Flame retardancy prediction device and flame retardancy prediction model generation device

Abstract

A flame resistance predicting device (100) comprises: an information acquiring unit (111) for receiving the input of material prescription information related to a material of a polymer composite material; and a predicting unit (113) which, using a flame resistance prediction model (prediction model (121)) for predicting information related to the flame resistance of the polymer composite material, predicts information related to the flame resistance of the polymer composite material from the material prescription information of the polymer composite material that has been input. Examples of the material prescription information include the type of resin, the type of an additive, the ratio of the resin and the additive, the structure of the resin, the structure of the additive, and a molding process for the polymer composite material.

Description

Technical Field

[0001] The present invention relates to a flame retardancy prediction device, a flame retardancy prediction model generation device, a flame retardancy prediction model, and a feature amount extraction device that predict the flame retardancy of a polymer composite material from the formulation information of the material.

Background Art

[0002] With the development of society and the advancement of technology, the replacement of metals with plastics is progressing in applications such as automobiles, electronic devices, and industrial materials. From the perspective of recent SDGs (Sustainable Development Goals) and safety, there is an increasing need to impart flame retardancy to plastics more than ever. In particular, since electrical and electronic devices potentially have a risk of ignition due to circuit short-circuiting, deterioration, etc., the requirements for flame retardancy of polymer composite materials (polymeric organic materials) used for device casings and the like are expected to become increasingly strict.

[0003] As the combustion mechanism of polymer composite materials, it is known that combustion continues in a cycle including the following (1) to (6) (source: the homepage of the Japan Flame Retardant Association). (1) Combustion of flammable gas: supply of flammable gas and oxygen (2) Generation of radiant heat by combustion: increase in the temperature of the organic material surface (3) Heat conduction into the organic material: increase in the temperature of the organic material (4) Thermal decomposition of the organic material: generation of flammable gas (5) Diffusion of flammable gas to the material surface: diffusion in the organic material (6) Diffusion of flammable gas to the combustion field: diffusion in the gas phase

[0004] Therefore, in order to stop combustion, it is sufficient to prevent the above cycle from continuing, and the imparting of flame retardancy only needs to act on one or more of (1) to (6). For example, flame retardants based on the combustion mechanism have been designed, such as halogen compounds that stabilize active OH radicals through a radical trapping effect or have an oxygen barrier effect, phosphorus compounds that promote the formation of char and insulating layers, metal hydroxides that have an endothermic effect through dehydration reactions, and drip inhibitors that suppress the dripping (drip) of resin material during combustion. The imparting of flame retardancy to composite materials is designed by adding these flame retardants.

[0005] On the other hand, composite materials used in enclosures and other applications require properties other than flame retardancy, such as impact resistance and moldability, so multiple additives are often used in combination. However, these properties often involve trade-offs. For example, it is known that adding flame retardants reduces the toughness and rigidity of the resin, and Patent Document 1 discloses a thermoplastic resin composition that solves this problem. [Prior art documents] [Patent Documents]

[0006] [Patent Document 1] Japanese Patent Publication No. 2011-144222 [Overview of the Initiative] [Problems that the invention aims to solve]

[0007] The thermoplastic resin composition described in Patent Document 1 exhibits excellent flame retardancy, prevents the adhesion of flame retardants to the mold during injection molding, and possesses rigidity, toughness, and heat resistance, but other properties are not mentioned. When designing the optimal formulation for composite materials that satisfy various properties, an experimental formulation is determined by subtly blending multiple additives based on past intuition and experience, then the mixture is actually kneaded and molded, flame retardancy and other evaluation tests are conducted, and the formulation is revised based on the results—a trial-and-error process. However, this method is inefficient for development.

[0008] Thus, developing flame-retardant composite materials is not easy. While optimizing a single property like flame retardancy allows for material design based on the cycles (1) to (6) described above, various additives are necessary to satisfy diverse properties in addition to flame retardancy. In composite materials where the resins, additives, formulations, and molding processes differ, the combustion behavior varies, making it difficult for humans to interpret what contributed to the flame retardancy. As a result, the current practice in developing flame-retardant composite materials involves trial and error through repeated prototyping to determine the optimal formulation.

[0009] This invention was made in view of the above background, and aims to provide a flame retardancy prediction device, a flame retardancy prediction model generation device, a flame retardancy prediction model, and a feature extraction device that enable the development of flame retardant composite materials in an efficient manner. [Means for solving the problem]

[0010] The above objectives of the present invention are achieved by the following means.

[0011] (1) The system comprises an information acquisition unit that receives input material formulation information relating to the materials of a polymer composite material, and a prediction unit that predicts information relating to the flame retardancy of the polymer composite material from the input material formulation information of the polymer composite material using a flame retardancy prediction model that predicts information relating to the flame retardancy of the polymer composite material. The flame retardancy prediction model consists of a first-stage prediction model that predicts information related to the combustion of the polymer composite material from material formulation information related to the polymer composite material, and a second-stage prediction model that predicts information related to the flame retardancy of the polymer composite material from information related to the combustion of the polymer composite material. A flame retardancy prediction device.

[0012] (2) The flame retardancy prediction device according to (1), characterized in that the material formulation information includes at least one piece of information from the following: the type of resin, the type of additive, the ratio of the resin to the additive, the structure of the resin, the structure of the additive, and information regarding the molding process of the polymer composite material.

[0013] (3) The flame retardancy prediction device according to (2), characterized in that the structure of the resin includes at least one of the chemical structure of the resin, weight-average molecular weight, number-average molecular weight, molecular weight distribution, degree of copolymerization, and degree of crosslinking.

[0014] (4) The additive includes at least one of a filler, a plasticizer, a colorant, a flame retardant, an ultraviolet absorber, an antioxidant, and an elastomer, and is characterized in that it is the flame retardancy prediction device according to (2).

[0015] (5) The information regarding the molding process includes at least one of kneading process conditions and molding conditions, and is characterized in that it is the flame retardancy prediction device according to (2).

[0016] (6) The information regarding the flame retardancy is the compliance of the standard regarding the flame retardancy, and is characterized in that it is the flame retardancy prediction device according to (1).

[0018] ( 7 ) The information regarding the combustion is the information obtained from the change over time during the combustion of the polymer composite material, and is characterized in that it is the flame retardancy prediction device according to ( 1 ).

[0019] ( 8 ) The The second part information regarding the combustion of the polymer composite material that serves as the input of the prediction model is the information selected according to the importance to the information regarding the flame retardancy that serves as the output of the The second part prediction model, and is characterized in that it is the flame retardancy prediction device according to ( 1 ).

[0021] ( 9 ) A flame retardancy prediction model generation device including a learning unit that generates a A first-stage prediction model that predicts information related to the combustion of the polymer composite material from the input material formulation information, using training data associated with information related to the combustion of the polymer composite material, and prediction model for predicting the information regarding the flame retardancy using teacher data in which the material formulation information regarding the material of the polymer composite material and the From the information relating to the combustion of the aforementioned polymer composite material information regarding the flame retardancy of the polymer composite material are associated. The second part prediction model , and a flame retardancy prediction model comprising the first-stage prediction model and the second-stage prediction model.

[0023] ( 10 ) The information regarding the combustion is the information obtained from the change over time during the combustion of the polymer composite material, and is characterized in that it is the flame retardancy prediction model generation device according to ( 9 ).

[0024] ( 11 ) For the information related to the combustion, it further includes an importance calculation unit that calculates the importance of the information related to the flame retardancy output by the prediction model, and the learning unit The second part selects the information related to the combustion that is input to the prediction model according to the importance, and The second part generates the prediction model, which is characterized in that the flame retardancy prediction model generation device described in The second part ( 9 ).

Advantages of the Invention

[0027] According to the present invention, it is possible to provide a flame retardancy prediction device, a flame retardancy prediction model generation device, a flame retardancy prediction model, and a feature quantity extraction device that can improve the efficiency of developing composite materials with flame retardancy.

Brief Description of the Drawings

[0028] [Figure 1] It is a functional block diagram of a flame retardancy prediction device according to the first embodiment. [Figure 2] It is a data configuration diagram of a material information database according to the first embodiment. [Figure 3] It is a flowchart of a learning process according to the first embodiment. [Figure 4] It is a flowchart of a prediction process according to the first embodiment. [Figure 5] It is a confusion matrix showing the relationship between the actual suitability of a composite material with UL94 V-2 grade according to the first embodiment and the suitability predicted by the prediction model. [Figure 6] It is a confusion matrix showing the relationship between the actual suitability of a composite material with UL94 V-0 grade according to the first embodiment and the suitability predicted by the prediction model. [Figure 7] It is a functional block diagram of a flame retardancy prediction device according to the second embodiment. [Figure 8] It is a data configuration diagram of a material information database according to the second embodiment. [Figure 9] It is a diagram showing a decision tree. [Figure 10] This is a decision tree for explaining importance according to the second embodiment. [Figure 11] This is a flowchart of the learning process according to the second embodiment. [Figure 12] This is a confusion matrix showing the relationship between the actual suitability of the UL94 V-2 grade composite material according to the second embodiment and the suitability predicted by the prediction model. [Modes for carrying out the invention]

[0029] The flame retardancy prediction device in an embodiment for carrying out the present invention is described below. The flame retardancy prediction device predicts information related to the flame retardancy of a polymer composite material (composite material) from the material formulation information (material formulation information) of the polymer composite material (composite material) using machine learning technology. Material formulation information refers to the ratio of resins and additives contained in the composite material. Information related to flame retardancy refers to, for example, whether or not it meets the standards related to flame retardancy. In order to improve the prediction accuracy, information related to specific combustion may be added to the input of the prediction process in addition to the material formulation information. By predicting flame retardancy from material formulation information, composite materials predicted to be flame-retardant can be prioritized for prototyping and flame retardancy testing. Prototyping and testing from the most promising candidates can reduce the number of prototypes and tests, thereby streamlining development.

[0030] <<First Embodiment: Configuration of Flame Retardancy Prediction Device>> Figure 1 is a functional block diagram of a flame retardancy prediction device 100 according to the first embodiment. The flame retardancy prediction device 100 is a computer and comprises a control unit 110, a storage unit 120, and an input / output unit 180. User interface devices such as a display, keyboard, and mouse are connected to the input / output unit 180. The input / output unit 180 may also include a communication device, enabling data transmission and reception with other devices. Furthermore, a media drive may be connected to the input / output unit 180, enabling data exchange using a recording medium.

[0031] The memory unit 120 is comprised of memory devices such as ROM (Read Only Memory), RAM (Random Access Memory), and SSD (Solid State Drive). The memory unit 120 stores a material information database 130 (see Figure 2 below), a prediction model 121, and a program 122. The prediction model 121 is a machine learning model that uses the material formulation information of the composite material as the explanatory variable and the suitability (conformity / non-conformity) of the composite material with respect to flame retardancy standards as the objective variable. The program 122 includes descriptions of the learning process (see Figure 3 below) and prediction process (see Figure 4 below) related to the prediction model 121.

[0032] ≪First Embodiment: Material Information Database≫ Figure 2 is a data structure diagram of the material information database 130 according to the first embodiment. The material information database 130 is, for example, tabular data, and each row (record) includes columns (attributes) for identification information 131 (labeled as ID in Figure 2), material formulation information 132, and flame retardancy 133. Identification information 131 is identification information for the composite material. Material formulation information 132 refers to the ratio of resins and additives that make up the composite material, for example, by weight. Note that there are generally multiple additives.

[0033] Flame retardancy 133 indicates the suitability of the composite material to the flame retardancy standard, with "Y" indicating suitability and "N" indicating unsuitability. Flame retardancy 133 is information obtained from the results of combustion experiments conducted on the composite material formed from the materials indicated in material formulation information 132. The flame retardancy standard is, for example, UL94 related to flame retardancy issued by Underwriters Laboratories. UL94 has grades such as V-0 and V-2. Flame retardancy 133 indicates the suitability of the composite material to, for example, the V-2 grade. Flame retardancy 133 may also indicate suitability to other standards or to the material's own flame retardancy standards.

[0034] The data (records) in the materials information database 130 are used for generating the prediction model 121 (see step S12 in Figure 3 below) and for evaluating the performance of the prediction model 121 (see step S13). For example, the data is appropriately divided into training data and validation data, the model is generated (trained) using the training data, and then its performance (prediction accuracy) is evaluated using the validation data. The data in the materials information database 130 is used, for example, for training and performance evaluation using the holdout method or cross-validation method.

[0035] <<First Embodiment: Control Unit>> Returning to Figure 1, the control unit 110 is configured to include a CPU (Central Processing Unit) and comprises an information acquisition unit 111, a learning unit 112, and a prediction unit 113. The information acquisition unit 111 acquires material formulation information and flame retardancy information of composite materials and stores them in the material information database 130.

[0036] The learning unit 112 trains a machine learning model using data from the materials information database 130 as training data to generate a prediction model 121. The prediction model 121 is, for example, a decision tree machine learning model, but it may also be a model of other machine learning techniques such as a random forest, a neural network, or an SVM (Support Vector Machine). The prediction unit 113 predicts the flame retardancy of the composite material using the prediction model 121 based on the material formulation information of the composite material acquired by the information acquisition unit 111.

[0037] <<First Embodiment: Learning Process>> Figure 3 is a flowchart of the learning process according to the first embodiment. In step S11, the information acquisition unit 111 acquires material formulation information and flame retardancy of the composite material and stores them in the material information database 130. The data in the material information database 130 is data in which material formulation information 132 is considered an explanatory variable (input) and flame retardancy 133 is considered an objective variable (output, correct label).

[0038] In step S12, the learning unit 112 acquires multiple data points from the material information database 130, and uses these data as training data to train a machine learning model and generate a predictive model 121. The training data is data in which material formulation information 132 is considered as an explanatory variable (input) and flame retardancy 133 is considered as an objective variable (output, correct label). In step S13, the learning unit 112 evaluates the performance (prediction accuracy) of the prediction model 121 using data from the material information database 130 that was not used in step S12.

[0039] In step S14, the learning unit 112 determines whether the performance of the prediction model in step S13 is above a predetermined threshold. If the performance is above the predetermined threshold (step S14 → YES), the learning unit 112 terminates the learning process. If the performance is below the threshold (step S14 → NO), the learning unit 112 returns to step S12.

[0040] <<First Embodiment: Prediction Processing>> Figure 4 is a flowchart of the prediction process according to the first embodiment. In step S21, the information acquisition unit 111 acquires material formulation information of the composite material to be predicted for flame retardancy. In step S22, the prediction unit 113 uses the material formulation information acquired in step S21 as input (explanatory variable) to the prediction model 121 and obtains flame retardancy as the output prediction result. The prediction unit 113 displays the prediction result on the display connected to the input / output unit 180.

[0041] <<First Embodiment: Performance Evaluation>> The performance evaluation results of the flame retardancy prediction device 100 (prediction model 121) are shown below. In the first evaluation experiment, 38 composite materials were fabricated and combustion tests were conducted to determine whether they met the UL94 V2 grade (compliant / uncompliant, compliant / uncompliant). The resin was polypropylene, and there were 10 types of additives. 27 of the 38 materials were used as training data to generate the decision tree prediction model 121. Subsequently, the performance (prediction accuracy) was evaluated using 11 validation data sets that were not used as training data.

[0042] Figure 5 is a confusion matrix showing the relationship between the actual suitability of the composite material to the UL94 V-2 grade according to the first embodiment and the suitability predicted by the prediction model. Of the four numbers, the "5" in the upper left is the number of composite materials that were predicted to be unsuitable for the V-2 grade, and whose combustion test results were also unsuitable, and which were correctly predicted (true negative). The "3" in the lower right is the number of composite materials that were predicted to be suitable for the V-2 grade, and whose combustion test results were also suitable, and which were correctly predicted (true positive).

[0043] The "2" in the upper right corner represents the number of composite materials that were predicted as suitable for V-2 grade but failed the combustion test, resulting in an incorrect prediction (false positive). The "1" in the lower left corner represents the number of composite materials that were predicted as unsuitable for V-2 grade but failed the combustion test, resulting in an incorrect prediction (false negative). The accuracy rate is (5+3) / (5+3+1+2)=72.7%, which is generally good prediction accuracy.

[0044] In the second evaluation experiment, 26 composite materials were fabricated and combustion tests were conducted to determine whether they met the V0 grade standard. The resin used was polypropylene, and 10 types of additives were used. Fifteen of the 26 materials were used as training data to generate a decision tree prediction model 121. Subsequently, performance evaluations were performed using 11 validation data sets that were not used as training data.

[0045] Figure 6 is a confusion matrix showing the relationship between the actual suitability of the UL94 V-0 grade composite material according to the first embodiment and the suitability predicted by the prediction model. The accuracy rate is 72.7%, indicating generally good prediction accuracy.

[0046] Features of the first embodiment The flame retardancy prediction device 100 predicts whether a composite material meets flame retardancy standards based on its material formulation information. By predicting flame retardancy from material formulation information, composite materials predicted to be flame retardant can be prioritized for prototyping and flame retardancy testing. Prototyping and testing can be performed starting with the most promising candidates, reducing the number of prototypes and tests and streamlining development.

[0047] ≪Second Embodiment: Overview≫ In the first embodiment, the only information (explanatory variables) for predicting flame retardancy was material formulation information. Information related to combustion may also be added to the explanatory variables. Figure 7 is a functional block diagram of the flame retardancy prediction device 100A according to the second embodiment. Compared to the flame retardancy prediction device 100 according to the first embodiment, the control unit 110 is equipped with an importance calculation unit 114, and the configuration of the learning unit 112A and the material information database 130A is different.

[0048] Figure 8 is a data configuration diagram of the material information database 130A according to the second embodiment. Combustion information 134, which is information related to the combustion of the composite material, is added as an attribute. Combustion information 134 is information obtained by manufacturing and burning the composite material, and includes information that changes over time during combustion (information obtained from changes over time). Combustion information includes the time from when the composite material is exposed to the flame until it ignites, the way the flame spreads, the color of the flame, the size of the flame, whether or not a char (carbonized layer) is formed, the time until the first drip occurs, the number of drips, the viscosity of the drip (high viscosity / low viscosity, etc.), the color of the drip flame, the amount of foaming (large / medium / small, etc.), etc. The way the flame spreads can be, for example, "heat is transferred by the burnt resin" or "the flame wraps around". The size of the flame may be expressed numerically as the length, or as a ratio to the length of the test piece of the composite material.

[0049] Returning to Figure 7, we will continue explaining the differences from the first embodiment. The importance calculation unit 114 calculates the importance (influence) that each explanatory variable (feature) of the prediction model 121 has on the prediction result. The decision tree used to calculate this importance will be explained below. Figure 9 shows a decision tree 310. Decision trees are generally used for regression and classification, but below we will explain classification (where the dependent variable is the classification result (category)) as an example.

[0050] Nodes 311-317 represent sets of items to be classified. Node 311, the root node, represents all items to be classified. Items included in node 311 are classified into the set of items shown in node 312 if the explanatory variable X is less than 3, and into the set of items shown in node 313 if the explanatory variable X is 3 or greater. Node 313 is the terminal node in decision tree 310, indicating that classification is complete.

[0051] The items to be classified in node 312 are classified into the set of items indicated by node 314 if the explanatory variable Y is less than 7.3, and into the set of items indicated by node 315 if the explanatory variable Y is 7.3 or greater. Node 314 is a terminal node in decision tree 310, indicating that classification has finished.

[0052] The items to be classified in node 315 are classified into the set of items indicated by node 316 if explanatory variable Z is 3, and into the set of items indicated by node 317 if explanatory variable Z is not 3. Nodes 316 and 317 are terminal nodes in decision tree 310, indicating that classification has finished.

[0053] As shown above, the decision tree 310 classifies the set of items to be classified, indicated by node 311, into four sets (categories), indicated by nodes 313, 314, 316, and 317. Note that although the items to be classified included in nodes 313, 314, 316, and 317 are all different, any of the nodes may be the same in terms of the classification result (category). For example, suppose the item to be classified is a composite material, and the classification is a determination of whether or not it is flame-retardant. Then the number of categories is 2, and nodes 313, 314, 316, and 317 are either conforming nodes or non-conforming nodes.

[0054] When constructing a decision tree 310, or in other words, when dividing a higher-level node (a collection of items to be classified) into two lower-level nodes, the explanatory variables and their values ​​are selected as the dividing condition (classification condition) such that the sum of the Gini coefficients of the lower-level nodes is minimized. The Gini coefficient indicates the degree of mixing (impurity) of the categories (target variables) to be classified that are included in a node. For example, if the categories to be classified that are included in a node are all the same, the Gini coefficient will be 0. Nodes with a Gini coefficient of 0 do not require classification and become terminal nodes of the decision tree 310. The depth of the decision tree (number of branches) can be determined arbitrarily, and the Gini coefficient of the terminal nodes does not necessarily have to be 0. The depth of the decision tree is determined by considering, for example, the number of terminal nodes, the number of data points included in each node, and the error rate.

[0055] Figure 10 shows a decision tree 330 for explaining importance according to the second embodiment. The decision tree 330 is a predictive model 121 where the explanatory variable is combustion information 134 and the objective variable is flame retardancy 133, which is whether or not it meets the V-2 grade standard. Nodes 331 to 335 contain the number of composite materials (number of data points) included in the node, the number of composite materials that meet or do not meet the V-2 grade standard (V2 compliant / V2 non-compliant), and the Gini coefficient (Gini impurity).

[0056] In Figure 10, the Gini coefficient indicates the degree of mixing (impurity) of flame retardancy suitability of the composite materials contained in the node. If the composite materials contained in the node are either all suitable or all unsuitable, the Gini coefficient (impurity) will be 0. The lower the Gini coefficient, the higher the degree of suitability / unsuitability, resulting in a more desirable (higher accuracy) judgment result. During the decision tree construction process (learning), the branching conditions (explanatory variables and their ranges for dividing the data from higher-level nodes into data from lower-level nodes) are determined so that the sum of the Gini coefficients of the lower-level nodes is minimized.

[0057] The importance calculation unit 114 calculates importance using the Gini coefficient when a decision tree is used as the machine learning model. For example, the importance is the difference between the Gini coefficient of the top node and the sum of the Gini coefficients of the lower nodes. The importance calculation unit 114 may also calculate importance by referring to metrics used in Boruta, Lasso, etc., in addition to decision trees. The learning unit 112A generates a predictive model 121 using explanatory variables with high importance.

[0058] The explanatory variables for which the importance calculation unit 114 calculates importance may be only the combustion information 134, or they may be both the material formulation information 132 and the combustion information 134. Furthermore, if an explanatory variable with high importance is included in the combustion information 134, the learning unit 112A may generate a predictive model using this explanatory variable and the material formulation information 132 as explanatory variables.

[0059] ≪Second Embodiment: Learning Process≫ Figure 11 is a flowchart of the learning process according to the second embodiment. Below, we will explain the process by which the importance calculation unit 114 calculates the importance of the explanatory variables, which are the combustion information 134, and the learning unit 112A generates a predictive model 121 using the explanatory variables of the combustion information 134 and the material formulation information 132, which have high importance, as explanatory variables.

[0060] In step S31, the information acquisition unit 111 acquires material formulation information of the composite material, combustion information of the composite material, and flame retardancy, and stores them in the material information database 130A. In step S32, the importance calculation unit 114 calculates the importance of the explanatory variables (attributes) included in the combustion information 134 (see Figure 8). In step S33, the learning unit 112A selects explanatory variables of high importance. In the repeated processing from steps S33 to S36, the learning unit 112A selects explanatory variables by adding them one by one in order of importance, for example, starting with the explanatory variables of high importance. In step S34, the learning unit 112A trains a machine learning model using training data in which the explanatory variables (attributes) selected in step S33 and the material formulation information 132 are explanatory variables, and generates a predictive model 121. The training data is obtained from the material information database 130A.

[0061] In step S35, the learning unit 112A evaluates the performance of the prediction model 121 using data from the material information database 130A that was not used in step S34. In step S36, the learning unit 112A returns to step S33 if the performance (prediction accuracy) calculated in step S35 is below a predetermined threshold (step S35 → NO), and terminates the learning process if it exceeds the threshold (step S35 → YES).

[0062] ≪Second Embodiment: Performance Evaluation≫ As part of the performance evaluation of the flame retardancy prediction device 100A (prediction model 121), after selecting explanatory variables related to combustion that were of high importance, the prediction model 121 was evaluated using the selected combustion-related information and material formulation information 132 as explanatory variables, with the V-2 grade as the dependent variable.

[0063] ≪Second Embodiment: Performance Evaluation: Selection of High-Importance Explanatory Variables Related to Combustion≫ In selecting explanatory variables related to combustion, 18 composite materials were fabricated, and combustion tests were conducted to obtain information on various combustion-related aspects and the suitability of each material for UL94 V2 grade. A decision tree was created with combustion-related information as the explanatory variable and flame retardancy as the dependent variable, and importance was calculated following Figure 9. The results are shown in Figure 10.

[0064] To divide the composite materials contained in node 331 based on the value of a single explanatory variable, focusing on the number of drips among the explanatory variables and dividing based on whether that value is less than 4 resulted in the minimum sum of the Gini coefficients for the lower nodes 332 and 333. Node 333 consists entirely of non-conforming materials (Gini coefficient is 0), so no division is necessary.

[0065] For node 332, dividing the node based on whether the drip viscosity was high or low resulted in the minimum sum of the Gini coefficients for the lower nodes 334 and 335. Based on these results, the number of drips and the viscosity of the drip were selected as important explanatory variables in this embodiment.

[0066] ≪Second Embodiment: Performance Evaluation: Evaluation Results≫ In the evaluation experiment, 38 composite materials were fabricated, and combustion tests were conducted to obtain the number of drips and whether they met the UL94 V2 grade. The material formulation information and flame retardancy were the same as in the first evaluation experiment in the first embodiment. A decision tree prediction model 121 was generated using 27 of the 38 materials as training data. Subsequently, the prediction accuracy was evaluated using 11 validation data that were not used as training data.

[0067] Figure 12 is a confusion matrix showing the relationship between the actual suitability of the UL94 V-2 grade composite material according to the second embodiment and the suitability predicted by the prediction model. The accuracy rate is 81.8%, which is an even better prediction accuracy compared to the first embodiment.

[0068] Features of the second embodiment The flame retardancy prediction device 100A predicts whether a composite material meets flame retardancy standards based on material formulation information and combustion information. Combustion information includes the time from flame contact to ignition, the way the flame spreads, and the color of the flame. By selecting explanatory variables with high importance, it becomes possible to predict flame retardancy with high accuracy using fewer explanatory variables. In the experiment conducted this time, it was confirmed that the number of drips was of high importance and that the prediction model was more accurate than the prediction model that used only material formulation information as an explanatory variable.

[0069] <<Variation: Prediction from combustion information>> In the above-described embodiment, the explanatory variables of the prediction model include material formulation information. The flame retardancy prediction device may also predict flame retardancy using a prediction model that does not include material formulation information 132 and uses only combustion information 134 as an explanatory variable.

[0070] ≪Variation: Multi-stage predictive model≫ The explanatory variables of the prediction model in the second embodiment described above are explanatory variables related to combustion information of high importance, and material formulation information. The flame retardancy prediction device may predict combustion information from material formulation information and predict flame retardancy from the predicted combustion information. Specifically, the flame retardancy prediction device may predict flame retardancy from material formulation information using two prediction models: a first-stage prediction model that predicts combustion information from material formulation information and a second-stage prediction model that predicts flame retardancy from combustion information.

[0071] For the combustion information, which is the dependent variable of the first-stage prediction model and the explanatory variable of the second-stage prediction model, combustion information that is highly important (influenced) on the predicted flame retardancy may be used. Examples of highly important combustion information include the number of drips and drip viscosity (see Figure 10). The first stage of the prediction model can be generated using training data in which material formulation information 132 from the material information database 130A (see Figure 8) is used as the explanatory variable and combustion information 134 is used as the objective variable. The second stage of the prediction model can be generated using training data in which combustion information 134 is used as the explanatory variable and flame retardancy 133 is used as the objective variable. By using such a flame retardancy prediction device, it is possible to predict flame retardancy from material formulation information without having to fabricate composite materials and conduct combustion experiments to obtain the number of drips or drip viscosity.

[0072] ≪Variation: Material Formula Information≫ The material formulation information in the above-described embodiment refers to the ratio of resins and additives that make up the composite material, but may also include information relating to other materials. For example, the material formulation information may include information on the type of resin, the type and amount of additives, and the material structure. For resins, the material formulation information may include weight-average molecular weight and number-average molecular weight, molecular weight distribution, degree of copolymerization, degree of crosslinking, chemical structure, and various other physical properties. The material formulation information for additives may also include size, structure, and various other physical properties. Additives are compounds other than polymer resins that affect the properties of the polymer composite material to be manufactured, such as fillers, plasticizers, colorants, flame retardants, UV absorbers, antioxidants, and elastomers. The material formulation information may also include information regarding the molding process of the composite material (molding conditions), such as kneading process conditions and molding conditions.

[0073] <<Modification: Prediction of other characteristics>> The flame retardancy prediction system described above predicts information related to flame retardancy, but it may also predict other properties (physical properties) at the same time. By using a prediction model generated from training data that includes not only flame retardancy but also properties of composite materials other than flame retardancy as the target variable, it is possible to predict other properties simultaneously with flame retardancy. For example, by outputting the toughness and stiffness of composite resins simultaneously with flame retardancy, it can be used to predict formulations that resolve the trade-off between these properties and flame retardancy.

[0074] <<Other variations>> Although several embodiments of the present invention have been described above, these embodiments are merely illustrative and do not limit the technical scope of the present invention. The flame retardancy prediction device 100A (see Figure 7) according to the second embodiment comprises a learning unit 112A, a prediction unit 113, and an importance calculation unit 114, but each of these may be provided by a separate device. For example, a flame retardancy prediction model generation device may comprise the learning unit, a feature extraction device may comprise the importance calculation unit, and a flame retardancy prediction device may comprise the prediction unit. In this case, the flame retardancy prediction model generation device generates a prediction model by referring to the importance of explanatory variables calculated by the feature extraction device, and the flame retardancy prediction device predicts flame retardancy using this prediction model.

[0075] The present invention can take on various other embodiments, and furthermore, various modifications such as omissions and substitutions can be made without departing from the spirit of the invention. These embodiments and their variations are included in the scope and spirit of the invention as described herein, and are also included in the scope of the invention and its equivalents as described in the claims.

[0076] Regarding resins and additives: In the embodiments described above, polypropylene was given as an example of a resin, but other resins and additives were not mentioned. Below, resins and additives, including fillers and flame retardants, will be described.

[0077] ≪Resin≫ The resin molded articles (composite materials) according to the first and second embodiments are thermoplastic resins and are the main materials constituting the resin molded articles. There may be one or more types of thermoplastic resins. Examples of such thermoplastic resins include polyolefins, polycarbonates, polybutylene terephthalate, polyethylene terephthalate, polyphenylene sulfite, polyamide-imide, polyetheretherketone, polyethersulfone, polyimide, polyvinyl chloride resins, styrene resins, polyamides, polyacetal resins, acrylic resins, cellulose resins, and thermoplastic elastomers.

[0078] Examples of polyolefins include polyethylene, ethylene-vinyl acetate copolymer, cycloolefin, and polypropylene. Examples of styrene-based resins include polystyrene, syndiotactic polystyrene, acrylonitrile-styrene copolymer, and acrylonitrile-butadiene-styrene copolymer (ABS resin). Examples of cellulose-based resins include cellulose acetate. Examples of thermoplastic elastomers include polystyrene-based thermoplastic elastomers, polyolefin-based thermoplastic elastomers, polyurethane-based thermoplastic elastomers, 1,2-polybutadiene-based thermoplastic elastomers, ethylene-vinyl acetate copolymer-based thermoplastic elastomers, fluororubber-based thermoplastic elastomers, and chlorinated polyethylene-based thermoplastic elastomers.

[0079] <Fillers, Flame Retardants> The filler can be used without particular limitations, such as fibrous fillers, needle-shaped fillers, or plate-shaped fillers. Examples of fibrous fillers include various inorganic fibers such as glass fibers, various organic fibers such as aramid fibers, and carbon fibers. Examples of needle-shaped fillers include whiskers such as potassium titanate and mineral-based needle-shaped fillers such as wollolastonite. Examples of plate-shaped fillers include mica and talc. Of these, glass fibers, aramid fibers, and talc are preferred from the viewpoint of improving the rigidity of the foamed molded article.

[0080] The flame retardant may be an organic flame retardant or an inorganic flame retardant. Organic flame retardants include, for example, phosphorus compounds and halogen compounds. Inorganic flame retardants include, for example, antimony compounds and metal hydroxides. Phosphorus compounds readily impart high flame retardancy to resin compositions and are environmentally toxic. Phosphorus compounds are typically phosphate ester compounds. Examples of phosphate ester compounds include phosphite esters, phosphate esters, and phosphonic acid esters. Phosphate esters are particularly preferred.

[0081] <<Plasticizers, UV absorbers, antioxidants>> Examples of plasticizers include aromatic carboxylic acid esters (such as dibutyl phthalate), aliphatic carboxylic acid esters (such as methylacetyl ricinolate), aliphatic dialbonate esters (such as adipic acid-propylene glycol polyesters), aliphatic tricarboxylic acid esters (such as triethyl citrate), phosphate triesters (such as triphenyl phosphate), epoxy fatty acid esters (such as epoxybutyl stearate), and petroleum resins. Examples of antioxidants include hindered phenol-based, sulfur-containing organic compound-based, and phosphorus-containing organic compound-based antioxidants. Examples of UV absorbers include benzotriazole-based, benzophenone-based, and salicylate-based UV absorbers.

[0082] <<Other ingredients>> Other components that can be used include reinforcing agents, dispersants, pigments and colorants, fillers, crystallization accelerators, clarifiers, anti-bubble agents, flame retardant additives, antistatic agents, processing aids, lubricants, organic peroxides, plasticizers, and photocatalysts. These other components may be used individually or in combination of two or more. Reinforcement materials such as carbon black, silica, glass fibers, carbon fibers, cellulose fibers, and aramid fibers can be used. As dispersants, waxes, modified waxes, metal soaps, low molecular weight polyethylene, low molecular weight polypropylene, etc., can be used.

[0083] As for the wax, paraffin wax, polyolefin wax, etc., can be used. As modified waxes, for example, waxes having acidic groups in their molecular structure can be used. Specific examples include maleic acid modified wax, fumaric acid modified wax, acrylic acid modified wax, methacrylic acid modified wax, and crotonic acid modified wax. As metal soaps, sodium salts, calcium salts, zinc salts, magnesium salts, and aluminum salts of fatty acids such as stearic acid, hydroxystearic acid, behenic acid, montanic acid, lauric acid, and sebacic acid can be used.

[0084] Inorganic pigments and colorants can be used. Specific examples of inorganic pigments include red iron oxide, titanium dioxide, cadmium red, cadmium yellow, zinc oxide, ultramarine, cobalt blue, calcium carbonate, titanium yellow, lead white, red lead, lead yellow, and Prussian blue. Specific examples of organic pigments include quinacridone, polyazo yellow, anthraquinone yellow, polyazo red, azo lake yellow, perylene, phthalocyanine blue, phthalocyanine green, and isoindolinone yellow. [Explanation of symbols]

[0085] 100 Flame retardancy prediction device (flame retardancy prediction model generation device) 100A Flame Retardancy Prediction Device (Flame Retardancy Prediction Model Generator, Feature Extraction Device) 111 Information Acquisition Department 112,112A Learning Department 113 Prediction Section 114 Importance calculation part 121 Predictive Model (Flame Retardancy Predictive Model) 130,130A Material Information Database 132 Material Formulation Information 133 Flame retardant 134 Combustion Information

Claims

1. An information acquisition unit that accepts input of material formulation information related to polymer composite materials, The system includes a prediction unit that predicts information related to the flame retardancy of a polymer composite material from input material formulation information of the polymer composite material, using a flame retardancy prediction model that predicts information related to the flame retardancy of the polymer composite material. The flame retardancy prediction model described above is: A first-stage predictive model that predicts information related to the combustion of the polymer composite material from material formulation information related to the polymer composite material, and This consists of a second-stage predictive model that predicts information regarding the flame retardancy of the polymer composite material from information regarding the combustion of the polymer composite material. Flame retardancy prediction device.

2. The material formulation information includes at least one of the following: the type of resin, the type of additive, the ratio of the resin to the additive, the structure of the resin, the structure of the additive, and the molding conditions of the polymer composite material. The flame retardancy prediction device according to feature 1.

3. The structure of the resin includes at least one of the following: the chemical structure of the resin, the weight-average molecular weight, the number-average molecular weight, the molecular weight distribution, the degree of copolymerization, and the degree of crosslinking. The flame retardancy prediction device according to feature 2.

4. The additive includes at least one of fillers, plasticizers, colorants, flame retardants, ultraviolet absorbers, antioxidants, and elastomers. The flame retardancy prediction device according to feature 2.

5. The molding conditions include at least one of the kneading process conditions and molding conditions. The flame retardancy prediction device according to feature 2.

6. The aforementioned information relating to flame retardancy is whether or not it meets the standards related to flame retardancy. The flame retardancy prediction device according to feature 1.

7. The information relating to combustion is obtained from the changes over time during the combustion of the polymer composite material. The flame retardancy prediction device according to feature 1.

8. The information relating to the combustion of the polymer composite material, which serves as input to the second-stage prediction model, is selected according to its importance to the information relating to flame retardancy, which serves as output to the second-stage prediction model. The flame retardancy prediction device according to feature 1.

9. A first-stage predictive model that predicts information related to the combustion of a polymer composite material from input material formulation information, using training data that associates material formulation information related to the polymer composite material with information related to the combustion of the polymer composite material. A second-stage predictive model that predicts the flame retardancy information from the combustion information of the polymer composite material using training data that associates the combustion information of the polymer composite material with the flame retardancy information of the polymer composite material, and The system includes a learning unit that generates a flame retardancy prediction model composed of the first-stage prediction model and the second-stage prediction model. Flame retardancy prediction model generation device.

10. The information relating to combustion is obtained from the changes over time during the combustion of the polymer composite material. The flame retardancy prediction model generation apparatus according to feature 9.

11. The system further includes an importance calculation unit that calculates the importance of the information relating to combustion to the information relating to flame retardancy, which is the output of the second-stage prediction model. The learning unit selects the combustion-related information, which serves as input to the second-stage prediction model, according to its importance, and generates the second-stage prediction model. The flame retardancy prediction model generation apparatus according to feature 9.

Images