JP7416175B1 - Prediction program, information processing device, and prediction method - Google Patents

Abstract

The present invention provides a new technique for predicting adhesive properties of adhesive materials. [Solution] A prediction program for predicting the adhesive properties of an adhesive material is configured to input the viscoelastic properties of the adhesive material to a computer and output the adhesive properties of the adhesive material. A step of obtaining a learned predictive model and a step of inputting viscoelastic properties into the predictive model to predict adhesive properties of other adhesive materials are executed. [Selection diagram] Figure 2

Description

The present disclosure relates to a prediction program, an information processing device, and a prediction method.

JP 2021-038344A (Patent Document 1) discloses a search method aimed at easily searching for a curable composition having excellent adhesive strength. The search method searches for the adhesive strength of a curable composition using a learned model. The trained model receives inputs such as the molecular weight of the main agent, the molecular weight of the curing agent, the molar ratio of the content of amino groups in the amine curing agent to the content of glycidyl groups in the main agent, and the curing temperature. and is configured to output the adhesive strength of the curable composition.

Japanese Patent Application Publication No. 2021-038344

With conventional prediction methods, it has been difficult to accurately predict the adhesive properties of adhesive materials. Therefore, new techniques for predicting the adhesive properties of adhesive materials are desired.

In one example of the present disclosure, a prediction program is provided for predicting adhesive properties of a pressure-sensitive adhesive material. The above prediction program includes the steps of: receiving an input of viscoelastic properties of an adhesive material to a computer, and acquiring a prediction model trained to output the adhesive properties of the adhesive material; A step of inputting the viscoelastic properties and predicting the adhesive properties of other adhesive materials is executed.

In one example of the present disclosure, the step of predicting includes a step of receiving input of desired adhesive properties, and a step of receiving input of desired adhesive properties, and a step of receiving input of desired adhesive properties, and a step of receiving input of desired adhesive properties, and a step of receiving input of desired adhesive properties, and a step of receiving input of desired adhesive properties, and a step of receiving input of desired adhesive properties, and a step of receiving input of desired adhesive properties, and when adhesive properties outputted from the prediction model become the desired adhesive properties, the predicted adhesive properties are input to the prediction model. and searching for viscoelastic properties.

In an example of the present disclosure, the prediction model is generated by a learning process using a plurality of learning data. The plurality of learning data include a plurality of first data. Each of the plurality of first data associates the adhesive properties of the adhesive material with the viscoelastic properties of the adhesive material as a label.

In an example of the present disclosure, the plurality of learning data further includes a plurality of second data. Each of the plurality of pieces of second data associates the adhesive properties of the polymeric material as a label with the viscoelastic properties of the polymeric material other than the adhesive material.

In one example of the present disclosure, the viscoelastic properties input to the prediction model are at least one of the storage modulus of the adhesive material, the loss modulus of the adhesive material, and the loss tangent of the adhesive material. Contains one.

In one example of the present disclosure, the viscoelastic properties input to the prediction model include a feature extracted from the storage modulus of the adhesive material, a feature extracted from the loss modulus of the adhesive material, and a feature extracted from the loss tangent of the adhesive material.

In one example of the present disclosure, the adhesive properties output from the prediction model include a property indicating the strength required for peeling the adhesive material and the adherend, and a property indicating the difficulty of the adhesive material slipping from the adherend. and a property indicating ease of adhesion of the adhesive material and the adherend.

In one example of the present disclosure, the adhesive properties output from the prediction model include a property indicating the strength required for peeling the adhesive material and the adherend, and a property indicating the difficulty of the adhesive material slipping from the adherend. and a property indicating ease of adhesion of the adhesive material and the adherend.

In another example of the present disclosure, an information processing device for predicting adhesive properties of an adhesive material is provided. The information processing device includes a control section for controlling the information processing device. The control unit receives input of the viscoelastic properties of the adhesive material, and performs a process of acquiring a predictive model trained to output the adhesive properties of the adhesive material, and adding the viscoelastic properties to the predictive model. and execute the process to predict the adhesive properties of other adhesive materials.

In another example of the present disclosure, a predictive method for predicting adhesive properties of adhesive materials is provided. The above prediction method includes the steps of receiving an input of the viscoelastic properties of an adhesive material, obtaining a prediction model trained to output the adhesive properties of the adhesive material, and adding the viscoelastic properties to the above prediction model. and predicting adhesive properties of other adhesive materials.

These and other objects, features, aspects and advantages of the present invention will become apparent from the following detailed description of the invention, taken in conjunction with the accompanying drawings.

It is a figure showing an example of adhesive material. FIG. 3 is a diagram conceptually showing a method for predicting adhesive properties. FIG. 1 is a diagram illustrating an example of a hardware configuration of an information processing device. FIG. 3 is a diagram showing an example of a learning data set. It is a figure which shows an example of the storage elastic modulus prescribed|regulated to the data for learning. It is a figure which shows an example of the loss elasticity modulus prescribed|regulated to the data for learning. It is a figure which shows an example of the loss tangent prescribed|regulated in the data for learning. FIG. 1 is a diagram illustrating an example of a functional configuration of an information processing device. FIG. 3 is a diagram conceptually showing learning processing by a learning section. FIG. 3 is a diagram conceptually showing prediction processing by a prediction unit. 3 is a flowchart showing the flow of learning processing. It is a flowchart which shows the flow of search processing. 7 is a diagram illustrating an example of a learning data set according to modification example 1. FIG. 7 is a diagram showing a prediction model according to modification example 1. FIG. 7 is a diagram conceptually illustrating search processing according to Modification 1. FIG. 7 is a diagram for explaining feature extraction processing in modification example 2. FIG. It is a figure which shows the result of a performance evaluation experiment of a prediction model.

Hereinafter, each embodiment according to the present invention will be described with reference to the drawings. In the following description, the same parts and components are given the same reference numerals. Their names and functions are also the same. Therefore, detailed explanations thereof will not be repeated. Note that each embodiment and each modification described below may be selectively combined as appropriate.

<A. Adhesive material＞
First, the adhesive material will be explained with reference to FIG. FIG. 1 is a diagram showing an example of an adhesive material.

The term "adhesive material" as used herein means a material that has the function of bonding objects together. Examples of adhesive materials include adhesive tapes and adhesives. In the following description, an example of an adhesive tape will be described as an adhesive material, but the adhesive material is not limited to an adhesive tape.

The adhesive tape may be a single-sided tape or a double-sided tape. Further, the use of the adhesive tape is not particularly limited, and may be for industrial use or for household use. Examples of the adhesive tape types include coreless tape, nonwoven fabric core tape, film core tape, foam core tape, metal foil core tape, and the like.

FIG. 1 shows a single-sided adhesive tape 10A that is an example of the adhesive material 10. The single-sided adhesive tape 10A includes, for example, a sheet-like core base material 12, an adhesive 14, and a release agent 16.

The type of core base material 12 is not particularly limited. The core base material 12 may be a plastic film, paper, foam, metal foil, or any other type of sheet. Good too.

An adhesive 14 is applied to one surface of the core base material 12. A release agent 16 is applied to the other surface of the core base material 12 .

The release agent 16 covers the adhesive 14 while the single-sided adhesive tape 10A is wound into a roll, and protects the adhesive 14 from decreasing its adhesive strength.

<B. Overview＞
Hereinafter, the characteristics related to at least one of the viscosity of the adhesive 14 and the elasticity of the adhesive 14 constituting the adhesive material 10 will also be referred to as "viscoelastic characteristics." As an example, the viscoelastic property is at least one of a physical quantity indicating the viscosity of the adhesive 14 constituting the adhesive material 10, a physical quantity indicating the elasticity of the adhesive, and a physical quantity indicating both the viscosity and the elasticity. Represented by one. Specific examples of viscoelastic properties will be described later.

Further, the adhesion-related properties of the adhesive material 10 are also referred to as "adhesive properties." The adhesive property is expressed by a physical quantity indicating the adhesive force of the adhesive material 10, etc. In the case of double-sided tapes, each side of the adhesive has adhesive properties. A specific example will be described later.

Below, with reference to FIG. 2, a method for predicting the adhesive properties of the adhesive material 10 on the corresponding side from the viscoelastic properties of the adhesive 14 of the adhesive material 10 will be described. FIG. 2 is a diagram conceptually showing a method for predicting adhesive properties.

The adhesive property prediction function is implemented in the information processing device 100, for example. The information processing device 100 is, for example, a desktop PC (Personal Computer), a notebook PC, a tablet terminal, a smartphone, or another computer.

The information processing device 100 predicts the adhesive properties of the adhesive material 10 using the learned prediction model 124. The prediction model 124 is generated in advance using a learning data set. The learning data set associates actual measurement data regarding adhesive properties as a label with actual measurement data regarding the viscoelastic properties of various adhesive materials.

By using such a training data set for learning, when the prediction model 124 receives data indicating the viscoelastic properties of the adhesive 14 constituting the adhesive material 10, the prediction model 124 Now outputs data showing adhesive properties. In other words, the prediction model 124 receives data related to the viscoelastic properties of the adhesive 14 constituting the adhesive material 10 as an explanatory variable, and uses data related to the adhesive properties of the adhesive material 10 as an objective variable. It will be output as .

As described above, by learning the correlation between the viscoelastic properties of the adhesive 14 constituting the adhesive material 10 and the adhesive properties of the adhesive material 10, the information processing device 100 can It becomes possible to predict the adhesive properties of the adhesive material 10 without depending on the composition, structure, molecular weight, etc. of the adhesive material 10.

<C. Hardware configuration of information processing device 100>
Next, with reference to FIG. 3, the hardware configuration of the information processing device 100 will be described. FIG. 3 is a diagram illustrating an example of the hardware configuration of the information processing device 100.

The information processing device 100 includes a control device 101 (control unit), a ROM (Read Only Memory) 102, a RAM (Random Access Memory) 103, a communication interface 104, a display interface 105, an input interface 107, and an auxiliary memory. device 120. These components are connected to bus 115.

Control device 101 is configured, for example, by at least one integrated circuit. The integrated circuit is, for example, at least one CPU (Central Processing Unit), at least one GPU (Graphics Processing Unit), at least one ASIC (Application Specific Integrated Circuit), at least one FPGA (Field Programmable Gate Array), or the like. It may be configured by a combination of the following.

The control device 101 controls the operation of the information processing device 100 by executing various programs. The control device 101 reads a program to be executed from the auxiliary storage device 120 or the ROM 102 to the RAM 103 based on receiving an execution command for various programs. The RAM 103 functions as a working memory and temporarily stores various data necessary for executing programs.

A LAN (Local Area Network), an antenna, and the like are connected to the communication interface 104. The information processing device 100 exchanges data with external devices via the communication interface 104. The external device includes, for example, a server.

A display device 106 is connected to the display interface 105 . The display interface 105 sends an image signal for displaying an image to the display device 106 according to a command from the control device 101 or the like. The display device 106 is, for example, a liquid crystal display, an organic EL (Electro Luminescence) display, or another display. Note that the display device 106 may be configured integrally with the information processing apparatus 100 or may be configured separately from the information processing apparatus 100.

An input device 108 is connected to the input interface 107. Input device 108 is, for example, a mouse, keyboard, touch panel, or other device capable of receiving user operations. Note that the input device 108 may be configured integrally with the information processing apparatus 100 or may be configured separately from the information processing apparatus 100.

The auxiliary storage device 120 is, for example, a storage medium such as a hard disk or flash memory. The auxiliary storage device 120 stores, for example, a learning data set 122, a prediction model 124, a learning program 126, and a prediction program 128. These storage locations are not limited to the auxiliary storage device 120, and may be stored in the storage area of the control device 101 (for example, cache memory, etc.), the ROM 102, the RAM 103, an external device (for example, a server), or the like.

The learning program 126 is a program for generating the predictive model 124 using the learning data set 122. The learning program 126 may be provided not as a standalone program but as a part of any program. In this case, the learning process by the learning program 126 is realized in cooperation with any program. Even if the program does not include such a part of the module, it does not deviate from the spirit of the learning program 126 according to this embodiment. Further, some or all of the functions provided by the learning program 126 may be realized by dedicated hardware. Furthermore, the information processing apparatus 100 may be configured in a form such as a so-called cloud service in which at least one server executes a part of the processing of the learning program 126.

The prediction program 128 is a program for predicting the adhesive properties of the adhesive material from the viscoelastic properties of the adhesive material using the learned prediction model 124. The prediction program 128 may be provided not as a standalone program but as a part of any program. In this case, the learning process by the prediction program 128 is realized in cooperation with an arbitrary program. Even if the program does not include some of these modules, it does not depart from the spirit of the prediction program 128 according to this embodiment. Furthermore, some or all of the functionality provided by the prediction program 128 may be realized by dedicated hardware. Furthermore, the information processing apparatus 100 may be configured in a form such as a so-called cloud service in which at least one server executes a part of the processing of the prediction program 128.

<D. Learning dataset 122>
Next, the learning data set 122 shown in FIG. 3 will be explained with reference to FIGS. 4 to 7. FIG. 4 is a diagram showing an example of the learning data set 122.

The learning data set 122 includes a plurality of learning data 123. The number of learning data 123 included in the learning data set 122 is arbitrary. As an example, the number of learning data 123 is from several tens to several tens of thousands.

The learning data 123 associates the adhesive properties of the adhesive material with the viscoelastic properties of the adhesive material as a label. The viscoelastic properties are defined as explanatory variables in the learning data 123. The adhesive property is defined as an objective variable in the learning data 123.

Further, the learning data 123 is associated with a tape ID (Identification). The tape ID is information for uniquely identifying the learning data 123. For example, the tape ID is input by the user so as not to be duplicated.

The viscoelastic properties defined in the learning data set 122 are actual values measured using a dynamic viscoelasticity measuring device. The viscoelastic properties include, for example, at least one of a storage modulus of the adhesive material, a loss modulus of the adhesive material, and a loss tangent of the adhesive material. In the example shown in FIG. 4, the viscoelastic properties include the storage modulus, the loss modulus, and the loss tangent.

Preferably, the viscoelastic properties defined in the learning data set 122 include a loss tangent. By defining the relationship between the loss tangent and the adhesive property in the learning data set 122, the prediction accuracy by the above-mentioned prediction model 124 is further improved.

The storage modulus corresponds to the energy component stored in the adhesive material as elastic energy when the adhesive material is deformed, and is an index representing the degree of hardness of the adhesive material.

FIG. 5 is a diagram showing an example of the storage elastic modulus defined in one learning data 123. In the example of FIG. 5, the temperature is indicated by "T", and the storage modulus at temperature T is indicated by "G1(T)".

The loss modulus corresponds to a component of loss energy that is dissipated due to internal friction when the adhesive material is deformed, and is an index representing the degree of viscosity of the adhesive material.

FIG. 6 is a diagram showing an example of the loss elasticity modulus defined in one learning data 123. In the example of FIG. 6, the temperature is indicated by "T", and the loss modulus at temperature T is indicated by "G2(T)".

The loss tangent is expressed as the ratio between the storage modulus "G1 (T)" and the loss modulus "G2 (T)". Typically, the loss tangent “tan δ(T)” is expressed by the following formula (1).

tanδ(T)=G2(T)/G1(T)...(1)
FIG. 7 is a diagram showing an example of a loss tangent defined in one learning data 123. In the example of FIG. 7, the temperature is indicated by "T", and the loss tangent at temperature T is indicated by "tan δ(T)".

The adhesive properties specified in the learning data 123 are actual values measured using various measuring devices such as a testing machine. The adhesive properties include the strength required for peeling the adhesive material 10 and the adherend (hereinafter also referred to as "adhesive force"), and the resistance to slipping of the adhesive material 10 to the adherend (hereinafter referred to as "adhesive force"). (also referred to as "holding force") and ease of adhesion between the adhesive material 10 and the adherend (hereinafter also referred to as "tack"). In the example of FIG. 4, the adhesive properties are defined by three values: adhesive force, holding force, and tack.

Examples of adhesive strength include peel adhesive strength, shear adhesive strength, and split adhesive strength.

The peel adhesive force is expressed, for example, by the force required to peel off the adhesive material in the direction perpendicular to the adhered surface (so-called 90° peel adhesive force). Alternatively, the peel adhesive force is expressed as the force required to peel off the adhesive material in a direction parallel to the adhered surface (so-called 180° peel adhesive force).

The shear adhesive force corresponds to the force when a force (that is, a shear stress) in the opposite direction is applied on a horizontal plane to adherends that are joined together using an adhesive material to cause the joint to break.

The splitting adhesive force corresponds to the force that occurs when adherends that are joined by an adhesive material are separated from each other by applying a force in the opposite direction in a vertical plane to the adherends.

Examples of the holding force include shear holding force and constant load holding force. The shear holding force is expressed as the distance by which the adhesive material deviates when a load is applied in the shearing direction to the adhesive material attached to the adherend for a certain period of time. Alternatively, the holding force is expressed as the time from when a load is applied in the shear direction to the adhesive material attached to the adherend until the adhesive material falls. On the other hand, constant load holding power is the distance that the adhesive material affixed to an adherend is displaced when a load is applied in the vertical direction, or the distance from which the adhesive material shifts after the load is applied. It is expressed as the time it takes for the adhesive material to fall.

An example of a tuck is a ball tuck. Ball tack is measured by sequentially rolling balls of different diameters against an inclined surface covered with an adhesive material. The ball tack is expressed by the maximum diameter of the ball stopped on the slope.

The learning data set 122 includes not only learning data 123 related to adhesive materials (hereinafter also referred to as "first data group"), but also learning data related to polymer materials other than adhesive materials. 123 (hereinafter also referred to as "second data group").

In this case, each learning data constituting the first data group is associated with an adhesive property indicating that it has an adhesive function as a label with respect to the viscoelastic property of the adhesive material. The adhesive property indicating that the adhesive has an adhesive function is defined as a value larger than "0" in the learning data 123, for example.

On the other hand, each learning data constituting the second data group is associated with an adhesive property indicating that it does not have an adhesive function as a label with respect to the viscoelastic property of a polymer material other than the adhesive material. A polymer material other than an adhesive material means a polymer material that is not used as an adhesive material and does not have an adhesive function. The adhesive property indicating that it does not have an adhesive function is defined as "0" in the learning data 123, for example.

Not only the learning data 123 related to the adhesive material but also the learning data 123 related to polymer materials other than the adhesive material are used in the learning described later, so that the prediction accuracy and prediction of the adhesive properties by the predictive model 124 can be improved. The scope of application will be improved.

<E. Functional configuration of information processing device 100>
Next, the functional configuration of the information processing device 100 will be described with reference to FIGS. 8 to 10. FIG. 8 is a diagram illustrating an example of the functional configuration of the information processing device 100.

As shown in FIG. 8, the information processing device 100 includes a learning section 152, a prediction section 154, and an output section 156 as functional configurations. Below, these functional configurations will be explained in order.

Note that the learning unit 152, the prediction unit 154, and the output unit 156 do not need to be installed in the information processing device 100. Some functional configurations of the learning unit 152, the prediction unit 154, and the output unit 156 may be implemented in the information processing device 100, and the remaining functional configurations may be implemented in another computer such as a server.

(E1. Learning section 152)
First, with reference to FIG. 9, the functions of the learning section 152 shown in FIG. 8 will be described. FIG. 9 is a diagram conceptually showing the learning process by the learning section 152.

The learning unit 152 executes a learning process using the above-mentioned learning data 123 (see FIG. 4) and generates a prediction model 124. The machine learning algorithm employed is not particularly limited, and various machine learning algorithms such as a neural network such as deep learning, a support vector machine, or a decision tree system may be employed. Below, learning processing using a neural network will be explained.

The prediction model 124 is composed of an input layer X, a middle layer H, and an output layer Y.

The input layer X receives input of viscoelastic properties defined in the learning data 123 as an explanatory variable. As an example, the input layer X includes a unit group x ₁ , a unit group x ₂ , and a unit group x ₃ .

The unit group x ₁ is configured to receive input of the storage elastic modulus “G1(T)” defined in the prediction model 124, for example. The number of units forming the unit group _x1 is the same as the number of data forming the storage elastic modulus "G1(T)". As an example, if the storage elastic modulus "G1 (T)" is composed of 190 pieces of data from "G1 (-40) to G1 (150)", unit group x ₁ is composed of 190 units. Ru. Each unit constituting the unit group _x1 outputs input data to each unit in the first layer of the intermediate layer H.

The unit group x ₂ is configured to receive input of the loss elasticity “G2(T)” defined in the prediction model 124, for example. The number of units forming the unit group _x2 is the same as the number of data forming the loss elasticity "G2(T)". As an example, if the loss modulus "G2(T)" is composed of 190 pieces of data from "G2(-40) to G2(150)", unit group x ₂ is composed of 190 units. Ru. Each unit constituting the unit group _x2 outputs input data to each unit in the first layer of the intermediate layer H.

The unit group x ₃ is configured to receive input of the loss tangent “tan δ(T)” defined in the prediction model 124, for example. The number of units forming the unit group _x3 is the same as the number of data forming the loss tangent "tan δ(T)". As an example, if the loss tangent "tan δ (T)" is composed of 190 pieces of data from "tan δ (-40) to tan δ (150)", the unit group x ₃ is composed of 190 units. . Each unit constituting the unit group _x3 outputs input data to each unit in the first layer of the intermediate layer H.

The intermediate layer H is composed of a plurality of layers. The number of intermediate layers H is arbitrary. In the example of FIG. 9, the intermediate layer H is composed of N layers (N is a natural number). Each layer of the intermediate layer H includes a plurality of units. In the example of FIG. 9, the first layer of the intermediate layer H is composed of units h _A1 , h _A2 . . . . The final layer of the intermediate layer H is composed of units h _N1 , h _N2 . . . .

Each unit constituting each layer of the intermediate layer H is connected to each unit of the previous layer and each unit of the next layer. Each unit in each layer receives each output value from each unit in the previous layer, multiplies each output value by a weight, integrates the multiplication results, and adds a predetermined bias to the integration result ( or subtraction), input the addition result (or subtraction result) to a predetermined function (for example, a sigmonite function), and output the output value of the function to each unit of the next layer.

The output layer Y outputs adhesive properties according to the viscoelastic properties input to the input layer X. As an example, the output layer Y is composed of units y ₁ to y ₃ . Below, units y ₁ to y ₃ are also referred to as unit y.

Each unit y is connected to each unit h _N1 , h _N2 . . . of the final layer of the intermediate layer H. Each unit y receives an output value from each unit in the final layer of the intermediate layer H, multiplies each output value by a weight, integrates the multiplication results, and applies a predetermined bias to the integration result. Addition (or subtraction) is performed, the addition result (or subtraction result) is input to a predetermined function (for example, a sigmonite function), and the output result of the function is output as an output value.

The number of units constituting the output layer Y is determined according to the number of objective variables defined in the learning data 123. As an example, when predicting the above-mentioned adhesive force, above-mentioned holding force, and above-mentioned tack, the number of units forming the output layer Y is three.

Unit _y1 is configured to output adhesive force as a prediction result. Unit _y2 is configured to output the conservative force as a prediction result. Unit _y3 is configured to output tack as a prediction result.

In the example of FIG. 9, one prediction model 124 is configured to output three predicted values of adhesive force, holding force, and tack; however, one prediction model 124 outputs one predicted value. You can. As an example, the first prediction model 124 is configured to receive input of the viscoelastic properties of the adhesive material and output the adhesive force of the adhesive material. The second prediction model 124 is configured to receive input of the viscoelastic properties of the adhesive material and output the holding force of the adhesive material. The third prediction model 124 is configured to receive input of the viscoelastic properties of the adhesive material and output the tack of the adhesive material.

Next, the updating process of the internal parameters of the prediction model 124 by the learning unit 152 will be described.

The learning unit 152 inputs the viscoelastic properties defined in the first learning data 123 to the prediction model 124. Thereby, the prediction model 124 outputs adhesive properties according to the inputted viscoelastic properties. Next, the learning unit 152 compares the output prediction results “s1” to “s3” with the target variable defined in the first learning data 123. The target variable is represented by, for example, scores "sA" to "sC".

The learning unit 152 calculates the error “Z” between the output results “s1” to “s3” of the prediction model 124 and the objective variables “sA” to “sC”. As an example, the error "Z" is calculated based on the following equation (2).

Z={(s1-sA) ² + (s2-sB) ² + (s3-sC) ² }/3...(2)
Next, the learning unit 152 updates various parameters (for example, weights and biases) included in the prediction model 124 so that the error "Z" becomes smaller. Updating of the parameters is realized, for example, by error backpropagation.

The learning unit 152 repeatedly updates the internal parameters of the prediction model 124 for each learning data 123 included in the learning data set 122. As a result, the prediction model 124 outputs accurate prediction results as learning progresses.

Note that the learning unit 152 does not need to use all of the learning data 123 included in the learning data set 122 for the learning process, and uses some of the learning data 123 included in the learning data set 122 to develop a predictive model. 124 may be generated. The remaining learning data 123 is used for evaluating the prediction model 124 and the like.

(E2. Prediction unit 154)
Next, with reference to FIG. 10, the function of the prediction unit 154 shown in FIG. 8 will be described. FIG. 10 is a diagram conceptually showing prediction processing by the prediction unit 154.

The prediction unit 154 receives the input of the desired adhesive property, and searches for the viscoelastic property input to the predictive model 124 when the adhesive property output from the predictive model 124 becomes the desired adhesive property. This makes it possible to search for viscoelastic properties with a desired balance of physical properties.

Desired adhesive properties are entered, for example, on an input screen displayed on the display device 106 described above. The user inputs desired adhesive properties into the input screen using, for example, the input device 108 described above.

Desired adhesive properties may be specified by a single numerical value or by a range of numerical values. As an example, the desired adhesive properties are specified by at least one of a numerical range Δsα for adhesive force, a numerical range Δsβ for holding force, and a numerical range Δsγ for tack.

The numerical range Δsα is specified, for example, at least one of the lower limit value of adhesive force and the upper limit value of adhesive force. The numerical range Δsβ is specified, for example, at least one of the lower limit value of the holding force and the upper limit value of the holding force. The numerical range Δsγ is specified, for example, at least one of a lower limit value of tack and an upper limit value of tack.

The prediction unit 154 searches for input viscoelastic properties such that the output of the prediction model 124 has the desired adhesive properties.

As an example, the prediction unit 154 generates viscoelastic property candidates (hereinafter also referred to as "viscoelastic property candidates") to be input to the prediction model 124, and sequentially inputs each viscoelastic property candidate to the prediction model 124. Then, the prediction result “s1” output from the prediction model 124 is included in the numerical range Δsα, and the prediction result “s2” output from the prediction model 124 is included in the numerical range Δsβ, and the prediction result “s2” output from the prediction model 124 is included in the numerical range Δsα. When the prediction result “s3” is included in the numerical range Δsγ, the prediction unit 154 stores the viscoelastic property candidate input to the prediction model 124. Otherwise, the prediction unit 154 discards the viscoelastic property candidate input to the prediction model 124 without storing it. Thereby, the prediction unit 154 can search for viscoelastic properties that have desired adhesive properties.

As another example, the prediction unit 154 may search for viscoelastic properties having desired adhesive properties using a search algorithm such as SMBO (Sequential Model-based Global Optimization).

Note that if there are multiple viscoelastic property candidates that satisfy the desired adhesive property, the prediction unit 154 may output all of the viscoelastic property candidates as the search results, or may output all of the viscoelastic property candidates as search results, or may A viscoelastic property candidate that minimizes the viscoelastic properties may be output as a search result.

As described above, the prediction unit 154 searches for viscoelastic properties that have desired adhesive properties. As an alternative to this search method, a method of predicting adhesive properties from the composition, structure, molecular weight, etc. of the adhesive 14 constituting the adhesive material 10 can be considered. However, with this method, the prediction accuracy of adhesive properties is low for materials whose composition, structure, molecular weight, etc. are unknown. On the other hand, the prediction method according to the present embodiment can predict the adhesive properties of the adhesive material 10 without depending on the composition, structure, molecular weight, etc. of the adhesive 14 constituting the adhesive material 10. It is possible to more accurately and flexibly search for a physical property balance that provides desired adhesive properties.

(E3. Output section 156)
Next, the function of the output section 156 shown in FIG. 8 will be explained.

The output unit 156 outputs the search result by the prediction unit 154. The search results can be output to any destination. As an example, the search result is displayed on the display device 106 of the information processing apparatus 100. As another example, the search results are stored as data.

Preferably, when there are a plurality of search results obtained by the prediction unit 154, the output unit 156 displays the search results that are closer to the desired adhesive properties with emphasis over other search results. The method of highlighting is arbitrary. As an example, the output unit 156 realizes highlighted display by displaying the search results in a predetermined color such as red. Alternatively, the output unit 156 realizes highlighted display by displaying a larger size.

<F. Flowchart related to learning process>
Next, with reference to FIG. 11, the flow of learning processing by the information processing apparatus 100 will be described. FIG. 11 is a flowchart showing the flow of learning processing.

The control device 101 of the information processing device 100 functions as the learning section 152 (see FIG. 8) described above by executing the learning program 126 (see FIG. 3) described above, and executes the learning process shown in FIG. 11. In other aspects, some or all of the learning process may be performed by circuit elements or other hardware.

In step S110, the control device 101 initializes a variable "i". As an example, variable "i" is initialized to "1".

In step S112, the control device 101 acquires the i-th learning data 123 included in the learning data set 122.

In step S114, the control device 101 inputs the viscoelastic properties defined in the learning data 123 acquired in step S112 to the prediction model 124. Thereby, the prediction model 124 outputs adhesive properties according to the inputted viscoelastic properties as a prediction result.

In step S116, the control device 101 calculates the error between the adhesive property as the correct value specified in the learning data 123 acquired in step S112 and the adhesive property as the prediction result obtained in step S114. Then, the internal parameters of the prediction model 124 are updated so that the error becomes smaller than the current one. The parameters are updated by, for example, error backpropagation.

In step S120, the control device 101 determines whether to end the learning process. As an example, the control device 101 determines to end the learning process when the variable "i" is larger than a predetermined value. Alternatively, the control device 101 determines to end the learning process when all of the learning data 123 included in the learning data set 122 has been used for learning. If the control device 101 determines to end the learning process (YES in step S120), it ends the process shown in FIG. 11. If not (NO in step S120), the control device 101 switches control to step S122.

In step S122, the control device 101 increments the variable "i". That is, the control device 101 increases the variable "i" by one. After that, the control device 101 returns control to step S112.

<G. Flowchart related to search processing>
Next, with reference to FIG. 12, the flow of search processing by the information processing apparatus 100 will be described. FIG. 12 is a flowchart showing the flow of search processing.

The control device 101 of the information processing device 100 functions as the above-described prediction unit 154 (see FIG. 8) by executing the above-mentioned prediction program 128 (see FIG. 3), and executes the search process shown in FIG. 12. In other aspects, some or all of the search processing may be performed by circuit elements or other hardware.

In step S210, the control device 101 initializes a variable "j". As an example, variable "j" is initialized to "1".

In step S212, the control device 101 receives input of desired adhesive properties. Desired adhesive properties are input by a user on an input screen displayed on display device 106, for example.

In step S214, the control device 101 generates viscoelastic property candidates that may have desired adhesive properties. Viscoelastic property candidates may be generated randomly or based on predetermined rules.

In step S216, the control device 101 acquires the prediction model 124 from the auxiliary storage device 120. Alternatively, if the prediction model 124 is stored in another computer, the control device 101 acquires the prediction model 124 from the other computer.

In step S218, the control device 101 acquires the j-th viscoelastic property candidate from among the viscoelastic property candidates generated in step S214, and inputs the viscoelastic property candidate into the prediction model 124. Thereby, the prediction model 124 outputs adhesive properties according to the inputted viscoelastic properties as a prediction result.

In step S220, the control device 101 determines whether the adhesive properties as the prediction result obtained in step S218 satisfy the desired adhesive properties obtained in step S212. When the control device 101 determines that the adhesive properties as a prediction result satisfy the desired adhesive properties (YES in step S220), the control device 101 switches the control to step S222. If not (NO in step S220), the control device 101 switches control to step S230.

In step S222, the control device 101 stores the j-th viscoelastic property candidate in the auxiliary storage device 120.

In step S230, the control device 101 determines whether to end the search process. As an example, the control device 101 determines to end the search process when viscoelastic properties that satisfy the desired adhesive properties are discovered. Alternatively, the control device 101 determines to end the search process when the variable "j" is larger than a predetermined value. If the control device 101 determines to end the search process (YES in step S220), the control device 101 switches control to step S232. Otherwise (NO in step S120), the control device 101 switches control to step S234.

In step S232, the control device 101 outputs the viscoelastic property candidates stored in step S222 as a search result. As an example, the search results are displayed on the display device 106 of the information processing apparatus 100. Alternatively, the search results are output as data.

In step S234, the control device 101 increments the variable "j". That is, the control device 101 increases the variable "j" by one. After that, the control device 101 returns control to step S218.

<H. Modification example 1>
Next, with reference to FIGS. 13 to 15, a first modification of the method for predicting adhesive properties will be described.

The prediction model 124 described above was configured to receive the viscoelastic properties of the adhesive material as input. On the other hand, the prediction model 124 according to this modification is configured to receive not only the viscoelastic properties of the adhesive material but also other information related to the adhesive material as input.

First, with reference to FIG. 13, the learning data set 122 according to modification example 1 will be described. FIG. 13 is a diagram illustrating an example of the learning data set 122 according to the first modification.

The learning data set 122 includes a plurality of learning data 123. The learning data 123 according to this modification differs from the learning data 123 shown in FIG. 4 described above in that additional information is defined as an explanatory variable. Since the other points are as described above, explanations other than the additional information will not be repeated below.

The additional information specified in the learning data 123 includes, for example, a base material ID, an adhesive thickness, and a base material thickness. Note that the additional information may further include information on additives such as particles and fillers. Furthermore, conditions for measuring each adhesive performance (for example, information on the adherend used when measuring adhesive strength, etc.) may be further included.

The base material ID is information indicating the type of the base material (for example, the above-mentioned core base material 12) that constitutes the adhesive material. The base material ID is predetermined according to the type of base material. As the base material ID, a characteristic value indicating the characteristics of the base material or the like may be used. The base material ID is input by the user, for example.

The adhesive thickness indicates the thickness of the adhesive (for example, the above-mentioned adhesive 14) constituting the adhesive material. The thickness is an actual value.

The base material thickness indicates the thickness of the base material (for example, the above-mentioned core base material 12) that constitutes the adhesive material. The thickness is an actual value.

Next, the prediction model 124 according to modification example 1 will be described with reference to FIG. 14. FIG. 14 is a diagram showing a prediction model 124 according to the first modification.

The prediction model 124 according to this modification is configured to receive not only the viscoelastic properties of the adhesive material but also the base material ID, the adhesive thickness, and the base material thickness. Different from 124.

Note that the prediction model 124 does not need to be configured to additionally accept all of the base material ID, adhesive thickness, and base material thickness; It may be configured to receive at least one of the following as an input.

By configuring the prediction model 124 to receive the base material ID as input, the information processing device 100 can learn the correlation between viscoelastic properties and adhesive properties of various adhesive materials. .

Further, by configuring the prediction model 124 to receive the adhesive thickness as input, the information processing device 100 can more accurately learn the correlation between the viscoelastic properties and the adhesive properties.

Furthermore, by configuring the prediction model 124 to receive the base material thickness as an input, the information processing device 100 can more accurately learn the correlation between the viscoelastic properties and the adhesive properties.

Next, search processing according to modification example 1 will be described with reference to FIG. 15. FIG. 15 is a diagram conceptually showing search processing according to modification 1.

The prediction unit 154 according to this modification searches for a combination of explanatory variables such that the output of the prediction model 124 has the desired adhesive properties.

As more specific processing, first, the prediction unit 154 generates viscoelastic property candidates to be input to the prediction model 124. The viscoelastic property candidates include the base material ID, the adhesive thickness, the base material thickness, the storage modulus of the adhesive material, the loss modulus of the adhesive material, and the loss tangent of the adhesive material. It is defined by a combination of

The prediction unit 154 sequentially inputs the generated viscoelastic property candidates to the prediction model 124. Then, the prediction unit 154 determines that the prediction result "s1" output from the prediction model 124 is included in the numerical range Δsα, the prediction result "s2" output from the prediction model 124 is included in the numerical range Δsβ, and Viscoelastic property candidates when the prediction result "s3" output from the model 124 is included in the numerical range Δsγ are stored. Otherwise, the prediction unit 154 discards the input viscoelastic property candidate without storing it.

Thereby, the prediction unit 154 can further specify not only the adhesive properties having the desired adhesive properties, but also the optimal type of base material, the optimal thickness of the adhesive, and the optimal thickness of the base material. .

<I. Modification example 2>
Next, with reference to FIG. 16, a second modification of the method for predicting adhesive properties will be described.

In the examples of FIGS. 9 and 10 described above, the viscoelastic properties were input directly to the prediction model 124. In contrast, in this modification, the extracted viscoelastic properties are input to the prediction model 124. Since the other points are as described above, the explanation thereof will not be repeated.

FIG. 16 is a diagram for explaining the feature extraction process of viscoelastic properties. Feature extraction for the viscoelastic properties is performed, for example, by the feature extraction unit 151 shown in FIG. 16.

The correlation value between the viscoelastic property and the adhesive property after the feature extraction by the feature extraction unit 151 is higher than the correlation value between the viscoelastic property and the adhesive property before the feature extraction by the feature extraction unit 151. Further, the number of dimensions representing the viscoelastic properties after the feature extraction by the feature extraction unit 151 is lower than the number of dimensions representing the viscoelastic properties before the feature extraction by the feature extraction unit 151. By extracting feature quantities, the prediction accuracy of the prediction model is further improved.

In one aspect, the feature extraction unit 151 performs feature extraction processing on the storage modulus “G1(T)” of the adhesive material, and inputs the extracted feature amount to the prediction model 124. Any algorithm may be adopted as the feature extraction algorithm for the storage modulus "G1(T)".

As an example, the feature extraction unit 151 selects the storage modulus at a specific temperature from the storage modulus "G1(T)" using the prediction accuracy of the prediction model as an index. As the selection method, for example, a filter method, a wrapper method, an embedding method, etc. can be used. As another example, the feature extraction unit 151 compresses information about the storage elastic modulus "G1(T)" by dimension reduction using a neural network, and inputs the compressed information to the prediction model 124 as a feature quantity. Good too. Note that a differential filter, a noise suppression filter, or the like may be provided before the feature extraction unit 151 as preprocessing to improve prediction accuracy.

In another aspect, the feature extraction unit 151 performs feature extraction processing on the loss modulus “G2(T)” of the adhesive material, and inputs the extracted feature amount to the prediction model 124. Any algorithm can be adopted as the feature extraction algorithm for the loss elasticity "G2(T)".

As an example, the feature extraction unit 151 selects the loss elasticity modulus at a specific temperature from the loss elasticity modulus "G2(T)" using the prediction accuracy of the prediction model as an index. As the selection method, for example, a filter method, a wrapper method, an embedding method, etc. can be used. As another example, the feature extraction unit 151 compresses information for the loss elasticity "G2(T)" by dimension reduction using a neural network, and inputs the compressed information to the prediction model 124 as a feature quantity. Good too. Note that a differential filter, a noise suppression filter, or the like may be provided before the feature extraction unit 151 as preprocessing to improve prediction accuracy.

In yet another aspect, the feature extraction unit 151 performs feature extraction processing on the loss tangent "tan δ (T)" of the adhesive material, and inputs the extracted feature amount to the prediction model 124. Any algorithm can be adopted as the feature extraction algorithm for the loss tangent "tan δ (T)".

As an example, the feature extraction unit 151 selects the loss tangent of a specific temperature from the loss tangent "tan δ (T)" using the prediction accuracy of the prediction model as an index. As a selection method, for example, a filter method, a wrapper method, an embedding method, etc. can be used. As another example, the feature extraction unit 151 may compress information by dimension reduction using a neural network for the loss tangent "tan δ (T)" and input the compressed information to the prediction model 124 as a feature quantity. good. Note that a differential filter, a noise suppression filter, or the like may be provided before the feature extraction unit 151 as preprocessing to improve prediction accuracy.

As described above, in this modification, the viscoelastic properties input to the prediction model 124 are the feature amount extracted from the storage elastic modulus "G1 (T)" of the adhesive material and the loss of the adhesive material. It includes at least one of a feature extracted from the elastic modulus "G2(T)" and a feature extracted from the loss tangent "tan δ(T)" of the adhesive material. Thereby, the information processing device 100 can further improve the prediction accuracy of adhesive properties.

<J. Evaluation experiment>
The inventor generated the predictive model 124 under various conditions in order to prove that there is a correlation between the viscoelastic properties of the adhesive material and the adhesive properties of the adhesive material, and the generated predictive model 124 We conducted performance evaluation experiments on the following.

FIG. 17 is a diagram showing the results of a performance evaluation experiment of the prediction model 124. Below, with reference to FIG. 17, each performance evaluation experiment and experiment results conducted by the inventor will be explained.

(J1. Example 1)
First, a performance evaluation experiment of the prediction model 124 according to the first embodiment will be described.

In this embodiment, the inventors generated three predictive models (hereinafter also referred to as "predictive models 124_1A to 124_1C") from the above-mentioned learning data set 122, and evaluated the performance of each of the predictive models 124_1A to 124_1C. We conducted an experiment. A neural network was used as a learning algorithm to generate the predictive models 124_1A to 124_1C.

(a) Prediction model 124_1A
335 pieces of learning data 123 included in the above-mentioned learning data set 122 (see FIG. 13) were used for learning the predictive model 124_1A. The prediction model 124_1A receives the adhesive thickness, base material thickness, base material type, storage modulus, loss modulus, and loss tangent as explanatory variables, and sets the 180° peel adhesive force as the objective variable. It was learned to output as . Moreover, the prediction model 124_1A was generated as a regression model.

For the storage modulus as an explanatory variable, values from -40°C to 150°C were used. For the storage modulus as an explanatory variable, values from -40°C to 150°C were used. Values from −40° C. to 150° C. were used for the loss tangent as an explanatory variable.

The coefficient of determination was used as an index representing the prediction accuracy of the prediction model 124_1A. The closer the value of the coefficient of determination is to "1.0", the higher the prediction accuracy of the prediction model 124_1A becomes.

The coefficient of determination of the prediction model 124_1A when using the 335 learning data 123 used during learning was "0.91". Furthermore, the coefficient of determination of the prediction model 124_1A when using the 18 pieces of learning data 123 that were not used during learning was "0.86".

(b) Prediction model 124_1B
For learning the predictive model 124_1B, 212 pieces of learning data 123 included in the above-mentioned learning data set 122 were used. The predictive model 124_1B is different from the above-described predictive model 124_1A in that it is configured to output the retention force as an objective variable and is configured as a binary classification model. The other configurations of the prediction model 124_1B are the same as the prediction model 124_1A.

AUC (Area Under the Curve) was used as an index representing the prediction accuracy of the prediction model 124_1B. The closer the value of ACU is to "1.0", the higher the prediction accuracy of the prediction model 124_1B.

The AUC of the prediction model 124_1B when using the 212 learning data 123 used during learning was "0.98". Furthermore, the AUC of the prediction model 124_1B when using the 12 pieces of learning data 123 that were not used during learning was "1.0".

(c) Prediction model 124_1C
For learning the predictive model 124_1C, 285 pieces of learning data 123 included in the above-mentioned learning data set 122 were used. The prediction model 124_1C differs from the above-described prediction model 124_1A in that it is configured to output ball tack as an objective variable. The other configurations of the prediction model 124_1C are the same as the prediction model 124_1A.

The coefficient of determination of the prediction model 124_1C when using the 285 learning data 123 used during learning was "0.96". Furthermore, the coefficient of determination of the prediction model 124_1C when using 16 pieces of learning data that were not used during learning was "0.88".

(J2. Example 2)
Next, a performance evaluation experiment of the prediction model 124 according to the second embodiment will be described.

In this embodiment, the inventors generated three other prediction models (hereinafter also referred to as "prediction models 124_2A to 124_2C") from the above-mentioned learning data set 122, and each of the prediction models 124_2A to 124_2C. A performance evaluation experiment was conducted. A neural network was used as a learning algorithm to generate the predictive models 124_2A to 124_2C.

(a) Prediction model 124_2A
The prediction model 124_2A was trained to receive the adhesive thickness, base material thickness, base material type, and storage modulus as explanatory variables and output the 180° peel adhesive force as the objective variable. The other points of the prediction model 124_2A are the same as the prediction model 124_1A described above.

The coefficient of determination of the prediction model 124_2A when using the 335 learning data 123 used during learning was "0.90". Furthermore, the coefficient of determination of the prediction model 124_2A when using the 18 pieces of learning data 123 that were not used during learning was "0.84".

(b) Prediction model 124_2B
The prediction model 124_2B was trained to receive the adhesive thickness, base material thickness, base material type, and storage modulus as explanatory variables, and output the holding force as an objective variable. The other points of the prediction model 124_2B are the same as the above-mentioned prediction model 124_1B.

The AUC of the prediction model 124_2B when using the 212 learning data 123 used during learning was "0.93". Furthermore, the AUC of the prediction model 124_2B when using the 12 pieces of learning data 123 that were not used during learning was "0.90".

(c) Prediction model 124_2C
The prediction model 124_2C was trained to receive the adhesive thickness, base material thickness, base material type, and storage modulus as explanatory variables, and to output ball tack as an objective variable. The other points of the prediction model 124_2C are the same as the prediction model 124_1C described above.

The coefficient of determination of the prediction model 124_2C when using the 285 learning data 123 used during learning was "0.80". Furthermore, the coefficient of determination of the prediction model 124_2C when using the 16 pieces of learning data 123 that were not used during learning was "0.79".

(J3. Example 3)
Next, a performance evaluation experiment of the prediction model 124 according to the third embodiment will be described.

In this embodiment, the inventors further generate three other prediction models (hereinafter also referred to as "prediction models 124_3A to 124_3C") from the above-mentioned learning data set 122, and each of the prediction models 124_3A to 124_3C We conducted performance evaluation experiments on the following. A neural network was used as a learning algorithm to generate the predictive models 124_3A to 124_3C.

(a) Prediction model 124_3A
The prediction model 124_3A was trained to receive inputs of adhesive thickness, base material thickness, base material type, and loss modulus as explanatory variables, and output 180° peel adhesive strength as an objective variable. The other points of the prediction model 124_3A are the same as the prediction model 124_1A described above.

The coefficient of determination of the prediction model 124_3A when using the 335 learning data 123 used during learning was "0.90". Furthermore, the coefficient of determination of the prediction model 124_3A when using the 18 pieces of learning data 123 that were not used during learning was "0.82".

(b) Prediction model 124_3B
The prediction model 124_3B was trained to receive the adhesive thickness, base material thickness, base material type, and loss modulus as explanatory variables, and output holding force as the objective variable. The other points of the prediction model 124_3B are the same as the prediction model 124_1B described above.

The AUC of the prediction model 124_3B when using the 212 pieces of learning data 123 used during learning was "0.95". Furthermore, the AUC of the prediction model 124_3B when using the 12 pieces of learning data 123 that were not used during learning was "0.90".

(c) Prediction model 124_3C
The prediction model 124_3C was trained to receive the adhesive thickness, base material thickness, base material type, and loss modulus as explanatory variables, and output ball tack as an objective variable. The other points of the prediction model 124_3C are the same as the prediction model 124_1C described above.

The coefficient of determination of the prediction model 124_3C when using the 285 learning data 123 used during learning was "0.86". Furthermore, the coefficient of determination of the prediction model 124_3C when using the 16 pieces of learning data 123 that were not used during learning was "0.77".

(J4. Example 4)
Next, a performance evaluation experiment of the prediction model 124 according to the fourth embodiment will be described.

In this embodiment, the inventors further generate three other prediction models (hereinafter also referred to as "prediction models 124_4A to 124_4C") from the above-mentioned learning data set 122, and each of the prediction models 124_4A to 124_4C We conducted performance evaluation experiments on the following. A neural network was used as a learning algorithm to generate the predictive models 124_4A to 124_4C.

(a) Prediction model 124_4A
The prediction model 124_4A was trained to receive inputs of adhesive thickness, base material thickness, base material type, and loss tangent as explanatory variables, and output 180° peel adhesive strength as an objective variable. The other points of the prediction model 124_4A are the same as the prediction model 124_1A described above.

The coefficient of determination of the prediction model 124_4A when using the 335 learning data 123 used during learning was "0.91". Furthermore, the coefficient of determination of the prediction model 124_4A when using the 18 pieces of learning data 123 that were not used during learning was "0.86".

(b) Prediction model 124_4B
The prediction model 124_4B was trained to receive the adhesive thickness, base material thickness, base material type, and loss tangent as explanatory variables and output the holding force as the objective variable. The other points of the prediction model 124_4B are the same as the prediction model 124_1B described above.

The AUC of the prediction model 124_4B when using the 212 pieces of learning data 123 used during learning was "0.98". Furthermore, the AUC of the prediction model 124_4B when using the 12 pieces of learning data 123 that were not used during learning was "1.0".

(c) Prediction model 124_4C
The prediction model 124_4C was trained to receive the adhesive thickness, base material thickness, base material type, and loss tangent as explanatory variables and to output ball tack as an objective variable. The other points of the prediction model 124_4C are the same as the prediction model 124_1C described above.

The coefficient of determination of the prediction model 124_4C when using the 285 learning data 123 used during learning was "0.90". Furthermore, the coefficient of determination of the prediction model 124_4C when using the 16 pieces of learning data 123 that were not used during learning was "0.89".

(J5. Example 5)
Next, a performance evaluation experiment of the prediction model 124 according to the fifth embodiment will be described.

In this embodiment, the inventors further generate three other prediction models (hereinafter also referred to as "prediction models 124_5A to 124_5C") from the above-mentioned learning data set 122, and each of the prediction models 124_5A to 124_5C We conducted performance evaluation experiments on the following. A neural network was used as a learning algorithm to generate the predictive models 124_5A to 124_5C.

(a) Prediction model 124_5A
In this example, feature amount extraction is performed for each of the storage modulus, loss modulus, and loss tangent by dimension reduction using a neural network in the feature extraction unit 151. Prediction model 124_5A received these extracted feature values, adhesive thickness, base material thickness, and base material type as explanatory variables, and was trained to output 180° peel adhesive strength as an objective variable. . The other points of the prediction model 124_5A are the same as the prediction model 124_1A described above.

The coefficient of determination of the prediction model 124_5A when using the 335 learning data 123 used during learning was "0.95". Furthermore, the coefficient of determination of the prediction model 124_5A when using the 18 pieces of learning data 123 that were not used during learning was "0.91".

(b) Prediction model 124_5B
In this example, feature amount extraction is performed for each of the storage modulus, loss modulus, and loss tangent by dimension reduction using a neural network in the feature extraction unit 151. The prediction model 124_5B was trained to receive these extracted feature amounts, adhesive thickness, base material thickness, and base material type as explanatory variables, and output holding force as an objective variable. The other points of the prediction model 124_5B are the same as the prediction model 124_1B described above.

The AUC of the prediction model 124_5B when using the 212 learning data 123 used during learning was "0.99". Furthermore, the AUC of the prediction model 124_5B when using the 12 pieces of learning data 123 that were not used during learning was "1.0".

(c) Prediction model 124_5C
In this example, feature amount extraction is performed for each of the storage modulus, loss modulus, and loss tangent by dimension reduction using a neural network in the feature extraction unit 151. The prediction model 124_5C was trained to receive these extracted feature amounts, adhesive thickness, base material thickness, and base material type as explanatory variables, and to output ball tack as an objective variable. The other points of the prediction model 124_5C are the same as the prediction model 124_1C described above.

The coefficient of determination of the prediction model 124_5C when using the 285 learning data 123 used during learning was "0.98". Furthermore, the coefficient of determination of the prediction model 124_5C when using the 16 pieces of learning data 123 that were not used during learning was "0.91".

(J6. Comparative example)
Next, a performance evaluation experiment of the prediction model 124 according to a comparative example will be described.

In this comparative example, the inventors further generate three other prediction models (hereinafter also referred to as "prediction models 124_5A to 124_5C") from the above-mentioned learning data set 122, and each of the prediction models 124_5A to 124_5C We conducted performance evaluation experiments on the following. A neural network was used as a learning algorithm to generate the predictive models 124_5A to 124_5C.

(a) Prediction model 124_5A
The prediction model 124_5A was trained to receive inputs of adhesive thickness, base material thickness, and base material type as explanatory variables, and output 180° peel adhesive strength as an objective variable. The other points of the prediction model 124_5A are the same as the prediction model 124_1A described above.

The coefficient of determination of the prediction model 124_5A when using the 335 learning data 123 used during learning was "0.65". Furthermore, the coefficient of determination of the prediction model 124_5A when using the 18 pieces of learning data 123 that were not used during learning was "0.60".

(b) Prediction model 124_5B
The prediction model 124_5B was trained to receive inputs of adhesive thickness, base material thickness, and base material type as explanatory variables, and output holding force as an objective variable. The other points of the prediction model 124_5B are the same as the prediction model 124_1B described above.

The AUC of the prediction model 124_5B when using the 212 pieces of learning data 123 used during learning was "0.20". Furthermore, the AUC of the prediction model 124_5B when using the 12 pieces of learning data 123 that were not used during learning was "0.18".

(c) Prediction model 124_5C
The prediction model 124_5C was trained to receive inputs of adhesive thickness, base material thickness, and base material type as explanatory variables, and output ball tack as an objective variable. The other points of the prediction model 124_5C are the same as the prediction model 124_1C described above.

The coefficient of determination of the prediction model 124_5C when using the 285 learning data 123 used during learning was "0.20". Furthermore, the coefficient of determination of the prediction model 124_5C when using the 16 pieces of learning data 123 that were not used during learning was "0.19".

(J7. Evaluation)
As shown in FIG. 17, the prediction accuracy of all the prediction models according to Examples 1 to 4 was better than the prediction accuracy of the prediction model according to the comparative example. This showed that storage modulus, loss modulus, and loss tangent all accurately predict adhesive properties when used as explanatory variables. In particular, the use of loss tangent as an explanatory variable improved the prediction accuracy of adhesive properties. Furthermore, when the three storage modulus, loss modulus, and loss tangent were used as explanatory variables, the prediction accuracy of adhesive properties was improved.

The embodiments disclosed this time should be considered to be illustrative in all respects and not restrictive. The scope of the present invention is indicated by the claims rather than the above description, and it is intended that all changes within the scope and meanings equivalent to the claims are included.

10 adhesive material, 10A single-sided adhesive tape, 12 core base material, 14 adhesive, 16 release agent, 100 information processing device, 101 control device, 102 ROM, 103 RAM, 104 communication interface, 105 display interface, 106 display device, 107 input interface, 108 input device, 115 bus, 120 auxiliary storage device, 122 training data set, 123 learning data, 124 prediction model, 124 second prediction model, 124 third prediction model, 124 first prediction model, 126 learning program, 128 prediction program, 151 feature extraction unit, 152 learning unit, 154 prediction unit, 156 output unit.

Claims (10)

A prediction program for predicting adhesive properties of an adhesive material, the prediction program comprising:
The prediction program causes the computer to
A step of receiving an input of the viscoelastic properties of the adhesive material and acquiring a predictive model learned to output the adhesive properties of the adhesive material is performed , and the predictive model is based on a plurality of learning data. The plurality of learning data includes a plurality of first data, and each of the plurality of first data is generated by a learning process using the adhesive material. The adhesive properties of the adhesive material are associated as a label, and the viscoelastic properties include a temperature-dependent loss tangent of the adhesive material,
inputting viscoelastic properties into the prediction model to predict the adhesive properties of other adhesive materials;
The viscoelastic properties input to the prediction model include a temperature-specific loss tangent of the adhesive material,
The step of predicting includes:
accepting input of desired adhesive properties;
A prediction program comprising the step of searching for viscoelastic properties input to the prediction model when the adhesive properties output from the prediction model become the desired adhesive properties. The plurality of learning data further includes a plurality of second data, and each of the plurality of second data sets the adhesion of the polymer material other than the adhesive material to the viscoelastic property of the polymer material. The prediction program according to claim 1 , wherein characteristics are associated as labels. The viscoelastic properties input to the prediction model further include at least one of a temperature-dependent storage modulus of the adhesive material and a temperature-dependent loss modulus of the adhesive material. The prediction program according to 1 or 2. The viscoelastic properties input to the prediction model further include a feature extracted from the storage modulus of the adhesive material, a feature extracted from the loss modulus of the adhesive material, and a feature extracted from the storage modulus of the adhesive material. 3. The prediction program according to claim 1, wherein the prediction program includes at least one feature extracted from a loss tangent of . The adhesive material includes an adhesive and a base material coated with the adhesive,
The prediction model is further configured to receive as input at least one of the thickness of the adhesive, the thickness of the base material, and the type of the base material. Forecasting program described. The adhesive properties output from the prediction model include a property that indicates the strength required for peeling the adhesive material and the adherend, a property that indicates the difficulty of the adhesive material slipping from the adherend, and a property that indicates the strength required for peeling the adhesive material and the adherend. The prediction program according to claim 1 or 2, comprising at least one of a characteristic indicating ease of adhesion of a material and an adherend. An information processing device for predicting adhesive properties of an adhesive material, the information processing device comprising:
comprising a control unit for controlling the information processing device,
The control unit includes:
Upon receiving the input of the viscoelastic properties of the adhesive material, a process is executed to obtain a predictive model trained to output the adhesive properties of the adhesive material, and the predictive model is based on a plurality of learning data. The plurality of learning data includes a plurality of first data, and each of the plurality of first data is generated by a learning process using the adhesive material. The adhesive properties of the adhesive material are associated as a label, and the viscoelastic properties include a temperature-dependent loss tangent of the adhesive material,
inputting the viscoelastic properties into the prediction model and performing a process of predicting the adhesive properties of other adhesive materials;
The viscoelastic properties input to the prediction model include a temperature-specific loss tangent of the adhesive material,
The prediction process is
a process of accepting input of desired adhesive properties;
An information processing device comprising: searching for a viscoelastic property input to the predictive model when the adhesive property output from the predictive model becomes the desired adhesive property. The viscoelastic properties input to the prediction model further include at least one of a temperature-dependent storage modulus of the adhesive material and a temperature-dependent loss modulus of the adhesive material. 7. The information processing device according to 7. A prediction method executed by an information processing device to predict adhesive properties of an adhesive material, the method comprising:
The predictive model includes a step of receiving an input of the viscoelastic properties of the adhesive material and acquiring a predictive model trained to output the adhesive properties of the adhesive material, the predictive model comprising a plurality of learning data. The plurality of learning data includes a plurality of first data, and each of the plurality of first data is generated by a learning process using the adhesive material. The adhesive properties of the adhesive material are associated as a label, and the viscoelastic properties include the temperature-dependent loss tangent of the adhesive material,
inputting viscoelastic properties into the prediction model to predict adhesive properties of other adhesive materials,
The viscoelastic properties input to the prediction model include a temperature-specific loss tangent of the adhesive material,
The step of predicting includes:
accepting input of desired adhesive properties;
A prediction method comprising the step of searching for viscoelastic properties input to the prediction model when the adhesive properties output from the prediction model become the desired adhesive properties. The viscoelastic properties input to the prediction model further include at least one of a temperature-dependent storage modulus of the adhesive material and a temperature-dependent loss modulus of the adhesive material. 9. The prediction method described in 9.

Images