the current time point is increased or decreased. In this way, the temperature control of the vertical fusion sintering is adaptively adjusted based on the structural thermodynamic diagram of the vertical fusion sintering strip to improve the sintering quality of the vertical fusion sintering strip.

Particularly, in the technical solution of the present application, when global average pooling is performed on each feature matrix of the spatially-structured thermodynamic change feature map along the channel dimension to obtain a spatially-structured thermodynamic change feature vector, since the feature value of each position of the spatially-structured thermodynamic change feature vector is obtained by global average pooling of the feature matrix of the corresponding channel position of the spatially-structured thermodynamic change feature map, the context correlation between the feature values of the spatially-structured thermodynamic change feature vector is weaker than the context correlation of the spatially-structured thermodynamic change feature map along the channel dimension, so that the context feature correlation expression capability of the spatially-structured thermodynamic change feature vector on the spatially-structured thermodynamic change feature map is weakened, and the distribution consistency of the spatially-structured thermodynamic change feature vector and the power feature vector in the time sequence dimension is poor, and the classification effect of the classification feature matrix is affected.

Therefore, in order to promote the context feature-dependent expression capability of the spatially-structured thermodynamic change feature vector for the spatially-structured thermodynamic change feature map, a local scene metric loss function for context statistics of the spatially-structured thermodynamic change feature vector is preferably introduced, expressed as:

Here, μ and σ are feature sets v _i Mean and variance of e V, V _i Is the eigenvalue of each position of the spatial structure thermodynamic change eigenvector V, and L is the length of the spatial structure thermodynamic change eigenvector V.

Here, the context statistical local scene metric loss function is based on global pooling of each feature matrix along a channel of the spatial structure thermodynamic change feature map, and feature values of each position of the spatial structure thermodynamic change feature vector V are regarded as separate channel feature descriptors to serve as a squeezing representation of a channel scene of a feature, so that the relevance of the local scene of each channel can be improved based on the context statistical metric of the class probability expression of the feature set as the loss function, and the context feature relevance expression capability of the spatial structure thermodynamic change feature vector for the spatial structure thermodynamic change feature map is improved. And the distribution consistency of the thermal variation feature vector of the space structure and the power feature vector in the time sequence dimension can be improved, so that the classification effect of the classification feature matrix is improved. That is, the accuracy of the temperature self-adaptive control of the vertical fusion sintering process is improved, so that the molding quality of the vertical fusion sintering strip is improved.

Based on this, this application provides a preparation facilities of high strength tungsten lanthanum silk, and it includes: the data monitoring and collecting unit is used for obtaining power values of the heater at a plurality of preset time points in a preset time period and thermodynamic diagrams of the vertical fusion sintering strips at the plurality of preset time points in the preset time period; a sintered bar thermodynamic diagram coding unit, configured to obtain a thermodynamic diagram of a sintered bar at each predetermined time point in the thermodynamic diagrams of sintered bars at the predetermined time points by using a first convolutional neural network model of spatial attention; the sintering bar thermodynamic diagram dynamic feature extraction unit is used for obtaining a thermodynamic change feature diagram of the space structure through a second convolution neural network model using a time attention mechanism; the power time sequence feature coding unit is used for arranging power values of the heaters at a plurality of preset time points into power input vectors according to a time dimension and then obtaining power feature vectors by using a third convolution neural network model of a one-dimensional convolution kernel; the dimension reduction unit is used for carrying out global mean value pooling on each feature matrix of the spatial structure thermodynamic change feature diagram along the channel dimension so as to obtain a spatial structure thermodynamic change feature vector; the responsiveness estimation unit is used for calculating responsiveness estimation of the power characteristic vector relative to the space structure thermodynamic change characteristic vector so as to obtain a classification characteristic matrix; and a control result generation unit for passing the classification feature matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the power value of the heater at the current time point should be increased or decreased.

Fig. 1 illustrates an application scenario diagram of a preparation device of a high-strength tungsten lanthanum wire according to an embodiment of the application. As shown in fig. 1, in this application scenario, power values of heaters at a plurality of predetermined time points within a predetermined period are acquired by a power meter (e.g., P1 as illustrated in fig. 1), and a thermodynamic diagram of a sintered strand of vertical fusion at a plurality of predetermined time points within the predetermined period is acquired by a thermal infrared camera (e.g., C as illustrated in fig. 1) (e.g., P2 as illustrated in fig. 1). Next, the above information is input to a server (e.g., S in fig. 1) in which a preparation algorithm for a high-strength tungsten lanthanum wire is deployed, wherein the server is capable of processing the above input information with the preparation algorithm for a high-strength tungsten lanthanum wire to generate a classification result indicating whether the power value of the heater at the current point in time should be increased or decreased.

Having described the basic principles of the present application, various non-limiting embodiments of the present application will now be described in detail with reference to the accompanying drawings.

Exemplary System

Fig. 2 illustrates a block diagram of a preparation apparatus for a high strength tungsten lanthanum wire according to an embodiment of the present application. As shown in fig. 2, a preparation apparatus 300 of a high-strength tungsten lanthanum wire according to an embodiment of the present application includes an inference module, where the inference module includes: a data monitoring and acquisition unit 310; a sintered bar thermodynamic diagram coding unit 320; a sintered bar thermodynamic diagram dynamic feature extraction unit 330; a power timing characteristic encoding unit 340; a dimension reduction unit 350; a responsiveness estimation unit 360; and a control result generation unit 370.

The data monitoring and collecting unit 310 is configured to obtain power values of the heater at a plurality of predetermined time points in a predetermined time period and thermodynamic diagrams of the vertical fusion sintering strips at a plurality of predetermined time points in the predetermined time period; the sintering bar thermodynamic diagram coding unit 320 is configured to obtain a thermodynamic diagram of a vertical fusion sintering bar at each predetermined time point in the thermodynamic diagrams of the vertical fusion sintering bars at the predetermined time points by using a first convolution neural network model of spatial attention so as to obtain a plurality of spatial structure thermodynamic characteristic diagrams; the sintering bar thermodynamic diagram dynamic feature extraction unit 330 is configured to obtain a thermodynamic change feature map of the spatial structure by using a second convolutional neural network model of a time attention mechanism; the power timing characteristic encoding unit 340 is configured to arrange power values of the heaters at the plurality of predetermined time points into power input vectors according to a time dimension, and obtain power characteristic vectors by using a third convolutional neural network model of a one-dimensional convolutional kernel; the dimension reduction unit 350 is configured to perform global averaging pooling on each feature matrix of the spatial structure thermodynamic change feature map along the channel dimension to obtain a spatial structure thermodynamic change feature vector; the responsiveness estimation unit 360 is configured to calculate a responsiveness estimation of the power feature vector relative to the spatially structured thermal variation feature vector to obtain a classification feature matrix; and the control result generating unit 370 is configured to pass the classification feature matrix through a classifier to obtain a classification result, where the classification result is used to indicate that the power value of the heater at the current time point should be increased or decreased.

Fig. 4 illustrates a system architecture diagram of an inference module in a preparation apparatus of a high strength tungsten lanthanum wire according to an embodiment of the present application. As shown in fig. 4, in the system architecture of the apparatus 300 for preparing high-strength tungsten lanthanum wire, in the process of inference, the data monitoring and collecting unit 310 is used to obtain power values of heaters at a plurality of predetermined time points in a predetermined time period and thermodynamic diagrams of vertical fusion sintering strips at a plurality of predetermined time points in the predetermined time period; the sintering bar thermodynamic diagram coding unit 320 obtains a plurality of space structure thermodynamic characteristic diagrams by using a first convolution neural network model of space attention from thermodynamic diagrams of the vertical fusion sintering bars at each preset time point in the thermodynamic diagrams of the vertical fusion sintering bars at the preset time points acquired by the data monitoring and collecting unit 310; secondly, the sintered bar thermodynamic diagram dynamic feature extraction unit 330 uses the plurality of spatial structure thermodynamic feature diagrams generated by the sintered bar thermodynamic diagram coding unit 320 to obtain a spatial structure thermodynamic change feature diagram through a second convolution neural network model using a time attention mechanism; meanwhile, the power timing characteristic encoding unit 340 arranges the power values of the heaters at a plurality of predetermined time points acquired by the data monitoring and collecting unit 310 into a power input vector according to a time dimension, and then obtains a power characteristic vector by using a third convolutional neural network model of a one-dimensional convolutional kernel; next, the dimension reduction unit 350 performs global averaging pooling on each feature matrix of the spatial structure thermodynamic change feature map along the channel dimension to obtain a spatial structure thermodynamic change feature vector; then, the responsiveness estimation unit 360 calculates responsiveness estimation of the power feature vector generated by the power timing feature encoding unit 340 with respect to the spatial structure thermal variation feature vector generated by the dimension reduction unit 350 to obtain a classification feature matrix; further, the control result generation unit 370 passes the classification feature matrix through a classifier to obtain a classification result indicating whether the power value of the heater at the current time point should be increased or decreased.

Specifically, during the operation of the apparatus 300 for preparing a high-strength tungsten lanthanum wire, the data monitoring and collecting unit 310 is configured to obtain power values of the heater at a plurality of predetermined time points within a predetermined time period and thermodynamic diagrams of the vertical fusion sintering strip at a plurality of predetermined time points within the predetermined time period. Considering that the sintering quality of the vertical smelting sintering is directly related to the thermodynamic distribution of the sintering bar in the vertical smelting sintering process, the thermodynamic diagram of the vertical smelting sintering bar is selected as monitoring data. In the technical scheme of the application, the power values of the heaters at a plurality of preset time points in a preset time period can be obtained through the power meter, and the thermodynamic diagrams of the vertical fusion sintering strips at the plurality of preset time points in the preset time period are obtained.

Specifically, during operation of the apparatus 300 for producing high-strength lanthanum tungsten wire, the sintered rod thermodynamic diagram encoding unit 320 is configured to obtain a plurality of spatial structure thermodynamic characteristic diagrams by using a first convolution neural network model of spatial attention from thermodynamic diagrams of the sintered rods at each predetermined time point among thermodynamic diagrams of the sintered rods at the predetermined time points. That is, a deep convolutional neural network model is used as a feature extractor to capture the thermodynamic distribution characteristics of the vertical fusion sintering bar at each predetermined point in time, considering that the vertical fusion sintering bar is an organic whole, and capturing the thermodynamic distribution salient features of the vertical fusion sintering bar in its spatial domain is expected when extracting the thermodynamic distribution features. Therefore, in the technical solution of the present application, a spatial attention mechanism is integrated in the first convolutional neural network model. More specifically, in the technical solution of the present application, the process of encoding the thermodynamic diagrams at the respective predetermined time points using the first convolutional neural network model with a spatial attention mechanism to obtain the spatial structural thermodynamic characteristic diagram includes: first convolutionally encoding the thermodynamic diagrams of the plurality of vertical fusion strips at predetermined points in time using a convolutionally layer of the spatial attention module to obtain a spatial attention diagram; inputting the spatial attention map into a Softmax activation function of the spatial attention module to obtain a spatial attention score map; and calculating the space attention score graph and the thermal force graph of the vertical fusion sintering strips at a plurality of preset time points, and multiplying the position by the thermal force graph of the vertical fusion sintering strips at the preset time points to obtain a plurality of space structure thermal characteristic graphs. Wherein convolutionally encoding the thermodynamic diagram of the vertical fusion sinters at the plurality of predetermined points in time using the convolution layer of the spatial attention module includes extracting local features of the thermodynamic diagram that need to be focused in a spatial domain by convolution kernels inside the convolution layer to obtain the spatial attention diagram. The spatial attention map is then input to a Softmax activation function of the spatial attention module, which functions to map the eigenvalues of the individual locations in the spatial attention map into a probability space of 0 to 1, and the sum of the eigenvalues of all locations of the spatial attention score map is 1.

Fig. 5 illustrates a block diagram of a sintered bar thermodynamic diagram coding unit in a manufacturing apparatus for high strength tungsten lanthanum filaments according to an embodiment of the present application. As shown in fig. 5, the sintered bar thermodynamic diagram coding unit 320 includes: a convolutional encoding subunit 321, configured to input a thermodynamic diagram of the vertical fusion sintered bar at each predetermined time point in the convolutional encoding subunit into a multi-layer convolutional layer of the first convolutional neural network model to obtain a thermodynamic distribution feature map; a spatial attention subunit 322, configured to pass the thermodynamic distribution feature map through a spatial attention module of the first convolutional neural network model to obtain a spatial attention feature matrix; and, an attention applying subunit 323, configured to calculate a multiplication of each feature matrix of the thermodynamic distribution feature map along the channel dimension and the spatial attention feature matrix by a position point to obtain the thermodynamic feature map of the spatial structure.

In a specific example of the present application, the encoding process of the first convolutional neural network includes: firstly, performing deep convolution coding on thermodynamic diagrams of the vertical fusion sintering strips at all preset time points by using a plurality of convolution layers of the first convolution neural network model to obtain thermodynamic distribution characteristic diagrams; the thermodynamic distribution profile is then subjected to an average pooling and maximum pooling operation along the channel, followed by a separation of the aggregated channel attention profile into two-dimensional maps: f (F) _MAX And F _Avg The average pooling feature and the maximum pooling feature in the channel are respectively represented, and are activated through a Sigmoid function after standard convolution, so that a two-dimensional space attention feature map is generated. And then multiplying the two-dimensional space attention feature map and each feature matrix of the thermodynamic distribution feature map along the channel dimension by position points to obtain the space structure thermodynamic feature map.

Fig. 6 illustrates a flowchart of a first convolutional neural network encoding process in a preparation apparatus for high strength tungsten lanthanum filaments according to an embodiment of the present application. As shown in fig. 6, in the first convolutional neural network coding process, it includes: s210, carrying out global average pooling and global maximum pooling along the channel dimension on the thermodynamic distribution feature map by using the spatial attention module to obtain a first feature matrix and a second feature matrix; s220, aggregating the first feature matrix and the second feature matrix to obtain a space feature map; s230, convolving the spatial feature map by using the spatial attention module to obtain a spatial attention matrix; and S240, performing nonlinear activation on the spatial attention moment array by using a Sigmoid function to obtain the spatial attention characteristic matrix.

Specifically, during operation of the apparatus 300 for preparing a high-strength tungsten lanthanum wire, the sintering bar thermodynamic diagram dynamic feature extraction unit 330 is configured to obtain a thermodynamic change signature of a spatial structure by using a second convolutional neural network model of a time attention mechanism. That is, the change in the thermal profile of the penumbra in the time dimension is captured by using a second convolutional neural network model of the time-attentive mechanism. More specifically, each layer of the second convolutional neural model using the temporal attention mechanism performs on input data in forward transfer of the layer: carrying out convolution processing on input data to obtain a convolution characteristic diagram; pooling the convolution feature map along a channel dimension to obtain a pooled feature map; performing nonlinear activation on the pooled feature map to obtain an activated feature map; the output of the last layer of the second convolution nerve using the time attention mechanism is the space structure thermodynamic change characteristic diagram, and the input of the first layer of the second convolution nerve using the time attention mechanism is the space structure thermodynamic change characteristic diagram.

Fig. 7 illustrates a block diagram of a sintered bar thermodynamic diagram dynamic feature extraction unit in a high strength tungsten lanthanum wire fabrication apparatus according to an embodiment of the present application. As shown in fig. 7, the sintered bar thermodynamic diagram dynamic feature extraction unit 330 includes: an extraction subunit 331, configured to extract a first spatial structure thermal feature map and a second spatial structure thermal feature map of two adjacent predetermined time points of the plurality of spatial structure thermal feature maps; a time accumulation subunit 332, configured to calculate a point-by-point multiplication of the first spatial structure thermodynamic characteristic map and the second spatial structure thermodynamic characteristic map to obtain a time accumulation spatial structure thermodynamic characteristic map; an activation subunit 333, configured to activate the time-cumulative spatial structure thermodynamic signature with a Softmax activation function to obtain a time attention signature; and a time attention applying subunit 334, configured to multiply the time attention profile and the second spatial structure thermodynamic profile by location points to obtain a time enhanced second spatial structure thermodynamic profile corresponding to the second spatial structure thermodynamic profile.

In a specific example of the present application, the encoding process of the second convolutional neural network model includes: firstly, extracting two space structure thermodynamic characteristic diagrams of two adjacent preset time points in the plurality of space structure thermodynamic characteristic diagrams, and recording the two space structure thermodynamic characteristic diagrams as a first space structure thermodynamic characteristic diagram and a second space structure thermodynamic characteristic diagram; then, calculating the position-wise product of the first space structure thermodynamic characteristic diagram and the second space structure thermodynamic characteristic diagram to obtain a time accumulation space structure thermodynamic characteristic diagram, and activating the time accumulation space structure thermodynamic characteristic diagram by using a Softmax activation function to obtain a time attention characteristic diagram; and then, multiplying the time attention characteristic diagram and the second space structure thermodynamic characteristic diagram according to position points to obtain a time enhancement second space structure thermodynamic characteristic diagram corresponding to the second space structure thermodynamic characteristic diagram. And then, carrying out cyclic iteration in the same mode, wherein a time-enhanced spatial structure thermodynamic characteristic diagram corresponding to the spatial structure thermodynamic characteristic diagram at the last preset time point in the plurality of spatial structure thermodynamic characteristic diagrams is the spatial structure thermodynamic change characteristic diagram.

Specifically, during the operation of the apparatus 300 for preparing a high-strength tungsten lanthanum wire, the power time sequence feature encoding unit 340 is configured to arrange power values of the heaters at the plurality of predetermined time points into a power input vector according to a time dimension, and obtain a power feature vector by using a third convolutional neural network model of a one-dimensional convolutional kernel. In the technical scheme of the application, firstly, the power values of the heaters at a plurality of preset time points are arranged into power input vectors according to a time dimension, then the power input vectors are used as the input of the third convolutional neural network model, and the mode features of the power distribution in the local time span in the power input vectors are extracted by the third convolutional neural network model, so that the power feature vectors are obtained.

Specifically, during the operation process of the apparatus 300 for preparing a high-strength tungsten lanthanum wire, the dimension reduction unit 350 is configured to perform global averaging pooling on each feature matrix of the spatial structure thermodynamic change feature map along the channel dimension to obtain a spatial structure thermodynamic change feature vector. It should be understood that the spatial structure thermodynamic change feature map and the power feature vector are first dimension unified, and in a specific example, each feature matrix of the spatial structure thermodynamic change feature map along the channel dimension is globally averaged and pooled to obtain a spatial structure thermodynamic change feature vector.

Specifically, during operation of the apparatus 300 for preparing a high-strength tungsten lanthanum wire, the responsiveness estimation unit 360 is configured to calculate a responsiveness estimation of the power feature vector relative to the spatially-structured thermodynamic variation feature vector to obtain a classification feature matrix. It should be understood that the power value of the heater is the cause of the change in the thermal distribution of the vertical fusion sintering bar, that is, the power value of the heater is the condition data, and the thermal distribution of the vertical fusion sintering bar is the ending index. The influence of the power of the expression heater on the thermal distribution of the final vertical fusion sintering strip can be optimized by fully utilizing the high-dimensional implicit response rule of the condition data on the ending indexes. In particular, a responsiveness estimate of the power feature vector relative to the spatially structured thermodynamic change feature map is calculated to represent a high-dimensional implicit response pattern between the condition data to the ending index. In a specific example, the responsiveness estimate of the power feature vector relative to the spatially structured thermodynamic change feature vector is calculated to obtain a classification feature matrix with the following formula;

wherein, the formula is:

V ₂ ＝M*V

Specifically, during the operation of the apparatus 300 for preparing a high-strength tungsten lanthanum wire, the control result generating unit 370 is configured to pass the classification feature matrix through a classifier to obtain a classification result, where the classification result is used to indicate that the power value of the heater at the current time point should be increased or decreased. That is, in this way, the temperature control of the green sintering is adaptively adjusted based on the structural thermodynamic diagram of the green sintering bar to improve the sintering quality of the green sintering bar. In one specific example, the classification feature matrix is processed using the classifier to obtain a classification result with the following formula;

Wherein Project (F) represents projecting the classification feature matrix as a vector, W ₁ To W _n Weight matrix for all the connection layers of each layer, B ₁ To B _n Representing the bias vector for each fully connected layer.

It should be appreciated that the first convolutional neural network model using spatial attention, the second convolutional neural network model using temporal attention mechanisms, and the third convolutional neural network model using one-dimensional convolution kernels need to be trained prior to the inference using the neural network models described above. That is, in the preparation device of the high-strength tungsten-lanthanum wire, the preparation device further comprises a training module, which is used for training the first convolutional neural network model using spatial attention, the second convolutional neural network model using a temporal attention mechanism and the third convolutional neural network model using a one-dimensional convolutional kernel.

Fig. 3 illustrates a block diagram of a training module in a preparation apparatus of a high strength tungsten lanthanum wire according to an embodiment of the present application. As shown in fig. 3, the apparatus 300 for preparing a high-strength tungsten lanthanum wire according to an embodiment of the present application further includes a training module, where the training module includes: a training data acquisition unit 3010; training a sintered bar thermodynamic diagram coding unit 3020; training segment semantic coding unit 3030; training the power timing characteristic encoding unit 3040; training the dimension reduction unit 3050; training the responsiveness estimation unit 3060; a classification loss unit 3070; a context loss unit 3080; and a training unit 3090.

The training data acquisition unit 3010 is configured to acquire training data, where the training data includes training power values of the heater at a plurality of predetermined time points in a predetermined time period, training thermodynamic diagrams of the vertical fusion sintering strip at a plurality of predetermined time points in the predetermined time period, and real tag values that power values of the heater at a current time point should be increased or should be decreased; the training sintered bar thermodynamic diagram coding unit 3020 is configured to pass the training thermodynamic diagrams of the vertical fusion sintered bars at each predetermined time point in the training thermodynamic diagrams of the vertical fusion sintered bars at the predetermined time points through the first convolutional neural network model using spatial attention to obtain a plurality of training spatial structure thermodynamic characteristic diagrams; the training sintering bar thermodynamic diagram dynamic feature extraction unit 3030 is configured to pass the plurality of training space structure thermodynamic feature diagrams through the second convolutional neural network model using the time attention mechanism to obtain a training space structure thermodynamic change feature diagram; the training power time sequence feature encoding unit 3040 is configured to arrange training power values of the heaters at the plurality of predetermined time points into training power input vectors according to a time dimension, and then obtain training power feature vectors through the third convolutional neural network model using a one-dimensional convolutional kernel; the training dimension reduction unit 3050 is configured to perform global mean pooling on each feature matrix of the training space structure thermodynamic change feature map along the channel dimension to obtain a training space structure thermodynamic change feature vector; the training responsiveness estimation unit 3060 is configured to calculate responsiveness estimation of the training power feature vector relative to the training space structure thermal variation feature vector to obtain a training classification feature matrix; and, the classification loss unit 3070 is configured to pass the training classification feature matrix through a classifier to obtain a classification loss function value; the context loss unit 3080 is configured to calculate a context-counted local scene metric loss function value of the training spatial structure thermodynamic change feature vector, where the context-counted local scene metric loss function value is related to statistical features of feature value sets of all positions of the training spatial structure thermodynamic change feature vector; and the training unit 3090 is configured to train the first convolutional neural network model using spatial attention, the second convolutional neural network model using temporal attention mechanism, and the third convolutional neural network model using one-dimensional convolution kernel with a weighted sum of the classification loss function value and the context-counted local scene metric loss function value as a loss function value.

Fig. 8 illustrates a system architecture diagram of a training module in a high strength tungsten lanthanum wire preparation apparatus according to an embodiment of the present application. As shown in fig. 8, in the system architecture of the apparatus 300 for preparing a high-strength tungsten lanthanum wire, during the training process, training data is firstly acquired by the training data acquisition unit 3010, where the training data includes training power values of the heater at a plurality of predetermined time points in a predetermined period of time, training thermodynamic diagrams of the vertical fusion sintering strip at a plurality of predetermined time points in the predetermined period of time, and real tag values that the power value of the heater at a current time point should be increased or decreased; the training sintered bar thermodynamic diagram coding unit 3020 passes the training thermodynamic diagrams of the vertical fusion sintered bars at each predetermined time point in the training thermodynamic diagrams of the vertical fusion sintered bars at the predetermined time points acquired by the training data acquisition unit 3010 through the first convolution neural network model using the spatial attention to obtain a plurality of training spatial structure thermodynamic characteristic diagrams; the training sintering bar thermodynamic diagram dynamic feature extraction unit 3030 obtains a training space structure thermodynamic variation feature diagram by passing the training space structure thermodynamic feature diagrams obtained by the training sintering bar thermodynamic diagram coding unit 3020 through the second convolutional neural network model using the time attention mechanism; the training power time sequence feature encoding unit 3040 arranges the training power values of the heaters at a plurality of preset time points acquired by the training data acquisition unit 3010 into training power input vectors according to a time dimension, and then obtains training power feature vectors through the third convolutional neural network model using a one-dimensional convolutional kernel; the training dimension reduction unit 3050 performs global mean pooling on each feature matrix of the training space structure thermodynamic change feature map along the channel dimension to obtain a training space structure thermodynamic change feature vector; then, the training responsiveness estimation unit 3060 calculates responsiveness estimation of the training power feature vector relative to the training space structure thermal variation feature vector to obtain a training classification feature matrix; and the classification loss unit 3070 passes the training classification feature matrix through a classifier to obtain a classification loss function value; then, the context loss unit 3080 calculates a context-counted local scene metric loss function value of the training spatial structure thermodynamic change feature vector, wherein the context-counted local scene metric loss function value is related to statistical features of feature value sets of all positions of the training spatial structure thermodynamic change feature vector; further, the training unit 3090 trains the first convolutional neural network model using spatial attention, the second convolutional neural network model using temporal attention mechanism, and the third convolutional neural network model using one-dimensional convolution kernel with a weighted sum of the classification loss function value and the context statistics local scene metric loss function value as a loss function value.

In summary, a high strength tungsten lanthanum wire fabrication apparatus 300 in accordance with embodiments of the present application is illustrated that improves the sintering quality of a sintered vertical strand by adaptively adjusting the temperature control of the vertical strand sintering based on the structural thermodynamic diagram of the vertical strand sintering.

Exemplary method

Fig. 9 illustrates a flow chart of a method of preparing a high strength tungsten lanthanum wire according to an embodiment of the present application. As shown in fig. 9, a method for preparing a high-strength tungsten lanthanum wire according to an embodiment of the present application includes the steps of: s110, obtaining power values of heaters at a plurality of preset time points in a preset time period and thermodynamic diagrams of vertical fusion sintering strips at the plurality of preset time points in the preset time period; s120, obtaining a plurality of space structure thermodynamic characteristic diagrams by using a first convolution neural network model of space attention from thermodynamic diagrams of the vertical fusion sintering strips at each preset time point in the thermodynamic diagrams of the vertical fusion sintering strips at the preset time points; s130, the plurality of space structure thermodynamic feature diagrams are subjected to a second convolution neural network model of a time attention mechanism to obtain space structure thermodynamic change feature diagrams; s140, arranging the power values of the heaters at a plurality of preset time points into power input vectors according to a time dimension, and obtaining power characteristic vectors by using a third convolution neural network model of a one-dimensional convolution kernel; s150, carrying out global mean value pooling on each feature matrix of the spatial structure thermodynamic change feature diagram along the channel dimension to obtain a spatial structure thermodynamic change feature vector; s160, calculating the response estimation of the power feature vector relative to the space structure thermodynamic change feature vector to obtain a classification feature matrix; and S170, passing the classification feature matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating that the power value of the heater at the current time point is increased or decreased.

In one example, in the above method for preparing a high-strength tungsten lanthanum wire, the step S120 includes: inputting the thermodynamic diagram of the vertical fusion sintering strips at each preset time point into a multi-layer convolution layer of the first convolution neural network model to obtain a thermodynamic distribution characteristic diagram; passing the thermodynamic distribution feature map through a spatial attention module of the first convolutional neural network model to obtain a spatial attention feature matrix; and calculating the position-wise point multiplication of each feature matrix of the thermodynamic distribution feature diagram along the channel dimension and the spatial attention feature matrix to obtain the thermodynamic feature diagram of the spatial structure. Wherein the passing the thermodynamic distribution feature map through the spatial attention module of the first convolutional neural network model to obtain a spatial attention feature matrix comprises: respectively carrying out global average pooling and global maximum pooling along the channel dimension on the thermodynamic distribution feature map by using the spatial attention module so as to obtain a first feature matrix and a second feature matrix; the first feature matrix and the second feature matrix are aggregated to obtain a space feature diagram; convolving the spatial feature map with the spatial attention module to obtain a spatial attention matrix; and performing nonlinear activation on the spatial attention moment matrix by using a Sigmoid function to obtain the spatial attention characteristic matrix.

In one example, in the above method for preparing a high-strength tungsten lanthanum wire, the step S130 includes: extracting a first space structure thermodynamic characteristic diagram and a second space structure thermodynamic characteristic diagram of two adjacent preset time points of the plurality of space structure thermodynamic characteristic diagrams; calculating the position-based point multiplication of the first space structure thermodynamic characteristic diagram and the second space structure thermodynamic characteristic diagram to obtain a time accumulation space structure thermodynamic characteristic diagram; activating the time accumulation space structure thermodynamic characteristic diagram by using a Softmax activation function to obtain a time attention characteristic diagram; and multiplying the time attention feature map and the second space structure thermodynamic feature map by position points to obtain a time enhancement second space structure thermodynamic feature map corresponding to the second space structure thermodynamic feature map.

In one example, in the above method for preparing a high-strength tungsten lanthanum wire, the step S160 includes: calculating the response estimation of the power feature vector relative to the space structure thermodynamic change feature vector by the following formula to obtain a classification feature matrix;

wherein, the formula is:

V ₂ ＝M*V

In one example, in the above method for preparing a high-strength tungsten lanthanum wire, the step S170 includes: processing the classification feature matrix by using the classifier in the following formula to obtain a classification result;

wherein, the formula is: o=softmax { (W) _n ,B _n ):…:(W ₁ ,B ₁ ) Project (F), where Project (F) represents projecting the classification feature matrix as a vector, W ₁ To W _n Weight matrix for all the connection layers of each layer, B ₁ To B _n Representing the bias vector for each fully connected layer.

In summary, a method of preparing a high strength tungsten lanthanum wire according to embodiments of the present application is illustrated that improves the sintering quality of a sintered vertical strand by adaptively adjusting the temperature control of the vertical strand sintering based on the structural thermodynamic diagram of the vertical strand sintering.

Claims

1. The utility model provides a preparation facilities of high strength tungsten lanthanum silk which characterized in that includes:

2. The apparatus for producing high-strength tungsten lanthanum wire according to claim 1, wherein the sintered bar thermodynamic diagram coding unit comprises:

A convolution coding subunit, configured to input a thermodynamic diagram of the vertical fusion sintering strips at each predetermined time point in the convolution coding subunit into a multi-layer convolution layer of the first convolution neural network model to obtain a thermodynamic distribution feature map;

a spatial attention subunit, configured to pass the thermodynamic distribution feature map through a spatial attention module of the first convolutional neural network model to obtain a spatial attention feature matrix; and

and the attention application subunit is used for calculating the position-wise dot multiplication of each feature matrix of the thermodynamic distribution feature diagram along the channel dimension and the spatial attention feature matrix to obtain the thermodynamic feature diagram of the spatial structure.

3. The apparatus for producing a high strength tungsten lanthanum wire according to claim 2, wherein the spatial attention subunit is further configured to:

respectively carrying out global average pooling and global maximum pooling along the channel dimension on the thermodynamic distribution feature map by using the spatial attention module so as to obtain a first feature matrix and a second feature matrix;

the first feature matrix and the second feature matrix are aggregated to obtain a space feature diagram;

convolving the spatial feature map with the spatial attention module to obtain a spatial attention matrix; and

And using a Sigmoid function to perform nonlinear activation on the spatial attention moment matrix to obtain the spatial attention characteristic matrix.

4. The apparatus for preparing high-strength tungsten lanthanum wire according to claim 3, wherein the sintered bar thermodynamic diagram dynamic feature extraction unit comprises:

an extraction subunit, configured to extract a first spatial structure thermodynamic characteristic map and a second spatial structure thermodynamic characteristic map of two adjacent predetermined time points of the plurality of spatial structure thermodynamic characteristic maps;

the time accumulation subunit is used for calculating the position-wise point multiplication of the first space structure thermodynamic characteristic diagram and the second space structure thermodynamic characteristic diagram to obtain a time accumulation space structure thermodynamic characteristic diagram;

an activation subunit, configured to activate the time-cumulative spatial structure thermodynamic feature map with a Softmax activation function to obtain a time attention feature map; and

and the time attention applying subunit is used for multiplying the time attention characteristic diagram and the second space structure thermodynamic characteristic diagram according to position points to obtain a time enhanced second space structure thermodynamic characteristic diagram corresponding to the second space structure thermodynamic characteristic diagram.

5. The apparatus for producing a high-strength lanthanum tungsten wire according to claim 4, wherein the responsiveness estimating unit is further configured to: calculating the response estimation of the power feature vector relative to the space structure thermodynamic change feature vector by the following formula to obtain a classification feature matrix;

Wherein, the formula is:

V ₂ ＝M*V

6. The apparatus for producing a high-strength lanthanum-tungsten wire according to claim 5, wherein the control result generating unit is further configured to: processing the classification feature matrix by using the classifier in the following formula to obtain a classification result;

7. The apparatus for preparing a high-strength tungsten lanthanum wire according to claim 1, further comprising a training module for training the first convolutional neural network model using spatial attention, the second convolutional neural network model using temporal attention mechanism, and the third convolutional neural network model using one-dimensional convolutional kernel.

8. The apparatus for producing high-strength tungsten lanthanum wire according to claim 7, wherein the training module comprises:

The training data acquisition unit is used for acquiring training data, wherein the training data comprises training power values of the heater at a plurality of preset time points in a preset time period, training thermodynamic diagrams of the vertical fusion sintering strips at a plurality of preset time points in the preset time period, and real label values of which the power values of the heater at the current time point are increased or reduced;

the training sintering bar thermodynamic diagram coding unit is used for enabling training thermodynamic diagrams of the vertical fusion sintering bars at each preset time point in the training thermodynamic diagrams of the vertical fusion sintering bars at the preset time points to pass through the first convolution neural network model using the space attention so as to obtain a plurality of training space structure thermodynamic characteristic diagrams;

the training sintering bar thermodynamic diagram dynamic characteristic extraction unit is used for enabling the training space structure thermodynamic characteristic diagrams to pass through the second convolution neural network model using the time attention mechanism so as to obtain training space structure thermodynamic variation characteristic diagrams;

the training power time sequence feature coding unit is used for arranging training power values of the heaters at a plurality of preset time points into training power input vectors according to time dimension and obtaining training power feature vectors through the third convolution neural network model using the one-dimensional convolution kernel;

The training dimension reduction unit is used for carrying out global averaging and pooling on each feature matrix of the training space structure thermodynamic change feature diagram along the channel dimension so as to obtain a training space structure thermodynamic change feature vector;

the training response estimation unit is used for calculating the response estimation of the training power feature vector relative to the training space structure thermodynamic change feature vector so as to obtain a training classification feature matrix; and

the classification loss unit is used for passing the training classification characteristic matrix through a classifier to obtain a classification loss function value;

a context loss unit, configured to calculate a context-counted local scene metric loss function value of the training spatial structure thermal variation feature vector, where the context-counted local scene metric loss function value is related to statistical features of feature value sets of all positions of the training spatial structure thermal variation feature vector; and

a training unit for training the first convolutional neural network model using spatial attention, the second convolutional neural network model using temporal attention mechanism, and the third convolutional neural network model using one-dimensional convolutional kernel with a weighted sum of the classification loss function value and the context statistical local scene metric loss function value as a loss function value.

9. The apparatus for producing a high-strength tungsten lanthanum wire according to claim 8, wherein the context loss unit is further configured to: calculating a local scene measurement loss function value of the context statistics of the training space structure thermodynamic change feature vector according to the following formula;

wherein, the formula is:

wherein v is _i Is the eigenvalue of each position of the thermodynamic change eigenvector of the space structure, and mu and sigma are v _i E means and variances of V, and V represents the spatially structured thermodynamic change feature vector, L is the length of the spatially structured thermodynamic change feature vector.

10. The preparation method of the high-strength tungsten lanthanum wire is characterized by comprising the following steps of: