Anomaly detection system and method for casting equipment

Abstract

To provide a system and method for detecting an abnormality in a casting apparatus with high accuracy by accurately extracting abnormal data without omissions and detecting an abnormality based on the extracted abnormal data.SOLUTION: An abnormality detection unit that detects an abnormality based on casting waveform data transmitted from a casting apparatus includes: a first abnormality determination step of determining abnormality data using an unsupervised learning model; a supervised data setting step of setting abnormality data as supervised data; and a second abnormality determination step of determining abnormalities in a casting apparatus using a supervised learning model.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an abnormality detection system and an abnormality detection method for a casting apparatus, comprising an abnormality detection unit that performs abnormality detection based on casting waveform data transmitted from a casting apparatus which injection-fills molten metal into a mold cavity formed by clamping a casting mold to produce and cast a cast product. [Background Art]

[0002] If casting is continued while an abnormality lurks in a casting apparatus that injection-fills molten metal such as aluminum alloy into a mold cavity, casting quality will fluctuate in the initial stage, and gradually casting defects that cannot satisfy the set quality standards will occur frequently, resulting in interruption of casting. In addition, when a stock shortage occurs due to casting defects, unnecessary time such as separate remanufacturing is required. At this point, it is preferable to perform fundamental maintenance of the casting apparatus. However, if production scheduling is prioritized and casting is forced to continue while casting defects are generated, it will eventually lead to damage to the casting apparatus including the casting mold, resulting in complete shutdown of casting.

[0003] For this reason, many abnormality detection means for casting apparatuses have been proposed, which can detect even minor abnormalities with high accuracy and enable stable production of high-quality cast products by performing appropriate maintenance at an early stage. For example, as disclosed in Patent Document 1, there has been proposed a means for acquiring internal and external state variables from an injection molding machine, and detecting an abnormality of the injection molding machine using an analysis method of a supervised learning program based on state variables (normal data) obtained from an injection molding machine in which no abnormality occurred and state variables (abnormal data) obtained from an injection molding machine in which an abnormality occurred. According to this, it is stated that high-accuracy abnormality detection can be performed regardless of the amount of knowledge and experience of an analyst who analyzes abnormalities. In addition, as disclosed in Patent Document 2, a means for detecting an abnormality of a mold using an analysis method of a supervised learning program based on state information indicating the amount of wear of the mold before molding has been proposed. According to this, it is stated that the wear state of the mold after molding can be accurately predicted. [Prior Art Documents] [Patent Documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2017-30221 [Patent Document 2] Japanese Patent Publication No. 2020-199706 [Overview of the project] [Problems that the invention aims to solve]

[0005] Here, Patent Document 1 describes acquiring abnormal and normal data from multiple injection molding machines, defining the abnormal data as training data, and performing anomaly detection using a supervised learning program. For this to work, it is assumed that the multiple injection molding machines have exactly the same equipment capacity, including the molding materials used, molds, and ancillary equipment such as temperature control means. However, this is not realistic as all equipment contains inherent errors. Even if all equipment had the same equipment capacity without errors, it would still be contradictory as the task of identifying the presence or absence of anomalies and analyzing the nature of the anomalies would be left to the analyst.

[0006] Furthermore, in Patent Document 2, defining clear information such as the amount of mold wear before molding as training data improves the accuracy of supervised learning programs, thereby improving the accuracy of anomaly detection and prediction of mold wear. However, in casting and casting equipment, anomalies appear as a result of a complex interplay of many uncertain factors, such as the temperature and composition of the molten metal, thermal expansion and deformation of injection means such as injection sleeves, thermal expansion and adhesion of impurities to the mold, the lubrication state of the drive unit of the casting equipment, and the influence of oil temperature in the case of hydraulic drive. Moreover, in many cases, there is little or no anomaly data for each individual element, making it difficult to define training data. Furthermore, it is anticipated that accurate collection of anomaly data may not be possible due to omissions or oversights in the recognition of operators when anomalies occur. As a result, it is difficult to accurately define training data for anomaly detection in casting equipment, and using supervised learning programs as a means of anomaly detection is considered an undesirable form.

[0007] Therefore, the present invention aims to provide a highly accurate abnormality detection system and method for a casting apparatus that can accurately extract abnormal data without any omissions, and perform abnormality detection based on the extracted abnormal data. [Means for solving the problem]

[0008] The abnormality detection system for casting apparatuses of the present invention is An anomaly detection system for a casting apparatus, which produces castings by injecting and filling molten metal into a mold cavity formed by clamping a casting mold, includes an anomaly detection unit that detects anomalies based on casting waveform data transmitted from the casting apparatus, The anomaly detection unit is characterized by comprising: a first learning unit that extracts reference waveform data from the casting waveform data and defines its features using an unsupervised learning program; a second learning unit that compares the reference waveform data with the casting waveform data and counts it as an anomaly score value; a first anomaly determination unit that determines the casting waveform data whose anomaly score value exceeds a preset threshold as an anomaly data; a teacher data setting unit that sets the anomaly data determined by the first anomaly determination unit as teacher data; and a second anomaly determination unit that uses a supervised learning program to determine anomalies in the casting apparatus from the casting waveform data based on the settings of the teacher data setting unit.

[0009] The abnormality detection method for casting apparatuses of the present invention is: The invention is characterized by comprising: a waveform data collection step for collecting the aforementioned casting waveform data; a first learning step for extracting reference waveform data from the casting waveform data and defining its features in a first learning unit using an unsupervised learning program; a second learning step for comparing the reference waveform data with the casting molding data and counting them as abnormal score values ​​in a second learning unit; a first abnormality determination step for determining the casting waveform data whose abnormal score value exceeds a preset threshold as abnormal data in a first abnormality determination unit; a teacher data setting step for setting the abnormal data determined in the first abnormality determination step as teacher data in a teacher data setting unit; and a second abnormality determination step for determining abnormalities in the casting apparatus from the casting waveform data using a supervised learning program based on the teacher data set in the teacher data setting step in a second abnormality determination unit.

[0010] In the method for detecting abnormalities in a casting apparatus of the present invention, Preferably, the reference waveform data is the casting waveform data from past production castings in which the casting quality of the casting product satisfied a predetermined quality standard value.

[0011] The abnormality detection system for casting apparatuses of the present invention is An anomaly detection system for a casting apparatus, which produces castings by injecting and filling molten metal into a mold cavity formed by clamping a casting mold, includes an anomaly detection unit that detects anomalies based on casting waveform data transmitted from the casting apparatus, The anomaly detection unit is characterized by comprising: a first learning unit that extracts reference waveform data from the casting waveform data and defines its features; a second learning unit that compares the reference waveform data with the casting waveform data and counts it as an anomaly score value; a first anomaly determination unit that determines the casting waveform data whose anomaly score value exceeds a preset threshold as an anomaly data; a teacher data setting unit that sets the anomaly data determined by the first anomaly determination unit as teacher data; and a second anomaly determination unit that determines anomalies in the casting apparatus from the casting waveform data based on the settings of the teacher data setting unit.

[0012] The abnormality detection method for casting apparatuses of the present invention is: The system is characterized by comprising: a waveform data collection step for collecting the aforementioned casting waveform data; a first learning step for extracting reference waveform data from the casting waveform data and defining its features in a first learning unit; a second learning step for comparing the reference waveform data with the casting molding data and counting them as abnormal score values ​​in a second learning unit; a first abnormality determination step for determining that the casting waveform data whose abnormal score value exceeds a preset threshold is abnormal data in a first abnormality determination unit; a teacher data setting step for setting the abnormal data determined in the first abnormality determination step as teacher data in a teacher data setting unit; and a second abnormality determination step for determining abnormalities in the casting apparatus from the casting waveform data based on the teacher data set in the teacher data setting step in a second abnormality determination unit.

[0013] In the method for detecting abnormalities in a casting apparatus of the present invention, Preferably, the reference waveform data is the casting waveform data from past production castings in which the casting quality of the casting product satisfied a predetermined quality standard value.

[0014] Furthermore, in the abnormality detection method for casting apparatus of the present invention, Preferably, the first learning step, the second learning step, and the first anomaly detection step are performed using an unsupervised learning program, while the training data setting step and the second anomaly detection step are performed using a supervised learning program. [Effects of the Invention]

[0015] According to the present invention, abnormal data can be accurately extracted without any omissions, and by performing abnormality detection based on the extracted abnormal data, a highly accurate abnormality detection system and method for casting equipment can be provided. [Brief explanation of the drawing]

[0016] [Figure 1] This is a conceptual diagram showing a casting apparatus according to an embodiment of the present invention. [Figure 2]It is a conceptual diagram showing an anomaly detection system according to an embodiment of the present invention. [Figure 3] It is a diagram showing an anomaly detection method according to an embodiment of the present invention. [Figure 4] It is a diagram showing the anomaly detection method according to an embodiment of the present invention, following FIG. 3.

Mode for Carrying Out the Invention

[0017] Preferred embodiments for carrying out the present invention will be described below with reference to the drawings. The following embodiments do not limit the invention claimed in each claim. Furthermore, not all combinations of features described in the embodiments are necessarily essential to the solution means of the invention claimed in each claim. In addition, in the present embodiment, the scale and dimensions of each component may be exaggerated in the drawings, and some components may be omitted.

[0018] (Casting Apparatus) First, a casting apparatus according to an embodiment of the present invention will be described with reference to FIG. 1. A casting apparatus 100 shown in FIG. 1 includes a casting mold 10, a mold clamping unit 20, an injection unit 30, a casting control unit 40, and an anomaly detection unit 50. Although FIG. 1 illustrates the casting apparatus 100 in which the casting mold 10 and the injection unit 30 are arranged horizontally (a horizontal clamping horizontal casting casting apparatus 100), the present invention is not limited thereto. For example, the casting apparatus may have a configuration in which the casting mold 10 is arranged horizontally and the injection unit 30 is arranged vertically (a horizontal clamping vertical casting casting apparatus 100), or may have a configuration in which both the casting mold 10 and the injection unit 30 are arranged vertically (a vertical clamping vertical casting casting apparatus 100). In any case, the components are unchanged, and only the combination of the arrangement of the casting mold 10 and the injection unit 30 is changed. Therefore, the following description will be given taking the horizontal clamping horizontal casting casting apparatus 100 as an example.

[0019] The casting mold 10 forms a mold cavity 13 and a mold gate 14 by operating the clamping section 20 to clamp the fixed platen 21 and the movable platen 22. It is preferable to apply a release agent to the mold cavity 13 and the mold gate 14 before injecting and filling the molten metal from the injection section 30 into the mold cavity 13. It is also preferable to provide mold temperature control means, including a temperature control circuit (not shown), in the fixed mold 11 and the movable mold 12 to adjust them to a predetermined temperature. Alternatively, the casting mold 10 may be provided with a vacuum suction means (not shown), and this vacuum suction means may be used to directly vacuum the inside of the mold cavity 13.

[0020] The clamping unit 20 comprises a fixed platen 21 that supports the fixed mold 11, a movable platen 22 that supports the movable mold 12, and a clamping platen 23 that supports the clamping drive unit 25. The clamping drive unit 25 is a hydraulic drive means such as a hydraulic cylinder, and the clamping drive unit 25 and the movable platen 22 are connected via a cylinder rod 26. The fixed platen 21 and the clamping platen 23 are connected by a plurality of tie bars 24 that pass through the movable platen 22. The clamping control unit 27 operates the clamping drive unit 25, causing the movable platen 22 to slide in the mold opening and closing direction via the cylinder rod 26, using the tie bars 24 as guides. Here, the sliding of the movable platen 22 and the movable mold 12 is defined as the movement in the direction approaching the fixed platen 21 and the fixed mold 11 as the clamping operation, and the movement in the direction moving away from them as the mold opening operation.

[0021] In Figure 1, the clamping drive unit 25 is shown as a hydraulic drive means such as a hydraulic cylinder, but it is not limited to this. For example, it may be an electric drive means using a ball screw mechanism that converts the rotational motion of an electric motor into linear motion, or a hybrid drive means that combines a hydraulic drive means and an electric drive means. Furthermore, multiple clamping drive units 25 may be arranged, or the clamping drive unit 25 may be arranged on the tie bar 24.

[0022] The injection unit 30 comprises a cylindrical injection sleeve 31 arranged horizontally, a plunger tip 32 that slides in the front-rear direction within the injection sleeve 31, a pouring port 34 for supplying molten metal M such as aluminum alloy into the injection sleeve 31, and an injection drive unit 36 ​​that operates the sliding of the plunger tip 32. The tip of the injection sleeve 31 opposite the pouring port 34 passes through the fixed platen 21 and the fixed mold 11 and is connected to the mold gate 14. Here, the sliding of the plunger tip 32 is defined as moving forward F when approaching the mold gate 14, and as moving forward F when moving forward R, and as moving backward R when moving away from the mold gate 14.

[0023] Here, the plunger rod 33, which is connected to the plunger tip 32, is connected via a connecting portion 35 to the drive rod 37 of the injection drive unit 36 ​​of a hydraulic drive means such as a hydraulic cylinder. While the plunger tip 32 is waiting behind the pouring port 34, molten metal M is supplied from the pouring port 34 into the injection sleeve 31 using a molten metal supply means (not shown). The injection control unit 38 operates the injection drive unit 36, and the forward and backward movement of the plunger tip 32 is controlled via the drive rod 37, the connecting portion 35, and the plunger rod 33. The forward movement of the plunger tip 32 presses the molten metal M supplied into the injection sleeve 31, and it is injected and filled into the mold cavity 13 via the mold gate 14.

[0024] Furthermore, the injection sleeve 31 and plunger tip 32 are provided with cooling means (not shown) including a channel through which a cooling medium such as cooling water flows. In addition, it is preferable to apply a lubricant to the sliding surfaces of the injection sleeve 31 and plunger tip 32 in order to prevent galling damage due to strong contact between the plunger tip 32 and the injection sleeve 31, stabilize the sliding state, and suppress the adhesion of molten metal residue. Alternatively, the vacuum suction of the mold cavity 13 by a vacuum suction means provided in the casting mold 10 and injection filling may be performed simultaneously. Alternatively, a vacuum suction means (not shown) may be provided in the injection sleeve 31 to vacuum the inside of the injection sleeve 31 and indirectly vacuum the inside of the mold cavity 13 through the mold gate 14, or a combination of direct and indirect vacuum suction may be performed.

[0025] In Figure 1, the injection drive unit 36 ​​is shown as a hydraulic drive means such as a hydraulic cylinder, but it is not limited to this. For example, it may be an electric drive means using a ball screw mechanism that converts the rotational motion of an electric motor into linear motion, or a hybrid drive means that combines a hydraulic drive means and an electric drive means. Furthermore, in the case of a hydraulic drive means, it may also be equipped with a pressure accumulation means such as an accumulator.

[0026] Furthermore, although the injection sleeve 31 is positioned horizontally in Figure 1, it may be positioned arbitrarily within a range from horizontal to vertically downward relative to the mold cavity 13 and mold gate 14. Also, although molten metal M is supplied to the injection sleeve 31 from the pouring port 34, for example, the injection sleeve 31 may be connected to a melting furnace (not shown) that holds the molten metal M by a connecting means such as a molten metal supply pipe, and the molten metal M may be supplied to the injection sleeve 31 via this connecting means. Also, although the injection sleeve 31 is shown as a combination of a plunger tip 32, for example, pressurized gas may be supplied into a sealed molten metal holding furnace (not shown), and the molten metal M may be injected and filled into the mold cavity 13 via a molten metal supply pipe, or a transport means such as a transport pump may be used to inject and fill the molten metal M instead of supplying pressurized gas.

[0027] The casting control unit 40 is connected to the mold clamping control unit 27 and the injection control unit 38. Based on preset casting conditions, the casting control unit 40 operates the mold clamping control unit 27 to perform the mold clamping and mold opening operations of the casting mold 10. Similarly, the casting control unit 40 operates the injection control unit 38 to inject and fill the mold cavity 13 with molten metal M. In addition, it sends operation commands to peripheral equipment such as molten metal supply means, cooling means, and vacuum suction means (not shown) to operate the casting apparatus 100 and manage the casting process.

[0028] Furthermore, the abnormality detection unit 50 is connected to the casting control unit 40 of the casting apparatus 100. This allows the abnormality detection unit 50 to receive casting waveform data transmitted from the clamping control unit 27, the injection control unit 38, and peripheral equipment via the casting control unit 40, and to perform abnormality detection of the casting apparatus 100 based on the casting molding data. This state in which the abnormality detection unit 50 and the casting control unit 40 of the casting apparatus 100 are connected is called the abnormality detection system 200, which will be explained in detail with reference to Figure 2. In Figure 1, the abnormality detection system 200 is shown as having the abnormality detection unit 50 and the casting apparatus 100 as separate components, but for example, the abnormality detection system 200 may also be in a form in which the abnormality detection unit 50 is incorporated into the casting control unit 40 of the casting apparatus 100.

[0029] (Anomaly detection system) Next, an anomaly detection system 200 according to an embodiment of the present invention will be described with reference to Figure 2. Figure 2(a) shows the anomaly detection unit 50 of the anomaly detection system 200, and Figure 2(b) shows the data setting unit 52 of the anomaly detection unit 50.

[0030] First, as shown in Figure 2(a), the anomaly detection unit 50 includes a data transmission / reception unit 51, a data setting unit 52, a first learning unit 53, a second learning unit 54, a first anomaly determination unit 55, a teacher data setting unit 56, a second anomaly determination unit 57, and a display unit 58.

[0031] The data transmission / reception unit 51 is connected to the casting control unit 40 of the casting apparatus 100 and receives casting waveform data from the injection control unit 38, the clamping control unit 27, and peripheral equipment 99 such as molten metal supply means, cooling means, and vacuum suction means via the casting control unit 40. It also transmits the determination data from the first abnormality determination unit 55 and the second abnormality determination unit 57 to the injection control unit 38, the clamping control unit 27, and the peripheral equipment 99 via the casting control unit 40.

[0032] The data setting unit 52 sets the conditions necessary for detecting abnormalities in the casting apparatus 100 based on the casting waveform data received by the data transmission / reception unit 51. Further details will be explained using Figure 2(b).

[0033] The first learning unit 53 extracts reference waveform data from the casting waveform data and defines the features of the extracted reference waveform data (referred to as reference features) using, for example, a known deep learning image diagnostic method. The second learning unit 54 collects the casting waveform data received by the data transmission / reception unit 51 and similarly defines the features of the collected casting waveform data (referred to as evaluation features) using a deep learning image analysis method. The reference features and evaluation features are compared, the degree of deviation between the features is counted as an anomaly score value, and stored in the second learning unit 54 as an anomaly score value count data. The first anomaly determination unit 55 determines that casting waveform data in the second learning unit 54 whose anomaly score value count data exceeds a preset threshold is an anomaly waveform data. The first learning unit 53, the second learning unit 54, and the first anomaly determination unit 55 use an unsupervised learning program to learn the definitions of reference features and evaluation features and to determine anomaly waveform data (referred to as unsupervised learning). This unsupervised learning is displayed on the display unit 58.

[0034] The training data setting unit 56 sets the abnormal waveform data determined by the first abnormality determination unit 55 as training data. The second abnormality determination unit 57, based on this training data, performs abnormality determination by extracting casting waveform data indicating abnormalities in the casting apparatus 100 from the casting waveform data received by the data transmission / reception unit 51 using a supervised learning program (this is called supervised learning). The determination data from the first abnormality determination unit 55 and the second abnormality determination unit 57 is transmitted from the data transmission / reception unit 51 to the casting control unit 40 of the casting apparatus 100. This supervised learning is displayed on the display unit 58.

[0035] The display unit 58 displays the casting waveform data received by the data transmission / reception unit 51, the reference waveform data from the first learning unit 53, the abnormal score value count data from the second learning unit 54, the judgment results from the first abnormal judgment unit 55 and the second abnormal judgment unit 57, the teacher data from the teacher data setting unit 56, and the setting items from the data setting unit 52.

[0036] Here, the casting waveform data transmitted from the mold clamping control unit 27 includes, with the time axis from the start to the end of the casting process, for example, the mold clamping force waveform, the drive torque waveform of the mold clamping drive unit 25, the vibration waveform of the mold clamping unit 20, the release force waveform of the mold clamping drive unit 25 at the initial stage of mold opening, the push force waveform of an extrusion means (not shown) built into the mold clamping unit 20, the mold opening and closing speed waveform, the stress waveform generated in the multiple tie bars 24, and so on.

[0037] Furthermore, the casting waveform data transmitted from the injection control unit 38 includes, with the time axis from the start to the end of the casting process, for example, the forward speed waveform of the plunger tip 32 (referred to as the injection speed waveform), the pressing force waveform of the molten metal M by the plunger tip 32 (referred to as the casting pressure waveform), the time waveform required for injection filling to fill the mold cavity 13 with molten metal M (referred to as the injection time waveform), the position waveform of the plunger tip 32 (referred to as the injection position waveform), the drive torque waveform of the injection drive unit 36, the temperature waveform of the injection sleeve 31 or plunger tip 32, the temperature waveform of the molten metal M, the vibration waveform or vibration acceleration waveform during the forward and backward movement of the plunger tip 32, the injection speed switching time waveform (referred to as the acceleration waveform), the operation command waveform of the injection control unit 38 and the execution waveform of the injection drive unit 36, etc.

[0038] Furthermore, the casting waveform data transmitted from the peripheral equipment 99 includes, on a time axis from the start to the end of the casting process, for example, the temperature waveform of the casting mold 10, the temperature and flow rate waveforms of the cooling medium used for the cooling means of the casting mold 10 and the injection unit 30, the vacuum waveform inside the mold cavity 13 or injection sleeve 31, the thermal expansion waveform of the casting mold 10, the waveform of the amount of release agent applied to the mold cavity 13 and the mold gate 14, the waveform of the deformation amount of the casting mold 10 during the clamping operation, the waveform of the molten metal transport time and transport amount of the molten metal supply means, the temperature and humidity, etc.

[0039] Next, the data setting unit 52 will be explained using Figure 2(b). First, the abnormality detection selection switch 521 selects an item to be detected as abnormal from the assumed abnormality list of the casting apparatus 100 that has been set in advance in the abnormality detection unit 50. The assumed abnormality list includes, for example, chip galling, which indicates a sliding abnormality of the plunger tip 32; component life, which indicates the replacement time of consumable parts such as the plunger tip 32 and the vacuum ring used in the vacuum suction means; maintenance prediction, which indicates the maintenance time of key components of the casting apparatus 100 such as the casting mold 10, the clamping drive unit 25 and the injection drive unit 36; lubrication abnormality, which indicates poor lubrication of the plunger tip 32 and the injection sleeve 31; molten metal supply abnormality of the molten metal supply means; mold retention abnormality, where a casting remains in the mold cavity 13; temperature abnormality of each part, which indicates an abnormality of the cooling means, etc. Note that the abnormality detection selection switch 521 may select only one item, or it may be possible to select multiple items and prioritize the items to perform abnormality detection. In addition, items may be added or deleted as the casting process continues.

[0040] The measurement waveform selection switch 522 selects and sets the casting waveform data to be used for anomaly detection from the casting waveform data transmitted from the injection control unit 38, the mold clamping control unit 27, and peripheral equipment 99, which are received by the data transmission / reception unit 51. Although the selection and setting are to be performed by the casting operator, for example, the casting waveform data deemed suitable for anomaly detection may be automatically set based on the item selected by the anomaly detection selection switch 521. Alternatively, the casting waveform data deemed suitable may be displayed preferentially, and the casting operator may select and set from the displayed data.

[0041] The reference waveform selection switch 523 extracts and sets the reference waveform data for which the reference features are defined in the first learning unit 53 from the casting waveform data transmitted by the data transmission / reception unit 51. Here, the reference waveform data is, for example, casting waveform data in which the casting quality of the casting product in past production castings satisfied a predetermined quality standard value. Alternatively, the analysis results obtained using CAE analysis means such as flow analysis may be used as the reference waveform data. Here, since the accuracy of the reference waveform data can be improved by learning from the first learning unit 53 to the second learning unit 54 and the second anomaly determination unit 57, the setting of the reference waveform data for the reference waveform selection switch 523 in the first learning unit 53 can be simple.

[0042] The axis selection switch 524 selects and sets the axis that will serve as the reference for the reference waveform data and the casting waveform data during unsupervised and supervised learning. For example, if chip galling is selected with the anomaly detection selection switch 521 and the injection velocity waveform is selected with the measurement waveform selection switch 522 and the reference waveform selection switch 523, it is preferable to set the elapsed time from the start to the end of the forward movement of the plunger tip 32 or the forward position of the plunger tip 32 with the axis selection switch 524. Alternatively, the axis selection switch 524 may be set automatically based on the selections of the anomaly detection selection switch 521, the measurement waveform selection switch 522, and the reference waveform selection switch 523.

[0043] The learning interval setting switch 525 sets the interval (referred to as the learning interval) that is the cause of an anomaly, based on the setting of the axis selection switch 524, from the reference waveform data and the casting waveform data. In particular, the data volume of the casting waveform data collected in production casting is enormous, requiring an anomaly detection unit 50 with an extremely large data memory capacity, which leads to an increase in the size and cost of the anomaly detection system 200. In addition, the analysis time required for anomaly detection becomes longer. Therefore, by setting only the necessary parts from the casting waveform data as the learning interval, the data volume can be compressed, making the anomaly detection system 200 smaller, reducing costs, and shortening the analysis time required for anomaly detection.

[0044] The setting of this learning interval may be done, for example, by a molding engineer operating the casting process, by selecting points in the casting waveform data where changes have been observed based on past casting performance. Alternatively, it may be set using analytical methods such as flow analysis. Or, it may be automatically set by finding points of change in the casting waveform data using unsupervised learning performed for each casting shot. The setting of this learning interval will be explained in detail with reference to Figures 3 and 4.

[0045] When the first learning start switch 526 is pressed, the first learning unit 53 starts unsupervised learning for selecting reference waveform data and defining reference features. When the second learning start switch 527 is pressed, the second learning unit 54 starts unsupervised learning for collecting casting waveform data and defining evaluation features. When the first anomaly judgment start switch 529 is pressed, the first anomaly judgment unit 55 starts unsupervised learning to compare reference features and evaluation features, count anomaly score values, and determine anomaly data. When the second anomaly judgment start switch 531 is pressed, the second anomaly judgment unit 57 starts supervised learning for anomaly judgment of the casting apparatus 100 based on the teacher data set in the teacher data setting unit 56. Note that the first anomaly judgment start switch 529 may be set to start automatically in conjunction with the second learning start switch 527.

[0046] The threshold setting switch 528 is used to set the threshold value when counting abnormal score values ​​by comparing the reference characteristics of the reference waveform data with the evaluation characteristics of the casting waveform data. The set threshold may be adjusted by subsequent unsupervised learning. This threshold may be set by, for example, a molding engineer operating the casting process, who calculates the threshold from past casting performance data. Alternatively, the threshold may be set using analytical methods such as flow analysis. Or, the threshold may be automatically set by finding the change points in the casting waveform data using unsupervised learning for each casting shot.

[0047] Furthermore, the teacher data setting switch 530 selects and sets the casting waveform data that will serve as the teacher data for supervised learning performed by the second anomaly determination unit 57. Alternatively, the anomaly data determined by the first anomaly determination unit 55 may be automatically set in the teacher data setting switch 530. Also, the second anomaly determination start switch 531 may be automatically started in conjunction with the teacher data setting switch 530.

[0048] Note that the switch labels are omitted in Figure 2(b). Each switch is also equipped with a display panel EP that displays the selected item, the progress of unsupervised and supervised learning, and the judgment result.

[0049] In this way, the anomaly detection system 200, which performs anomaly detection of the casting apparatus 100 using the casting waveform data of the casting apparatus 100 in the anomaly detection unit 50, can accurately extract anomaly data without any omissions, and by performing anomaly detection based on the extracted anomaly data, a highly accurate anomaly detection system 200 for the casting apparatus 100 can be provided. As a result, highly accurate preventive maintenance of the casting apparatus 100 is made possible, enabling the stable production of high-quality castings. Furthermore, by using a method of setting a learning interval, the memory capacity of the casting waveform data can be compressed, enabling miniaturization and cost reduction of the anomaly detection system 200. Moreover, the accuracy of anomaly detection can be further improved by using an unsupervised learning program to extract anomaly data from the casting waveform data, and then using the extracted anomaly data as training data to perform anomaly detection of the casting apparatus 100 using a supervised learning program.

[0050] (Anomaly detection method) Next, a method for detecting abnormalities in a casting apparatus 100 using the abnormality detection system 200 shown in Figures 1 and 2, according to an embodiment of the present invention, will be described with reference to Figures 3 and 4.

[0051] First, casting waveform data transmitted from the injection control unit 38, the clamping control unit 27, and the peripheral equipment 99 of the casting apparatus 100 is received by the data transmission / reception unit 51 of the abnormality detection unit 50 via the casting control unit 40 (waveform data acquisition process).

[0052] From the casting waveform data collected in this waveform data acquisition process, the first learning unit 53 extracts reference waveform data based on the settings in the data setting unit 52. For example, let's assume that the abnormality detection selection switch 521 of the data setting unit 52 is set to chip galling, the measurement waveform selection switch 522 is set to injection speed waveform, and the axis selection switch 524 is set to injection position. The following explanation will be based on these assumed settings. Figure 3(a) shows the reference waveform data H1 with injection position on the horizontal axis and injection speed on the vertical axis, based on these assumed settings. This reference waveform data H1 is stored in the first learning unit 53 and displayed on the display unit 58.

[0053] Furthermore, the first learning unit 53 sets a learning interval (learning interval) for the reference waveform data H1 based on the setting of the learning interval setting switch 525 of the data setting unit 52. It also starts the first learning start switch 526 to define the characteristics of the reference waveform data H1 (called reference characteristics) within the set learning interval, and saves this learning interval and reference characteristics as data in the first learning unit 53. In this way, the first learning unit 53 extracts the reference waveform data H1, sets the learning interval, and defines the reference characteristics using an unsupervised learning program (first learning process).

[0054] Here, the learning interval is set using the learning interval setting switch 525, as shown in Figure 3(a). Note that the learning interval (S1, S2) may be set by, for example, a molding engineer operating the casting process, selecting points where changes in the casting waveform data are observed based on past casting performance. Alternatively, the analysis results may be set using analytical means such as flow analysis. Alternatively, an unsupervised learning program may be used for each casting shot to automatically determine and set the change points in the casting waveform data. Furthermore, the learning interval (S1, S2) may be initially set wider, and the change points are searched for as casting waveform data is collected, gradually narrowing the learning interval (S1, S2). This setting of the learning interval (S1, S2) compresses the memory capacity of the casting waveform data, shortening the time of the first learning process, improving learning accuracy, and enabling miniaturization and cost reduction of the anomaly detection system 200.

[0055] Furthermore, the definition of the reference features involves, for example, using a known deep learning image diagnostic method to extract the features (referred to as reference features) of the reference waveform data H1. For example, in Figure 3(a), the reference features of the reference waveform data H1 are defined as having a peak in the learning interval (S1, S2), having one peak, and having an ejection velocity less than or equal to the maximum value HS, and are set in the first learning unit 53.

[0056] Next, the second learning start switch 527 is activated, and as shown in Figure 3(b), the second learning unit 54 performs a comparative evaluation of the casting waveform data during production casting (referred to as evaluation waveform data H2) collected in the waveform data acquisition process with the reference waveform data H1. First, in the learning interval (S1, S2), the features of the evaluation waveform data H2 (referred to as evaluation features) are defined, similar to the definition of the reference features, using, for example, a known deep learning image diagnostic method. For example, in Figure 3(b), the evaluation features of the evaluation waveform data H2 are defined as: a peak exists in the learning interval (S1, S2), there is one peak, and the injection speed is less than or equal to the maximum value HS, and these are set in the second learning unit 54. Subsequently, the reference features and evaluation features set in the first learning unit 53 are compared to extract the differences between the reference waveform data H1 and the evaluation waveform data H2. The second learning unit 54 performs this process using an unsupervised learning program (second learning process).

[0057] In Figure 3(b), no difference is found between the reference characteristics of the reference waveform data H1 and the evaluation characteristics of the evaluation waveform data H2. Therefore, the second learning unit 54 determines that the evaluation waveform data H2 is normal waveform data H2. As a result, the second learning unit 54 determines that the casting apparatus 100 is in a normal state. Note that in Figure 3(b), the normal waveform data H2 is shown in a simplified form, but in reality, multiple normal waveform data H2 are displayed on the display unit 58 depending on the number of shots in production casting.

[0058] Next, as shown in Figure 3(c), we will explain the form in which differences were found between the evaluation characteristics of the casting waveform data (evaluation waveform data H3) collected in the waveform data acquisition process during production casting and the reference characteristics of the reference waveform data H1. For example, the evaluation characteristics of evaluation waveform data H3 are defined as having a peak in the learning interval (S1, S2), having two peaks, and the injection speed being greater than or equal to the maximum value HS. As a result, the reference characteristics of reference waveform data H1 and the evaluation characteristics of evaluation waveform data H3 differ in two ways, so the second learning unit 54 determines that evaluation waveform data H3 is abnormal waveform data H3. As a result, the second learning unit 54 determines that the casting apparatus 100 is in an abnormal state and starts numerical management of the abnormal state.

[0059] Specifically, the difference between the evaluation features of the abnormal waveform data H3 and the reference features of the reference waveform data H1 is quantified based on a pre-set rule, and counted and stored as an abnormal score value in the second learning unit 54. This process, up to the counting of the abnormal score value, is called the second learning process. For example, in Figure 3(c), the difference between the maximum injection velocity shown by the abnormal waveform data H3 and the maximum value HS of the injection velocity used to define the reference features is set as the abnormal score value K1. Alternatively, the difference between the injection velocity of the abnormal waveform data H3 and the reference waveform data H1 at the injection position showing the maximum injection velocity of the abnormal waveform data H3 is set as the abnormal score value K2. Or, the abnormal score value may be added (K1 + K2) from the evaluation feature of the abnormal waveform data H3 where there are two peaks in the learning interval (S1, S2).

[0060] In this invention, an unsupervised learning program is used in the first learning step performed in the first learning unit 53 and the second learning step performed in the second learning unit 54. In casting using the casting apparatus 100, many uncertain factors are intricately intertwined, such as the temperature and composition of the molten metal M, thermal expansion and deformation of the injection unit 30 including the injection sleeve 31 and plunger tip 32, thermal expansion of the casting mold 10, adhesion of impurities to the mold cavity 13 and abnormal mold residue of the casting, lubrication status of the sliding parts of the casting apparatus 100, and the influence of oil temperature in the case of hydraulic drive means. As a result, abnormalities occur in the casting apparatus 100, and abnormal waveform data H3 is confirmed. Furthermore, even for each individual uncertain factor, there are many cases where the abnormal waveform data H3 is small or unclear. Moreover, depending on the knowledge and skill level of the casting engineer, it is difficult to accurately recognize minor abnormal waveform data H3, and it is anticipated that accurate collection of abnormal waveform data may not be possible due to omissions and other issues. Generally, it is common practice to set this abnormal waveform data H3 as training data and use a supervised learning program. However, in casting using a casting apparatus 100 where it is difficult to accurately collect abnormal waveform data H3, it is difficult to accurately set the training data, and a supervised learning program cannot be definitively said to be a preferable method.

[0061] Therefore, in this invention, abnormality detection of the casting apparatus 100 is performed using an unsupervised learning program, utilizing normal waveform data H2, which is easy to collect and can be clearly and accurately set, and represents a state in which no abnormality has been detected in the casting apparatus 100 or the casting. This eliminates the need to set up training data, which has issues with accuracy, and improves the accuracy of abnormality detection of the casting apparatus 100.

[0062] Next, the first abnormality detection start switch 529 is activated, and the first abnormality detection unit 55 uses the abnormality score value counted in the second learning process to detect abnormalities in the casting apparatus 100. For example, as shown in Figure 4(a), a waveform (called the abnormality score value waveform ER) is used, with the number of casting shots in production casting on the horizontal axis and the abnormality score value on the vertical axis, showing the cumulative number of abnormality score values. This abnormality score value waveform ER is displayed on the display unit 58. For example, in Figure 4(a), the abnormality score value waveform ER is not displayed until the number of casting shots ES1, indicating that the casting apparatus 100 is in a normal state. It is preferable to complete the first learning process by this number of casting shots ES1. The abnormality score value waveform ER is first observed at the number of casting shots ES1, and the abnormality score value waveform ER shows an increasing trend as the number of casting shots increases. At this point, the abnormality score value waveform ER is small, and the first abnormality detection unit 55 determines that the abnormality level of the casting apparatus 100 is small. Furthermore, the first abnormality detection unit 55 determines that there is no need to immediately interrupt production casting and perform maintenance on the casting apparatus 100, and therefore continues production casting.

[0063] Here, based on the setting of the threshold setting switch 528, two thresholds are set in the first abnormality determination unit 55. One of the thresholds is threshold ERA, which alerts the molding technician that if production casting continues as is, an abnormality (chip galling) in the casting apparatus 100, as set by the abnormality detection selection switch 521 of the data setting unit 52, will occur with a high probability. When the number of casting shots ES2 in which the abnormality score value waveform ER exceeds the threshold ERA, an alarm prompting a warning is issued, and at the same time, a warning message is displayed on the display unit 58. Upon receiving this warning, it is preferable for the casting technician to temporarily suspend production casting and perform maintenance on the casting apparatus 100, and the appropriate timing for maintenance of the casting apparatus 100 can be indicated.

[0064] Another threshold is threshold ERZ, which indicates that a critical malfunction has already occurred in the casting apparatus 100 and that production casting should be stopped immediately to carry out large-scale maintenance. For example, in Figure 4(a), at the timing of casting shot number ES3 when the abnormal score value waveform ER exceeds threshold ERZ, an alarm is issued indicating that production casting should be stopped immediately and maintenance of the casting apparatus 100 should be performed, and at the same time, a warning is displayed on the display unit 58. It is preferable to set the thresholds (ERA, ERZ) by referring to, for example, past maintenance records, analysis results from analysis means such as flow analysis, and the history of malfunctions of the casting apparatus 100 stored in the abnormality detection unit 50. The first abnormality determination unit 55 detects abnormalities in the casting apparatus 100 using the abnormal score value waveform ER and thresholds (ERA, ERZ) (first abnormality determination step).

[0065] Next, as shown in Figure 4(b), following the waveform data acquisition step of the data transmission / reception unit 51, the first learning step of the first learning unit 53, the second learning step of the second learning unit 54, and the first abnormality determination step of the first abnormality determination unit 55, the process proceeds to the teacher data setting step. In this teacher data setting step, in the first abnormality determination step, the casting waveform data of the number of casting shots (ES2, ES3) where the abnormal score value waveform ER exceeds the threshold (ERA, ERZ) is registered as abnormal data in the teacher data setting unit 56. Furthermore, in the present invention, the teacher data setting unit 56 is characterized by setting the registered abnormal data as teacher data. Although two thresholds (ERA, ERZ) are used for registering abnormal data, it is preferable to use threshold ERA to register abnormal data and set the teacher data, as this indicates a suitable maintenance time for the casting apparatus 100.

[0066] Furthermore, the present invention is characterized by comprising a second abnormality determination step in which the second abnormality determination unit 57 performs an abnormality determination of the casting apparatus 100 using a supervised learning program based on the training data set in the training data setting step. Specifically, the casting waveform data collected in the waveform data acquisition step is compared with the training data set in the training data setting step, and if the casting waveform data ≥ the training data, the second abnormality determination unit 57 determines that an abnormality (for example, chip galling) has occurred in the casting apparatus 100, interrupts production casting, and performs maintenance on the casting apparatus 100.

[0067] Thus, in production casting using a casting apparatus 100 where the recognition and extraction of abnormal data is difficult, abnormal data is extracted using an unsupervised learning program based on casting waveform data of a normal state that can be selected with high accuracy and clarity. This ensures that abnormal data is extracted accurately without any omissions. Furthermore, the extracted abnormal data is set as training data, and abnormality detection of the casting apparatus 100 is performed using a supervised learning program. This provides a highly accurate method for detecting abnormalities in the casting apparatus 100. As a result, highly accurate preventive maintenance of the casting apparatus 100 is possible, enabling the stable production of high-quality castings. In addition, by setting a learning interval, the memory capacity of the casting waveform data can be compressed, enabling miniaturization and cost reduction of the abnormality detection system 200.

[0068] Although preferred embodiments of the present invention have been described above, the technical scope of the present invention is not limited to the embodiments described above. Various modifications or improvements can be made to the above embodiments. [Explanation of Symbols]

[0069] 100 Casting apparatus 200 Anomaly Detection Systems 10 Casting molds 11 Fixed mold 12. Movable molds 13 Mold Cavity 14 Mold Gate 20 Mold clamping part 21 Fixed plate 22 Movable plate 23 Mold clamping board 24 Tie Bar 25-type clamping drive unit 26 Cylinder rod 27 Type clamping control unit 30 Injection part 31 Injection Sleeve 32 plunger tips 33 Plunger Rod 34 pouring spouts 35 Connecting part 36 Injection drive unit 37 Drive Rod 38 Injection control unit 40 Casting Control Unit 50 Anomaly detection unit 51 Data transmission / reception unit 52 Data Setting Section 521 Anomaly detection selection switch 522 Measurement waveform selection switch 523 Reference waveform selection switch 524 Axis selection switch 525 Learning interval setting switch 526 First learning start switch 527 Second learning start switch 528 Threshold setting switch 529 First Anomaly Detection Start Switch 530 Teacher Data Setting Switch 531 Second Anomaly Detection Start Switch EP display panel 53. First Learning Department 54 Second Learning Department 55 1st abnormality determination section 56 Training Data Setting Unit 57 2nd abnormality determination section 58 Display section 99 Peripheral facilities M molten metal F forward R rear H1 Reference waveform data H2 Evaluation waveform data (normal waveform data) H3 Evaluation waveform data (abnormal waveform data) HS Maximum S1, S2 learning interval K1, K2 Abnormal Score Values ES1, ES2, ES3 number of casting shots ER abnormal score value waveform ERA, ERZ threshold

Claims

1. An anomaly detection system for a casting apparatus, which produces castings by injecting and filling molten metal into a mold cavity formed by clamping a casting mold, and which includes an anomaly detection unit that detects anomalies based on casting waveform data transmitted from the casting apparatus from the start to the end of the casting process, The casting apparatus comprises an injection sleeve having a pouring port for supplying the molten metal, and a plunger tip that slides within the injection sleeve, and is an apparatus that presses the molten metal supplied into the injection sleeve by the forward movement of the plunger tip and injects and fills it toward the mold cavity. The anomaly detection unit is configured to perform processing using an unsupervised learning program and determination using training data. The anomaly detection unit, in processing using the unsupervised learning program, In past production casting, the characteristics of the reference waveform data, which is the casting waveform data where the casting quality of the casting product satisfies a predetermined quality standard value, are defined. Defining the characteristics of the casting waveform data collected from the casting apparatus, The characteristics related to the aforementioned reference waveform data and the characteristics related to the aforementioned casting waveform data are compared and counted as abnormal score values, The cast waveform data whose abnormal score value exceeds a preset threshold is determined to be abnormal data, and the following is performed: The anomaly detection unit sets the anomaly data determined by processing using the unsupervised learning program as the training data. An anomaly detection system for a casting apparatus, characterized in that the anomaly detection unit performs an anomaly determination of the casting apparatus based on the comparison result between the casting waveform data collected from the casting apparatus and the training data in the determination using the training data.

2. The casting apparatus further comprises a vacuum suction means for vacuum suction within the mold cavity or the injection sleeve, An abnormality detection system for a casting apparatus according to claim 1, wherein the casting apparatus performs vacuum suction using the vacuum suction means and injects and fills the molten metal toward the mold cavity by the forward movement of the plunger tip.

3. An abnormality detection method for a casting apparatus using the abnormality detection system described in claim 1 or 2, A waveform data acquisition step for collecting the aforementioned casting waveform data, The process includes a step for detecting abnormalities in the casting apparatus, The process for detecting abnormalities in the casting apparatus includes a process of performing processing using an unsupervised learning program and a process of making a determination using training data. The process of performing processing using the aforementioned unsupervised learning program is: In past production casting, the characteristics of the reference waveform data, which is the casting waveform data where the casting quality of the casting product satisfies a predetermined quality standard value, are defined. Defining the characteristics of the casting waveform data collected from the casting apparatus, The characteristics related to the aforementioned reference waveform data and the characteristics related to the aforementioned casting waveform data are compared and counted as abnormal score values, This includes determining the casting waveform data whose abnormal score value exceeds a preset threshold as abnormal data, The step of detecting an anomaly in the casting apparatus includes setting the anomaly data determined by processing using the unsupervised learning program as the training data, A method for detecting abnormalities in a casting apparatus, wherein the step of making a determination using the training data includes making an abnormality determination of the casting apparatus based on the result of comparing the casting waveform data collected from the casting apparatus with the training data.

4. An anomaly detection system for a casting apparatus, which produces castings by injecting and filling molten metal into a mold cavity formed by clamping a casting mold, and which includes an anomaly detection unit that detects anomalies based on casting waveform data transmitted from the casting apparatus from the start to the end of the casting process, The casting apparatus comprises an injection sleeve having a pouring port for supplying the molten metal, and a plunger tip that slides within the injection sleeve, and is an apparatus that presses the molten metal supplied into the injection sleeve by the forward movement of the plunger tip and injects and fills it toward the mold cavity. The aforementioned abnormality detection unit, In past production casting, the characteristics of the reference waveform data, which is the casting waveform data where the casting quality of the casting product satisfies a predetermined quality standard value, are defined. Defining the characteristics of the casting waveform data collected from the casting apparatus, The characteristics related to the aforementioned reference waveform data and the characteristics related to the aforementioned casting waveform data are compared and counted as abnormal score values, The casting waveform data whose abnormal score value exceeds a preset threshold is determined to be abnormal data. The aforementioned abnormal data is set as training data, An abnormality detection system for a casting apparatus, characterized in that it is configured to perform the following: determine an abnormality in the casting apparatus based on the result of comparing the casting waveform data collected from the casting apparatus with the training data.

5. The casting apparatus further comprises a vacuum suction means for vacuum suction within the mold cavity or the injection sleeve, The abnormality detection system for a casting apparatus according to claim 4, wherein the casting apparatus performs vacuum suction using the vacuum suction means and injects and fills the molten metal toward the mold cavity by the forward movement of the plunger tip.

6. An abnormality detection method for a casting apparatus using the abnormality detection system described in claim 4 or 5, A waveform data acquisition step for collecting the aforementioned casting waveform data, In past production casting, a process of defining the characteristics of reference waveform data, which is the casting waveform data, in which the casting quality of the casting product satisfies a predetermined quality standard value, A step of defining the characteristics of the casting waveform data collected from the casting apparatus, A step of comparing the characteristics of the reference waveform data with the characteristics of the casting waveform data and counting them as abnormal score values, A step of determining the casting waveform data, in which the abnormal score value exceeds a preset threshold, as abnormal data, The process of setting the aforementioned abnormal data as training data, A method for detecting abnormalities in a casting apparatus, comprising the step of determining an abnormality in the casting apparatus based on the result of comparing the casting waveform data collected from the casting apparatus with the training data.

Images