Failure cause analysis system, failure cause analysis program, and failure cause analysis method

Abstract

To provide a failure cause analysis system, a failure cause analysis program and a failure cause analysis method which enable failure rates to be efficiently reduced without depending on empirical values and gut feeling, and enable countermeasures for reducing the failure rates to be standardized.SOLUTION: A failure cause analysis system 60, which is applied to a casting molding line, collects various kinds of numeric data for manufacturing a casting mold or a casting mold poured with molten metal, from a sand processing device 10 and a casting mold-making device 20 in a casting-mold making step and a pouring device 43 in a pouring step. A failure cause analysis device 61 stores, in a data base 70, the collected various kinds of numeric data as one data group while associating the data group with each casting mold or each casting mold poured with molten metal. A data calculating unit 62b analyzes many stored data groups to extract factors showing abnormality, and calculates a degree of influence of other kinds of data on the extracted abnormal factors, for each kind of data.SELECTED DRAWING: Figure 4

Description

Technical Field

[0005]

[0001] The present invention relates to a defect cause analysis system, a defect cause analysis program, and a defect cause analysis method.

Background Art

[0002] In a series of casting lines where mold sand is molded into a mold, molten metal is poured into the molded mold, and after cooling, the mold sand is broken up to obtain a cast product, defects may occur in the cast product due to various operations performed in each process. Conventionally, the identification of factors causing such defects has been performed based on the experience and intuition of line workers.

[0003] By the way, there is a known system that analyzes product design data and defective data of products produced therefrom, and predicts the defective rate of an evaluation target product from the design data of the product, rather than factors causing defects (see Patent Document 1). However, this system predicts the defective rate itself and does not estimate what factors are causing the defects.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] When relying on the experience and intuition of line workers for factors causing defects in cast products, it is not possible to efficiently reduce the defective rate. Also, countermeasures for reducing the defective rate become personnel-dependent, and when there are no people with experience or excellent intuition, the countermeasures for reducing the defective rate become insufficient.

[0006] This invention has been made in view of the above problems, and its main objective is to provide a defect cause analysis system, a defect cause analysis program, and a defect cause analysis method that can efficiently reduce the defect rate without relying on experience or intuition, and that can standardize measures to reduce the defect rate. [Means for solving the problem]

[0007] To achieve the above objective, the defect cause analysis system of the first invention is: A defect cause analysis system applied to a casting line that performs a molding process of shaping mold sand to form a mold, a pouring process of melting a molten material to generate molten metal and pouring it into the molded mold to form a poured mold, a cooling process of cooling the poured mold to form a cooled mold after pouring, and a mold dismantling process of dismantling the cooled mold after pouring to obtain a cast product, wherein the system analyzes the cause of defects in the cast product that occurred in the molding process or the pouring process, The mold obtained by the molding process or the poured mold obtained by the pouring process is used as an intermediate product. Regarding the various manufacturing information obtained when manufacturing the aforementioned intermediate product, a data collection means collects numerical data for each intermediate product that is manufactured sequentially, A storage means for storing each collected numerical data as a single data group linked to the individual intermediate product, An abnormality factor extraction means analyzes a large number of data sets stored in the storage means and extracts factors indicating abnormalities from the various manufacturing information, An influence calculation means that analyzes the aforementioned large number of data sets and calculates the degree of influence of other types of manufacturing information on the extracted abnormal factors for each type of manufacturing information, It is characterized by having the following features.

[0008] The defect cause analysis system of the second invention is: The storage means stores the defect rate of the group of cooled molds after pouring, which was determined by inspecting the cast products obtained from the multiple cooled molds after pouring, and associates it with the intermediate product that served as the basis for each individual cooled mold that constitutes the group of cooled molds after pouring. The abnormal factor extraction means is characterized by analyzing the data set from the large number of data sets in which the defect rate exceeds the defect threshold, and extracting abnormal factors.

[0009] The third invention's defect cause analysis system is: The abnormal factor extraction means is characterized by classifying the numerous data sets into groups according to the defect rate, analyzing the data sets belonging to each group that exceeds the defect threshold, and extracting abnormal factors from the various manufacturing information.

[0010] The fourth invention's defect cause analysis system is: The anomaly factor extraction means is characterized by extracting specific manufacturing information as anomaly factors, where the number of occurrences exceeding a first anomaly discrimination threshold exceeds a second anomaly discrimination threshold, through unsupervised learning using the large number of data sets.

[0011] The fifth invention's defect cause analysis system is: The influence calculation means is characterized by using a model generated by supervised learning with respect to the manufacturing information extracted as the abnormal factor as training data, and calculating the influence of other types of manufacturing information on the abnormal factor.

[0012] The defect cause analysis system of the sixth invention is: When the intermediate product is the pre-pouring mold, the various manufacturing information used when manufacturing the pre-pouring mold is characterized by including information on the composition of the molten metal poured into the pre-pouring mold, information on the receiving of the molten metal when a ladle receives the molten metal, information on the pouring of the molten metal from the ladle to a single mold, and information on the remaining molten metal after the pouring of the molten metal from one ladle to a set number of molds has been completed.

[0013] The seventh invention's defect cause analysis system is: When the intermediate product is the mold, the various manufacturing information used when manufacturing the mold includes sand characteristic information relating to the properties of the mold sand, sand generation device information relating to a mold sand generation device that generates the mold sand, and molding device information relating to a molding device used when molding one mold.

[0014] The eighth invention is a defect cause analysis system, The storage means is characterized in that, for each mold, in addition to mold sand information relating to the mold sand used as the material of the mold and sand generation device information relating to the production of the mold sand, it also stores and includes in the data group mold sand information relating to another mold sand used as the material of a mold made before the mold and sand generation device information relating to the production of that other mold sand.

[0015] The failure cause analysis program of the ninth invention is: This invention relates to a casting line that performs the following steps: a molding step of shaping mold sand to create a mold; a pouring step of melting a molten material to generate molten metal and pouring it into the molded mold to create a poured mold; a cooling step of cooling the poured mold to create a cooled mold after pouring; and a mold dismantling step of dismantling the cooled mold after pouring to obtain a cast product. The invention relates to a defect cause analysis program that analyzes the cause of defects in the cast product that occurred in the molding step or the pouring step. The mold obtained by the molding process or the poured mold obtained by the pouring process is used as an intermediate product. The steps include: collecting various manufacturing information obtained when manufacturing the aforementioned intermediate product, and collecting numerical data for each individual intermediate product that is manufactured sequentially; The steps include: storing each collected numerical data as a single data set linked to the individual intermediate product; The steps include analyzing a large number of stored data sets and extracting factors indicating abnormalities from the various manufacturing information, The steps include analyzing the aforementioned large number of data sets and calculating the degree of influence of other types of manufacturing information on the extracted abnormal factors for each type of manufacturing information, It is characterized by causing the execution of the following:

[0016] The defect cause analysis method of the tenth invention is applied to a casting line that performs a molding process of molding mold sand into a mold, a melting process of melting a melting material to generate molten metal, a pouring process of pouring the molten metal into the molded mold to obtain a poured mold, a cooling process of cooling the poured mold to obtain a cooled mold after pouring, and a shake-out process of shake-out the cooled mold after pouring to obtain a casting product, and is a defect cause analysis method for analyzing the cause of generating a defect in the casting product in the molding process or the pouring process, using the mold obtained by the molding process or the poured mold obtained by the pouring process as an intermediate product, a data collection process of collecting respective numerical data when manufacturing each of the sequentially manufactured individual intermediate products for various manufacturing information obtained when manufacturing the intermediate product, a storage process of associating and storing each collected numerical data as a data group with the individual intermediate products, an abnormal factor extraction process of analyzing a large number of stored data groups and extracting factors indicating abnormalities from among the various manufacturing information, an influence degree calculation process of analyzing the large number of data groups and calculating the influence degree of other types of manufacturing information on the extracted abnormal factors for each type of the manufacturing information, characterized by comprising the above.

Effect of the Invention

[0017] According to the first invention, in the molding process or the pouring process, by analyzing the data groups obtained each time an intermediate product is manufactured, abnormal factors are extracted from the manufacturing information obtained when manufacturing the intermediate product.

[0018] In a casting line, specific manufacturing information for producing a single intermediate product is influenced by various other types of manufacturing information. For example, if we consider the pouring time, which is the time from the start to the completion of pouring into a mold, it's not as simple as just adjusting the drive settings of the pouring device to manipulate the numerical data for the pouring time. Numerical data from various other types of manufacturing information, such as pouring temperature, received weight, and remaining weight, influence the numerical data for the pouring time, and these can all affect the numerical data for the pouring time. Therefore, simply extracting abnormal factors from various types of manufacturing information is insufficient for analyzing the cause of defects.

[0019] Therefore, in the first invention, after extracting abnormal factors, the degree of influence of other types of manufacturing information on those abnormal factors is calculated. By estimating the manufacturing information that is causing the defect rate of cast products from this influence, the defect rate can be reduced by improving the estimated manufacturing information. As a result, the defect rate can be reduced efficiently without relying on experience or intuition, and measures to reduce the defect rate can be standardized.

[0020] According to the second invention, the defect rate of the cast products, determined for each group of molds that have cooled after pouring, is linked to a data set used when manufacturing a single intermediate product. Therefore, by analyzing the data set in which the defect rate exceeds the defect threshold, it is possible to extract the abnormal factors that are causing the defect rate to exceed that threshold. As a result, the analysis of data sets with low defect rates is not performed, and the process of extracting abnormal factors can be carried out efficiently.

[0021] According to the third invention, among the groups classified by defect rate, abnormal factors are extracted by analyzing the data set belonging to each group that exceeds the defect threshold. This allows for the extraction of abnormal factors that could not be extracted when analyzing the entire dataset together. This helps to further reduce the defect rate.

[0022] According to the fourth invention, in the various manufacturing information obtained each time an intermediate product is manufactured, it is unclear which information is abnormal or not. However, by using an unsupervised learning method for anomaly detection, abnormal factors can be suitably extracted.

[0023] According to the fifth invention, the influence of other types of manufacturing information can be suitably calculated for each type of manufacturing information using a model generated by supervised learning with the extracted anomaly factors as training data.

[0024] According to the sixth invention, since it covers manufacturing information in the pouring process that can affect the molten metal defect rate (defect rate related to molten metal), the accuracy of extracting abnormal factors in the pouring process and calculating the degree of influence of those abnormal factors can be improved. This makes it possible to improve the accuracy of analyzing the causes of defects.

[0025] According to the seventh invention, since it covers manufacturing information in the molding process that may affect the sand defect rate (defect rate related to sand), the accuracy of extracting abnormal factors in the molding process and calculating the degree of influence of those abnormal factors can be improved. This makes it possible to improve the accuracy of defect cause analysis.

[0026] According to the eighth invention, mold sand information relating to another mold sand used as the material for a previously molded mold, and sand generation device information relating to the generation of said other mold sand are also included in the data set and used for extracting abnormal factors and calculating their impact. When molding a mold, there is a possibility that not only the mold sand used for molding that mold is mixed in, but also other mold sand used as the material for a previous mold. Therefore, by including information on that other mold sand in the analysis, the accuracy of extracting abnormal factors in the molding process and calculating the impact of said abnormal factors can be further improved. This makes it possible to further improve the accuracy of analyzing the causes of defects.

[0027] According to the ninth invention, by operating a failure cause analysis system incorporating this program, the same effects as those obtained by the failure cause analysis system of the first invention can be obtained.

[0028] According to the tenth invention, by performing the defect cause analysis method using a defect cause analysis system, the same effects as those obtained by the defect cause analysis system of the first invention can be obtained. [Brief explanation of the drawing]

[0029] [Figure 1] A schematic diagram showing the general layout of the casting line. [Figure 2] A schematic diagram showing the general process of manufacturing mold sand. [Figure 3] A schematic diagram showing the general structure of the molding part in a molding machine. [Figure 4] A block diagram showing the overall configuration of the defect cause analysis system. [Figure 5] A block diagram showing the database structure. [Figure 6] A flowchart illustrating the analysis process using a defect cause analysis device. [Figure 7] A schematic diagram showing an example of a binding process for sand with poor quality. [Figure 8] A schematic diagram showing an example of a binding process for defective molten metal. [Modes for carrying out the invention]

[0030] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[0031] The defect cause analysis system, defect cause analysis program, and defect cause analysis method of this embodiment are applied to a casting line that manufactures cast products.

[0032] [Overview of the casting line] First, let's explain the outline of the casting line. As shown in Figure 1, the casting line performs four main processes. First, a molding process is performed in which the mold sand is shaped to form a mold M. Next, a pouring process is performed in which the molten metal is generated by melting the molten material and poured into the molded mold M to form a poured mold Ma. Here, the generation of molten metal and the pouring are described together as the pouring process. Next, a cooling process is performed in which the poured mold Ma is cooled to form a post-cooled mold Mb. After that, a mold dismantling process is performed in which the post-cooled mold Mb is dismantled to obtain the cast product C.

[0033] In the molding process, a sand processing device 10 is used, as shown in Figure 2. The sand processing device 10 includes a mixing device 11, a device for inputting sand, water, additives, etc. (not shown), etc. The recovered sand obtained from mold dismantling in the mold dismantling process, along with water, additives, and newly supplied sand, is put into the mixing device 11 and mixed to produce mold sand S. The sand processing device 10 corresponds to a mold sand generation device. The mold sand S produced in the sand processing device 10 is sent to the conveying device 13 via the tank 12, and then transported by the conveying device 13 to the tank 21 of the molding device 20.

[0034] The molding apparatus 20 can be of any type, but for example, a vertical molding method is used. Figure 3(a) shows a schematic of the molding space 22 in a molding apparatus 20 of this type. As shown in Figure 3(a), the molding apparatus 20 has a front mold 23 and a rear mold 24, which are molds for molding the molding sand S, arranged horizontally opposite each other, and a molding space 22 is formed between the front mold 23 and the rear mold 24. A tank 21 (see Figure 2) for storing the molding sand S is provided above the molding space 22, and the molding sand S is introduced from the tank 21 into the molding space 22 by applying air pressure to the tank 21 (air blow). Then, as shown in Figure 3(b), the front mold 23 and the rear mold 24 are brought closer to each other to compress the introduced molding sand S and generate a partial mold P. Next, the two molds 23 and 24 are separated to obtain a single partial mold P (here, the middle partial mold P2, which will be described later) as shown in Figure 3(c).

[0035] Here, a partial mold P refers to a part that constitutes the mold M. Multiple partial molds P are stacked together to form the completed mold M. Figure 3(d) shows an example of a mold M having a casting space 31 for casting three cast products C. In this case, the mold M has four partial molds P: a lower partial mold P1, a middle partial mold P2, an upper partial mold P3, and a lid partial mold P4. As shown in Figure 3(c), each formed partial mold P is rotated 90 degrees and stacked sequentially within the frame 32 to form the mold M as an intermediate product in the molding process.

[0036] Of the partial molds P, those located between the lower partial mold P1, which is positioned at the bottom, and the lid partial mold P4, which is positioned at the top (in the example of Figure 3, the middle partial mold P2 and the upper partial mold P3) are molded one at a time. The lower partial mold P1 and the lid partial mold P4 are molded simultaneously by providing a sprue cutting plate (not shown) in the molding space 22 between the front mold 23 and the rear mold 24.

[0037] If the amount of mold sand S produced at one time by the sand processing device 10 is considered as one batch, then in the molding process, multiple pre-set molds M are molded using one batch of mold sand S, and the generated molds M are sequentially discharged from the molding device 20. Once the sand processing of one batch by the sand processing device 10 is complete, the sand processing of the next batch is carried out, and the sand is transported to the molding device 20, where molds M are molded using the next batch of mold sand S. The molds M sequentially discharged from the molding device 20 are transported sequentially to the pouring process that follows the molding process by a conveying device such as a conveyor (not shown).

[0038] It should be noted that after using up one batch of mold sand S, the sand processing device 10 and the molding device 20 are not completely cleaned to generate new mold sand S. Therefore, when molding multiple molds M using one batch of mold sand S, and then molding another mold M using the next batch of mold sand S, the mold M will contain mold sand S from previous batches, such as the previous batch and the batch before that. In Figure 2, to schematically illustrate this, the mold sand S of the current batch is shown as a circle, the mold sand S of the previous batch is shown as a square, and the mold sand S of the batch before that is shown as a triangle. This illustration does not exclude the case where mold sand S from batches older than the batch before that is included.

[0039] Next, the pouring process includes a process of generating molten metal and a process of pouring the generated molten metal into the mold M. As shown in Figure 1, in the process of generating molten metal, as is generally known, molten metal is generated by melting molten materials such as pig iron and scrap iron. A cupola 41 is used as a melting furnace to melt the molten materials. Coke, which is the combustion material, and the molten materials are sequentially fed into the cupola 41 from the top, and molten metal is generated at the bottom of the cupola 41 by burning the coke with a combustion burner. If the unit of molten metal generated at one time by the cupola 41 is defined as one batch, then in the process of generating molten metal, the molten metal from one batch is used to pour the molten metal into the ladle 42 a predetermined number of times.

[0040] In the process of pouring molten metal into the mold M, a pouring device 43 is used. The pouring device 43 has a ladle 42, and the molten metal generated in the cupola 41 is poured into the ladle 42 after impurities have been separated. The ladle 42, filled with molten metal, is transported to the pouring position. At the pouring position, the ladle 42 is tilted and poured into the mold M, which has been transported from the molding device 20 of the molding process to the pouring position. If the mold M after pouring is considered a poured mold Ma as an intermediate product in the pouring process, the poured mold Ma is sent out from the pouring position by a transport device (not shown), and the next mold M is introduced to the pouring position. In this way, the mold Ms that have been transported from the molding device 20 of the molding process are sequentially transported to the pouring position, where molten metal is poured from the ladle 42, and when pouring is complete, they are sent out. After the ladle 42, filled with molten metal from the cupola 41 in a single transfer, is transported to the pouring position, the number of molds M that will receive molten metal from the ladle 42 is predetermined. Once the pouring of molten metal into the predetermined number of molds M is complete, the ladle 42 returns to the receiving position and receives molten metal from the cupola 41 again. After receiving the molten metal, it is transported back to the pouring position and poured into the molds M.

[0041] The poured mold Ma, sent from the pouring position, is transported to the dismantling position by a transport device (not shown) while being cooled. The transport process is the cooling process, during which the molten metal solidifies, resulting in the cooled mold Mb after pouring.

[0042] In the mold dismantling process, the mold Mb, which has cooled after pouring, is dismantled using a mold dismantling device 50 to obtain the cast product C. Note that the mold dismantling device 50 shown in Figure 1 is just one example. The mold sand S dismantled in this mold dismantling process is recovered and, as mentioned above, reused as mold sand S for creating a new mold M.

[0043] [Defect Cause Analysis System] Next, we will describe a defect cause analysis system 60 that applies to the above-described casting line and analyzes the causes of defects in the cast product C from among several factors involved in the manufacturing of the mold M in the molding process and the poured mold Ma in the pouring process.

[0044] As shown in Figure 4, the defect cause analysis system 60 includes a defect cause analysis device 61. The defect cause analysis device 61 has a system control unit 62 that controls the system, a memory 63, a data storage device 64, a data input unit 65, an information display unit 66, and a network interface 67. These are connected via a data bus 68, enabling them to send and receive information from each other. The system control unit 62 is a microcomputer composed of a CPU and the like. The memory 63 is a non-volatile storage unit that stores the defect cause analysis program 69 executed by the defect cause analysis system 60. The data storage device 64 stores a database 70 as a storage means. The network interface 67 is an interface unit for connecting the defect cause analysis device 61 to the sand processing device 10 and the molding device 20 provided in the molding process, and to the cupola 41 and the pouring device 43 provided in the pouring process. Therefore, the network interface 67 corresponds to the data collection means.

[0045] The sand processing device 10 is equipped with a weight measuring device for measuring the weight of the material to be fed into the mixing device 11, a moisture meter for measuring the moisture content of the mold sand S, a temperature measuring device for measuring the temperature of the mold sand S, a CB value measuring device for measuring the compactability (CB value) of the mold sand S, and a timer for measuring the mixing time, etc. Illustrations of each of these devices are omitted. Numerical data is measured by these devices each time a batch of mold sand S is produced, and this measured numerical data is transmitted to the defect cause analysis device 61. The information transmitted from the sand processing device 10 to the defect cause analysis device 61 includes not only the numerical data measured by the various measuring devices, but also numerical data that are preset values ​​in the sand processing device 10, such as the target value of the mold sand S to be fed in and the target value of the moisture content.

[0046] The molding apparatus 20 is equipped with an air pressure measuring device for measuring the air pressure when introducing the molding sand S, a hydraulic pressure measuring device for measuring the squeeze pressure that presses the front mold 23 and rear mold 24, a level sensor for measuring the displacement (compression ratio) of the front mold 23 and rear mold 24 and the thickness of the partial mold P, and a timer for measuring various times. Illustrations of these devices are omitted. Each time multiple partial molds P are molded by the molding apparatus 20, numerical data is measured by these devices and transmitted to the defect cause analysis device 61. The numerical data transmitted from the molding apparatus 20 to the defect cause analysis device 61 includes not only numerical data measured by the various measuring devices, but also numerical data that are preset values ​​in the molding apparatus 20, such as the time of air blowing to introduce the molding sand S and the holding time of the pressing state by the front mold 23 and rear mold 24.

[0047] The pouring device 43 is equipped with a weight measuring device for measuring the weight of the ladle 42, a thermal camera for photographing the pouring process from the ladle 42, and a timer for measuring various times. These devices are not shown in the diagram. Numerical data is measured by these devices not only when receiving molten metal from the cupola 41, but also each time molten metal is poured into each mold M, and this measured numerical data is transmitted to the defect cause analysis device 61. The information transmitted from the pouring device 43 to the defect cause analysis device 61 includes not only numerical data measured by the various measuring devices, but also numerical data that are preset values ​​in the pouring device 43, such as the angular velocity when tilting the ladle 42 and the pouring angle.

[0048] The data processing unit 62a of the system control unit 62 stores numerical data of various information collected from the sand processing device 10, molding device 20, cupola 41, and pouring device 43 via the network interface 67 into the database 70 of the data storage device 64. This process corresponds to the data collection and storage steps in the defect cause analysis method. As shown in Figure 5, the database 70 is composed of a mold sand database 71, a molding database 72, a molten metal database 73, a pouring metal database 74, a sand defect rate database 75, and a molten metal defect rate database 76.

[0049] The mold sand database 71 stores numerical data related to the production of molds M in the molding process, specifically information about mold sand S and its generation. Information related to the generation of mold sand S includes, for example, the weight of the recovered sand added, the weight of the replenishment sand added, the weight of the water added, the weight of the additives added, the moisture content of the mold sand S, the temperature, the CB value, and the mixing time by the mixing device 11. This information is called sand generation device information. For each batch of sand processing, the individual numerical data of sand generation device information obtained in that batch of sand processing is successively stored in the mold sand database 71.

[0050] Furthermore, information regarding the mold sand S includes various characteristic information of the mold sand S obtained from one batch of sand processing, such as sand particle size, total clay content, pressure resistance, and shear force. Numerical data for each sand characteristic is obtained by analyzing the generated mold sand S in batch units, and after the analysis results are known, the data is entered by the operator using the data input unit 65 or obtained from a sand characteristic analysis device connected to the network interface 67. In this way, the mold sand database 71 stores individual numerical data of sand generation device information and sand characteristic information obtained in the sand processing of a batch, on a batch-by-batch basis.

[0051] The molding database 72 stores numerical data related to the molding of the mold M, which is part of the manufacturing information involved in the production of the mold M in the molding process. Information related to the molding of the mold M includes, for example, the air blow pressure when introducing the molding sand S, the air blow time, the squeeze pressure of the front mold 23 and the rear mold 24, the displacement (compression ratio) of the front mold 23 and the rear mold 24, the forward time and backward time of the front mold 23 and the rear mold 24, and the pressing and holding time by the mold. In addition, there is the total mold displacement, which is the average value of the front mold displacement and the rear mold displacement, the thickness from the front-to-back center X between the molds to the front or rear end of the partial mold P (the former is called the front mold thickness, and the latter is called the rear mold thickness), the partial thickness of the partial mold P, and the overall height of the completed mold M. This information is called molding device information. The molding device information includes measured values ​​and set values, such as air blow pressure.

[0052] Each time a partial mold P is fabricated, individual numerical data of the fabrication equipment information obtained during that fabrication is successively stored in the fabrication database 72. Then, each numerical data from the fabrication of the partial mold P is stored together as a single value for one mold M that is fabricated using that partial mold P. In this way, the fabrication database 72 stores individual numerical data of the fabrication equipment information obtained during the fabrication of one mold M (finished mold) unit.

[0053] Incidentally, the numerical data for the displacement of the front mold 23 and rear mold 24, and the mold thickness, are not stored directly from the numerical data obtained from the level sensor; rather, predetermined calculations using the numerical data are required. For example, for numerical data of the front mold displacement and rear mold displacement, it is necessary to calculate the difference between the retracted position data and the forward position data obtained from the level sensor. For numerical data of the total mold displacement, it is necessary to calculate the average value of the numerical data for the front mold displacement and the numerical data for the rear mold displacement. Also, for numerical data of the front mold thickness and rear mold thickness, it is necessary to obtain the position data of the front or rear end of the partial mold P from the level sensor, using the front-to-rear center X between the molds as a reference, and calculate the distance between the reference and the reference. For numerical data of the partial thickness of the partial mold P, it is necessary to calculate the sum of the numerical data for the front mold thickness and the numerical data for the rear mold thickness information. The data processing unit 62a performs these calculations and then stores each numerical data in the molding database 72.

[0054] The molten metal database 73 stores numerical data on molten metal generation, which occurs during the pouring process when the molten metal material is melted to produce molten metal. This information on molten metal generation is one of the pieces of information related to the manufacturing of the poured mold Ma. This information includes, for example, component information about the molten metal, the temperature of the generated molten metal, and information about the cupola 41, such as the weight of the generated molten metal. Numerical data on the molten metal's component information is obtained by analyzing the molten metal in batches. After the analysis results are known, the data is entered by an operator using the data input unit 65 or obtained from a component analyzer connected to the network interface 67. Information on the molten metal temperature and the cupola 41 is also obtained in batches. Therefore, the molten metal database 73 stores numerical data on a batch basis, including molten metal component information, molten metal temperature, and information about the cupola 41.

[0055] The pouring database 74 stores numerical data of information regarding the pouring of molten metal into the mold M during the pouring process. This pouring information is also part of the information related to the manufacturing of the poured mold Ma. This information includes, for example, the pouring time from the start to the completion (draining) of pouring into one mold M, the pouring weight (the weight of molten metal poured into one mold M), and the pouring temperature (the temperature of the molten metal poured into one mold M). Numerical data for pouring time and pouring temperature are obtained from image information captured by a thermal camera. When the image information is transmitted to the defect cause analysis device 61, the data processing unit 62a analyzes the image information and obtains numerical data for pouring time and pouring temperature from the analysis results. Alternatively, the pouring device 43 may perform image analysis and transmit the resulting numerical data for pouring time and pouring temperature to the defect cause analysis device 61. Other examples of information related to the manufacturing of the poured mold Ma include various setting values ​​for the tilting of the ladle 42 for each pour, such as the angular velocity and draining angle when tilting the ladle 42. This information is referred to as pouring information.

[0056] Another example of information during pouring is the weight of molten metal received by the ladle 42 from the cupola 41. The received weight corresponds to the received weight information. Yet another example is the remaining weight of molten metal, which is the weight of the molten metal remaining in the ladle 42 after the pouring of molten metal from the ladle 42 into a predetermined number of molds M has been completed. The remaining weight is one of the remaining weight information. Since the numerical data for the received weight and the remaining weight are common to the predetermined number of molds M, the data processing unit 62a assigns the same numerical data to each of the predetermined number of molds M. In this way, the pouring database 74 stores the numerical data of the information during pouring obtained in the pouring of molten metal into each mold M.

[0057] Furthermore, the information stored in database 70 includes not only measured values ​​but also set values, such as various setting values ​​related to the tilting of the ladle 42. Even with set values, operators may change the set values ​​depending on the line conditions at the time, so the same set values ​​are not always used during line operation. For this reason, information of the type in which numerical data is set in advance is also stored in database 70 as one of the manufacturing information that may cause defects, and is included in the analysis of the cause of defects.

[0058] The sand defect rate database 75 and the molten metal defect rate database 76 store numerical data on the defect rate of cast products C manufactured by the casting line. In the casting line, a group of post-cooled molds Mb consisting of multiple post-cooled molds is disassembled, and the resulting cast products C are made into one lot. Each cast product C in each lot is inspected to determine whether or not it is defective. The number of post-cooled molds Mb that make up one group of post-cooled molds (one lot) is set in advance to be a least common multiple of the number of molds M made using one batch of mold sand S and the number of molds M poured with one batch of molten metal.

[0059] The types of defects include sand defects caused by the mold M before pouring and pouring defects caused by the pouring process, and each is inspected separately. Once the number of cast products C with sand defects and the number of cast products C with pouring defects for a given lot are determined, the operator inputs these numbers into the data input unit 65. The data processing unit 62a uses these input values ​​to calculate the defect rate for sand defects from the ratio of the number of cooled molds Mb after pouring to the number of sand defects in one lot, and similarly calculates the defect rate for molten metal defects from the ratio of the number of cooled molds Mb after pouring to the number of molten metal defects in one lot. The data processing unit 62a stores the calculated defect rate data for each lot in the sand defect rate database 75 and the molten metal defect rate database 76.

[0060] The system control unit 62 has a data calculation unit 62b in addition to the data processing unit 62a. The data calculation unit 62b performs calculations for failure cause analysis using a large amount of numerical data stored in the database 70.

[0061] [Defect cause analysis process using defect cause analysis equipment] Next, we will explain the defect cause analysis process, which analyzes the causes of sand defects and molten metal defects in the cast product C based on numerical data of various information collected during the molding and pouring processes. As shown in Figure 6, the defect cause analysis process performs data preprocessing in step S11, abnormal factor extraction in step S12, influence calculation in step S13, and influence display in step S14. This process is mainly performed by the system control unit 62 of the defect cause analysis device 61, according to the defect cause analysis program 69 stored in memory 63.

[0062] [Data preprocessing] The data preprocessing in step S11 is performed by the data processing unit 62a of the system control unit 62. Here, the data calculation unit 62b organizes the numerical data of various manufacturing information stored in the database 70 into a state in which it can perform calculation processing using the numerical data.

[0063] Data preprocessing includes the linking of numerical data. The linking process in step S111 is divided into two parts: a linking process for sand defects, which is performed to analyze the causes of sand defects, and a linking process for molten metal defects, which is performed to analyze the causes of molten metal defects.

[0064] The process of linking defective sand is performed on the numerical data of each piece of information stored in the mold sand database 71, the molding database 72, and the sand defect rate database 75.

[0065] In the mold sand database 71, numerical data for each piece of information is stored in batch units. On the other hand, in the molding database 72, numerical data for each piece of information is stored in units of one mold. Therefore, the units in which data is stored differ between the numerical data in the mold sand database 71 and the numerical data in the molding database 72. However, the number of molds M that are molded using one batch of mold sand S is predetermined, and the numerical data for the sand generation equipment information and sand property information is common to that predetermined number of molds M. Therefore, the data processing unit 62a assigns and links the numerical data for the sand generation equipment information and sand property information of the batch that produced the mold sand S that became the material for each mold M (mold M before pouring).

[0066] Figure 7 illustrates the data linking process using the example of fabricating three molds M from one batch of mold sand S. As shown in Figure 7, the sand generation equipment information and sand property information for one batch are the same for the three molds M fabricated using it. Therefore, three cells are created for each batch and the same numerical value is assigned to each. Then, these are linked to the numerical data of the fabrication equipment used to fabricate each of the three molds M stored in the fabrication database 72. This creates a series of fabrication data sets ZD in which the numerical data of sand generation equipment information, sand property information, and fabrication equipment information are linked for each mold.

[0067] Furthermore, as explained at the beginning regarding the casting line, when creating a mold M, it is possible that not only the mold sand S from the batch that made up the mold M, but also the mold sand S from batches prior to that batch may be mixed in. Therefore, the data processing unit 62a associates each mold M with not only the numerical data of the sand generation equipment information and mold sand information from the batch that made up the mold M, but also the numerical data of the sand generation equipment information and mold sand information from batches prior to that batch, on a mold-by-mold basis. In the example in Figure 7, the numerical data of the sand generation equipment information and mold sand information from up to two batches prior are associated. Note that the number of batches prior to be associated can be arbitrarily changed by setting.

[0068] The sand defect rate is the same for all molds M that formed the basis of each cooled mold Mb after pouring that constitutes the lot from which the defect rate was calculated. Therefore, the data processing unit 62a assigns and links the numerical data of the sand defect rate to each mold M that formed the basis of each cooled mold Mb after pouring that constitutes the lot from which the sand defect rate was calculated. As shown in Figure 7, if three molds M are made from one batch of mold sand S, and a total of 12 molds M are made from four batches of mold sand S, and one lot consists of these molds, then suppose the sand defect rate of the cast products C obtained from each cooled mold Mb after pouring in that lot is 5%. In this case, for the 12 molds M that formed the basis of each cooled mold Mb after pouring, the numerical data of a 5% sand defect rate is also linked to the series of numerical data of sand generation equipment information, sand characteristic information, and molding equipment information that are linked on a mold-by-mold basis.

[0069] Through the above-described linking process for sand defects, numerical data such as sand generation equipment information, sand properties information, molding equipment information, and sand defect rate (defect rate related to mold sand S) are linked to each molded mold M. As a result, a molding data group ZD is formed for each molded mold M, consisting of a series of linked numerical data. As the casting line operates, this molding data group ZD is accumulated in the database 70, and the database 70 stores big data consisting of a large amount of molding data group ZD.

[0070] The linking process for molten metal defects is performed on the numerical data of each piece of information stored in the molten metal database 73, the pouring database 74, and the molten metal defect rate database 76.

[0071] In the molten metal database 73, numerical data of each piece of information during molten metal generation is stored on a batch basis. On the other hand, in the pouring database 74, numerical data of each piece of information during pouring is stored on a mold basis. Therefore, the units in which the numerical data in the molten metal database 73 and the numerical data in the pouring database 74 are stored are different. However, the number of molds M into which one batch of molten metal is poured is predetermined, and for that predetermined number of molds M, the numerical data of the molten metal generation information, such as the molten metal component information, is common. Therefore, the data processing unit 62a assigns and links the numerical data of the molten metal generation information of the batch that generated the molten metal poured into each mold M (mold M before pouring).

[0072] Figure 8 illustrates the data linking process using the example of pouring molten metal from one batch into three molds M. As shown in Figure 8, the information at the time of molten metal generation in one batch is the same for the three molds M into which it is poured. Therefore, three cells are created for each batch and the same numerical value is assigned to each. Then, this is linked to the numerical data of the pouring information stored in the pouring database 74 for each of the three molds M. This creates a series of pouring data CDs in which the numerical data of molten metal generation information and pouring information are linked for each mold.

[0073] The defect rate for pouring defects is the same for all molds M that served as the basis for each cooled mold Mb after pouring in a lot in which the defect rate occurred. Therefore, the data processing unit 62a assigns and links numerical data for the pouring defect rate to each mold M that served as the basis for each cooled mold Mb after pouring in a lot in which the molten metal defect rate was calculated. As shown in Figure 8, if a lot consists of 12 molds M in total, where molten metal is poured into 3 molds M from 1 batch of molten metal, and poured using 4 batches of molten metal, then suppose the molten metal defect rate of the cast product C obtained from each cooled mold Mb after pouring in that lot is 3%. In this case, for the 12 molds M that served as the basis for each cooled mold Mb after pouring, the numerical data of a 3% molten metal defect rate is also linked to the series of numerical data such as pouring information, received weight information, remaining weight information, and molten metal composition information, which are linked on a mold-by-mold basis.

[0074] Through the above linking process for molten metal defects, numerical data such as molten metal generation information, pouring information, received weight information, remaining weight information, and molten metal defect rate (defect rate related to molten metal) are linked for each poured mold Ma. As a result, a pouring data group CD is formed for each poured mold Ma, consisting of a series of linked numerical data. As the casting line operates, this pouring data group CD is accumulated in the database 70, and the database 70 stores big data consisting of a large amount of pouring data group CD.

[0075] In addition to the data linking process described above, data preprocessing also includes the data organization process in step S112. Data organization is performed on error values ​​and missing data. Numerical data collected during line operation in the molding and molten metal processes may include error data, such as unexpected values ​​or data that cannot be obtained at all. Upper and lower thresholds are set for each type of information, and if the numerical data exceeds the upper threshold or falls below the lower threshold, it is determined to be error data. Error data arises from errors during data transmission and reception, errors in measuring devices, etc. Also, in the molding and pouring processes, the mold M before pouring or the mold Ma after pouring may be missing due to malfunctions of the transport device (not shown), etc., in which case the mold M or mold Ma corresponding to the collected numerical data does not exist. Therefore, in order to improve the accuracy of the analysis of the cause of defects, the data processing unit 62a deletes data groups containing these error values ​​and data groups corresponding to missing mold M, as shown in Figures 7 and 8. Note that the order of the data linking process in step S111 and the data organization process in step S112 may be reversed.

[0076] [Abnormal Factor Extraction Processing] Once the data preprocessing in step S11 is completed, the data calculation unit 62b of the system control unit 62 performs the abnormal factor extraction process in step S12. In the abnormal factor extraction process, a large amount of molding data group ZD or pouring data group CD is used to extract specific information indicating abnormalities (abnormal factors) from various information related to the manufacture of the mold M or various information related to the manufacture of the poured mold Ma. Therefore, the data calculation unit 62b corresponds to the abnormal factor extraction means. Furthermore, this abnormal factor extraction process corresponds to the abnormal factor extraction step in the defect cause analysis method.

[0077] In the abnormality factor extraction process, first, in step S121, the molding data group ZD or the pouring data group CD is classified into groups according to the defect rate. In this case, one group may be formed based on an absolute value, or one group may be formed based on a predetermined numerical range. For example, if the defect threshold is set to 0.1, the data is classified into groups with a defect rate of 0.1% or less, groups with a defect rate greater than 0.1% and less than or equal to 1%, groups with a defect rate greater than 1% and less than or equal to 2%, and groups with a defect rate greater than 2%.

[0078] In the subsequent step S122, machine learning is performed using an unsupervised learning model on the molded data group ZD or pouring data group CD classified into each group. In this embodiment, a method using the Mahalanobis distance is adopted as the unsupervised learning model. A defect threshold is set, and only groups exceeding the defect threshold are targeted for unsupervised learning. For example, in the previous example, a defect rate of 0.1% is set as the defect threshold, so groups with a defect rate of 0.1% or less are not targeted for unsupervised learning. However, this does not exclude groups below the defect threshold from being targeted for unsupervised learning; unsupervised learning may be performed on all groups, including those below the defect threshold.

[0079] This unsupervised learning process extracts numerical data from each of the molded data group ZD or poured data group CD belonging to each group that exceeds a predetermined anomaly detection threshold (i.e., does not belong to the unit space), for each group that exceeds the defect threshold. Simultaneously, the attributes of the extracted numerical data are identified, and the anomaly score (Mahalanobis distance) is calculated.

[0080] In the following step S123, an anomaly detection process is performed. In the anomaly detection process, within a specific group that has been subjected to unsupervised learning, attributes in which numerical data exceeding a pre-set first anomaly detection threshold are found in numbers exceeding a pre-set second anomaly detection threshold are identified as anomaly factors. In other words, from the various information related to the manufacture of mold M or the manufacture of poured mold Ma, information (attributes) consisting of numerical data exceeding two anomaly detection thresholds is identified as an anomaly factor, and that information is extracted as an anomaly factor. The anomaly information extracted here may not only be one type of information, but may also include multiple types of information. Information extracted as an anomaly factor indicates that the numerical data has a large variation.

[0081] [Influence Calculation Process] After the anomaly factor extraction process, step S13 performs an influence calculation process. In the influence calculation process, the large amount of molding data group ZD or pouring data group CD is analyzed again. Through this analysis, the influence of other types of information on the information that has become an anomaly factor is calculated from among the various types of information related to the manufacturing of the mold M or the manufacturing of the poured mold Ma. Therefore, the data calculation unit 62b corresponds to the influence calculation means. Furthermore, the process of this influence calculation process corresponds to the influence calculation step in the defect cause analysis method. The influence is calculated using a supervised learning model in machine learning. In this embodiment, a method using regularized regression is adopted as the supervised learning model.

[0082] In this learning model using regularized regression, information extracted as anomalies regarding the manufacturing of mold M or the poured mold Ma is used as the target variable (training data), and other types of information are used as explanatory variables to create a linear predictive model equation. The resulting predictive model equation is expressed by the following equation (1). y=β0+β1x1+β2x2+···+βnxn ···(1) In Equation (1), y is the target variable, that is, the information extracted as an abnormal factor, x is the explanatory variable, that is, the information other than the information extracted as an abnormal factor. Also, βn is the partial regression coefficient for the explanatory variable, and β0 is the intercept.

[0083] In the created prediction model equation, the magnitude of the partial regression coefficient βn corresponds to the magnitude of the influence degree on the abnormal factor taken as the target variable. Therefore, by creating the prediction model equation, the partial regression coefficient βn, that is, the influence degree, for each piece of information taken as the explanatory variable is calculated. The influence degree calculated by this process indicates how much the variation in the numerical data of a specific piece of information taken as the abnormal factor is affected by the variation in the numerical data of other information. And the calculation result of this influence degree is the analysis result of the root cause analysis performed by the root cause analysis device 61.

[0084] [Influence Degree Display] In the subsequent step S14, an influence degree display process is performed. In the influence degree display process, for each piece of information taken as the explanatory variable, the numerical value of the partial regression coefficient βn, that is, the numerical value of the influence degree, is displayed on the information display unit 66. In this case, it is possible to display the respective influence degrees for all types of information taken as the explanatory variable, or it is also possible to display only those whose influence degrees exceed a predetermined threshold value. As a display method on the information display unit 66, for example, if a graph is created with each piece of information taken as the explanatory variable on the horizontal axis and the numerical value of the influence degree of each piece of information on the vertical axis and displayed, the influence degrees for each piece of information can be displayed in a list.

[0085] [Estimation of Root Cause of Defect and Countermeasure Based on Analysis] Based on the analysis result obtained by the root cause analysis device 61, that is, the respective influence degrees of other types of information on a specific piece of information taken as the abnormal factor, the operator estimates that the information with a relatively high influence degree is the root cause of the defect. Then, from those estimated as the root cause of the defect, information for which a countermeasure is possible is selected, and a countermeasure is implemented to prevent variation in the numerical data of the selected information.

[0086] For example, in an analysis of sand defects, suppose "pre-molding displacement (compression ratio)" is identified as an abnormal factor, and the influence of "sand particle size" on "pre-molding displacement" is relatively high, leading to the estimation that this is one of the causes of defects. In this case, it is analyzed that there is an abnormal variation in the numerical data of "pre-molding displacement," and that the variation in the numerical data of "sand particle size" is greatly influencing this variation. Therefore, the worker implements measures to suppress the variation in "sand particle size," such as changing the supply sand to one with fewer impurities. This improves the variation in "pre-molding displacement" and contributes to reducing the rate of sand defects.

[0087] Furthermore, in the analysis of molten metal defects, "pouring time" was identified as an abnormal factor, and it was estimated that the influence of "receiving weight" on "pouring time" was relatively high, and that this was one of the causes of the defects. In this case, it was analyzed that there was an abnormal variation in the numerical data of "pouring time," and that the variation in the numerical data of "receiving weight" greatly influenced this variation. Therefore, in order to suppress the variation in "receiving weight," the operator will take measures to ensure that the receiving weight remains constant, for example, by adjusting the amount of molten metal dispensed from the cupola 41 according to the remaining weight of molten metal. This will improve the variation in "pouring time" and contribute to reducing the rate of molten metal defects.

[0088] According to the defect cause analysis system 60, defect cause analysis program 69, and defect cause analysis method of this embodiment described above, the following effects and advantages can be obtained.

[0089] (1) In a casting line, specific manufacturing information when producing intermediate products such as mold M and poured mold Ma is influenced by various other types of manufacturing information. For example, the pouring time is not as simple as being able to manipulate the numerical data of the pouring time by adjusting the drive setting of the pouring device 43. Numerical data of various other types of manufacturing information, such as pouring temperature, received weight, and remaining weight, influence the numerical data of the pouring time, and these can affect the numerical data of the pouring time. Therefore, simply extracting abnormal factors from various types of manufacturing information is insufficient for analyzing the cause of defects.

[0090] Therefore, the defect cause analysis device 61 extracts abnormal factors from various types of manufacturing information and then calculates the degree of influence of other types of manufacturing information on those abnormal factors. By estimating the manufacturing information that is causing the defect rate of cast product C from this degree of influence, the defect rate can be reduced by improving the estimated manufacturing information. As a result, the defect rate can be reduced efficiently without relying on experience or intuition, and measures to reduce the defect rate can be standardized.

[0091] (2) The defect rate is linked to the molding data group ZD or the pouring data group CD, and anomaly extraction processing is performed on data groups that exceed the defect threshold. Since data groups with low defect rates are not analyzed, the process of extracting anomaly factors can be performed efficiently.

[0092] (3) The defect rate is linked to the molding data group ZD or the pouring data group CD. Among the groups classified by defect rate, the data group belonging to each group that exceeds the defect threshold is analyzed to extract abnormal factors. This allows for the extraction of abnormal factors that could not be extracted when analyzing the entire dataset together, which helps to further reduce the defect rate.

[0093] (4) The pouring data CD contains pouring information, received weight information, remaining weight information, and molten metal component information, encompassing manufacturing information that may affect the molten metal defect rate. Therefore, the accuracy of extracting abnormal factors in the pouring process and calculating the degree of influence of those abnormal factors can be improved, thereby improving the accuracy of defect cause analysis.

[0094] (5) The molding data set ZD includes sand generation equipment information, sand properties information, and molding equipment information, and comprehensively covers manufacturing information that may affect the sand defect rate. Therefore, the accuracy of extracting abnormal factors in the molding process and calculating the degree of influence of those abnormal factors can be improved, thereby improving the accuracy of defect cause analysis.

[0095] (6) The molding data group ZD is linked to mold sand information related to another mold sand S used as the material for a previously molded mold M, and sand generation equipment information related to the generation of that other mold sand S, and is used for extracting abnormal factors and calculating their impact. When molding a mold M, there is a possibility that not only the mold sand S used for molding that mold M, but also another mold sand S used as the material for a previous mold M may be mixed in. Therefore, by including information on that other mold sand S in the analysis, the accuracy of extracting abnormal factors in the molding process and calculating the impact of those abnormal factors can be further improved. This makes it possible to further improve the accuracy of analyzing the causes of defects.

[0096] (7) Numerical data is collected for each piece of mold sand S in batch units, for each piece of molding apparatus 20 in single mold units, and for the sand defect rate in lot units, with each data collection unit being different. Similarly, numerical data is collected for each piece of molten metal generation in batch units, for each piece of molten metal pouring in single mold units, and for the molten metal defect rate in lot units, with each data collection unit being different. Therefore, the data processing unit 62a performs data preprocessing to assign and link the batch-unit numerical data and lot-unit numerical data to a single mold unit. This makes it possible to analyze the defect rate on a single mold unit basis.

[0097] <Other embodiments> The embodiment is not limited to the one described above; for example, it may be implemented as follows.

[0098] (a) In the above embodiment, the data group classified as below the defective threshold in the anomaly factor extraction process is not included in the machine learning using an unsupervised learning model, but it may also be included in the unsupervised learning.

[0099] (b) In the above embodiment, a method using the Mahalanobis distance was adopted as the learning model for unsupervised learning in the anomaly factor extraction process. The learning model for unsupervised learning is not limited to this, and for example, methods such as graphical modeling and support vector machines (SVM) may be adopted.

[0100] (c) In the above embodiment, a regularized regression method was adopted as the learning model for supervised learning in the influence calculation process. The learning model for supervised learning is not limited to this, and for example, decision trees, multiple regression, deep learning (DNN), etc. may be adopted.

[0101] (d) In the above embodiment, the defect cause analysis system 60 analyzes the causes of defects in the molding process and pouring process in the casting line, but it may also analyze the causes of defects in the mold dismantling process. In this case, the pressing force of the mold pressing part of the mold dismantling device 50 used in the mold dismantling process, and other set values ​​and measured values ​​that may cause defects during mold dismantling are used as manufacturing information, and a system is constructed to collect numerical data of this information during mold dismantling and transmit it to the defect cause analysis device 61. The cast product C obtained through the mold dismantling process becomes the intermediate product here. By performing the defect cause analysis process in the above embodiment using the collected numerical data, it is possible to analyze the causes of defects caused by the mold dismantling process.

[0102] (e) In the above embodiment, in the data preprocessing in the defect cause analysis process, data groups containing error values ​​are deleted as part of the data organization process. Instead of deleting in this way, the data may be organized by inserting arbitrary numerical data or surrounding numerical values ​​into the error value portion.

[0103] (f) In the above embodiment, the failure cause analysis system 60 performs the analysis of the degree of influence of other types of information on the extracted abnormal factors and displays it, and the estimation of the failure cause based on the analysis result is performed by a person. In addition to this, a failure cause estimation system may be constructed in which the system control unit 62 estimates the failure cause. In this case, the memory 63 stores a failure cause estimation program which adds a step of estimating the failure cause to the failure cause analysis program 69. The system control unit 62 executes the failure cause estimation process according to the failure cause estimation program.

[0104] In the defect cause estimation process, the degree of influence of other types of information on abnormal factors is calculated, and then the cause of the defect is estimated based on predetermined defect discrimination criteria. As a defect discrimination criterion in this case, for example, if some types of information can be immediately addressed on-site and others require detailed investigation and consideration, the former type of information can be set in advance, and among the information whose degree of influence exceeds a threshold, those that fall under the set information can be identified as the cause of the defect.

[0105] (g) In the above embodiment, in the anomaly determination process of step S123, information (attributes) consisting of numerical data exceeding two anomaly determination thresholds is determined to be an anomaly factor. Alternatively, information (attributes) consisting of numerical data exceeding one anomaly determination threshold may be determined to be an anomaly factor. For example, an attribute in which numerical data exceeding a predetermined anomaly determination threshold (not belonging to a unit space) is present in a number exceeding one predetermined anomaly determination threshold may be determined to be an anomaly factor. [Explanation of Symbols]

[0106] 11...Sand processing device (mold sand generation device), 20...Molding device, 42...Ladle, 60...Defect cause analysis system, 62b...Data calculation unit (means for extracting abnormal factors, means for calculating impact), 67...Network interface (means for collecting data), 70...Database (means for storing data), C...Casting product, M...Mold (intermediate product), Ma...Mold after pouring (intermediate product), Mb...Mold after pouring and cooling, ZD...Molding data group, CD...Pouring data group.

Claims

1. A defect cause analysis system applied to a casting line that performs a molding process of shaping mold sand to form a mold, a pouring process of melting a molten material to generate molten metal and pouring it into the molded mold to form a poured mold, a cooling process of cooling the poured mold to form a cooled mold after pouring, and a mold dismantling process of dismantling the cooled mold after pouring to obtain a cast product, wherein the system analyzes the cause of defects in the cast product that occurred in the molding process or the pouring process, The mold obtained by the molding process or the poured mold obtained by the pouring process is used as an intermediate product. Regarding the various manufacturing information obtained when manufacturing the aforementioned intermediate product, a data collection means collects numerical data for each intermediate product that is manufactured sequentially for mass production, A storage means for storing each collected numerical data as a single data group linked to the individual intermediate product, An abnormality factor extraction means analyzes a large number of data sets stored in the storage means and extracts factors that indicate abnormalities during the mass production of the intermediate product from among the various manufacturing information, An influence calculation means that analyzes the aforementioned large number of data sets and calculates the degree of influence of other types of manufacturing information on the extracted abnormal factors for each type of manufacturing information, Equipped with, The aforementioned intermediate product is the pre-molded mold and the mold. The various manufacturing information used when manufacturing the pre-pouring mold includes information on the composition of the molten metal poured into the pre-pouring mold, information on the receiving of the molten metal when a ladle receives the molten metal, information on the pouring of the molten metal from the ladle into a single mold, and information on the remaining molten metal after the pouring of the molten metal from one ladle into a set number of molds has been completed. The various manufacturing information used when manufacturing the mold includes sand properties information relating to the properties of the mold sand, sand generation device information relating to the mold sand generation device, and molding device information relating to the molding device used when molding one mold. The storage means includes, for each mold, mold sand information relating to the mold sand used as the material of the mold and sand generation device information relating to the generation of the mold sand, as information of the mold sand used in the current molding, in a single data group, and further includes in the single data group mold sand information relating to the mold sand used as the material of the mold molded immediately before the current mold and sand generation device information relating to the generation of the sand sand, as information of the mold sand used in the immediately preceding molding.

2. The storage means stores the defect rate of the group of cooled molds after pouring, which was determined by inspecting the cast products obtained from the multiple cooled molds after pouring, and associates it with the intermediate product that served as the basis for each individual cooled mold that constitutes the group of cooled molds after pouring. The defect cause analysis system according to claim 1, characterized in that the abnormal factor extraction means analyzes data sets from the numerous data sets in which the defect rate exceeds a defect threshold, and extracts abnormal factors.

3. The defect cause analysis system according to claim 2, characterized in that the abnormal factor extraction means classifies the numerous data sets into groups according to the defect rate, analyzes the data sets belonging to each group that exceeds the defect threshold, and extracts abnormal factors from the various manufacturing information.

4. The defect cause analysis system according to any one of claims 1 to 3, characterized in that the abnormal factor extraction means extracts specific manufacturing information in which the number exceeding a first abnormality discrimination threshold exceeds a second abnormality discrimination threshold by unsupervised learning using the large number of data sets, as the abnormal factor.

5. The defect cause analysis system according to any one of claims 1 to 4, characterized in that the impact calculation means uses a model generated by supervised learning with the manufacturing information extracted as the abnormal factor as training data to calculate the impact of other types of manufacturing information on the abnormal factor.

6. This invention relates to a casting line that performs the following steps: a molding step of shaping mold sand to create a mold; a pouring step of melting a molten material to generate molten metal and pouring it into the molded mold to create a poured mold; a cooling step of cooling the poured mold to create a cooled mold after pouring; and a mold dismantling step of dismantling the cooled mold after pouring to obtain a cast product. The invention relates to a defect cause analysis program that analyzes the cause of defects in the cast product that occurred in the molding step or the pouring step. The mold obtained by the molding process or the poured mold obtained by the pouring process is used as an intermediate product. On the computer, Regarding the various manufacturing information obtained when manufacturing the aforementioned intermediate product, the first step is to collect numerical data for each intermediate product that is manufactured sequentially for mass production, The second step involves associating each of the collected numerical data with the individual intermediate products as a single data group and storing it in a storage means, A third step involves analyzing a large number of stored data sets and extracting factors that indicate abnormalities during the mass production of the intermediate product from among the various manufacturing information. A fourth step involves analyzing the aforementioned large number of data sets and calculating the degree of influence of other types of manufacturing information on the extracted abnormal factors for each type of manufacturing information, Make it run, The aforementioned intermediate product is the pre-molded mold and the mold. The various manufacturing information used when manufacturing the pre-pouring mold includes information on the composition of the molten metal poured into the pre-pouring mold, information on the receiving of the molten metal when a ladle receives the molten metal, information on the pouring of the molten metal from the ladle into a single mold, and information on the remaining molten metal after the pouring of the molten metal from one ladle into a set number of molds has been completed. The various manufacturing information used when manufacturing the mold includes sand properties information relating to the properties of the mold sand, sand generation device information relating to the mold sand generation device, and molding device information relating to the molding device used when molding one mold. In the second step, for each mold, information on the mold sand used as the material for the mold and information on the sand generating apparatus for producing the mold sand are included in a single data group as information on the mold sand used in the current molding process and stored in the storage means. In addition, the single data group is also stored in the storage means as information on the mold sand used in the most recent molding process, including information on the mold sand used as the material for the mold mold produced immediately before the current mold and information on the sand generating apparatus for producing the mold sand. A failure cause analysis program characterized by the following features.

7. This invention relates to a casting line that performs the following steps: a molding step of shaping mold sand to form a mold; a pouring step of melting a molten material to generate molten metal and pouring it into the molded mold to form a poured mold; a cooling step of cooling the poured mold to form a cooled mold after pouring; and a mold dismantling step of dismantling the cooled mold after pouring to obtain a cast product, and provides a defect cause analysis method for analyzing the cause of defects in the cast product that occurred in the molding step or the pouring step. The mold obtained by the molding process or the poured mold obtained by the pouring process is used as an intermediate product. Regarding the various manufacturing information obtained when manufacturing the aforementioned intermediate product, a data collection process is performed to collect numerical data for each individual intermediate product that is manufactured sequentially for mass production, A storage process in which each of the collected numerical data is associated with the individual intermediate products and stored as a single data group, An abnormality factor extraction step involves analyzing a large number of stored data sets and extracting factors that indicate abnormalities during the mass production of the intermediate product from among the various manufacturing information, An influence calculation step which involves analyzing the aforementioned large number of data sets and calculating the degree of influence of other types of manufacturing information on the extracted abnormal factors for each type of manufacturing information, Equipped with, The aforementioned intermediate product is the pre-molded mold and the mold. The various manufacturing information used when manufacturing the pre-pouring mold includes information on the composition of the molten metal poured into the pre-pouring mold, information on the receiving of the molten metal when a ladle receives the molten metal, information on the pouring of the molten metal from the ladle into a single mold, and information on the remaining molten metal after the pouring of the molten metal from one ladle into a set number of molds has been completed. The various manufacturing information used when manufacturing the mold includes sand properties information relating to the properties of the mold sand, sand generation device information relating to the mold sand generation device, and molding device information relating to the molding device used when molding one mold. In the memory step, for each mold, mold sand information relating to the mold sand used as the material for the mold and sand generation equipment information relating to the generation of the mold sand are stored in a single data group as information of the mold sand used in the current molding, and the single data group is also stored as information of the mold sand used in the most recent molding, including mold sand information relating to the mold sand used as the material for the mold mold made immediately before the current mold and sand generation information relating to the generation of the mold sand. A method for analyzing the cause of defects, characterized by the above.

Images