JP7409270B2 - Crystallinity estimation device, crystallinity estimation method, crystallinity estimation program, and recording medium - Google Patents

Description

The present invention relates to a technique that is effective when applied to, for example, a crystallinity estimation technique that estimates the crystallinity of a material using machine learning.

Non-Patent Document 1 describes a simulation technique using Avrami's equation as a technique for estimating the degree of crystallinity for an arbitrary temperature history.

Nakamura, Kasai, Nagumo: Crystallization of amorphous Cu50Ti50 alloy produced by mechanical gliding method, Journal of the Japan Institute of Metals, Vol.54 No.12 1320-1328 (1990)

For example, when a material such as metal or plastic is solidified from a molten state, the degree of crystallinity of the material differs depending on the cooling temperature history. This difference in crystallinity may affect physical properties such as elastic modulus and ductility after the material is solidified. Therefore, in order to predict the final physical properties of a material after solidification, simulation techniques are used to estimate the degree of crystallinity. Here, "crystallinity" refers to the ratio of the crystallized region to the total region of the material.

In conventional simulation techniques, first, measured data indicating the temperature dependence of heat flow is obtained by using a differential scanning calorimetry (DSC, hereinafter referred to as "DSC"). Then, for example, we propose a model for crystal nucleation and crystal growth, and based on this proposed model, we hypothesize a functional form of the temperature function that indicates the temperature dependence of heat flow. Then, the temperature function of the heat flow is obtained by fitting and determining the parameters included in the assumed temperature function to match the actual measurement data. Next, the degree of crystallinity for an arbitrary temperature history is estimated by incorporating the specific temperature function of the obtained heat flow into Avrami's equation and simulating crystal growth.

However, with conventional simulation techniques, models of crystal nucleation and crystal growth are artificially proposed to determine the functional form of the temperature function of heat flow. For this reason, a large number of man-hours were required to determine the functional form of the temperature function.

An object of the present invention is to provide a technique for estimating the crystallinity of a material that can reduce the number of man-hours.

In one embodiment, the crystallinity estimating device estimates the crystallinity of a material based on a heat flow that indicates the amount of heat dissipated from the material per unit time and has temperature dependence. In the crystallinity estimation device configured, a temperature function acquisition unit configured to acquire a temperature function of heat flow based on actual measurement data obtained by measuring a heat flow value corresponding to a temperature value; , a crystallinity estimation section configured to estimate the crystallinity of the material based on the temperature function obtained by the temperature function obtaining section, and the temperature function obtaining section uses the measured data as training data. The temperature function is acquired by learning a neural network that inputs temperature values and outputs heat flow values.

A method for estimating crystallinity in one embodiment includes estimating the degree of crystallinity of a material based on a heat flow that indicates the amount of heat dissipated from a material per unit time and that has temperature dependence. In the temperature estimation method, there is a temperature function acquisition step of acquiring a temperature function of heat flow based on actual measurement data obtained by measuring a heat flow value corresponding to a temperature value, and a temperature function acquisition step of acquiring a temperature function of heat flow, and a temperature and a crystallinity estimation step that estimates the crystallinity of the material based on the function, and the temperature function acquisition step uses actual measurement data as training data, inputs the temperature value, and outputs the heat flow value. The temperature function is obtained by training a neural network with

A crystallinity estimation program in one embodiment is a crystallinity estimation program for estimating the crystallinity of a material based on a heat flow that indicates the amount of heat dissipated from a material per unit time and that has temperature dependence. Temperature function acquisition that obtains the temperature function of heat flow based on the actual measurement data obtained by measuring the heat flow value corresponding to the temperature value in the crystallinity degree estimation program for causing the computer to execute the crystallinity estimation process. and a crystallinity estimation process that estimates the crystallinity of the material based on the temperature function acquired in the temperature function acquisition process, and the temperature function acquisition process uses actual measurement data as training data. This process acquires a temperature function by training a neural network that inputs temperature values and outputs heat flow values.

This crystallization estimation program is recorded on a computer-readable recording medium.

According to one embodiment, it is possible to provide a technique for estimating the crystallinity of a material that can reduce the number of man-hours.

3 is a table showing examples of numerical data of heat flow corresponding to temperature values. FIG. 3 is a graph of numerical data of heat flow corresponding to temperature values. It is a flowchart explaining the method of estimating the degree of crystallinity by a general simulation technique. It is a flowchart explaining the estimation method of crystallinity degree based on a basic idea. FIG. 2 is a diagram showing an example of the hardware configuration of a crystallinity estimation device. FIG. 2 is a functional block diagram showing the functions of a crystallinity estimation device. FIG. 1 is a diagram schematically showing an example of a configuration of a neural network. It is a flow chart explaining operation of a crystallinity degree estimating device. It is a graph showing actual measurement data of heat flow measured by "DSC". It is a graph showing a temperature function of heat flow obtained by approximating measured data with a polynomial. It is a graph showing a temperature function of heat flow obtained by training a neural network using actually measured data as training data. It is a graph showing other evaluation items. It is a table showing the results of evaluating comparative examples and examples. FIG. 1 is a schematic diagram showing an experimental system. It is a graph showing an example of a change in actually measured resin temperature. It is a graph showing a comparison result between a calculated value and an actual value. FIG. 2 is a functional block diagram showing the configuration of an application example.

In all the drawings for explaining the embodiment, the same members are designated by the same reference numerals in principle, and repeated explanations thereof will be omitted. Note that, in order to make the drawings easier to understand, hatching may be added even in a plan view.

<Crystallinity estimation principle>
"Heat flow" is a physical quantity that indicates the amount of heat dissipated from a material per unit time, and is a physical quantity that is temperature dependent. In this embodiment, the crystallinity of the material is estimated based on this heat flow. This principle will be explained below.

When the target material is melted and then cooled and solidified, heat is released from the material. This is because the internal energy of the solidified state is lower than the internal energy of the molten state, so when the phase transition occurs from the molten state with high internal energy to the solidified state with low internal energy, excess energy is released as heat. This is because that.

Therefore, by examining the amount of heat released from a material that undergoes a phase transition from a molten state to a solidified state, it is possible to estimate how much the material has crystallized (solidified).

Specifically, the crystallinity of the material is estimated based on the heat flow, which indicates the amount of heat dissipated per unit time, rather than the amount of heat dissipated itself.

Heat flow is measured using, for example, "DSC". "DSC" measures the heat flow when a target material is once melted and then solidified at a constant cooling rate.

For example, FIG. 1 is a table showing examples of numerical data of heat flow corresponding to temperature values. As shown in FIG. 1, heat flow values corresponding to temperature values are obtained as discrete data.

FIG. 2 is a graph of numerical data of heat flow corresponding to temperature values. In FIG. 2, the horizontal axis represents temperature (° C.), while the vertical axis represents heat flow (W/g).

As mentioned above, when a material is cooled, crystallization of the material progresses, and heat is released due to changes in the microstructure due to crystallization. In "DSC", this heat is measured as heat flow, which is the amount of heat dissipated per unit time.

The following relationship (Equation 1) holds true between the amount of heat dissipation and the heat flow.

ここで、
ｔ：時間
Ｔ：温度
Ｑ：放熱量
Ｊ（Ｔ）：ヒートフロー

here,
t: Time T: Temperature Q: Heat radiation amount J(T): Heat flow

Since heat is released when a material crystallizes from a molten state, it is thought that the greater the crystallinity of the material, the greater the amount of heat released up to that point. That is, it is reasonable to consider that the crystallinity of a material is proportional to the amount of heat dissipation. From this, it is assumed that the relationship shown in (Equation 2) below holds true.

ここで、
Ｘ：結晶化度
Ｃ：比例定数
Ｑ：放熱量

here,
X: Crystallinity C: Constant of proportionality Q: Amount of heat dissipation

From the above-mentioned (Formula 1) and (Formula 2), the crystallization rate, which is the time differential of the degree of crystallinity, is expressed by the following (Formula 3).

ここで、
ｄＸ／ｄｔ：結晶化速度
Ｃ：比例定数
Ｊ：ヒートフロー

here,
dX/dt: Crystallization rate C: Proportionality constant J: Heat flow

It can thus be seen that the crystallization rate is proportional to the heat flow. That is, if the functional form of the heat flow is known, the crystallization rate can be determined by solving the differential equation expressed by (Equation 3).

Here, the proportionality constant "C" is set as a constant, but as the crystallization of the material progresses, the uncrystallized region that is not crystallized decreases, and if the uncrystallized region decreases, it is thought that the crystallization rate also decreases. . That is, considering that the proportionality constant "C" actually depends on the crystallinity "X", the crystallization rate "dX/dt" is expressed by the following (Equation 4).

ここで、
ｄＸ／ｄｔ：結晶化速度
Ｊ（Ｔ）：ヒートフロー
Ｃ（Ｘ）：係数

here,
dX/dt: Crystallization rate J(T): Heat flow C(X): Coefficient

This (Equation 4) is called "Avrami's equation". For example, considering the phenomenon of crystallinity saturation, "C(X)" included in (Formula 4) can be specifically expressed. As a result, the crystallinity "dX/dt" is expressed by the following (Equation 5).

ここで、
ｄＸ／ｄｔ：結晶化速度
Ｊ（Ｔ）：ヒートフロー
Ｘ：結晶化度
Ｘ_∞：飽和結晶化度

here,
dX/dt: Crystallization rate J(T): Heat flow X: Crystallinity X _∞ : Saturated crystallinity

A qualitative explanation of the fact that "C(X)" can be expressed as "X·(X _∞ -X)" is as follows. That is, it is considered that as the degree of crystallinity approaches a saturated state, the area to be crystallized decreases, and thus the crystallization rate decreases. In other words, it is considered that the closer the crystallinity is to the saturated state, the lower the crystallization rate becomes. From this, it is considered that the crystallization rate “dX/dt” is proportional to “(X _∞ −X)”. Further, in the initial state, it is considered that the more nuclei for crystal growth there are, the higher the crystallization speed becomes. The more crystal regions that serve as nuclei for crystal growth, the easier it is for crystal growth to proceed from those regions as starting points. Therefore, the crystallization rate "dX/dt" is considered to be proportional to the crystallinity "X".

From the above, it is considered appropriate to represent "C(X)" as X・(X _∞ -X). In this way, the crystallization rate "dX/dt" can be expressed by, for example, (Formula 5) by embodying the "Abrami's equation" expressed by (Formula 4).

Then, by solving the differential equations expressed by (Equation 4) and (Equation 5), the degree of crystallinity "X" can be calculated. In other words, if we know the temperature function of the heat flow "J(T)" included in (Formula 4) and (Formula 5), we can specifically solve the differential equation and calculate the degree of crystallinity "X". be able to. For example, if a temperature function that faithfully reproduces the curve showing the temperature dependence of heat flow shown in FIG. 2 is known, the accuracy of the crystallinity "X" can be improved. Therefore, it is important to accurately determine the temperature function of heat flow "J(T)".

It should be noted that it can be understood that the degree of crystallinity "X" changes depending on the cooling history for crystallizing the target material from the molten state, for example, by transforming the "Avrami equation" (Equation 4). It becomes easier to do. For example, when (Equation 4) is transformed to solve it, (Equation 6) is obtained.

Here, on the right side of (Equation 6), even though the temperature function of the heat flow "J(T)" itself does not change depending on the temperature history, the integrand function includes "dT/dt". This "dT/dt" represents the change in temperature over time, and the integral value on the right side will be a different value if the change in temperature over time is different. In other words, the integral value on the right side of (Equation 6) depends on the temporal change in temperature. A different temporal change in temperature means a different temperature history during the cooling process. Therefore, the integral value on the right side of (Equation 6) will differ depending on the temperature history. As a result, the crystallinity "X" calculated by solving (Equation 6) differs depending on the temperature history.

<General simulation technology>
First, a general simulation technique for estimating the degree of crystallinity based on the above-mentioned principle will be explained. FIG. 3 is a flowchart illustrating a method of estimating crystallinity using a general simulation technique.

In FIG. 3, first, measured data showing the temperature dependence of heat flow is obtained by using "DSC" (S101). Thereafter, for example, a researcher proposes a model for crystal nucleation and crystal growth, and based on this proposed model, a functional form of a temperature function indicating the temperature dependence of heat flow is assumed (S102). Then, by fitting and determining parameters included in the assumed temperature function to match the measured data (S103), a temperature function of heat flow is obtained (S104). Next, the specific temperature function of the obtained heat flow is incorporated into Avrami's equation (Equation 4 or Equation 5) (S105). Then, the environmental conditions (temporal changes in temperature) indicating the cooling history are input into Avrami's equation (S106), and a crystal growth simulation is performed to estimate the degree of crystallinity for an arbitrary temperature history (S107).

However, in such general simulation techniques, researchers artificially propose models for crystal nucleation and crystal growth, and assume a functional form of the temperature function of heat flow. For this reason, researchers' sense and experience are required to determine the functional form of the temperature function of heat flow so as to match the measured data. Furthermore, it takes a lot of time to determine the functional form of the temperature function. In other words, with general simulation techniques, determining the functional form of the temperature function of heat flow requires advanced knowledge and experience, as well as a large amount of time. A technology that can determine a temperature function that matches actual measurement data is desired. Therefore, in this embodiment, a device is devised to easily obtain a heat flow temperature function that matches actual measurement data without relying on advanced knowledge or experience.

Below, the technical idea of this embodiment in which this device is applied will be explained.

<Basic idea of embodiment>
The basic idea of this embodiment is to obtain a heat flow temperature function that matches actual measurement data using a neural network without human intervention.

Specifically, the basic idea of this embodiment is to acquire a temperature function by using actually measured data as training data and training a neural network that inputs temperature values and outputs heat flow values. It is a thought.

According to this basic idea, the degree of crystallinity of a material can be estimated by substituting the temperature function of heat flow obtained by a neural network into Avrami's equation.

At the root of this basic idea is the universality theorem, which states that neural networks can basically approximate any complex functional form, without relying on the researchers' sense or experience. , the idea is that a neural network can obtain a complex heat flow temperature function. In other words, the basic idea is that the degree of crystallinity of a material can be estimated with high precision, even without analyzing the detailed mechanisms of crystal nucleation and crystal growth, as long as a temperature function that matches the measured data with high precision can be obtained. It is based on this idea.

According to this basic idea, it is possible to obtain a heat flow temperature function that matches actual measurement data in a short time without requiring advanced knowledge or experience. In other words, the basic idea is excellent in that it allows you to obtain the temperature function of heat flow without human intervention, and it is useful in that it allows you to estimate the degree of crystallinity with high accuracy even without advanced knowledge or experience. This is a technical idea.

FIG. 4 is a flowchart illustrating a method of estimating the degree of crystallinity based on the basic idea.

In FIG. 4, first, measured data indicating the temperature dependence of heat flow is obtained by using "DSC" (S201). Thereafter, the actually measured data is used as teacher data to train a neural network that inputs temperature values and outputs heat flow values (S202). Then, as a result of learning by the neural network, a temperature function of heat flow is obtained (S203). Next, the specific temperature function of the obtained heat flow is incorporated into Avrami's equation (Equation 4 or Equation 5) (S204). Then, the environmental conditions (temporal change in temperature) indicating the cooling history are input into Avrami's equation (S205), and a crystal growth simulation is performed to estimate the degree of crystallinity for an arbitrary temperature history (S206).

As described above, the degree of crystallinity of a material can be estimated based on the basic idea of obtaining the temperature function of heat flow using a neural network.

<Configuration of crystallinity estimation device>
<<Hardware configuration>>
In the following, first, a hardware configuration of a crystallinity degree estimating device in this embodiment that embodies the above-mentioned basic idea will be described.

FIG. 5 is a diagram showing an example of the hardware configuration of crystallinity estimation device 100 in this embodiment. Note that the configuration shown in FIG. 5 merely shows an example of the hardware configuration of the crystallinity estimation device 100, and the hardware configuration of the crystallinity estimation device 100 is limited to the configuration shown in FIG. However, other configurations are also possible.

In FIG. 5, a crystallinity estimation device 100 includes a CPU (Central Processing Unit) 101 that executes a program. The CPU 101 is electrically connected to, for example, a ROM (Read Only Memory) 102, a RAM (Random Access Memory) 103, and a hard disk device 112 via a bus 113, and controls these hardware devices. It is configured as follows.

Further, the CPU 101 is also connected to an input device and an output device via a bus 113. Examples of input devices include the keyboard 105, the mouse 106, the communication board 107, and the scanner 111. On the other hand, examples of output devices include the display 104, the communication board 107, and the printer 110. Further, the CPU 101 may be connected to, for example, a removable disk device 108 or a CD/DVD-ROM device 109.

The crystallinity estimation device 100 may be connected to a network, for example. For example, when the crystallinity estimating device 100 is connected to other external devices via a network, the communication board 107 that constitutes a part of the crystallinity estimating device 100 can be connected to a LAN (Local Area Network), a WAN ( wide area network) or the Internet.

The RAM 103 is an example of volatile memory, and the recording media of the ROM 102, removable disk device 108, CD/DVD-ROM device 109, and hard disk device 112 are examples of nonvolatile memory. These volatile memories and nonvolatile memories constitute a storage device of the crystallinity estimation device 100.

The hard disk device 112 stores, for example, an operating system (OS) 201, a program group 202, and a file group 203. The programs included in the program group 202 are executed by the CPU 101 while using the operating system 201. Further, the RAM 103 temporarily stores at least a portion of the operating system 201 programs and application programs to be executed by the CPU 101, and also stores various data necessary for processing by the CPU 101.

The ROM 102 stores a BIOS (Basic Input Output System) program, and the hard disk drive 112 stores a boot program. When the crystallinity estimating device 100 is started, the BIOS program stored in the ROM 102 and the boot program stored in the hard disk drive 112 are executed, and the operating system 201 is started by the BIOS program and the boot program.

Program group 202 stores programs that implement the functions of crystallinity estimation device 100, and this program is read and executed by CPU 101. Further, the file group 203 stores information, data, signal values, variable values, and parameters indicating the results of processing by the CPU 101 as each item of the file.

The file is stored in a recording medium such as the hard disk device 112 or memory. Information, data, signal values, variable values, and parameters stored in recording media such as the hard disk device 112 and memory are read by the CPU 101 to the main memory and cache memory, and are extracted, searched, referenced, compared, calculated, processed, and It is used for operations of the CPU 101, such as editing, outputting, printing, and displaying. For example, during the operation of the CPU 101 described above, information, data, signal values, variable values, and parameters are temporarily stored in main memory, registers, cache memory, buffer memory, and the like.

The functions of the crystallinity estimating device 100 may be realized by firmware stored in the ROM 102, or may be realized by only software, only hardware such as elements, devices, substrates, and wiring, or by combining software and hardware. It may also be realized in combination with firmware. The firmware and software are stored as programs in a recording medium such as a hard disk device 112, a removable disk, a CD-ROM, a DVD-ROM, and the like. The program is read and executed by the CPU 101. That is, the program causes the computer to function as the crystallinity estimation device 100.

In this way, the crystallinity estimation device 100 includes a CPU 101 as a processing device, a hard disk device 112 and memory as storage devices, a keyboard 105, a mouse 106, a communication board 107 as input devices, a display 104 as an output device, and a printer. 110, a computer equipped with a communication board 107; The functions of the crystallinity estimation device 100 are realized using a processing device, a storage device, an input device, and an output device.

<<Functional block configuration>>
Next, the functional block configuration of the crystallinity estimation device 100 will be explained.

FIG. 6 is a functional block diagram showing the functions of the crystallinity estimation device.

As shown in FIG. 6, the crystallinity estimation device 100 includes an input section 301, a temperature function acquisition section 302, a crystallinity estimation section 303, an output section 304, and a data storage section 305. .

This crystallinity estimating device 100 includes the above-mentioned components, and is capable of determining the crystallization of a material based on a heat flow that indicates the amount of heat dissipated from a material per unit time and that has temperature dependence. is configured to estimate the degree of

The input unit 301 is configured to input actual heat flow data measured by, for example, "DSC". The measured data input to the input section 301 is then stored in the data storage section 305. For example, the measured data shown in FIG. 1 is input to the input unit 301.

The temperature function acquisition unit 302 is configured to acquire a temperature function of heat flow based on actual measurement data stored in the data storage unit 305.

For example, the temperature function acquisition unit 302 is configured to acquire a temperature function by using actually measured data as training data and training a neural network that inputs temperature values and outputs heat flow values. ing.

FIG. 7 is a diagram schematically showing a configuration example of a neural network that constitutes the temperature function acquisition unit 302. The neural network shown in FIG. 7 has, for example, a single intermediate layer 400, and this intermediate layer 400 includes, for example, six nodes 401. In this neural network, functions are considered as a combination of linear and nonlinear operations. That is, each of the six nodes 401 configuring the intermediate layer 400 performs a linear operation on the input data using weights and biases, and then substitutes the result of the linear operation into the activation function to generate a nonlinear Perform calculations. Then, the intermediate layer 400 outputs output data by combining the results of the nonlinear operations at each of the six nodes 401.

In this way, in neural networks, the function for obtaining output data from input data is considered as a combination of linear and nonlinear operations. Learning of a neural network means optimizing a function provided by a neural network so that input data and output data are related. For example, the learning of the neural network used in this embodiment involves optimizing the function given by the neural network so that the input data, temperature, and the output data, heat flow, are related. Become. As a result, if the function given by the neural network is optimized, this optimized function becomes a heat flow temperature function that matches the actual measurement data.

Here, the optimization of the function provided by the neural network is performed, for example, as shown below. That is, a data set is prepared using actual measurement data obtained by "DSC" as training data. Using this data set, if a function given by the neural network can be found that matches all the data contained in the data set, then the function given by the neural network can be optimized. It turns out. In this regard, for example, the difference between actual measured data, which is teaching data, and the output value from the neural network is introduced as an error function, and this error function is required to be small.

This requirement can be achieved, for example, by setting the weights and biases to appropriate initial values and gradually updating them using a gradient descent method, which ultimately minimizes the error function. I can do that. This is machine learning, which optimizes functions given by neural networks. As a result, in this embodiment, the neural network included in the temperature function acquisition unit 302 can acquire a heat flow temperature function that matches actual measurement data.

Note that in this embodiment, an example in which a neural network having a single intermediate layer 400 is used has been described, but the present invention is not limited to this, and for example, a neural network having a plurality of intermediate layers may also be used.

However, from the viewpoint of speeding up calculation, it is desirable that the intermediate layer be a single layer, and in consideration of errors from actual measurement data, it is desirable that the single layer has six or more nodes.

Subsequently, the crystallinity estimation unit 303 is configured to estimate the crystallinity of the material based on the temperature function acquired by the temperature function acquisition unit 302. Specifically, the crystallinity estimation unit 303 calculates the crystallinity “X” by substituting the temperature function acquired by the temperature function acquisition unit 302 into (Equation 4) or (Equation 5) and solving a differential equation. is configured to estimate. For example, differential equations can be solved using the Runge-Kutta method.

The output unit 304 is configured to output the crystallinity “X” estimated by the crystallinity estimation unit 303.

<Operation of crystallinity estimation device>
The crystallinity estimation device 100 is configured as described above, and the operation of the crystallinity estimation device 100 will be described below with reference to the drawings.

FIG. 8 is a flowchart illustrating the operation of the crystallinity estimation device.

First, it is assumed that the relational expressions shown in (Formula 1) to (Formula 6), actual measurement data, etc. are stored in the data storage unit 305 in advance.

In FIG. 8, the input unit 301 of the crystallinity estimating device 100 inputs actual measurement data indicating the temperature dependence of heat flow measured by using "DSC", for example (S301). The input actual measurement data is stored in the data storage unit 305.

Next, the temperature function acquisition unit 302 uses the actual measurement data stored in the data storage unit 305 as training data to train a neural network that inputs the temperature value and outputs the heat flow value (S302 ). Thereby, the temperature function acquisition unit 302 acquires the temperature function of the heat flow as a result of learning by the neural network (S303).

Next, the crystallinity estimation unit 303 incorporates the specific temperature function of the heat flow acquired by the temperature function acquisition unit 302 into the Avrami equation (Equation 4 or Equation 5) (S304). Then, environmental conditions (temporal changes in temperature) indicating the cooling history are input into Avrami's equation (S305), and a crystal growth simulation is performed. Thereby, the crystallinity estimation unit 303 estimates the crystallinity for an arbitrary temperature history (S306). Then, the output unit 304 outputs the degree of crystallinity estimated by the degree of crystallinity estimation unit 303 (S307).

As described above, according to the crystallinity estimation device 100 in this embodiment, the crystallinity of a material can be estimated using a neural network.

<Physical quantity estimation program>
The crystallinity estimation method carried out by the crystallinity estimation apparatus 100 described above can be realized by a crystallinity estimation program that causes a computer to execute crystallinity estimation processing.

For example, in the crystallinity estimation apparatus 100 consisting of a computer shown in FIG. 5, the crystallinity estimation program according to the present embodiment can be installed as one of the program group 202 stored in the hard disk drive 112. The crystallinity estimation method according to the present embodiment can be implemented by causing the computer serving as the crystallinity estimation apparatus 100 to execute this crystallinity estimation program.

A crystallinity estimation program that causes a computer to execute each process for creating data related to crystallinity estimation processing can be recorded on a computer-readable recording medium and distributed. Examples of recording media include magnetic storage media such as hard disks and flexible disks, optical storage media such as CD-ROMs and DVD-ROMs, and hardware devices such as non-volatile memories such as ROMs and EEPROMs. included.

<Verification of effectiveness>
Below, the results of verifying the effects of this embodiment will be explained.

<<Verification result 1>>
First, a description will be given of the verification results of whether the heat flow temperature function obtained by training the neural network closely matches the measured data.

FIG. 9 is a graph showing actual heat flow data measured by "DSC". The actual measurement data here is data obtained by measuring linear low density polyethylene using "DSC".

In FIG. 9, the horizontal axis indicates temperature (° C.), and the vertical axis indicates heat flow (W/g). Note that, as can be seen from a comparison with FIG. 2, in FIG. 9, the heat flow caused by heat going in and out of the container is removed.

FIG. 10 is a graph showing a temperature function of heat flow obtained by approximating the measured data shown in FIG. 9 using a polynomial. In FIG. 10, Comparative Example 1 is a temperature function obtained by approximating measured data with a 10th-order polynomial. Further, Comparative Example 2 is a temperature function obtained by approximating the measured data using a 100th degree polynomial. As shown in FIG. 10, it can be seen that in Comparative Example 1 and Comparative Example 2, there is a significant mismatch between the temperature function and the measured data.

Next, FIG. 11 is a graph showing a temperature function of heat flow obtained by learning a neural network using the actual measurement data shown in FIG. 9 as teacher data. In FIG. 11, Example 1 is a temperature function obtained by learning a three-node neural network. Further, Example 2 is a temperature function obtained by training a six-node neural network. As shown in FIG. 11, it can be seen that in Examples 1 and 2, the temperature function and the measured data match well.

Below, the results of quantitatively evaluating the comparative examples and examples will be explained.

The "error", which is one of the evaluation items, is the average value of the difference (absolute value) between the measured data and the calculated data, and is expressed by the following (Formula 7).

ここで、
ε：誤差
ｘ_ｉ：計算データ
ｘ´_ｉ：実測データ
Ｎ：データ数

here,
ε: Error x _i : Calculated data x' _i : Actual data N: Number of data

Moreover, FIG. 12 is a graph showing other evaluation items.

FIG. 12 shows "difference in peak position", "difference in maximum value", and "difference in minimum value" between the measured data and the calculated data.

FIG. 13 is a table showing the results of evaluating the comparative example and the example according to the evaluation items described above.

In FIG. 13, Comparative Examples 1 and 2 use the polynomial fitting function of Mathlab (Mathworks), and determine the coefficients of the polynomial by the least squares method using 15,500 data points. As shown in FIG. 13, it can be seen that Comparative Example 1 and Comparative Example 2 failed in all evaluation items. In particular, as shown in FIG. 10, as a result of the oscillating waveform at around -50° C. and 200° C., the error increases and it is found that this is inappropriate.

In FIG. 13, Example 1 and Example 2 are trained using the neural network fitting tool of Mathlab (Mathworks). The learning conditions are "Bayesian Regularization" for training, and "Mean Squared Error" for determining the end of learning. As shown in FIG. 13, it can be seen that both Example 1 and Example 2 passed in all evaluation items. However, as can be seen by comparing Example 1 and Example 2, the error is larger in the 3-node neural network than in the 6-node neural network. Therefore, it can be seen that it is desirable to use a neural network with six or more nodes from the viewpoint of obtaining a temperature function that matches actual measurement data with higher accuracy. In particular, in Examples 1 and 2, the temperature function does not have an oscillatory waveform, but has a smooth shape, so it can be seen that the function form is easy to handle without increasing errors.

The above evaluation results confirm that the heat flow temperature function obtained by training the neural network closely matches the measured data.

<<Verification result 2>>
Next, a description will be given of the verification results of whether the degree of crystallinity estimated based on the temperature function of heat flow obtained by training a neural network closely matches the degree of crystallinity that was actually measured. That is, here, the effectiveness of the crystallinity estimation method in this embodiment will be verified. Verification involves comparing the degree of crystallinity estimated based on the heat flow temperature function obtained from neural network learning with the degree of crystallinity of samples actually made by changing the cooling conditions during wire forming. It was done by

1. Experimental System FIG. 14 is a schematic diagram showing the experimental system.

In FIG. 14, when raw material pellets 10, which are raw materials for the coating material, are put into an extruder 11 and kneaded, a molten resin material is extruded from a die 13 via a crosshead 12. The extruded resin material is applied to the surface of the conductor 14 moving along the running line. Immediately after being extruded from the die 13, the resin material applied to the surface of the conductor 14 is air-cooled, and then water-cooled in a water tank 15. In this way, the resin material extruded from the die 13 is crystallized during the cooling process by air cooling and water cooling. Below, this degree of crystallinity will be measured.

The conductor 14 is made of stranded copper wire, for example. Further, the raw material pellets (plastic material) 10 are, for example, chain low density polyethylene. Using this conductor 14 and raw material pellets 10, an electric wire having an outer diameter of 2.63 mm is produced.

The extruder 11 has a cylinder diameter of 40 mm, and the cylinder temperature is set at 100°C to 180°C. Also, the screw is in full flight. The temperature of the crosshead (mold) 12 is set to 200° C., and the die 13 has a hole diameter of 2.6 mm.

In the running line, the air cooling distance is 380 mm and the linear speed is 20 m/min.

2. Preparation of evaluation sample In order to change the degree of crystallinity, the conductor 14 inserted into the crosshead 12 is heated to a temperature of 25°C to 200°C at the position 16 shown in FIG. Multiple evaluation samples with different history) were prepared.

3. Measurement of Resin Temperature The resin temperature necessary for calculating the degree of crystallinity was obtained by measuring the surface temperature during air cooling with a non-contact laser thermometer 17, and the temperature change of the resin material was determined.

4. Actual measurement of crystallinity degree Actual measurement of crystallinity degree was performed using "DSC". Approximately 5 mg of the coated electric wire was collected as an evaluation sample. The temperature change rate of "DSC" is 10° C./min, and the temperature is raised/lowered. The temperature change range was -50°C to 200°C.

First, as a first run, the temperature was raised from -50°C to 200°C, and then as a second run, the temperature was lowered from 200°C to -50°C. Furthermore, as a third run, the temperature was raised from -50°C to 200°C, and heat flow was measured in each of the first run, second run, and third run. The integrated value (integral value) of the heat flow value is latent heat.

Here, the latent heat obtained by performing a first run on the evaluation sample reflects the degree of crystallinity of each evaluation sample having a different temperature history (at the time of production). On the other hand, when a second run is performed on the evaluation samples, all the evaluation samples have the same temperature history (second run), so the crystallinity degrees of all the evaluation samples are the same. The latent heat obtained by performing the third run on the evaluation sample reflects the degree of crystallinity at the time of the second run. Therefore, the crystallinity "X" of each evaluation sample can be expressed, for example, by (Equation 8) shown below.

5. Experimental Results FIG. 15 is a graph showing an example of actually measured changes in resin temperature.

In FIG. 15, the horizontal axis indicates the distance (mm) from the exit of the die, and the vertical axis indicates the temperature (° C.) of the resin surface. The temperature history can be understood from the change in resin temperature shown in FIG. Then, the differential equation (Abrami's equation) in which the temperature function of the heat flow obtained by training the neural network is substituted, is solved by inputting the different temperature changes (temperature history) for each actually measured evaluation sample. This makes it possible to estimate the degree of crystallinity for each evaluation sample. Specifically, the degree of crystallinity was calculated for all of the prepared evaluation samples having different temperature histories.

On the other hand, for all of the prepared evaluation samples having different temperature histories, the heat flow was actually measured using "DSC", and then the crystallinity degree was actually measured based on (Equation 8).

FIG. 16 is a graph showing the comparison results between calculated values and actually measured values.

In FIG. 16, the horizontal axis indicates conductor temperature (° C.), and the vertical axis indicates crystallinity (%). As shown in Figure 16, the crystallinity degree (calculated value) estimated based on the heat flow temperature function obtained by training the neural network is in good agreement with the measured crystallinity degree (actual value). It is proven that this is the case.

<Application example>
For example, as shown in "<<Verification Result 1>>" above, the heat flow temperature function obtained by learning the neural network closely matches the measured data. Based on this, the following application examples can be considered.

For example, actual measurement data measured by "DSC" covers a data string of 10,000 to 20,000 points. Furthermore, in order to carry out the first run, second run, and third run, the number of actual measurement data points reaches tens of thousands of points, so a large amount of storage capacity is required to save the actual measurement data. Conceivable.

In this regard, the heat flow temperature function obtained by training the neural network closely matches the measured data. For this reason, it is possible to store a temperature function instead of actual measurement data and, for example, to output a heat flow value based on the temperature function when a temperature value is input. Thereby, it is possible to substantially reproduce the measured data. In this case, since it is only necessary to store the temperature function itself, it is possible to reduce the amount of data to be stored.

Specifically, as shown in FIG. 17, the crystallinity estimation device 100 is configured to include a data reproduction section 310 that reproduces actually measured data.

Here, the data reproduction section 310 is configured to include a temperature function storage section 311, a temperature value input section 312, a heat flow value acquisition section 313, and a heat flow value output section 314.

The temperature function storage unit 311 is configured to store the temperature function acquired by the temperature function acquisition unit 302 instead of actual measurement data, and the temperature value input unit 312 is configured to input temperature values. There is. Further, the heat flow value acquisition unit 313 is configured to acquire a heat flow value by substituting the temperature value input to the temperature value input unit 312 into a temperature function. Further, the heat flow value output section 314 is configured to output the heat flow value acquired by the heat flow value acquisition section 313.

For example, by having the data reproducing unit 310 instead of the measured data storage unit configured to store measured data, the crystallinity estimation device 100 reduces the amount of data to be stored compared to storing measured data. be able to. This is because in order to memorize the actual measurement data itself, it is necessary to memorize tens of thousands of points of data, whereas in order to realize the data reproduction section 310, it is only necessary to memorize the temperature function itself. It is.

Furthermore, while the actual measurement data is discrete data, the data reproduced by the data reproduction unit 310 is continuous data. Therefore, by providing the data reproduction section 310, it is possible to reproduce data that does not exist in the actual measurement data. In other words, by providing the data reproduction section 310 in place of the measured data storage section, not only can the amount of data to be stored be reduced, but also the data density of the data reproduced by the data reproduction section 310 can be reduced from the data density of the measured data. The advantage is that it can be made larger than the density.

The invention made by the present inventor has been specifically explained based on the embodiments thereof, but the present invention is not limited to the embodiments described above, and can be modified in various ways without departing from the gist thereof. Needless to say.

In the embodiment described above, linear low-density polyethylene was used as an example, but the technical idea in the embodiment is not limited to this, and any material whose heat flow can be measured by "DSC" may be used. It can be applied to both plastic and metal materials.

Furthermore, the technical concept of the embodiment described above is not limited to application to differential equations for estimating the degree of crystallinity, but since the functional form is smooth, it is flexible for calculation of other differentials and integrals, and for other uses. can correspond to

Furthermore, from the perspective of data storage, the data capacity can be reduced by storing functions acquired by training a neural network instead of actual measurement data, so it is possible to reduce the data capacity by storing functions acquired by training a neural network instead of actual measurement data. It is also easy to apply.

10 Raw material pellet 11 Extruder 12 Crosshead 13 Die 14 Conductor 15 Water tank 16 Position 17 Laser thermometer 100 Crystallinity estimation device 101 CPU
102 ROM
103 RAM
104 Display 105 Keyboard 106 Mouse 107 Communication board 108 Removable disk device 109 CD/DVD-ROM device 110 Printer 111 Scanner 112 Hard disk device 113 Bus 201 Operating system 202 Program group 203 File group 301 Input section 302 Temperature function acquisition section 303 Crystallinity Estimation unit 304 Output unit 305 Data storage unit 310 Data reproduction unit 311 Temperature function storage unit 312 Temperature value input unit 313 Heat flow value acquisition unit 314 Heat flow value output unit 400 Middle layer 401 Node

Claims (12)

A crystallinity estimation device configured to estimate the crystallinity of the material based on the heat flow, which indicates the amount of heat dissipated from the material per unit time and has temperature dependence,
a temperature function acquisition unit configured to acquire a temperature function of the heat flow based on actual measurement data obtained by measuring a heat flow value corresponding to a temperature value;
a crystallinity estimator configured to estimate the crystallinity of the material based on the temperature function obtained by the temperature function obtainer;
Equipped with
The temperature function acquisition unit is configured to acquire the temperature function by using the measured data as training data and training a neural network that inputs temperature values and outputs heat flow values. crystallinity estimation device. The crystallinity estimation device according to claim 1,
The middle layer of the neural network is composed of a single layer,
The crystallinity estimation device, wherein the intermediate layer has six or more nodes. The crystallinity estimation device according to claim 1 or 2,
In the crystallinity estimation device, the heat flow is expressed by Equation 1 below.

here,
J(T): Heat flow Q: Amount of heat T: Temperature t: Time The crystallinity estimation device according to any one of claims 1 to 3,
The crystallinity estimating unit solves the equation 4 by incorporating the temperature function of the heat flow obtained by the temperature function obtaining unit into the relational expression of the crystallization rate shown by the following equation 4, A crystallinity estimation device configured to estimate a crystallinity of the material.

here,
X: Crystallinity t: Time J(T): Heat flow T: Temperature C: Coefficient The crystallinity estimation device according to claim 4,
The crystallization rate included in Equation 4 is expressed by Equation 5 below.

here,
X: Crystallinity t: Time J(T): Heat flow T: Temperature X _∞ : Saturated crystallinity The crystallinity estimation device according to any one of claims 1 to 5,
The crystallinity estimation device includes a data reproduction unit that reproduces the measured data,
The data reproduction unit is
a temperature function storage unit configured to store the temperature function acquired by the temperature function acquisition unit instead of the actual measurement data;
a temperature value input portion configured to input a temperature value;
a heat flow value acquisition unit configured to acquire a heat flow value by substituting the temperature value input into the temperature value input unit into the temperature function stored in the temperature function storage unit;
a heat flow value output unit configured to output the heat flow value acquired by the heat flow value acquisition unit;
A crystallinity estimation device including: The crystallinity estimation device according to claim 6,
The crystallinity estimation device has the data reproducing unit instead of the measured data storage unit configured to store the measured data, so that the amount of data to be stored can be reduced compared to when storing the measured data. A crystallinity estimation device consisting of: The crystallinity estimation device according to claim 6 or 7,
The crystallinity estimation device, wherein the data density of the data reproduced by the data reproduction section is higher than the data density of the actually measured data. The crystallinity estimation device according to any one of claims 6 to 8,
The measured data is discrete data,
The crystallinity estimation device, wherein the data reproduced by the data reproduction unit is continuous data. In a crystallinity estimation method for estimating the crystallinity of the material based on the heat flow that indicates the amount of heat dissipated from the material per unit time and has temperature dependence,
a temperature function acquisition step of acquiring a temperature function of the heat flow based on actual measurement data obtained by measuring a heat flow value corresponding to the temperature value;
a crystallinity estimation step of estimating the crystallinity of the material based on the temperature function obtained in the temperature function obtaining step;
Equipped with
In the temperature function acquisition step, the temperature function is acquired by using the measured data as training data and training a neural network that inputs temperature values and outputs heat flow values. Estimation method. A method for causing a computer to execute a crystallinity estimation process of estimating the crystallinity of the material based on the heat flow, which indicates the amount of heat dissipated from the material per unit time and has temperature dependence. In the crystallinity estimation program,
a temperature function acquisition process of acquiring a temperature function of the heat flow based on actual measurement data obtained by measuring a heat flow value corresponding to the temperature value;
Crystallinity estimation processing that estimates the crystallinity of the material based on the temperature function obtained in the temperature function obtaining processing;
Equipped with
The temperature function acquisition process is a process of acquiring the temperature function by using the measured data as teacher data to learn a neural network that inputs temperature values and outputs heat flow values. Crystallinity estimation program. A computer-readable recording medium recording the crystallinity estimation program according to claim 11.

Images