Press die analysis device and data analysis method using the same, as well as press die analysis system and press die analysis device used therefor

Abstract

To provide a press die analysis device capable of separating and extracting factors of changes in processing conditions to obtain multifaceted information, a press die analysis method using the press die analysis device, a press die analysis system, and a press die analysis device using the press die analysis system.SOLUTION: A press die analysis device comprises a storage unit 7 that stores compressive loads measured by a plurality of load sensors 5 for each of multiple pressing cycles as load data. The press die analysis device also comprises an analysis unit 8 that analyzes the multiple load data stored in the storage unit 7. The analysis unit 8 performs processing to extract first feature quantities based on the multiple load data measured by the respective load sensors 5 for one pressing cycle. The analysis unit 8 extracts second feature quantities corresponding to change in processing conditions by applying a statistical method or machine learning analysis method to the multiple first feature quantities obtained by repeating the processing for the multiple pressing cycles.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a press die analysis device that enables long-period, wide-range, and complex analysis, a data analysis method using the press die analysis device, a press die analysis system, and the press die analysis device used therein. [Background technology]

[0002] In processing using press molding machines, various factors such as wear on mold parts due to repeated press processing and aging of the equipment can cause changes in molding quality and lead to defective products. Furthermore, if the deterioration of parts progresses and the press mold is damaged, it will require a great deal of cost to repair and readjust the press mold and equipment. BACKGROUND ART Conventionally, a press die analyzing device is known in which a plurality of load sensors (load cells) are arranged on a side surface facing a slide die of a press molding machine (see Patent Document 1, etc.). These load sensors are connected to a data analysis device, which analyzes the waveform data sent from the load sensors to estimate defective events for each part of the mold. The results of the estimation of defective events for each part of the mold by the data analysis device are stored in a data collection device and are used to determine signs of defective events.

[0003] In such a system, the load acting on the die during press working is measured by each load sensor and sent as waveform data to a data analysis device. This allows for early detection of die abnormalities caused by aging and other factors from changes in the load applied during press working, and aims to reduce costs by preventing defects. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 7017775 Summary of the Invention [Problem to be solved by the invention]

[0005] However, the condition of a die can change gradually over a long period of time. Furthermore, changes in processing conditions are influenced by a variety of complex factors (parameters), including temperature increases associated with press processing in addition to aging. For this reason, it is not easy to capture such minute changes in a die from data measured instantaneously before and after several press processes. Therefore, in order to analyze changes in processing conditions during press processing and accurately capture the state of the die, it is necessary to obtain more information.

[0006] The present invention aims to provide a press die analysis device that can separate and extract the causes of complex changes in processing conditions and obtain multifaceted information, thereby preventing product defects and die abnormalities and enabling low-cost, high-quality production, a data analysis method using the press die analysis device, a press die analysis system, and a press die analysis device used therein. [Means for solving the problem]

[0007] The press die analyzing device of the present invention is a press die analyzing device that analyzes changes in processing conditions during press working using a die. The press die analyzing device is equipped with a plurality of load sensors that are arranged on the rear side of the die and measure the compressive load applied to various locations on the die during press working. The press die analyzing device is equipped with a storage unit that stores the compressive loads measured by the plurality of load sensors as load data for each of multiple press workings. The press die analyzing device also includes an analysis unit that analyzes the plurality of load data stored in the storage unit. The analysis unit performs a process to extract first feature values ​​based on the plurality of load data measured by each load sensor for one press working. The analysis unit then applies a statistical method or a machine learning analysis method to the plurality of first feature values ​​obtained by repeating the process for multiple press workings, thereby extracting second feature values ​​corresponding to changes in processing conditions. [Effects of the Invention]

[0008] According to the present invention, there are provided a press die analysis device, a data analysis method using the press die analysis device, a press die analysis system, and a press die analysis device used therein, which can separate and extract the causes of long-period, wide-ranging, and complex changes in processing conditions and obtain multifaceted information, thereby preventing the occurrence of product defects and die abnormalities and enabling low-cost, high-quality production. [Brief explanation of the drawings]

[0009] [Figure 1] 1 is a side view illustrating an overall configuration in which a press die analyzing device according to an embodiment of the present invention is applied to a press machine. [Figure 2] FIG. 2 is a plan view showing an example of load data processed by the press die analyzing device. [Figure 3] FIG. 10 is a diagram schematically illustrating a state in which accumulated load data is converted into point cloud data. [Figure 4] 10 is a graph showing the change in the first principal component separated and extracted by the press die analysis device. [Figure 5] 10 is a graph showing the change in the second principal component separated and extracted by the press die analysis device. [Figure 6] 10 is a graph showing the temperature trend in a press die used for comparison. [Figure 7] 10 is a graph showing an example of eigenvectors obtained in the process of principal component analysis. [Figure 8] FIG. 10 is a plan view visualized by back-projecting the change in the first principal component onto the die using a press die analysis device. [Figure 9] FIG. 10 is a plan view visualized by back-projecting the change in the second principal component onto the die using a press die analysis device. [Figure 10] FIG. 10 is a plan view showing a mask pattern used for filtering. DETAILED DESCRIPTION OF THE INVENTION

[0010] An embodiment of the present invention will be described below with reference to the accompanying drawings. The same components are designated by the same reference numerals, and redundant description will be omitted. The press die analysis device 1 shown in Fig. 1 is provided in a press machine 10 that performs press working on a workpiece 11 using a die 2. The die 2 has an upper die 3 and a lower die 4 that apply a load in the vertical direction to press the placed workpiece 11.

[0011] For this reason, the press machine 10 is provided with a slide 12 that supports the upper die 3 of the mold 2, and a bolster (base) 13 on which the lower die 4 is disposed. The press machine 10 then lowers the upper die 3 together with the slide 12 toward the lower die 4, and presses the workpiece 11 between the upper die 3 and the lower die 4 into a desired workpiece shape 11a.

[0012] The press die analysis device 1 of the present invention and the data analysis method using the same analyze long-term, wide-ranging, and complex changes in processing conditions caused by aging of the die 2 used in such a press processing machine 10, temperature changes, and material changes in the workpiece 11. For this reason, the press die analyzing device 1 of the embodiment is provided with a plurality of load sensors 5 on the back surface 3a side, which is the side opposite to the processing surface of the upper die 3. The press die analyzing device 1 also includes a control unit 6 connected to the load sensors 5 and a monitor 14 connected to the control unit 6.

[0013] Of these, the load sensors 5 in the embodiment are dispersedly disposed at various locations within the same plane between the rear surface 3a of the upper mold 3 and the underside of the slide 12. In the embodiment, 54 load sensors 5 are arranged in a grid pattern (a lattice pattern) in the vertical and horizontal directions at equal intervals. However, the number of load sensors 5 may be more than one, and the arrangement positions of the load sensors 5 are not limited to this.

[0014] Each load sensor 5 measures the compressive load over time during press working, and the analysis unit 8 extracts the maximum load from the measured loads. That is, the load generated from when the slide 12 starts to descend until it reaches near the lowest point is measured. Each load sensor 5 then outputs the measured load data to the control unit 6. The press die analysis device 1 collects the compressive loads measured by each load sensor 5 every time press working is performed by the press machine 10.

[0015] 2 shows a schematic diagram of the load data 20 for each press working collected by the control unit 6. In reality, the maximum load of each load sensor 5 for each press working is collected as a numerical value in the form of a list. For ease of understanding, the collected maximum loads are shown in m sheets in FIG. 2, with one sheet for each press working, but in reality, such a chart is not created. Here, data on the maximum load measured by each of the 54 load sensors 5 in one press working is collected as one piece of load data 20. For example, if press working is performed m times, the load data 20 for each press working is collected as a numerical value, from the first sheet to the mth sheet.

[0016] 2 conceptually shows that the compressive load measured by each load sensor 5 varies depending on the workpiece shape 11a during one press working. Furthermore, the compressive load measured by the load sensor 5 at the same location varies for each press working performed multiple times due to changes in the working conditions.

[0017] The control unit 6 of the press die analyzing device 1 of the embodiment includes an accumulation unit 7, an analysis unit 8, and an image output unit 9, as shown in FIG. The accumulation unit 7 is an internal storage device such as a memory or a hard disk drive. Alternatively, the accumulation unit 7 may be configured as an external storage device such as a storage device connected to the control unit 6. The accumulation unit 7 stores and accumulates the compressive loads measured by the load sensors 5 as load data 20 for each of a plurality of press working operations.

[0018] The analysis unit 8 performs analysis based on the plurality of load data 20 stored in the storage unit 7. In the embodiment, for the sake of explanation, the concept of analysis is expressed as a single point P1, P2 to Pm in a multidimensional space as shown in FIG. 3. Hereinafter, adding a point or a circle to a graph in the figure will be expressed as "plotting." Such plotting is not performed in the analysis process by the analysis unit 8, and a process for creating such a graph is not required.

[0019] In the storage unit 7 of the embodiment, 54 pieces of load data 20 indicating the maximum values ​​measured by the load sensors 5 in one press working operation are stored. The analysis unit 8 analyzes the plurality of load data 20 stored in the storage unit 7 and extracts a first feature amount corresponding to a change in processing conditions.

[0020] For ease of explanation, a conceptual diagram expressing the principal component analysis performed in the embodiment as a point cloud is displayed in the multidimensional space of Fig. 3. Fig. 3 is shown for ease of understanding, and such a diagram is not actually expressed as a point cloud in the actual calculation processing by the analysis unit 8. However, even if such a conceptual diagram is used, it is not possible to plot and represent all of the intersections of the 54 load data 20 measured for each press processing in 54-dimensional space as points P1, P2 to Pm.

[0021] For this reason, in Fig. 3, for ease of explanation, a simplified concept will be explained in three dimensions (X-axis, Y-axis, Z-axis) corresponding to three representative load sensors 5. In Fig. 3, the intersections of the maximum values ​​of the three load sensors 5 are plotted as points P1, P2 to Pm, each of which is a circle, for one press working operation. Here, each of the points P1, P2 to Pm corresponds to the extracted first feature amount.

[0022] For example, when the analysis unit 8 plots the load data of the first press working, which is the first load data 20 in Figure 2, on the 54-dimensional space of Figure 3, it becomes a conceptual point P1. Here, the same process is repeated up to m times, which is the number of press workings. This corresponds to the process of extracting the first feature amount being repeated m times in the actual process by the analysis unit 8. As a result, the load data of the mth press working, which becomes the mth load data 20 in Fig. 2, is conceptually plotted as points P1, P2 to Pm in the three-dimensional space shown in Fig. 3.

[0023] In this manner, the analysis unit 8 of the embodiment performs calculations and processes the plurality of load data 20 stored in the storage unit 7 as variables to conceptually create point cloud data in the multidimensional space of FIG. In the control unit 6 of this embodiment, when analysis is performed by the calculation processing of the analysis unit 8, a large amount of data can be processed using the same concept as plotting the same hundreds to thousands of points P1, P2 to Pm (m = hundreds to thousands) in a 54-dimensional space using load data 20 measured through hundreds to thousands of press workings.

[0024] The distribution of points P1, P2 to Pm plotted in the 54-dimensional space obtained in this way represents the change in processing conditions. However, it is generally known that it is difficult to grasp the factors and characteristics of trends in high-dimensional data. Therefore, the press die analysis device 1 of the embodiment actually uses principal component analysis, which is one of the regression methods, as a statistical method. The principal component analysis performed by the analysis unit 8 is one of the regression methods that aggregates data having many variables to create principal components. Principal component analysis is not so-called supervised learning that predicts something, but unsupervised learning. In principal component analysis, up to 54 principal components can be extracted from 54-dimensional data, and each principal component corresponds to various changes in processing conditions.

[0025] Here, the calculation process of the load data 20 executed by the analysis unit 8 of this embodiment will be described along the steps of principal component analysis. First, the control unit 6 performs data standardization on the load data 20 of each load sensor 5 (setting the average of each variable to 0 and the variance to 1).

[0026] Next, a variance-covariance matrix is ​​calculated from the standardized load data 20. The variance-covariance matrix can be expressed as shown in Equation 1, using multiple sets of load data 20 for each of the multiple original press operations, with the number of press operations and the load measured by the load sensor 5 as a matrix.

[0027] [Formula 1] TIFF2025186972000002.tif49166…Formula 1 Here, V[So] is the variance where o=1 to n, and Cov(Sp,Sq) is the covariance where p=1 to n, q=1 to n, excluding p=q.

[0028] The analysis unit 8 calculates the eigenvalues ​​and the eigenvectors, which are the second feature amounts, from the variance-covariance matrix of Equation 1. The eigenvalues ​​are values ​​that indicate the magnitude of the eigenvectors, and the magnitude indicates the variance (degree of contribution) of the principal components. The number of eigenvalues ​​calculated corresponds to the number of load sensors 5. The eigenvector is a proportional coefficient between the principal component and the original variable. The number of eigenvectors obtained corresponds to the number of load sensors 5. The eigenvectors serve as reference information for grasping the machining condition change region.

[0029] That is, the analysis unit 8 calculates the variance-covariance matrix of Equation 1 for the first feature values ​​of multiple press processing operations, and further calculates the eigenvalues ​​and eigenvectors thereof, thereby extracting the principal components, which are the second feature values ​​corresponding to changes in processing conditions.

[0030] The principal component analysis method of the embodiment can derive new variables (principal components) using all variables. Principal components are elements that represent the characteristics of data. In the embodiment, they are expressed in the form of "first principal component PC1, second principal component PC2, ... nth principal component PCn" in order of decreasing contribution rate. As shown in Figure 2, up to 54 principal components can be extracted from load data 20 measured on hundreds to thousands of sheets through m press processes. However, principal components with low contribution rates are often unnecessary for analysis because their influence on changes in processing conditions is small. Therefore, in some cases, the first principal component PC1, the second principal component PC2, …, the z-th principal component PCz (z < n) may be sequentially extracted from the principal components that are presumed to be highly correlated with the changes in the processing conditions of the mold 2, that is, those with a large contribution rate.

[0031] In FIG. 3 showing the concept, as in the case of a point group plotted in a multi-dimensional space, it is illustrated that data with high correlation gather as elliptical (rugby ball-shaped) point group data. Among these, the major axis direction of the elliptical shape, that is, the eigenvector having the largest eigenvalue becomes the first principal component PC1.

[0032] The analysis unit 8 of the embodiment can calculate the principal component for each shot by calculating the inner product of the load data 20 for each press process (also referred to as a shot) of each load sensor 5 with each eigenvector. Through the above processing, the analysis unit 8 can convert the transition of the load data for each shot obtained by the fifty-four sensors into the transition of the fifty-four principal components.

[0033] The information separated and extracted by the analysis unit 8 is displayed on the monitor 14 connected to the control unit 6. FIG. 4 plots the number of press processes (also referred to as the number of shots) on the horizontal axis and the first principal component PC1 for each shot on the vertical axis. The principal component is extracted for each press process by projecting the load data 20 for each press process onto each eigenvector. Thereby, the user can grasp from the graph displayed on the monitor 14 how the principal component changes as the number of press processes increases. From FIG. 4 of the embodiment, it can be seen that the first principal component PC1 increases as the number of press processes increases, and the increase rate decreases after exceeding a certain number of times.

[0034] In Figure 5, the horizontal axis is the number of press operations and the vertical axis is the second principal component PC2. This shows how much the second principal component PC2 changes with each press operation. By looking at the graph in Figure 5 displayed on the monitor 14, the user can see how the second principal component changes as the number of press operations increases.

[0035] From FIG. 5 of the embodiment, it can be seen that the second principal component PC2 increases sharply once when the number of press workings is small, and then starts to decrease once the number of press workings exceeds a certain number. In this way, the press die analyzing apparatus 1 of the embodiment separates and extracts the first principal component PC1 to the n-th principal component PCn, which show some kind of trend, by the analysis unit 8, and visually displays the changes in each principal component, which can be used as information when the user performs analysis.

[0036] Next, it is estimated what kind of machining condition change the first principal component PC1 or the second principal component PC2 corresponds to. In the control unit 6 of this embodiment, the analysis unit 8 can calculate the proportionality coefficient between the first principal component PC1 and the temperature change to quantify the degree of match. This allows an objective degree of match to be obtained automatically. However, such calculation of the proportionality coefficient does not have to be performed.

[0037] For example, Figure 6 shows the temperature of the same die 2 measured by the load sensor 5 plotted against the number of presses. In Figure 6, the temperature tends to increase as the number of presses increases. The first principal component PC1 shown in Figure 4 is thought to represent the change in load due to temperature change, and it shows the same trend as the change in temperature. Therefore, the first principal component PC1 that was separated and extracted is thought to be due to changes in temperature.

[0038] Furthermore, the analysis unit 8 can obtain eigenvectors in the process of calculating the principal component analysis. The eigenvector indicates a proportionality coefficient that indicates how proportional each load sensor 5 is to the first principal component PC1. In other words, by referring to the eigenvector, it is possible to know how much each load sensor 5 contributes to a specific component, for example, the first principal component PC1.

[0039] However, for example, as shown in FIG. 7, even if the values ​​of the eigenvectors of each load sensor 5 are graphed with the first principal component PC1 on the horizontal axis and the second principal component PC2 on the vertical axis, it is difficult to consider the relationship between the variables from this alone.

[0040] In contrast to this, in the press die analyzing device 1 of the embodiment, the image output unit 9 back-projects each eigenvector of the first principal component PC1 and the second principal component PC2 onto the positions of the corresponding load sensors 5. As shown in FIG. 8, in the image output on the display screen of the monitor 14, the positions where the changes in the machining conditions due to the first principal component PC1 occur are displayed and visualized, superimposed on the workpiece shape 11a.

[0041] In Fig. 8, the first principal component PC1 is estimated to be a temperature change, so the user can know and visually grasp that the load is changing significantly in the left region of the workpiece shape 11a due to the influence of the temperature change. Furthermore, as shown in FIG. 9, in the image output on the display screen of the monitor 14, the positions where the changes in the machining conditions due to the second principal component PC2 occur are displayed and visualized superimposed on the workpiece shape 11a.

[0042] 5, it can be seen that the second principal component PC2 increases sharply when the number of press workings is small, and then decreases as the number of press workings increases. However, there are cases where it is difficult to estimate what kind of component this is.

[0043] Even in such a case, by referring to Figure 9 together with Figure 5, the user can visually understand that the change in the processing conditions is small on the right side of the workpiece shape 11a and near the center where the amount of material is large, which provides the user with more hints for considering what the change in the processing conditions is. For example, it can be assumed that the change on the right side near the center is due to the changes in the shape and material of the die 2 and workpiece 11. It can also be assumed that the area where the change in processing conditions is relatively large on the periphery is due to changes in the sliding resistance of the part that stabilizes the die 2 due to the eccentricity of the slide 12.

[0044] In this way, the press die analyzing device 1 of the embodiment can separate and extract factors that cause complex changes in processing conditions, and obtain a large amount of information on the long-cycle, wide-range, and complex state of the die 2 and information on the progression of changes, thereby enabling the user to accurately grasp changes in processing conditions. That is, in the embodiment, the change in the first principal component PC1 shown in Fig. 4 has a similar tendency to the rate of increase in the temperature of the mold 2 shown in Fig. 6. From this, it can be estimated that the first principal component PC1 is a temperature change.

[0045] Furthermore, the press die analysis device 1 can back-project the second principal component PC2 shown in Fig. 5 so that it overlaps with the workpiece shape 11a shown in Fig. 9. This allows the user to infer from visual information whether the second principal component PC2 is related to the shape. Then, various measures can be taken based on the analysis results obtained by the press die analyzing device 1 and the data analysis method using the press die analyzing device 1.

[0046] For example, the user can understand that mold 2 is gradually undergoing long-term, wide-ranging, and complex changes in processing conditions. This makes it possible to carry out maintenance work before the dimensional errors of the product increase. In addition, the information can be used to determine the time-dependent deterioration of mold 2, and various countermeasures can be taken, such as determining when to replace it.

[0047] The press die analysis device 1 of this embodiment performs processing to extract first and second feature amounts from multiple load data, and can separate and extract factors that cause complex changes in processing conditions, thereby obtaining multifaceted information, even when the die changes gradually over a long period of time. This makes it possible to predict the next change in processing conditions. This provides practically beneficial effects, such as enabling various measures to be taken, such as using information on the deterioration of the die 2 over time as a guide for when to replace it.

[0048] As described above, the press die analyzing device 1 of the present invention is provided with a plurality of load sensors 5 that are arranged on the rear side of the die 2 and measure the compressive load applied to various parts of the die 2 during press working. The press die analyzing device 1 includes a storage unit 7 that stores, as load data, compressive loads measured by the plurality of load sensors 5 for each of a plurality of press workings. The press die analyzing device 1 also includes an analyzing unit 8 that analyzes the plurality of load data stored in the storage unit 7.

[0049] The analysis unit 8 performs a process of extracting first feature amounts based on multiple load data measured by each load sensor 5 during one press working. The analysis unit 8 then applies a statistical method or a machine learning analysis method to the multiple first feature amounts obtained by repeating this process for multiple press workings. The analysis unit 8 then extracts eigenvectors as second feature amounts corresponding to changes in the working conditions.

[0050] This allows us to separate and extract the causes of long-cycle, wide-ranging, and complex changes in processing conditions, and by obtaining multifaceted information, we can accurately grasp the state of the mold. In detail, the first feature extracted by the processing of the analysis unit 8 is extracted from the load data measured at various points of the mold 2 repeatedly and continuously for multiple press processes.

[0051] For this reason, the analysis unit 8 of the press die analysis device 1 applies a statistical method or a machine learning analysis method to these multiple first feature quantities. The eigenvectors as second feature quantities separated and extracted in this way become multifaceted information when analyzing the causes of long-period, wide-ranging, and complex changes in processing conditions. Therefore, the user can accurately grasp the state of the die.

[0052] The storage unit 7 stores a plurality of load data measured by each load sensor 5 in one press working. The analysis unit 8 then analyzes the plurality of load data 20 measured for each of the press workings. During the analysis, the analysis unit 8 applies principal component analysis as a statistical method to extract a second feature value corresponding to a change in the working conditions. Principal component analysis derives new variables (principal components). Furthermore, eigenvectors can be obtained during the process of calculating principal component analysis. Therefore, by aggregating a large amount of load data20 to create principal components, it is possible to separate and extract factors even for long-period, wide-ranging, and complex changes in processing conditions for which variables are not identified.

[0053] Furthermore, the maximum value measured by each load sensor 5 is used as the load data. For this reason, the storage unit 7 stores the maximum value required for analysis among the compressive loads measured by the multiple load sensors 5 as the load data 20. The analysis unit 8 can efficiently collect and analyze a large amount of load data 20 over a long period and a wide range without reducing the amount of information obtained from the load data 20. Furthermore, the press die analysis device 1 can obtain the required load data using a general load sensor 5. Therefore, there is no need for special measuring equipment to monitor the die 2 over a long period of time. This makes it possible to suppress increases in manufacturing costs.

[0054] The press die analysis device 1 then visualizes the position where the change in processing conditions occurs by back-projecting the eigenvector as the second feature onto each corresponding load sensor 5. This makes it possible to visualize multifaceted information in accordance with the position of the die 2, and to more accurately grasp the state of the die 2. For example, the press die analysis device 1 visualizes the location where a change in processing conditions, such as a temperature change, occurs by displaying it on the screen of the monitor 14 together with the workpiece shape 11a. This allows the user to visually grasp which part of the die 2 is most affected by the temperature change.

[0055] The data analysis method using the press die analysis device 1 performs a long-term, wide-ranging, and complex analysis of changes in processing conditions during press processing using the die 2. As a result, the press die analysis device 1 can separate and extract the causes of complex changes in processing conditions of the die 2. By obtaining multifaceted information, the state of the die can be accurately grasped. Therefore, even if the state of the mold 2 changes gradually over a long period of time, it is possible to visualize each factor of the change in processing conditions, for example. This allows the user to grasp the exact state of the mold and take appropriate action.

[0056] Furthermore, the press die analysis system arranges multiple load sensors 5 on the rear side of the die 2 to measure the compressive load applied to various locations on the die 2 during press forming. The press die analysis system accumulates the compressive loads measured by the multiple load sensors 5 as load data in an accumulation unit 7 for each of multiple press forming operations. The press die analysis system then analyzes changes in processing conditions during press forming using the die 2. To this end, the press die analysis system includes an analysis unit 8 that analyzes the multiple load data accumulated in the accumulation unit 7. The analysis unit 8 performs a process of extracting first feature values ​​based on the multiple load data measured by each load sensor 5 for each press forming operation. This process is repeated for multiple press forming operations. In this way, the press die analysis system applies a statistical method or a machine learning analysis method to the multiple first feature values ​​obtained by repeating the process. This allows the analysis unit 8 to extract second feature values ​​corresponding to changes in processing conditions.

[0057] Furthermore, for example, the press die analysis system may be provided with the analysis unit 8 and the storage unit 7 separately. In this case, the analysis unit 8 analyzes the plurality of load data stored in the storage unit 7 via a network connecting the analysis unit 8 and the storage unit 7. Therefore, even if the analysis unit 8 and the accumulation unit 7 are installed separately in different companies located in remote locations, the causes of changes in the processing conditions of the complex mold 2 can be separated and extracted, and multifaceted information can be obtained.

[0058] In this way, the press die analysis device equipped with the analysis unit 8 can statistically analyze the plurality of load data 20 stored in the storage unit 7 via a network. Therefore, it can exert practically beneficial effects, such as being suitable for use by analysis companies that perform analysis.

[0059] The present invention is not limited to the above-described embodiments, and various modifications are possible. The above-described embodiments are provided as examples to facilitate understanding of the present invention, and are not necessarily limited to those including all of the configurations described. Furthermore, it is possible to replace part of the configuration of one embodiment with the configuration of another embodiment, or to add the configuration of another embodiment to the configuration of one embodiment. Furthermore, it is possible to delete part of the configuration of each embodiment, or to add or replace other configurations. Possible modifications of the above-described embodiments include, for example, the following.

[0060] In this embodiment, the analysis was performed using the maximum value of the load data 20. However, the load data used for the analysis is not limited to the maximum value, and for example, by performing the analysis using a load value at a specific timing during processing, it is possible to perform a more detailed analysis of defects that frequently occur at a specific timing during press processing.

[0061] In this embodiment, principal component analysis is used as the statistical method. However, the statistical method is not limited to principal component analysis. For example, a method such as factor analysis, which creates new variables (factors) using only variables assumed in advance, may also be used. That is, in addition to creating principal components by aggregating data with many variables, once it is known that one of the principal components is mold temperature change, as in the embodiment, factor analysis may be performed assuming the mold temperature change as a variable in advance. Furthermore, by assuming a variable as a statistical method, a support vector machine (SVM), which is a pattern recognition model using supervised learning, may be used. In this way, any regression method may be used as a statistical method.

[0062] Furthermore, for example, parameter estimation using a neural network may be used as a machine learning analysis method. In this case, by linking specific defects with load data and having the system learn the data, it is possible to visualize the area causing the defect, thereby improving productivity and stabilizing quality.

[0063] In this embodiment, a backprojected image is obtained from the eigenvectors. However, this procedure can also be performed in the reverse direction: that is, a corresponding eigenvector can be created using a mask pattern prepared by the user instead of the backprojected image, and the corresponding second feature can be obtained by calculating the dot product of the eigenvector and the weight data. In this case, the analysis unit 8 first calculates the eigenvectors corresponding to the mask pattern defined by the user. Alternatively, the user can directly set each component of the eigenvector while imagining the corresponding mask pattern. Next, the analysis unit 8 can obtain a second feature amount corresponding to a principal component by calculating the inner product of the user-defined eigenvector and the weight data.

[0064] FIG. 10 illustrates an example of a mask pattern and how to use it. The mask pattern represents the magnitude of the proportionality coefficient of each sensor, and in FIG. 10, the hatched area represents positive proportionality, and the non-hatched area represents negative proportionality. In the mask pattern in Fig. 10, the product shape area is set as positive and other areas as negative. When using such a mask, the value of the second feature increases as the load on the product shape area increases, and it is therefore expected that components corresponding to the temperature change and material denaturation obtained in principal component analysis can be extracted. For example, in the early stages of mold production, the amount of data is small, making it difficult to apply statistical or machine learning techniques. However, in analysis using mask patterns, it is possible to prepare unique mask patterns in advance using know-how from past molds. This makes it easier to identify issues and implement countermeasures, improving mold production efficiency and product quality.

[0065] The vector dot product calculation described so far in this embodiment is mathematically equivalent to solving an equation with each component of the vector as a coefficient, as shown in Equation 2. Therefore, the coefficients corresponding to each component of the eigenvector may be calculated individually and the calculation may be performed using an equation.

[0066] [Formula 2] TIFF2025186972000003.tif60166…Formula 2 In Equation 2, vector A is an eigenvector, and X is a vector whose components are the measurement data of each sensor during one press working.

[0067] Furthermore, the number and arrangement pattern of the load sensors 5 are not limited to those shown in the embodiment. For example, it is sufficient if multiple load sensors 5 are arranged on the back side of the die 2 so as to measure the compressive load applied to various parts of the die 2 during press working. 1, the embodiment has been described with a plurality of load sensors 5 provided on the back surface 3a side of the upper die 3, which is the opposite side to the machining surface, but this is not particularly limited. For example, the load sensors 5 may be disposed between the bolster 13 and the back surface side of the lower die 4, which is the opposite side to the machining surface.

[0068] In this way, the multiple load sensors 5 may be provided on the back side of at least one of the upper die 3 or the lower die 4, as long as they measure the compressive load applied to various locations on the die 2 during press working. That is, the number, type and arrangement pattern of the load sensors 5 are not particularly limited as long as there are a plurality of load sensors 5.

[0069] Furthermore, in the embodiment, the press die analyzing device 1 including the accumulation unit 7 and the analysis unit 8 has been described as an example, but the present invention is not particularly limited to this. For example, the present invention may be configured as a press die analysis system in which the analysis unit 8 and the storage unit 7 are located at separate locations, such as a press die analysis device 1 equipped with an analysis unit 8 and a storage unit 7 provided on a network cloud. In this case, the press die analysis device 1 equipped with the analysis unit 8 is connected via a network to a storage unit 7 provided in a remote data center or in another company or department within the company. Then, by operating a terminal on the analysis unit 8 side, the plurality of load data 20 stored in the storage unit 7 can be transmitted and received and analyzed.

[0070] In addition, in such a press die analysis system, the software for the storage unit 7 and the software for the analysis unit 8 are provided separately. Therefore, the same software for the analysis unit 8 can be installed on multiple terminals to analyze multiple pieces of load data 20 stored in the storage unit 7.

[0071] Furthermore, it is easy to change the statistical method or machine learning analysis method by simultaneously replacing the software on the analysis unit 8 side, which is frequently updated, on multiple terminals via a network, or by installing different software individually.

[0072] The network is not limited to the Internet using a VPN, but may be directly connected using other communication means such as a dedicated line or wirelessly. Moreover, instead of or in addition to the network, the load data 20 may be transferred using a storage device such as a portable USB memory.

[0073] Furthermore, even if the press processing machine 10 is installed in a remote location, such as another company or another department within a company, the press die analysis system to which the present invention is applied exhibits practically beneficial effects, such as being able to separate and extract the causes of complex changes in the processing conditions of the die 2 using the analysis unit 8 of the press die analysis device connected via a network, and being able to obtain multifaceted information when analyzing long-period, wide-ranging, and complex changes in processing conditions. [Explanation of symbols]

[0074] 1 Press die analysis equipment 2. Mold 3a back 5 Load Sensor 6 Control Unit 7. Storage section 8 Analysis Department 9 Image output section

Claims

1. A press die analysis device for analyzing changes in processing conditions during press processing using a die, a plurality of load sensors arranged on the rear side of the die and measuring the compressive load applied to each part of the die during press working; a storage unit that stores the compressive loads measured by the plurality of load sensors as load data for each of a plurality of press workings; an analysis unit that analyzes the plurality of load data stored in the storage unit, The analysis unit performs a process of extracting first feature quantities based on multiple load data measured by each load sensor in one press working operation, and extracts second feature quantities corresponding to changes in the working conditions by applying a statistical method or a machine learning analysis method to the multiple first feature quantities obtained by repeating the process for multiple press working operations.

2. 2. The press die analysis device according to claim 1, wherein the storage unit stores a plurality of load data measured by each load sensor in one press working, and the analysis unit, when analyzing the plurality of load data measured for each of a plurality of press workings, applies principal component analysis as the statistical method to extract the second feature amount corresponding to the change in the working conditions.

3. 2. The press die analyzing device according to claim 1, wherein the load data is a maximum value measured by each of the load sensors.

4. 2. The press die analyzing device according to claim 1, wherein the second feature amount is back-projected onto each of the corresponding load sensors to visualize the position where a change in processing conditions occurs.

5. 2. The press die analysis device according to claim 1, wherein the machine learning analysis method uses parameter estimation by a neural network.

6. 2. The press die analyzing device according to claim 1, wherein the analysis unit performs filtering using a mask of an arbitrary pattern.

7. 2. A data analysis method using the press die analysis device described in claim 1, characterized in that, in one press working, a point in multidimensional space is determined based on multiple load data applied to various parts of the die, point cloud data for multiple press workings is created, and a statistical method is applied to the point cloud data to extract the second feature amount corresponding to changes in the working conditions, thereby performing a long-term, wide-ranging, and comprehensive analysis of changes in working conditions in the press working of the die.

8. A press die analysis system that arranges a plurality of load sensors on a rear side of a die to measure compressive loads applied to various locations of the die during press working, accumulates the compressive loads measured by the plurality of load sensors as load data in an accumulation unit for each of a plurality of press workings, and analyzes changes in working conditions during press working using the die, an analysis unit that analyzes the plurality of load data stored in the storage unit, The analysis unit performs a process of extracting first feature quantities based on multiple load data measured by each load sensor in one press working operation, and extracts second feature quantities corresponding to changes in the working conditions by applying a statistical method or a machine learning analysis method to the multiple first feature quantities obtained by repeating the process for multiple press working operations.

9. 9. The press die analysis device used in the press die analysis system according to claim 8, further comprising: an analysis unit connected to the storage unit via a network and configured to analyze the plurality of load data stored in the storage unit.

Images