Method and apparatus for predicting the quality of die-cast molded products

Abstract

To provide a shrinkage cavity prediction method and a shrinkage cavity prediction apparatus for a die-cast molded product which can predict the conditions of shrinkage cavities based on data outputted from a die casting apparatus, a quality prediction method and a quality prediction apparatus, and an inspection method.SOLUTION: A correlation between the wave shapes showing the conditions of shrinkage cavities in a molded product and the time change of an injection pressure value upon molding the molded product is found, and using the correlation, the conditions of the shrinkage cavities in a new molded product are predicted. Further, a cluster in which the conditions of the shrinkage cavities satisfy inspection standards and a cluster which does not satisfy the inspection standards are beforehand created from the feature amount of the injection pressure value, and using the clusters, whether the conditions of the shrinkage cavities in the new molded product satisfy the inspection standards or not is predicted.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] The present invention relates to a method and apparatus for predicting shrinkage cavities in die-cast molded products. The present invention also relates to a method and apparatus for predicting the quality of die-cast molded products, and an inspection method for die-cast molded products using the quality prediction method.

Background Art

[0002] When pressure-injecting a molten alloy into the cavity of a mold, the cavity is divided into a plurality of volume elements, the solidification time when the flow of the molten metal stops for each volume element is calculated based on the temperature change of the molten metal, the pressure action time when the pressure of the pressure injection acts on each volume element is calculated based on the solidification time, an output value is calculated for each volume element by comparing the solidification time and the pressure action time, and a casting analysis method for determining the range in which shrinkage cavities occur based on the output value is known (Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in the above prior art, processes for dividing the cavity shape into a plurality of volume elements and calculating the solidification time and the pressure action time for each volume element are required, and the implementation mainly uses software-based simulation. Therefore, the above prior art has a problem that the state of shrinkage cavities cannot be predicted based on the data output from the die-casting apparatus without using software-based simulation.

[0005] The problem that the present invention aims to solve is to provide a method and apparatus for predicting shrinkage cavities in die-cast molded products, a method and apparatus for predicting the quality of die-cast molded products, and a method for inspecting die-cast molded products, which can predict the state of shrinkage cavities based on data output from a die-casting apparatus. [Means for solving the problem]

[0006] The first invention of this invention determines the correlation between the state of shrinkage cavities in a molded product and the injection pressure waveform, which shows the time change of the injection pressure value when the molded product was formed, and uses this correlation to predict the state of shrinkage cavities in a new molded product. Furthermore, if the die-casting apparatus injects molten metal into the mold cavity at a first pressure, and then injects molten metal into the cavity at a second pressure higher than the first pressure, the waveform showing the time variation of the first pressure is used as the injection pressure waveform. The above problems are solved by this. Furthermore, the second invention of the present invention solves the above problems by pre-generating clusters of shrinkage cavities that satisfy inspection criteria and clusters that do not satisfy the inspection criteria from the characteristic quantities of the injection pressure value, and using these clusters to predict whether or not the shrinkage cavities of a new molded product will satisfy the inspection criteria. [Effects of the Invention]

[0007] According to the present invention, the state of shrinkage cavities can be predicted based on data output from a die-casting device. [Brief explanation of the drawing]

[0008] [Figure 1] This is a block diagram showing an example of an embodiment of the molding system according to the present invention. [Figure 2] This figure shows an example of an injection pressure waveform displayed on a display device. [Figure 3] This figure shows another example of an injection pressure waveform displayed on a display device. [Figure 4] This data shows an example of the total diameter of shrinkage cavities in die-cast molded products. [Figure 5] This graph shows an example of a predicted value for the sum of the diameters of the crevice, output from a neural network trained using the data shown in Figure 4. [Figure 6]This is an example of a feature map showing features extracted from injection pressure data. [Figure 7] Figure 6 shows an example of a dendrogram illustrating the process of generating clusters from the features shown. [Figure 8] Figure 6 shows an example of a feature map for classification obtained from the features shown. [Figure 9] Figure 1 is a flowchart illustrating an example of the processing procedure in the shrinkage cavity prediction device. [Figure 10] Figure 1 is a flowchart showing an example of the processing procedure in the quality prediction device. [Figure 11] Figure 1 is a flowchart showing an example of the processing procedure in the inspection device. [Modes for carrying out the invention]

[0009] Embodiments of the present invention will be described below with reference to the drawings.

[0010] [Configuration of the molding system] Figure 1 is a block diagram illustrating the molding system 10 according to the present invention. The molding system 10 is a group of devices for molding molded products by die casting, and as shown in Figure 1, it comprises a die-casting device 1, a display device 2, and a control device 3. The control device 3 is connected to a server S which stores information necessary for processing in the control device 3. Each device constituting the molding system 10 is connected by a communication interface that supports various communication standards such as wired LAN standards and wireless LAN standards, and can exchange information with each other.

[0011] The die-casting apparatus 1 is a device that casts metal molded products by injecting molten metal (hereinafter also referred to as "molten metal") into a mold at high speed and high pressure. In this embodiment, the metal molded product formed using the die-casting apparatus 1 is referred to as a die-cast molded product. The die-casting apparatus 1 includes, for example, a heating furnace that melts the metal to be used as the material for the die-cast molded product into molten metal, a cylindrical sleeve for injecting the molten metal into a mold, a tip that slides against the inner surface of the sleeve and pushes the molten metal into the mold, a rod for moving the tip, a mold clamping mechanism for the mold and a hydraulic device for applying pressure to the rod, a cooling device for cooling the mold and the like, and a toggle device for moving the movable side of the mold. In addition, the die-casting apparatus 1 may also include a device for removing burrs from the die-cast molded product using media, a device for transporting the die-cast molded product to the next process (for example, a robotic arm and an automated guided vehicle), etc.

[0012] The metals used in die-cast products are non-ferrous metals such as aluminum, magnesium, zinc, and copper, and their alloys. Examples include Al-Si-Cu based ADC12, Mg-Al-Zn based MDC1D, and Zn-Al-Cu based ZDC1. The molds are made of tool steel such as SKD61 and are manufactured, for example, by cutting steel material.

[0013] The display device 2 is a device that provides information to operators engaged in the operation of the molding system 10, and examples include a liquid crystal display installed in the die-casting apparatus 1, a display on a wearable terminal worn by the operator, etc. The display device 2 may also be equipped with an input device for the operator to input instructions to the control device 3. Examples of input devices include a touch panel that receives input by the user's finger or stylus pen, and a microphone that acquires instructions by the user's voice. The display device 2 may also be equipped with a speaker as an output device.

[0014] The control device 3 is a device that controls the operations of the die-casting device 1 and the display device 2 to cooperate with each other to form die-cast molded products. The control device 3 is, for example, a computer, and includes a CPU (Central Processing Unit) 31 that is a processor, a ROM (Read Only Memory) 32 in which programs are stored, and a RAM (Random Access Memory) 33 that functions as an accessible storage device. The CPU 31 executes the programs stored in the ROM 32 and is an operation circuit for realizing the functions of the control device 3. Note that the ROM 32 may be an HDD (Hard Disk Drive).

[0015] In order to control the operation of the die-casting device 1, the control device 3 acquires information indicating the state of the device from sensors provided in the die-casting device 1. For example, the control device 3 acquires information on the injection pressure for injecting the molten metal into the mold from a pressure sensor provided on the rod. In addition to this, the control device 3 may acquire information on the injection speed for injecting the molten metal into the mold from a sensor provided in the hydraulic device that moves the rod. In addition to this, the control device 3 may acquire information on the injection position indicating the position of the chip and information on the degree of vacuum in the cavity of the mold. Note that the injection position is set such that the initial position of the chip or the rod before injecting the molten metal is 0 mm, and the direction approaching the mold is the positive direction.

[0016] In order to notify the operator of the state of the die-casting device 1, the control device 3 causes the display device 2 to display the information acquired from the die-casting device 1. In this case, the control device 3 may convert the values acquired from the sensors of the die-casting device 1 into a graph and output the graph to the display device for display so that the operator can easily check the state of the die-casting device 1. For example, the control device 3 causes the display device 2 to display a waveform of the injection pressure value as shown in FIG. 2.

[0017] Figure 2 shows an example of the injection pressure waveform of the injection pressure value displayed on the display device 2. As shown in Figure 2, the die-casting molding process includes a low-speed injection process in which molten metal is injected into the mold at a low speed and low pressure, a high-speed injection process in which molten metal is injected at a higher speed and higher pressure than the low-speed injection process, and a pressure boosting process to suppress the occurrence of casting defects after the molten metal has been injected. In the low-speed injection process, the molten metal is injected at an injection speed of about 0.1 to 0.7 m / s to avoid entrapping air inside the sleeve, and in the high-speed injection process, after most of the mold cavity (approximately 90 to 95%) has been filled with molten metal, the molten metal is pushed in at an injection speed of about 2 to 3 m / s to fill the limited unfilled space in the cavity. The injection pressure in the low-speed injection process is approximately 5 MPa or less, and the injection pressure in the high-speed injection process is about 10 MPa.

[0018] In the pressurization process, the tip hardly moves (i.e., the injection speed is almost 0 m / s), and a pressure higher than that of the high-speed injection process (specifically, around 30-70 MPa) is continuously applied to the molten metal. This suppresses deformation of the molded product that occurs when the molten metal solidifies and shrinks in the cavity, and prevents the formation of casting defects (especially shrinkage cavities) inside the molded product. This process of injecting additional molten metal after filling the cavity of the mold is also called riser injection.

[0019] Among the information acquired by the control device 3 from the die-casting apparatus 1, the injection pressure value is known to be related to the state of shrinkage cavities in the die-cast molded product. A shrinkage cavity, also called a shrinkage hole, is a solidification shrinkage defect that occurs during the molding process of a die-cast molded product. When molten metal changes from a molten state to a solid state, if molten metal is not supplied to the part where shrinkage occurs due to the phase change, the shrunken part remains as a void inside the molded product, becoming a shrinkage cavity. Shrinkage cavities always occur in die-cast molded products, but if the state of the shrinkage cavity exceeds the inspection standard, it is recognized as a defect.

[0020] If die-cast products are continuously molded using the same die-casting apparatus 1, shrinkage cavities that do not meet inspection standards may occur in the die-cast products due to insufficient cooling of the mold, variations in product shape, and variations in molten metal temperature. When shrinkage cavities occur in die-cast products, a waveform of the injection pressure value, such as that shown in Figure 3, is displayed on the display device 2. In other words, when shrinkage cavities that do not meet inspection standards occur inside the die-cast product, the injection pressure waveform of the low-speed injection process changes first, and the pressure value fluctuates up and down, causing the waveform to become distorted. Then, if die-cast products are continued to be molded in this state, the pressure value fluctuates up and down in the part where the pressure of the pressure-boosting process is high (specifically, the area X enclosed by the dashed line in Figure 3), causing the injection pressure waveform to become distorted.

[0021] The change in injection pressure in range X is larger than the change in the low-speed injection process, making it relatively easy for the operator or control device 3 to detect shrinkage cavities in the die-cast product. However, in the low-speed injection process, where shrinkage cavities begin to occur in the die-cast product, the change in injection pressure is smaller than the change in range X, making it difficult for the operator to visually detect the occurrence of defects, and conventional anomaly detection methods were unable to detect the occurrence of shrinkage cavities.

[0022] Therefore, the control device 3 of this embodiment is equipped with a shrinkage cavity prediction device 4 and an inspection device 5, as shown in Figure 1, in order to automatically determine whether the state of shrinkage cavities in the die-cast molded product meets the inspection criteria. The shrinkage cavity prediction device 4 is a device for predicting the state of shrinkage cavities (especially the sum of the diameters of the shrinkage cavities) from the injection pressure waveform of the injection pressure value, and is also used for selecting data to be used for machine learning in the inspection device 5. The inspection device 5 includes a quality prediction device as part of it, and is a device for determining whether the die-cast molded product meets the inspection criteria (especially the inspection criteria for the state of shrinkage cavities).

[0023] The shrinkage cavity prediction device 4 and inspection device 5 are, for example, computers similar to the control device 3. In this embodiment, the shrinkage cavity prediction device 4 and inspection device 5 are included in the control device 3, and the processing in the shrinkage cavity prediction device 4 and inspection device 5 is performed using the CPU 31, ROM 32, and RAM 33 of the control device 3. The program stored in ROM 32 contains the program for the shrinkage cavity prediction device 4 and inspection device 5 to realize their respective functions, and the CPU 31 executes the program stored in ROM 32 to realize these functions.

[0024] Figure 1 shows, for convenience, the relationship acquisition unit 41 and the state prediction unit 42 as functional blocks that realize the functions of the shrinkage cavity prediction device 4, and for convenience, the acquisition unit 51, extraction unit 52, classification unit 53, suitability prediction unit 54, recording unit 55, and output unit 56 as functional blocks that realize the functions of the inspection device 5. The functions of each functional block shown in Figure 1 will be described below.

[0025] First, the relationship acquisition unit 41 and the state prediction unit 42 of the shrinkage cavity prediction device 4 will be explained.

[0026] The relationship acquisition unit 41 has the function of determining the correlation between the state of shrinkage cavities in a die-cast molded product and the injection pressure waveform, which shows the change in the injection pressure value over time. The shrinkage cavity prediction device 4 uses the function of the relationship acquisition unit 41 to determine the correlation between the state of shrinkage cavities and the injection pressure waveform using machine learning, based on data that associates the injection pressure waveform output by the die-casting device 1 when it molds a die-cast molded product with the state of shrinkage cavities in the die-cast molded product. The data that associates the injection pressure waveform with the state of shrinkage cavities is stored in the server S in advance, and the shrinkage cavity prediction device 4 acquires this data from the server S as needed.

[0027] The condition of shrinkage cavities in a die-cast molded product refers to the size, location, shape, and number of shrinkage cavities. Information on the condition of shrinkage cavities can be obtained, for example, by scanning the die-cast molded product using computed tomography (CT) and processing the scan results with appropriate defect analysis software (e.g., VG Studio Max from Volume Graphics). Alternatively, or in addition to this, the die-cast molded product may be cut and the condition of the shrinkage cavities (shape, location, size, etc.) that have formed inside the molded product may be measured. In this embodiment, the condition of shrinkage cavities in the die-cast molded product was specifically expressed as the sum of the diameters of the shrinkage cavities (unit: mm) that have formed in the die-cast molded product. A larger sum of the diameters of the shrinkage cavities indicates a larger number of shrinkage cavities and / or a larger size of shrinkage cavities.

[0028] The injection pressure waveform, which shows the time change of the injection pressure value, is, for example, the injection pressure waveform shown in Figures 2 and 3. In the shrinkage cavity prediction device 4, when determining the correlation between the state of the shrinkage cavity and the injection pressure waveform, machine learning is performed using, for example, image information of the injection pressure waveform. As an example, the shrinkage cavity prediction device 4 acquires an image of the range Y shown in Figure 2 as the injection pressure waveform. In other words, if the die casting device 1 injects molten metal into the cavity of the mold at low pressure (first pressure) in the low-speed injection process, and then injects molten metal into the cavity at high pressure (second pressure higher than the first pressure) in the high-speed injection process, the injection pressure waveform showing the time change of the first pressure in the low-speed injection process may be used.

[0029] To determine the correlation between the state of shrinkage cavities and the injection pressure waveform using machine learning, a predetermined number of sets of information on the state of shrinkage cavities and image information on the injection pressure waveform are required. The predetermined number can be set to an appropriate number within the range in which machine learning can be performed appropriately, but for example, it is 50 to 500 sets or more. The correlation sought by the shrinkage cavities prediction device 4 is not particularly limited as long as it can appropriately predict the state of shrinkage cavities. An example of such a correlation is a convolutional neural network trained by machine learning.

[0030] For example, the shrinkage cavity prediction device 4 generates a neural network comprising an input layer into which image information of the injection pressure waveform is input, an intermediate layer that extracts features defined by a kernel from the image information input to the input layer through convolutional operations, and an output layer that outputs information on the state of shrinkage cavities corresponding to the extracted features. In the intermediate layer, the parameter weighting is adjusted so that the information on the state of shrinkage cavities output from the output layer corresponds to the information on the state of shrinkage cavities measured in advance.

[0031] As mentioned above, the change in injection pressure during the low-speed injection process when shrinkage cavities occur is relatively small. Therefore, if image information of the injection pressure waveform during the low-speed injection process is used, simply setting the kernel and performing pooling as usual may result in the loss of features in the intermediate layers, making it impossible to properly train the neural network. In this case, the shrinkage cavity prediction device 4 performs padding processing using the relationship acquisition unit 41 to prevent the size of the output data after the convolution operation using the kernel from becoming too small. For example, padding processing is performed so that the size of the data does not change before and after the convolution operation.

[0032] Alternatively, or in addition to the above, if only image information of the injection pressure waveform in the low-speed injection process is used, the stride for applying the kernel to the input data may be set smaller than when image information of the injection pressure waveform in processes other than the low-speed injection process is used. In this way, the number of hidden layers can be adjusted (for example, increased) by adjusting the size of the output data, and the neural network can be trained appropriately.

[0033] The state prediction unit 42 has the function of predicting the state of shrinkage cavities in a new molded product when a new molded product is formed using the injection pressure waveform output when a new molded product is formed and the correlation relationship obtained by the relationship acquisition unit 41. The shrinkage cavity prediction device 4 acquires image information of the injection pressure waveform using the function of the state prediction unit 42 and, for example, inputs this image information into a trained neural network to predict the sum of the diameters of the shrinkage cavities.

[0034] Figure 4 shows the total diameter of shrinkage cavities inside die-cast molded parts, measured with VG Studio Max. The horizontal axis shows the identification number assigned to each die-cast molded part, and the vertical axis shows the total measured diameter (unit: mm). The total diameter shown in Figure 4 was extracted using VG Studio Max from data obtained by scanning the die-cast molded parts with CT, and the smallest circumscribing sphere diameter among the detected cavities was 2.0 to 2.0 × 10⁻⁶. 3 Voids with a size of mm were extracted as shrinkage cavities. Standard settings (e.g., default settings) were used in VG Studio Max when extracting shrinkage cavities, with a defect value of 1.0 or higher and an edge distance calculation of mesh 1. The defect value is a threshold for determining whether a void is truly a shrinkage cavity (i.e., whether a shrinkage cavity is being misidentified from CT image data), and the edge distance calculation is the setting of the distance between the edge of the die-cast molded product and the void to avoid detecting blowholes as shrinkage cavities. The total number of data shown in Figure 4 is 300, and in this embodiment, a convolutional neural network was trained using data that corresponds to the image information of the injection pressure waveform corresponding to the sum of the diameters of these 300 data points. The image information of the injection pressure waveform is image information when molten metal is injected at low pressure in a low-speed injection process.

[0035] The trained neural network was evaluated using the data used for training. Specifically, image information of the injection pressure waveform used to train the neural network was input into the network, and the predicted value of the sum of diameters output was compared with the actual sum of diameters corresponding to the input image information. The results of the comparison are shown in Figure 5. In Figure 5, the horizontal axis shows the sum of diameters that were actually measured, and the vertical axis shows the sum of diameters that were predicted. As shown in Figure 5, the predicted sum of diameters are distributed close to the regression line L, indicating that the sum of diameters predicted by the trained neural network has a strong correlation with the actual sum of diameters.

[0036] Furthermore, when the correlation coefficient (for example, Pearson's product-moment correlation coefficient, which is obtained by dividing the covariance between the actual sum of diameters and the predicted sum of diameters by the standard deviation of each sum of diameters) was calculated for the results shown in Figure 5, the correlation coefficient r = 0.985 was obtained. Generally, a correlation coefficient of 0.8 or higher is interpreted as a strong correlation, so it can be seen that the predicted sum of diameters has a strong correlation with the actual sum of diameters.

[0037] Furthermore, the correlation was confirmed using data not used in training the neural network. Comparing the actual sum of diameters corresponding to this data with the sum of diameters predicted from the image information of the data, the two values ​​were close, resulting in the point Z shown in Figure 5. Therefore, it can be said that the convolutional neural network generated in this embodiment has a correlation that can appropriately predict the state of the nest.

[0038] Information on the state of shrinkage cavities in die-cast molded products is obtained by scanning the die-cast molded product with CT and processing the results with appropriate defect analysis software. However, the CT scan results and defect analysis processing are not always accurate, and the size and shape of the shrinkage cavities may be misrecognized. Therefore, it is preferable that the data used when determining the correlation in the shrinkage cavity prediction device 4 excludes inappropriate data that misrecognizes the state of the shrinkage cavities.

[0039] While it is not necessary to exclude all such inappropriate data, in this embodiment, data is deemed usable without problems for determining correlation if the correlation coefficient between the sum of diameters corresponding to the injection pressure waveform and the sum of diameters predicted from the injection pressure waveform using the correlation relationship is 0.8 or higher.

[0040] Next, the acquisition unit 51, extraction unit 52, classification unit 53, suitability prediction unit 54, recording unit 55, and output unit 56 of the inspection device 5 will be described.

[0041] The acquisition unit 51 has the function of acquiring injection pressure values ​​previously output from the die-casting apparatus 1 from the server S shown in Figure 1. When the inspection apparatus 5 acquires injection pressure values ​​from the server S using the function of the acquisition unit 51, it acquires the injection pressure value when the shrinkage cavity prediction apparatus 4 determines that the correlation coefficient between the sum of diameters corresponding to the injection pressure waveform and the sum of diameters predicted using the correlation relationship is 0.8 or higher. This is because if inappropriate data that misidentifies the state of shrinkage cavities is included, it will not be possible to generate appropriate clusters in the clustering described later.

[0042] The injection pressure value acquired by the inspection device 5 from the server S may be associated with an identification number assigned to each molded product and the injection pressure value used when that product was molded. The association between the identification number of the die-cast molded product and the injection pressure value may be performed using the functions of the server S after the injection pressure value has been stored in the server S, or it may be performed by the inspection device 5 when the product is output from the die-casting device 1. Alternatively, the association may be performed by the functions of the die-casting device 1 when the product is output from the die-casting device 1.

[0043] The extraction unit 52 has the function of extracting feature quantities from the injection pressure values ​​output from the die-casting apparatus 1. The method of extracting feature quantities is not particularly limited, but for example, the inspection apparatus 5 uses the function of the extraction unit 52 to train an autoencoder using the injection pressure values ​​output from the die-casting apparatus 1, and then uses the encoder of the trained autoencoder to extract feature quantities from the injection pressure values.

[0044] A feature is information specifically necessary to reconstruct a given injection pressure value. That is, if a feature is extracted from a given injection pressure value, the injection pressure value can be determined from that feature. However, the reconstruction of the injection pressure value from the feature does not necessarily have to be a complete reconstruction; a partial reconstruction is acceptable. An autoencoder (autoencoder) consisting of an encoder (feature extractor) and a decoder (reconstructor) can be used to extract feature quantities.

[0045] Specifically, the injection pressure value is input to the encoder (input layer) of the autoencoder, and the weighting of the node connections in the hidden layer is changed so that the input injection pressure value is output from the decoder (output layer). Through learning, the weighting is changed so that the injection pressure value input to the autoencoder matches the output, thereby extracting only the features necessary for reconstructing the injection pressure value, and a network is formed that efficiently generates the original injection pressure value from the extracted features. In other words, an autoencoder that can more accurately reconstruct the input injection pressure value can more accurately extract the features of the injection pressure value than an autoencoder that cannot reconstruct it accurately.

[0046] Figure 6 shows an example of features extracted using the trained autoencoder. The autoencoder was trained using 340 sets of injection pressure data, each set containing 3000 points. In this embodiment, the features are vectors, specifically vectors on the xy-plane as shown in Figure 6. Each point in Figure 6 corresponds to a feature extracted from the injection pressure values ​​obtained for a particular die-cast product. In this embodiment, 3000 injection pressure data points were obtained for each die-cast product, and feature extraction was performed for a total of 340 die-cast products. Therefore, Figure 6 shows 340 features.

[0047] In the case shown in Figure 6, the injection pressure data used for training the autoencoder and the injection pressure data used for feature extraction are the same. However, the injection pressure data used for training the autoencoder and the injection pressure data used for feature extraction may be different. Furthermore, the number of injection pressure data used for training the autoencoder is not limited to the number described above, and an appropriate value can be set within the range in which the autoencoder can appropriately extract features. The injection pressure data used for training the autoencoder is, for example, stored in advance on server S.

[0048] Furthermore, in the case of the shrinkage cavity prediction device 4, the inspection device 5 may use the low-pressure value as the injection pressure value when the die-casting device 1 injects molten metal into the cavity of the mold at low pressure (first pressure) and then injects molten metal into the cavity at high pressure (second pressure, which is higher than the first pressure). Note that the characteristic quantities shown in Figure 6 use data of the injection pressure value in the low-speed injection process (i.e., the low-pressure injection pressure value) as the injection pressure value.

[0049] The classification unit 53 has the function of classifying the feature quantities extracted by the function of the extraction unit 52 into multiple clusters. The inspection device 5 classifies the feature quantities into multiple clusters using known clustering methods such as the single-link method, full-link method, Ward's method, and centroid method, based on the function of the classification unit 53.

[0050] Figure 7 shows a specific method for generating clusters. Figure 7 is an example of a dendrogram showing the process of generating clusters from the features shown in Figure 6. The horizontal axis represents the identification number of each die-cast molded product, and the vertical axis represents the distance between the features shown in Figure 6 (the distance between clusters after clusters have been generated). In generating the clusters shown in Figure 7, Ward's method was used, and clustering was performed with a cluster threshold of 4. As a result, 10 clusters, C1 to C10, were generated.

[0051] Note that the clustering method is not limited to Ward's method; other methods may also be used. Furthermore, the cluster threshold can be set to an appropriate value within a range that allows multiple clusters to be classified into those that meet the inspection criteria and those that do not.

[0052] Furthermore, the classification unit 53 has the function of classifying multiple clusters into clusters that meet the inspection criteria and clusters that do not meet the inspection criteria, based on information on whether the molded products formed using the die-casting apparatus 1 meet the inspection criteria for the state of shrinkage cavities. The inspection apparatus 5 obtains information on the state of shrinkage cavities of the molded products corresponding to the identification number, and information on whether the state of shrinkage cavities meets the inspection criteria, from the server S using the function of the classification unit 53. Then, it compares the information on whether the inspection criteria are met with the identification number of the molded product belonging to the cluster and determines whether the molded product belonging to that cluster meets the inspection criteria. By performing this determination for each cluster, multiple clusters can be classified into clusters that meet the inspection criteria and clusters that do not meet the inspection criteria.

[0053] For example, for cluster C1 shown in Figure 7, the identification numbers of the molded products belonging to cluster C1 and information on whether or not the molded products with those identification numbers meet the inspection criteria are obtained from server S. If information is obtained that the molded products belonging to cluster C1 meet the inspection criteria, cluster C1 is classified as a cluster that meets the inspection criteria. Conversely, the identification numbers of the molded products belonging to cluster C10 and information on whether or not the molded products with those identification numbers meet the inspection criteria are obtained from server S. If information is obtained that the molded products belonging to cluster C10 do not meet the inspection criteria, cluster C10 is classified as a cluster that does not meet the inspection criteria.

[0054] By repeating the above determination, let's assume that clusters C1 to C6 in Figure 7 are classified as clusters that meet the inspection criteria, and clusters C7 to C10 are classified as clusters that do not meet the inspection criteria. When this cluster classification result is reflected in the feature map shown in Figure 6, a determination feature map like the one shown in Figure 8 is generated. Range A in Figure 8 is the range that contains the features of molded products that meet the inspection criteria and belong to clusters C1 to C6, and range B is the range that contains the features of molded products that do not meet the inspection criteria and belong to clusters C7 to C10. The generated determination feature map is stored in server S.

[0055] Furthermore, when clusters are generated using data that excludes inappropriate data where the state of shrinkage cavities is misidentified, as shown in Figure 8, a judgment feature map is obtained that appropriately classifies clusters into those that meet the inspection criteria and those that do not. However, when clusters are generated using data that includes data where the state of shrinkage cavities is misidentified, a judgment feature map that appropriately classifies clusters into those that meet the inspection criteria and those that do not, as shown in Figure 8, may not be obtained. If an appropriate judgment feature map is not obtained, the injection pressure value data used for cluster generation is reviewed, and cluster generation is performed again.

[0056] Using the feature map for determination shown in Figure 8, when a new molded product is formed using the die-casting apparatus 1, it is possible to predict whether the new molded product will meet the inspection criteria by determining whether the feature quantities of the new molded product belong to range A or range B. When a new die-cast molded product is formed, the inspection apparatus 5 uses the function of the extraction unit 52 to extract new feature quantities from the new injection pressure value output from the die-casting apparatus 1, and uses the function of the classification unit 53 to acquire the feature map for determination from the server S as needed.

[0057] The conformance prediction unit 54 has the function of predicting whether the condition of shrinkage cavities in the molded product meets the inspection criteria based on the cluster classification results. The inspection criteria are standards that prevent damage when the die-cast molded product is used in the assumed usage environment, and depend on the shape and material of the die-cast molded product. Examples of inspection criteria include that the location of shrinkage cavities does not exist between holes to be bolted, that the shrinkage cavities do not exist near the surface of the contour that characterizes the shape of the die-cast molded product, that no leakage occurs when a leakage test is performed, and that the sum of the diameters of the shrinkage cavities is below a predetermined range. The inspection device 5 uses the function of the conformance prediction unit 54 to identify the cluster to which the new molded product's characteristic quantities belong, and predicts whether the condition of shrinkage cavities in the new molded product meets the inspection criteria based on the classification results of the identified cluster.

[0058] For example, if the characteristic quantities of a new molded product fall within range A shown in Figure 8, it is predicted that the shrinkage cavity condition will meet the inspection criteria. In this case, the recording unit 55 records that the molded product meets the inspection criteria in the server S, corresponding it to an identification number. At the same time, the output unit 56 outputs an instruction to the display device 2 to discharge the molded product as a passing product, notifying the operator. Alternatively, or in addition to this, an instruction to discharge the die-cast molded product as a passing product may be output to equipment that discharges die-cast molded products to the next process, such as a robot arm or an automated guided vehicle. Note that the recording to the server S and the output to the display device 2 may occur at the same time, or one may occur earlier than the other.

[0059] Conversely, if the new feature quantities of the molded product fall within range B shown in Figure 8, it is predicted that the shrinkage cavity state will not meet the inspection criteria. In this case, the recording unit 55 records that the molded product does not meet the inspection criteria in the server S, corresponding it with an identification number. At the same time, the output unit 56 outputs an instruction to the display device 2 to analyze the shrinkage cavity state of the molded product, notifying the operator. Alternatively, or in addition to this, an instruction may be output to the device that transports the die-cast molded product to the next process, instructing it to transport the die-cast molded product to the process for analyzing the shrinkage cavity state. The process for analyzing the shrinkage cavity state may be, for example, non-destructive testing using CT scanning. Note that the recording to the server S and the output to the display device 2 may occur at the same time, or one may occur earlier than the other.

[0060] [Processing in molding systems] Referring to Figure 9, the procedure for information processing by the shrinkage cavity prediction device 4 will be explained. Figure 9 is an example of a flowchart showing the information processing performed in the molding system 10 of this embodiment. The processing described below is executed by the CPU 31, which is the processor of the control device 3, when the control device 3 receives an instruction to perform processing by the shrinkage cavity prediction device 4.

[0061] First, in step S1, the relationship acquisition unit acquires a waveform showing the time change of the injection pressure value output from the die-casting apparatus 1 from the server S. In the following step S2, information on the sum of the diameters of shrinkage cavities formed inside the molded product is acquired from the server S. In step S3, the molded product identification number, the waveform of the injection pressure value, and the information on the sum of the diameters of the shrinkage cavities are linked. In the following step S4, data corresponding the sum of the diameters of the shrinkage cavities and the waveform of the injection pressure value is output. In step S5, a neural network is trained using the data output in step S4. In the following step S6, the trained neural network is used to predict the sum of the diameters of the shrinkage cavities. In the following step S7, the state prediction unit 42 calculates the correlation coefficient between the measured value and the predicted value of the sum of the diameters of the shrinkage cavities.

[0062] In step S8, it is determined whether the correlation coefficient is above a predetermined value. If it is determined that the correlation coefficient is above a predetermined value, the process proceeds to step S9, and the trained neural network is output. On the other hand, if it is determined that the correlation coefficient is below a predetermined value, the process proceeds to step S5, and the neural network is trained again.

[0063] Next, with reference to Figure 10, the procedure for information processing by the quality prediction device will be explained. Figure 10 is an example of a flowchart showing the information processing performed in the molding system 10 of this embodiment. The processing described below is executed by the CPU 31, which is the processor of the control device 3, when the control device 3 receives an instruction to perform processing by the quality prediction device.

[0064] First, in step S11, the injection pressure value output from the die-casting device is acquired by the function of the acquisition unit 51. In the following step S12, the autoencoder is trained using the acquired injection pressure value by the function of the extraction unit 52. In the following step S13, the feature quantities of the injection pressure value are extracted using the encoder of the trained autoencoder. In step S14, a feature quantity map is generated by the function of the classification unit 53. In the following step S15, the feature quantities are clustered, and in the following step S16, they are classified into clusters that meet the inspection criteria and clusters that do not.

[0065] In step S17, the extraction unit 52 acquires a new injection pressure value output from the die-casting device, and in the following step S18, the encoder of the trained autoencoder is used to extract feature quantities of the new injection pressure value. In step S19, the fitting prediction unit 54 identifies the cluster to which the extracted feature quantities belong, and in the following step S20, it is predicted whether or not the inspection criteria are met based on the classification of the identified cluster.

[0066] In step S20, the system predicts whether the identified clusters meet the inspection criteria based on their classification. In the following step S21, it determines whether the prediction is correct based on the data used for training. If the prediction is correct, the system proceeds to step S22, where the output unit 56 outputs a feature map. Conversely, if the prediction is incorrect, the system proceeds to step S11, where feature extraction and cluster generation are performed again.

[0067] Next, with reference to Figure 11, the procedure for information processing by the inspection device 5 will be described. Figure 10 is an example of a flowchart showing the information processing performed in the molding system 10 of this embodiment. The process described below is executed by the CPU 31, which is the processor of the control device 3, each time a die-cast product is formed by the die-casting device 1.

[0068] First, in step S31, the acquisition unit 51 acquires the injection pressure value output from the die-casting device. In the following step S32, the extraction unit 52 extracts feature quantities of the injection pressure value using the encoder of the trained autoencoder. In the following step S33, the fitting prediction unit 54 identifies the cluster to which the extracted feature quantities belong. In the following step S34, the classification of the identified cluster is used to predict whether or not the inspection criteria are met.

[0069] In step S35, it is determined whether the prediction result from step S34 was a prediction that the product would meet the inspection criteria. If it is predicted that the product will meet the inspection criteria, the process proceeds to step S36, where the recording unit 55 records that the molded product meets the inspection criteria in association with the identification number, and in the following step S37, the output unit 56 outputs an instruction to the display device 2 to ship out the molded product as a passing product.

[0070] If it is predicted that the product will not meet the inspection standards, the process proceeds to step S38, where the recording unit 55 records that the molded product does not meet the inspection standards, corresponding to an identification number. In the following step S39, the output unit 56 outputs an instruction to the display device to analyze the state of shrinkage cavities in the molded product.

[0071] [Embodiments of the present invention] As described above, according to this embodiment, in a method for predicting shrinkage cavities in a die-cast molded product executed by a processor, the processor uses data that associates the state of shrinkage cavities in a molded product formed using a die-casting apparatus 1 with the injection pressure waveform that shows the time change of the injection pressure value output by the die-casting apparatus 1 when the molded product is formed, to determine the correlation between the state of shrinkage cavities and the injection pressure waveform, and when a new molded product is formed using the die-casting apparatus 1, it uses the injection pressure waveform output when the new molded product is formed and the correlation to predict the state of shrinkage cavities in the new molded product. This provides a method for predicting shrinkage cavities based on data output from the die-casting apparatus 1.

[0072] Furthermore, according to the shrinkage cavity prediction method of this embodiment, if the die-casting apparatus 1 injects molten metal into the cavity of the mold at a first pressure, and then injects the molten metal into the cavity at a second pressure higher than the first pressure, the processor may use a waveform showing the time change of the first pressure as the injection pressure waveform. This makes it possible to detect the occurrence of defects at an early stage when shrinkage cavity defects begin to occur in the die-cast molded product.

[0073] Furthermore, according to the shrinkage cavity prediction method of this embodiment, the processor may also determine the correlation using data that correlates the sum of the diameters of the shrinkage cavities generated inside the molded product with the injection pressure waveform. This makes it possible to obtain a correlation with a higher correlation coefficient.

[0074] Furthermore, according to the shrinkage cavity prediction method of this embodiment, the data relating the sum of the diameters of shrinkage cavities formed inside the molded product to the injection pressure waveform may be data in which the correlation coefficient between the sum of the diameters corresponding to the injection pressure waveform and the sum of the diameters predicted from the injection pressure waveform using the correlation relationship is 0.8 or higher. This makes it possible to suppress the use of data that incorrectly detects the state of shrinkage cavities.

[0075] Furthermore, according to this embodiment, in a quality prediction method for die-cast molded products executed by a processor, the processor pre-extracts feature quantities from the injection pressure value output from the die-casting apparatus 1, pre-classifies the feature quantities into a plurality of clusters, pre-classifies the plurality of clusters into clusters that satisfy the inspection criteria and clusters that do not satisfy the inspection criteria based on information as to whether the molded product molded using the die-casting apparatus 1 satisfies the inspection criteria, and when a new molded product is molded using the die-casting apparatus 1, it extracts new feature quantities from the new injection pressure value output from the die-casting apparatus 1, identifies the cluster to which the new feature quantities belong, and predicts whether the state of the shrinkage cavities of the new molded product satisfies the inspection criteria based on the classification results of the identified clusters. This provides a quality prediction method that makes it possible to predict the state of shrinkage cavities based on data output from the die-casting apparatus 1.

[0076] Furthermore, according to the quality prediction method of this embodiment, the processor may train an autoencoder using the injection pressure value output from the die-casting apparatus 1, and extract the feature quantities from the injection pressure value using the encoder of the autoencoder. This allows for more accurate extraction of feature quantities.

[0077] Furthermore, according to the quality prediction method of this embodiment, if the die-casting apparatus 1 injects molten metal into the cavity of the mold at a first pressure, and then injects the molten metal into the cavity at a second pressure higher than the first pressure, the processor may use the value of the first pressure as the injection pressure value. This makes it possible to detect the occurrence of defects at an early stage when shrinkage cavities begin to form in the die-cast product.

[0078] Furthermore, according to the quality prediction method of this embodiment, the processor may use data that correlates the sum of the diameters of shrinkage cavities formed inside the molded product with an injection pressure waveform that shows the time change of the injection pressure value to determine the correlation between the state of the shrinkage cavities and the injection pressure waveform, and generate a plurality of clusters using the injection pressure value in which the correlation coefficient between the sum of the diameters corresponding to the injection pressure waveform and the sum of the diameters predicted from the injection pressure waveform using the correlation is 0.8 or more. This makes it possible to generate appropriate clusters.

[0079] Furthermore, according to this embodiment, an inspection method for die-cast molded products is provided, which is executed by the processor, in which an identification number assigned to each molded product and an injection pressure value at the time the molded product was molded, output in association with the identification number are acquired, and the quality prediction method described above is used to predict whether or not the state of the shrinkage cavity meets the inspection criteria, and if it is predicted that the state of the shrinkage cavity meets the inspection criteria, the fact that the molded product meets the inspection criteria is recorded in association with the identification number and an instruction is output to ship the molded product out as a passing product, and if it is predicted that the state of the shrinkage cavity does not meet the inspection criteria, the fact that the molded product does not meet the inspection criteria is recorded in association with the identification number and an instruction is output to analyze the state of the shrinkage cavity of the molded product. This makes it possible to suppress the shipment of die-cast molded products that are predicted not to meet the inspection criteria as passing products.

[0080] Furthermore, according to this embodiment, a shrinkage cavity prediction device 4 is provided, which includes a relationship acquisition unit 41 that determines the correlation between the state of shrinkage cavities and the injection pressure waveform, using data that correlates the state of shrinkage cavities of a molded product molded using a die-casting device 1 with the injection pressure waveform, which shows the time change of the injection pressure value output when the die-casting device 1 molds the molded product, and the state prediction unit 42 that predicts the state of shrinkage cavities of a new molded product using the injection pressure waveform output when a new molded product is molded using the new molded product and the correlation. This makes it possible to predict the state of shrinkage cavities based on data output from the die-casting device 1.

[0081] Furthermore, according to this embodiment, a quality prediction device is provided which includes: an extraction unit 52 that extracts feature quantities from injection pressure values ​​output from a die-casting device 1; a classification unit 53 that classifies the feature quantities into a plurality of clusters and, based on information on whether or not the molded product formed using the die-casting device 1 satisfies the inspection criteria for shrinkage cavities, classifies the plurality of clusters into clusters that satisfy the inspection criteria and clusters that do not satisfy the inspection criteria; and a conformance prediction unit 54 that predicts whether or not the shrinkage cavities of the molded product satisfy the inspection criteria based on the classification results of the clusters. When a new molded product is formed using the die-casting device 1, the extraction unit 52 extracts new feature quantities from the new injection pressure values ​​output from the die-casting device 1; the conformance prediction unit 54 identifies the cluster to which the new feature quantities belong and predicts whether or not the shrinkage cavities of the new molded product satisfy the inspection criteria based on the classification results of the identified clusters. This makes it possible to predict the state of shrinkage cavities based on data output from the die-casting device 1. [Explanation of symbols]

[0082] 10…Molding System 1…Die casting equipment 2...Display device 3…Control device 31…CPU (Processor) 32...ROM 33…RAM 4. Nesting prediction device 41...Relationship Acquisition Department 42... State prediction unit 5…Inspection equipment (quality prediction equipment) 51…Acquisition part 52...Extraction part 53...Classification section 54...Suitability prediction unit 55... Records Department 56…Output section S... Server X, Y... range

Claims

1. In a method for predicting shrinkage cavities in die-cast molded products, which is performed by a processor, The aforementioned processor, Using data that correlates the state of shrinkage cavities in a molded product formed using a die-casting apparatus with the injection pressure waveform, which shows the time change in the injection pressure value output by the die-casting apparatus when it molded the product, the correlation between the state of shrinkage cavities and the injection pressure waveform is determined. When forming a new molded product using the die-casting apparatus, the state of shrinkage cavities in the new molded product is predicted using the injection pressure waveform output when the new molded product is formed and the correlation relationship. A shrinkage cavity prediction method in which, when the die-casting apparatus injects molten metal into the cavity of a mold at a first pressure, and then injects the molten metal into the cavity at a second pressure higher than the first pressure, the waveform showing the time change of the first pressure is used as the injection pressure waveform.

2. The shrinkage cavity prediction method according to claim 1, wherein the processor determines the correlation using data that correlates the sum of the diameters of shrinkage cavities formed inside the molded product with the injection pressure waveform.

3. The shrinkage cavity prediction method according to claim 2, wherein the data relating the sum of the diameters of shrinkage cavities formed inside the molded product to the injection pressure waveform is data in which the correlation coefficient between the sum of the diameters corresponding to the injection pressure waveform and the sum of the diameters predicted from the injection pressure waveform using the correlation relationship is 0.8 or more.

4. In a method for predicting the quality of die-cast molded products, which is performed by a processor, The aforementioned processor, Feature quantities are pre-extracted from the injection pressure values ​​output from the die-casting machine, The aforementioned features are pre-classified into multiple clusters, Based on information regarding whether the molded product formed using the die-casting apparatus meets the inspection criteria for shrinkage cavities, the clusters are pre-classified into clusters that meet the inspection criteria and clusters that do not meet the inspection criteria. When forming a new molded product using the die-casting apparatus, a new characteristic quantity is extracted from the new injection pressure value output from the die-casting apparatus. Identify the cluster to which the new feature belongs, A quality prediction method that predicts whether the state of shrinkage cavities in a new molded product meets the inspection criteria based on the classification results of the identified clusters.

5. The aforementioned processor, The autoencoder is trained using the injection pressure value output from the die-casting apparatus. The quality prediction method according to claim 4, wherein the feature quantity is extracted from the injection pressure value using the encoder of the autoencoder.

6. The quality prediction method according to claim 4, wherein the processor, after the die-casting apparatus injects molten metal into the cavity of the mold at a first pressure, injects the molten metal into the cavity at a second pressure higher than the first pressure, uses the value of the first pressure as the injection pressure value.

7. The aforementioned processor, Using data that correlates the sum of the diameters of shrinkage cavities formed inside the molded product with the injection pressure waveform showing the time change of the injection pressure value, the correlation between the state of the shrinkage cavities and the injection pressure waveform is determined. The quality prediction method according to claim 4, wherein a plurality of clusters are generated using injection pressure values ​​such that the correlation coefficient between the sum of diameters corresponding to the injection pressure waveform and the sum of diameters predicted from the injection pressure waveform using the correlation relationship is 0.8 or more.

8. The identification number assigned to each molded product and the injection pressure value used when the molded product was formed, which is output in association with the identification number, are obtained. The quality prediction method described in any one of claims 4 to 7 predicts whether the state of the shrinkage cavity meets the inspection criteria, If it is predicted that the shrinkage cavity condition meets the inspection criteria, the system records that the molded product meets the inspection criteria in correspondence with the identification number, and outputs an instruction to ship the molded product as a passing product. A method for inspecting a die-cast molded product, which is executed by the processor, wherein if it is predicted that the state of the shrinkage cavity does not meet the inspection criteria, the processor records that the molded product does not meet the inspection criteria in correspondence with the identification number, and outputs an instruction to analyze the state of the shrinkage cavity of the molded product.

9. A relationship acquisition unit that uses data that correlates the state of shrinkage cavities in a molded product formed using a die-casting device with the injection pressure waveform that shows the time change of the injection pressure value output by the die-casting device when it molded the product, to determine the correlation between the state of shrinkage cavities and the injection pressure waveform, When forming a new molded product using the die-casting apparatus, the system includes a state prediction unit that predicts the state of shrinkage cavities in the new molded product using the injection pressure waveform output when the new molded product is formed and the correlation relationship, The injection pressure waveform is a waveform that shows the time change of the first pressure when the die-casting apparatus injects molten metal into the cavity of the mold at a first pressure, and then injects the molten metal into the cavity at a second pressure higher than the first pressure, in the case of a shrinkage cavity prediction device.

10. An extraction unit that extracts feature quantities from the injection pressure value output from the die-casting machine, The aforementioned features are classified into multiple clusters, A classification unit that classifies a plurality of clusters into clusters that meet the inspection criteria and clusters that do not meet the inspection criteria, based on information as to whether the molded product formed using the die-casting apparatus meets the inspection criteria. The system includes a conformity prediction unit that predicts whether the state of shrinkage cavities in the molded product meets the inspection criteria based on the classification results of the clusters, When forming a new molded product using the die-casting apparatus, the extraction unit extracts a new feature quantity from the new injection pressure value output from the die-casting apparatus. The fitting prediction unit identifies the cluster to which the new feature belongs, A quality prediction device that predicts whether the state of shrinkage cavities in a new molded product meets the inspection criteria based on the classification results of the identified clusters.

Images