Prediction model

Abstract

To provide a prediction model that can predict purification performance of an exhaust gas purification catalyst, with high accuracy.SOLUTION: A prediction model according to the present invention predicts purification performance of an exhaust gas purification catalyst. The exhaust gas purification catalyst comprises a honeycomb base material having a bulkhead defining a plurality of cells, and a catalyst layer having a plurality of layers laminated on surfaces at the cell sides of the bulkhead. The prediction model is constructed by performing machine learning using information having information expressing a catalyst constitution as an explanatory variable and information expressing the purification performance as an objective variable. The information expressing the catalyst constitution includes data expressing coat states for every position of a plurality of portions of layers of the catalyst layer represented by a plurality of blocks.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The entire contents of Japanese Patent Application No. 2023-007974 filed with the Japan Patent Office on January 23, 2023, and the entire contents of Japanese Patent Application No. 2023-081090 filed with the Japan Patent Office on May 16, 2023, are incorporated by reference in the disclosure of this application. The present invention relates to a prediction model for predicting the performance of an exhaust gas purification catalyst. [Background technology]

[0002] Various prediction models for predicting the purification performance of exhaust gas purification catalysts have been known. Specifically, for example, Patent Document 1 describes a prediction model that is trained in advance using teacher data of the ratios of elements contained on the surfaces of multiple types of catalysts and the specific surface area values ​​on the catalyst surfaces. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2022-185940 Summary of the Invention [Problem to be solved by the invention]

[0004] The purification performance of an exhaust gas purification catalyst is determined by, for example, the catalyst configuration, the test conditions of the purification performance evaluation test, the durability history of the catalyst, etc. When the catalyst layer of an exhaust gas purification catalyst has multiple layers stacked on a substrate, a prediction method for predicting the purification performance of the exhaust gas purification catalyst using machine learning may use three pieces of information representing the catalyst configuration used as explanatory variables for machine learning: information on the support powder contained in each layer of the catalyst layer (the support powder constituting the slurry used to form each layer of the catalyst layer), information on the coating of each layer of the catalyst layer, and information on the honeycomb substrate. In this case, the coating information of each layer of the catalyst layer is data representing the stacking order of each layer of the catalyst layer as well as the width and position of the coating. Conventionally, when expressing this, the multiple layers of the catalyst layer are identified in stacking order as first layer, second layer, third layer, etc., and then the width and position of the coating of each layer of the first layer, second layer, third layer, etc. are expressed as data. However, with this representation method, for example, when predicting the catalytic performance of a prediction sample exhaust gas purification catalyst in which the stacking order of two adjacent layers in the catalyst layer of the learning sample exhaust gas purification catalyst (the stacking order of support powder C and support powder D) is reversed, as shown in Figure 2(a), the actual coating state of each layer of the catalyst layer may not be reflected in the coating information of each layer of the catalyst layer in the explanatory variables sufficiently and appropriately. Specifically, in this case, depending on the width and position of the coating of the two layers, as shown in Figure 2(a), even though the actual coating state of each layer of the catalyst layer of the prediction sample is almost the same as that of the learning sample, information that the actual coating state of each layer of the catalyst layer is different may be erroneously reflected in the coating information of each layer of the catalyst layer in the explanatory variables due to the influence of over-learning regarding the stacking order, resulting in a predicted value that is far from the actually measured value.

[0005] The present invention has been made in view of the above points, and an object of the present invention is to provide a prediction model that can predict the purification performance of an exhaust gas purification catalyst with high accuracy. [Means for solving the problem]

[0006] In order to solve the above-mentioned problems, the prediction model of the present invention is a prediction model that predicts the purification performance of an exhaust gas purification catalyst, wherein the exhaust gas purification catalyst comprises a honeycomb substrate having partition walls that define a plurality of cells, and a catalyst layer having a plurality of layers laminated on the cell-side surface of the partition walls, and the prediction model is constructed by performing machine learning using information that has information that represents the catalyst configuration as an explanatory variable and information that represents the purification performance as a target variable, and is characterized in that the information that represents the catalyst configuration includes data that represent the coating state at each of a plurality of positions in each layer of the catalyst layer, each layer being represented by a plurality of blocks. [Effects of the Invention]

[0007] According to the present invention, the purification performance of an exhaust gas purification catalyst can be predicted with high accuracy. [Brief explanation of the drawings]

[0008] [Figure 1] (a) is a schematic cross-sectional view showing an enlarged view of the partition walls and catalyst layer of the substrate in each catalyst sample. (b) is a diagram for explaining a method for obtaining information representing durability history. (c) is a diagram for explaining a method for obtaining the NOx 50% conversion time. (d) is a diagram for explaining a method for obtaining the NOx 50% conversion temperature. [Figure 2] 1A is a diagram illustrating the problem to be solved by the present invention, and FIGS. 1B and 1C are diagrams illustrating the main parts of a method for acquiring information representing a catalyst configuration used as an explanatory variable in machine learning according to a comparative example and an example, respectively. [Figure 3] The left graph in (a) shows the relationship between the predicted values ​​of the NOx 50% conversion time calculated by the prediction models of the comparative example and the example and the actual measured values, while the right graph in (a) shows the average error for those comparative example and the example. The left graph in (b) shows the relationship between the predicted values ​​of the NOx 50% conversion temperature calculated by the prediction models of the comparative example and the example and the actual measured values, while the right graph in (b) shows the average error for those predicted values ​​and the actual measured values. DETAILED DESCRIPTION OF THE INVENTION

[0009] The prediction model according to the embodiment is a prediction model used in place of the prediction model in a prediction device according to an embodiment of Patent Application No. 2023-007974, and predicts the purification performance of an exhaust gas purification catalyst (for example, the catalyst shown in Figure 1(a) described below) similar to the prediction model in the prediction device according to an embodiment of Patent Application No. 2023-007974.

[0010] The data representing the coating state at each of the multiple positions in each layer of the catalyst layer, which is included in the information representing the catalyst configuration used as an explanatory variable in the machine learning performed to construct the predictive model of the embodiment, is not particularly limited as long as the data representing the coating state at each of the multiple positions in each layer is represented by multiple blocks.For example, it may be intensity data at each diffraction angle of the composite XRD spectrum of each layer (support powder) at each of the multiple positions in each layer, or it may be multiple latent variables at each of the multiple positions in each layer obtained by dimensionally compressing the data, or it may be a predetermined number of latent variables obtained by further dimensionally compressing the latent variables. [Example]

[0011] The prediction model according to the embodiment will be described in more detail below with reference to comparative examples and examples.

[0012] [Comparative Example] An example of a prediction model for a comparative example was created.

[0013] 1. Obtaining a data set of explanatory variables and target variables to be used in building a predictive model (1) Preparation of exhaust gas purification catalyst samples First, 800 samples of exhaust gas purification catalysts (catalysts) for study were prepared in the same manner as in the examples of Japanese Patent Application No. 2023-007974. As shown in FIG. 1(a), each sample CS was a straight-flow type catalyst and included a honeycomb substrate (substrate) 10 integrally formed with a frame (not shown) and partition walls 14 that partition the inner space into a honeycomb shape to define multiple cells 12, and a catalyst layer 20 including at least a first layer and a second layer among first layers 20a to fifth layers 20e stacked on the cell-12-side surface 14c of the partition walls 14. Each layer of the catalyst layer 20 contained a mixed powder containing powders P1 to Px (x is a predetermined integer of 2 or greater) as a support powder, and a catalytic metal supported on at least one of the powders P1 to Px. The catalytic metal was at least one of catalytic metals M1 to My (y is a predetermined integer of 2 or greater).

[0014] (2) Obtaining explanatory variables and target variables for each catalyst sample Next, from each catalyst sample, a group consisting of information representing the catalyst configuration, information representing the durability history, and information representing the purification performance evaluation test was obtained as explanatory variables, and information representing the purification performance was obtained as a target variable. In obtaining the information representing the catalyst configuration, first, from each catalyst sample, information before dimensionality reduction regarding each layer of the catalyst layer and information regarding the substrate were obtained as information before dimensionality reduction. As shown in Table 1 below, the information before dimension reduction for each layer of the catalyst layer was obtained by synthesizing the XRD spectra of powders P1 to Px at the composition ratio ratio. N data points were obtained for the intensity [au] at each diffraction angle (N (4250) diffraction angles at 0.02° intervals in the range of 2θ = 5° to 90°) of the composite XRD spectrum of each layer, as in the example of Japanese Patent Application No. 2023-007974. The amounts [g] of each type of P1 to Px, the amounts [g] of each type of catalytic metal M1 to My supported on each type of P1 to Px, and coating information (data regarding the coating area of ​​each layer, data representing the width and position of the coating of each layer). Information regarding the substrate was obtained, including weight, cell shape, and wall thickness.

[0015] [Table 1]

[0016] Next, as illustrated in Figure 2(b) for a catalyst layer having first to fourth layers of one catalyst sample, the information representing the catalyst composition of each catalyst sample was obtained by performing dimensionality compression on N pieces of data representing the intensities at each diffraction angle of the composite XRD spectrum of each layer for the pre-dimensionality reduction information of each catalyst sample. To do this, as illustrated in Figure 2(b), first, for each layer of the catalyst layer, N pieces of data representing the intensities at each diffraction angle of the composite XRD spectrum of each layer were prepared, each represented by N blocks of a block assembly with N blocks arranged in the depth direction. Next, an autoencoder was used to perform dimensionality compression on the N pieces of data representing the intensities at each diffraction angle of the composite XRD spectrum of each layer, obtaining four latent variables for each layer, each represented by four blocks. This resulted in four latent variables for the multiple layers of the catalyst (layers 1 to 4 in the example of Figure 2(b)). Next, by combining the four latent variables of each layer with information other than the N data points of the intensity at each diffraction angle of the synthetic XRD spectrum of each layer in the information before dimensionality reduction, information expressing the catalyst composition of each catalyst sample (information after dimensionality reduction) was obtained.

[0017] To obtain information representing the durability history, first, a durability test was performed on each catalyst sample in the same manner as in the examples of Japanese Patent Application No. 2023-007974. Next, as shown in FIG. 1(b), the catalyst bed temperature (endurance temperature) and the air-fuel ratio (A / F sensor value) containing the catalyst were measured throughout each durability test and saved as data for each durability test. Next, as in the examples of Japanese Patent Application No. 2023-007974, durability history 2D histogram data (30 pixel x 30 pixel image data where pixel values ​​represent the frequency of combinations of air-fuel ratio and catalyst bed temperature) representing the durability history of each catalyst sample was created from the data of each durability test. Next, the durability history 2D histogram data was subjected to dimensional compression using a convolutional autoencoder in the same manner as in Japanese Patent Application No. 2023-081090, thereby obtaining 16-dimensional latent variable data representing the durability history as information representing the durability history. To obtain information representing the purification performance evaluation test (warm-up characteristics test and temperature characteristics test) and information representing purification performance (warm-up characteristics and temperature characteristics), a warm-up characteristics test was first conducted on each catalyst sample degraded in each durability test. For the warm-up characteristics test, a degraded catalyst was installed in the main exhaust system of a 2000cc gasoline engine, and a bypass line was attached to the engine side of the catalyst. The engine output was maintained constant, and exhaust gas was first diverted into the bypass line to allow the catalyst to reach room temperature. Next, the exhaust gas flow was switched to the main line. Test condition data (exhaust gas flow rate, catalyst inlet temperature, catalyst inlet CO concentration, catalyst inlet NOx concentration, catalyst inlet HC concentration, and catalyst inlet air-fuel ratio (A / F sensor value) per second) were obtained as information representing the warm-up characteristics test. Second-by-second data on the catalyst inlet temperature, catalyst inlet NOx concentration, and catalyst outlet NOx concentration were also obtained as information representing the warm-up characteristics. From this information, the time when the NOx purification amount (NOx concentration at the catalyst entrance - NOx concentration at the catalyst exit) reaches half of the NOx concentration at the catalyst entrance can be calculated as the NOx 50% purification time, as shown in Figure 1(c).In addition, a purification performance evaluation test (temperature characteristic test) was conducted for each sample of catalyst deteriorated in each durability test, similar to the example of Japanese Patent Application No. 2023-007974, and test condition data was obtained as information representing the temperature characteristic test. As information representing the temperature characteristic, data was obtained every second of the temperature at the inlet of the catalyst, the NOx concentration at the inlet of the catalyst, and the NOx concentration at the outlet of the catalyst during the temperature characteristic test. From this information, the NOx 50% purification temperature can be calculated, as shown in Figure 1(d).

[0018] (3) Obtaining a data set of explanatory variables and target variables For the 800 samples of exhaust gas purification catalysts, by obtaining the explanatory variables and response variables for each sample of catalyst as described above, a data set consisting of 800 sets of data for the explanatory variables and response variables was obtained.

[0019] 2. Building a predictive model Next, a predictive model was constructed by machine learning using the datasets of explanatory variables and objective variables. First, the dataset was randomly divided into three: a training dataset, a test dataset, and a validation dataset. Next, a multilayer perceptron model with two hidden layers was constructed. Random values ​​were assigned to the hyperparameters (the number of dimensions of the latent variables in the first hidden layer, the number of dimensions of the latent variables in the second hidden layer, the learning rate, dropout, and epsilon) to construct the model. Next, the multilayer perceptron model was trained on the training dataset while verifying the accuracy of the model on the test dataset. The model was trained to minimize the root mean squared error (RMSE) for both the training and test datasets. The iteration count was 1000. Next, predictions were made on the validation dataset using the trained model, and the RMSE was calculated relative to the actual measurements. This series of steps—model construction, training, and calculation of the RMSE for the validation dataset using the trained model—was repeated 100 times, changing the values ​​assigned to the hyperparameters. This resulted in a dataset consisting of 100 sets of hyperparameters (explanatory variables) and RMSE (objective variable). Next, the hyperparameter values ​​that minimized the RMSE were selected from the dataset. Next, a multilayer perceptron model was constructed with the selected values ​​assigned to the hyperparameters, and this was used as the prediction model.

[0020] [Example] An example of a prediction model according to the embodiment was created.

[0021] First, for each sample of the catalyst similar to that in the comparative example, a data set of explanatory variables and objective variables similar to those in the comparative example was obtained, except for the information representing the catalyst configuration. To obtain the information representing the catalyst configuration, first, the same pre-dimensionality reduced information as in the comparative example was obtained, except that the pre-dimensionality reduced information for each layer of the catalyst layer did not include coating information. Next, as illustrated in FIG. 2(c) for a catalyst layer having layers 1 to 4 of one catalyst sample, the pre-dimensionality reduced information for each catalyst sample was subjected to dimensionality reduction on N pieces of data representing the coating state at 20 positions in the width direction (cell extension direction) of each layer, thereby obtaining information representing the catalyst configuration. In this case, as illustrated in FIG. 2(c), first, for each layer of the catalyst layer, N pieces of data representing the coating state at 20 positions in the width direction of each layer (data representing the coating state at multiple positions in each layer) were prepared, each represented by a 20 × N block assembly in which 20 blocks are arranged in the width direction and N blocks are arranged in the depth direction for each width direction block. In the block assembly, 20 blocks aligned in the width direction represent the presence or absence of layer coating at 20 positions in the width direction (filled blocks: present, empty blocks: absent), and N blocks aligned in the depth direction represent N pieces of data representing the intensity at each diffraction angle of the synthetic XRD spectrum of each layer. Of the N pieces of data representing the coating state at each of the 20 positions in the width direction of each layer, the N pieces of data representing the layer coating at each of the 20 positions in the width direction of each layer are N pieces of data representing the intensity at each diffraction angle of the synthetic XRD spectrum of each layer, while the N pieces of data representing the layer non-coated at each of the 20 positions in the width direction of each layer are all set to the value "0." Next, an autoencoder was used to perform dimensionality compression on the N pieces of data representing the coating state at each of the 20 positions in the width direction of each layer, thereby obtaining four latent variables (data representing the coating state at multiple positions in each layer) for each of the 20 positions in the width direction of each layer.Next, a convolutional autoencoder was used to reduce the dimension of the four latent variables for each of the 20 positions in the width direction of each layer and convert them into features, thereby extracting a predetermined number of latent variables (eight latent variables for each of Filter 1 and Filter 2 in the example of Figure 2(c)) (data representing the coating state at each of the multiple positions in each layer). Note that these predetermined number of latent variables reflect the data content representing the coating width and position of each layer. Next, by combining these predetermined number of latent variables with information other than the N pieces of data on the intensity at each diffraction angle of the composite XRD spectrum of each layer in the information before dimensionality reduction, information representing the catalyst composition of each catalyst sample (information after dimensionality reduction) was obtained.

[0022] Next, a prediction model was constructed by performing machine learning using a data set of explanatory variables and response variables in the same manner as in the comparative example.

[0023] [Evaluation of prediction accuracy] For four samples for prediction of catalysts not used in constructing the prediction model, explanatory variables were obtained in the same manner as in the comparative example. Then, using the prediction model of the comparative example, information (objective variables) representing purification performance (warm air characteristics and temperature characteristics) was obtained from the obtained explanatory variables. Then, the NOx 50% purification time (predicted value) was calculated from the information representing the warm air characteristics, and the NOx 50% purification temperature (predicted value) was calculated from the information representing the temperature characteristics. Furthermore, for four similar samples for prediction, explanatory variables were obtained in the same manner as in the example. Then, using the prediction model of the example, information (objective variables) representing purification performance (warm air characteristics and temperature characteristics) was obtained from the obtained explanatory variables. Then, the NOx 50% purification time (predicted value) was calculated from the information representing the warm air characteristics, and the NOx 50% purification temperature (predicted value) was calculated from the information representing the temperature characteristics. As shown in FIG. 3(a), the average errors of the NOx 50% purification time (predicted value) and the NOx 50% purification time (measured value) for the four samples were significantly smaller for the example than for the comparative example. 3(b), the average error of the NOx 50% conversion temperature (predicted value) and the NOx 50% conversion temperature (measured value) for the four samples was smaller for the Example than for the Comparative Example. It is considered that the prediction model of the Example suppresses overlearning regarding the stacking order, and is therefore able to predict conversion performance with high accuracy.

Claims

[Claim 1] A prediction model for predicting the purification performance of an exhaust gas purification catalyst, The exhaust gas purification catalyst includes a honeycomb substrate having partition walls that define a plurality of cells, and a catalyst layer having a plurality of layers laminated on the cell-side surfaces of the partition walls, the prediction model is constructed by performing machine learning using information representing a catalyst configuration as an explanatory variable and information representing purification performance as a target variable; A prediction model characterized in that the information representing the catalyst configuration includes data representing the coating state at each of multiple positions in each layer of the catalyst layer, each layer being represented by a plurality of blocks.

Images