JP7683199B2 - Anomaly detection system, anomaly detection device, anomaly detection method, program, and trained model - Google Patents

Description

The present invention relates to an anomaly detection system, an anomaly detection device, an anomaly detection method, a program, and a trained model.

There is a known injection molding device that supplies a molten liquid of a molding material to a cavity formed between multiple molds to form a molded product. For example, Patent Document 1 discloses a technology that detects the injection speed using an encoder attached to an injection servo motor, and stops the injection operation if the injection speed does not reach the monitored speed during the monitoring time of the injection process.

JP 2009-000929 A

When a gate becomes clogged in a molding device, the molding material does not flow into the cavity as designed, which can lead to defects in the molded product. The technology in Patent Document 1 detects gate clogging by taking advantage of the fact that when the gate becomes clogged, the load pressure during injection becomes higher than the load pressure when molding a good product, and the injection speed becomes slower than the injection speed when molding a good product.

However, the inventors discovered that even if a clog occurs in the gate, the load pressure during injection may not change much from when a good product is molded. In such cases, the technology in Patent Document 1 cannot detect a clogged gate.

The present invention aims to provide an anomaly detection system, an anomaly detection device, an anomaly detection method, a program, and a trained model that can more reliably detect gate blockages.

(1) The abnormality detection system disclosed herein is an abnormality detection system including a molding apparatus that molds a molded product and an abnormality detection device that detects an abnormality in the molding apparatus, wherein the molding apparatus includes a mold section having a plurality of gates and a cavity connected to the plurality of gates formed therein, an injection section that fills the cavity with molten molding material via the plurality of gates, a first sensor provided in a region of the plurality of gates closest to a first gate and that detects the pressure or temperature of the molding material in the cavity, and a second sensor provided in a region of the plurality of gates closest to a second gate different from the first gate and that detects the pressure or temperature of the molding material in the cavity, wherein the abnormality detection device includes a data acquisition section that acquires a first evaluation value calculated based on first time series data of pressure or temperature detected by the first sensor and a second evaluation value calculated based on second time series data of pressure or temperature detected by the second sensor, and an abnormality detection section that detects blockage of the first gate or the second gate based on the first evaluation value and the second evaluation value.

When the first gate or the second gate is clogged, a difference occurs between the first time series data and the second time series data. To make it easier to compare this difference, the first time series data is quantified as a first evaluation value, and the second time series data is quantified as a second evaluation value, and a clog in the first gate or the second gate is detected based on these first evaluation value and second evaluation value. This makes it possible to detect clogs in some gates that could not be detected in the past. As a result, a clog in the gate can be detected more reliably.

(2) Preferably, a first distance traveled by the molding material from the first gate to the first sensor is equal to a second distance traveled by the molding material from the second gate to the second sensor.

By configuring in this way, when the gate is "not" clogged, the time at which the molding material reaches the first sensor and the second sensor can be aligned. This makes it possible to ensure that a difference occurs between the first time series data and the second time series data only when some of the gates are clogged. As a result, the accuracy of clog detection can be improved.

(3) Preferably, the first evaluation value includes a first rising time point of the first time series data or a first peak time point at which the pressure or temperature of the first time series data is at a maximum value, and the second evaluation value includes a second rising time point of the second time series data or a second peak time point at which the pressure or temperature of the second time series data is at a maximum value, and the abnormality detection unit detects a blockage of the first gate when the first rising time point is later than the second rising time point or when the first peak time point is later than the second peak time point.

(4) Preferably, the first sensor and the second sensor detect the pressure of the molding material, the first evaluation value includes a first peak value that is the maximum pressure value of the first time series data, the second evaluation value includes a second peak value that is the maximum pressure value of the second time series data, and the abnormality detection unit detects a blockage of the first gate when the first peak value is lower than the second peak value.

(5) Preferably, the first sensor and the second sensor detect the pressure of the molding material, the first evaluation value includes a first integral value which is a time integral from a first rising point of the pressure of the first time series data to a first peak point at which the pressure of the first time series data is at its maximum value, the second evaluation value includes a second integral value which is a time integral from a second rising point of the pressure of the second time series data to a second peak point at which the pressure of the second time series data is at its maximum value, and the abnormality detection unit detects a blockage of the first gate when the first integral value is smaller than the second integral value.

According to the configurations (3) to (5) above, the first evaluation value includes at least one of the first rising time, the first peak time, the first peak value, and the first integral value, and the second evaluation value includes at least one of the second rising time, the second peak time, the second peak value, and the second integral value corresponding to the first evaluation value. By comparing the first rising time and the second rising time, the first peak time and the second peak time, the first peak value and the second peak value, or the first integral value and the second integral value, it is possible to identify that the first gate is clogged out of the first and second gates. In other words, the abnormality detection system not only detects clogging of some gates that could not be detected in the past, but also identifies which gate is clogged.

(6) Preferably, the anomaly detection unit detects blockage of at least one of the first gate and the second gate by inputting the first evaluation value and the second evaluation value into a trained model that has been machine-learned to determine the correlation between the first evaluation value and the second evaluation value and the blockage of the multiple gates.

With this configuration, once the trained model is generated, it becomes possible to detect blockages in some gates from the first evaluation value and the second evaluation value. By using the trained model, blockages in some gates can be detected more accurately even when there is variation in molding conditions.

(7) The abnormality detection device disclosed herein is an abnormality detection device that detects an abnormality in a molding device that molds a molded product, the molding device having a mold section having a plurality of gates and a cavity connected to the plurality of gates formed therein, an injection section that fills the cavity with a molten molding material via the plurality of gates, a first sensor provided in a region of the plurality of gates closest to a first gate and detecting the pressure or temperature of the molding material in the cavity, and a second sensor provided in a region of the plurality of gates closest to a second gate different from the first gate and detecting the pressure or temperature of the molding material in the cavity, the abnormality detection device having a data acquisition section that acquires a first evaluation value calculated based on first time series data of the pressure or temperature detected by the first sensor and a second evaluation value calculated based on second time series data of the pressure or temperature detected by the second sensor, and an abnormality detection section that detects clogging of at least one of the first gate and the second gate based on the first evaluation value and the second evaluation value.

When the first gate or the second gate is clogged, a difference occurs between the first time series data and the second time series data. To make it easier to compare this difference, the first time series data is quantified as a first evaluation value, and the second time series data is quantified as a second evaluation value, and a clog in the first gate or the second gate is detected based on these first evaluation value and second evaluation value. This makes it possible to detect clogs in some gates that could not be detected in the past. As a result, a clog in the gate can be detected more reliably.

(8) The abnormality detection method disclosed herein is an abnormality detection method for detecting an abnormality in a molding device that includes a mold section having a plurality of gates and a cavity connected to the plurality of gates formed therein, and an injection section that fills the cavity with a molten molding material via the plurality of gates, and includes a data acquisition process for acquiring a first evaluation value calculated based on first time series data of the pressure or temperature of the molding material in the cavity detected by a first sensor provided in an area of the plurality of gates that is closest to a first gate, and a second evaluation value calculated based on second time series data of the pressure or temperature of the molding material in the cavity detected by a second sensor provided in an area of the plurality of gates that is closest to a second gate different from the first gate, and an abnormality detection process for detecting a blockage of at least one of the first gate and the second gate based on the first evaluation value and the second evaluation value.

When the first gate or the second gate is clogged, a difference occurs between the first time series data and the second time series data. To make it easier to compare this difference, the first time series data is quantified as a first evaluation value, and the second time series data is quantified as a second evaluation value, and a clog in the first gate or the second gate is detected based on these first evaluation value and second evaluation value. This makes it possible to detect clogs in some gates that could not be detected in the past. As a result, a clog in the gate can be detected more reliably.

(9) The program disclosed herein is a program for detecting an abnormality in a molding device having a mold section in which a plurality of gates and a cavity connected to the plurality of gates are formed, and an injection section that fills the cavity with a molten molding material via the plurality of gates, and causes a computer device to execute a data acquisition process for acquiring a first evaluation value calculated based on first time series data of the pressure or temperature of the molding material in the cavity detected by a first sensor provided in an area of the plurality of gates that is closest to a first gate, and a second evaluation value calculated based on second time series data of the pressure or temperature of the molding material in the cavity detected by a second sensor provided in an area of the plurality of gates that is closest to a second gate different from the first gate, and an abnormality detection process for detecting a blockage of at least one of the first gate and the second gate based on the first evaluation value and the second evaluation value.

When the first gate or the second gate is clogged, a difference occurs between the first time series data and the second time series data. To make it easier to compare this difference, the first time series data is quantified as a first evaluation value, and the second time series data is quantified as a second evaluation value, and a clog in the first gate or the second gate is detected based on these first evaluation value and second evaluation value. This makes it possible to detect clogs in some gates that could not be detected in the past. As a result, a clog in the gate can be detected more reliably.

(10) The trained model of the present disclosure is a trained model for detecting an abnormality in a molding device including a mold section having a plurality of gates and a cavity connected to the plurality of gates formed therein, and an injection section that fills the cavity with molten molding material via the plurality of gates, wherein the explanatory variables include a first evaluation value calculated based on first time series data of the pressure or temperature of the molding material in the cavity detected by a first sensor provided in an area of the plurality of gates that is closest to a first gate, and a second evaluation value calculated based on second time series data of the pressure or temperature of the molding material in the cavity detected by a second sensor provided in an area of the plurality of gates that is closest to a second gate different from the first gate, and the objective variable is a trained model including a value related to a clogged gate among the plurality of gates.

When the first gate or the second gate is clogged, a difference occurs between the first time series data and the second time series data. To make it easier to compare this difference, the first time series data is quantified as a first evaluation value, and the second time series data is quantified as a second evaluation value, and a clog in the first gate or the second gate is detected based on these first evaluation value and second evaluation value. This makes it possible to detect clogs in some gates that could not be detected in the past. As a result, a clog in the gate can be detected more reliably.

The present invention makes it possible to more reliably detect gate blockages.

1 is a block diagram illustrating a schematic diagram of an anomaly detection system according to a first embodiment. FIG. 2 is an explanatory diagram conceptually showing the molding device shown in FIG. 1 . FIG. 2 is an explanatory diagram conceptually showing the molding device shown in FIG. 1 . FIG. 4 is a cross-sectional view showing the mold part of FIG. 3 . 5 is a cross-sectional view showing a mold part cut along the cutting line of arrow V in FIG. 4 . FIG. 13 is a diagram illustrating gate clogging. 1 is an example of a graph showing time series data of pressure. 1 is an example of a graph showing time series data of temperature. 1 is a block diagram showing a functional configuration of a learning device according to a first embodiment. 1 is a block diagram showing a functional configuration of an abnormality detection device according to a first embodiment; FIG. 11 is a block diagram illustrating an anomaly detection system according to a second embodiment. FIG. 11 is a block diagram showing the functional configuration of an abnormality detection device according to a second embodiment.

First Embodiment
Hereinafter, a first embodiment of the present invention will be described with reference to the drawings.

<Overall configuration of anomaly detection system>
1 is a block diagram illustrating a schematic diagram of an abnormality detection system 10 according to a first embodiment. The abnormality detection system 10 includes a plurality of molding devices 20, a learning device 30, an abnormality detection device 40, an input unit 50, and a display unit 60.

The molding device 20, the learning device 30, the anomaly detection device 40, the input unit 50, and the display unit 60 are each configured to be able to communicate wirelessly or wired. The learning device 30 and the anomaly detection device 40 are configured by an information processing device (computer device) having a calculation unit (e.g., a CPU, a GPU, etc.) and a storage unit (e.g., a HDD, an SSD, etc.). The learning device 30 and the anomaly detection device 40 may be configured by the same information processing device, or may be configured by separate information processing devices.

In this embodiment, multiple molding devices 20 are connected to one learning device 30 and one abnormality detection device 40, and the learning device 30 and the abnormality detection device 40 perform learning and abnormality detection based on various data transmitted from the multiple molding devices 20. Note that the molding devices 20 may have a one-to-one correspondence with the learning devices 30 and the abnormality detection devices 40. In other words, the abnormality detection system 10 may have one molding device 20, or may have multiple learning devices 30 and abnormality detection devices 40.

The input unit 50 is, for example, a keyboard or a mouse, and accepts various inputs from the operator. The display unit 60 is, for example, a display or a speaker, and displays various information in the anomaly detection system 10. The input unit 50 and the display unit 60 may be integrated, for example, as a touch panel. In addition, the input unit 50 and the display unit 60 may be provided as a portable terminal device that can be moved to a location away from the molding device 20, the learning device 30, and the anomaly detection device 40.

<General configuration of molding device>
Figures 2 and 3 are explanatory diagrams conceptually showing the molding apparatus 20. In Figures 2 and 3, cross-sectional areas are hatched. The molding apparatus 20 has a bed 21, an injection section 22, a mold clamping section 23, a mold section 24, a plurality of sensors 25, a temperature sensor 26, and a control panel 27. Figure 2 shows the molding apparatus 20 with the mold section 24 open, and Figure 3 shows the molding apparatus 20 with the mold section 24 assembled. The molding apparatus 20 is an apparatus that performs mold clamping type injection molding.

The control panel 27 has a control unit 271 and a communication unit 272. The control unit 271 is electrically connected to each drive unit (motor 237, etc.) of the molding device 20, and outputs operation commands to each drive unit. The control unit 271 is also electrically connected to each sensor (sensor 25, etc.) of the molding device 20, and signals detected by each sensor are input to the control unit 271. The control unit 271 is composed of an information processing device having a calculation unit (e.g., CPU, GPU, etc.) and a storage unit (e.g., HDD, SSD, etc.).

The communication unit 272 communicates with other units (such as the learning device 30) of the anomaly detection system 10. For example, the communication unit 272 transmits signals detected by the sensors to the learning device 30 or the anomaly detection device 40. The communication unit 272 also receives detection information, described below, from the anomaly detection device 40.

The mold clamping unit 23 has a fixed platen 231, a movable platen 232, tie bars 233, a ball screw 234, a support platen 235, a mold clamping force sensor 236, and a motor 237. The fixed platen 231 and the support platen 235 are fixed to the bed 21. The support platen 235 supports the ball screw 234. The ball screw 234 is connected to the motor 237. When the motor 237 is rotated by an operation command from the control unit 271, the ball screw 234 moves. The movable platen 232 is fixed to the end of the ball screw 234 opposite the end connected to the motor 237.

Here, in the molding device 20, the direction in which the ball screw 234 moves is referred to as the "axial direction." The side on which the motor 237 is located relative to the ball screw 234 is referred to as "one side" in the axial direction, and the side on which the movable platen 232 is located relative to the ball screw 234 is referred to as "the other side" in the axial direction.

The movable platen 232 moves in the axial direction in accordance with the movement of the ball screw 234. A through hole 232a is formed in the movable platen 232, penetrating it in the axial direction. One axial end of the tie bar 233 is fixed to the support platen 235, and the other axial end is fixed to the fixed platen 231. The tie bar 233 is inserted into the through hole 232a of the movable platen 232. In this way, the tie bar 233 guides the axial movement of the movable platen 232.

The clamping force sensor 236 detects the pressure (reaction force of the clamping force) applied to the support platen 235 from the ball screw 234. The clamping force sensor 236 outputs a detection signal related to the pressure to the control unit 271. The clamping force sensor 236 may be installed in another position as long as it is a position where the clamping force in the mold section 24 described below can be detected. The fixed platen 231 is formed with a through hole 231a whose diameter expands on the other axial side. A cylinder 222 described below is inserted into the through hole 231a.

The mold section 24 has multiple molds 241, 242. The mold 241 is fixed to the movable platen 232. When the ball screw 234 moves in the axial direction, the mold 241 also moves in the axial direction together with the movable platen 232. In other words, the mold 241 is a movable mold. The mold 242 is fixed to the fixed platen 231. In other words, the mold 242 is a fixed mold.

Refer to Figure 3. When the mold 241 moves to the other axial direction by the ball screw 234 and contacts the mold 242 (i.e., when the multiple molds 241, 242 are combined), a cavity C1 is formed between the molds 241, 242. In this embodiment, the cavity C1 is an annular space that is filled with the molding material.

Figure 4 is an enlarged cross-sectional view of the mold part 24 shown in Figure 3. The mold 242 has formed therein a spool 243, a plurality of runners 244 (four in this embodiment) branching off from the spool 243, and a plurality of gates 245 (four in this embodiment) each connected to the plurality of runners 244. The cavity C1 is connected to the plurality of gates 245.

The spool 243 is a passage through which the molding material injected from the injection section 22 into the mold section 24 first flows. The end 243a of the spool 243 on the injection section 22 side opens so as to communicate with the nozzle 224 described below. The inner diameter of the spool 243 is smallest at the end 243a, and the inner diameter increases from the end 243a to the runner 244.

The multiple runners 244 branch off slightly toward the end 243a of the spool 243, away from the end 243b on the cavity C1 side (i.e., the side opposite the end 243a). The space on the spool 243 closer to the end 243b than the position where the runners 244 branch off is a resin reservoir for storing cold slug, and is also called a "cold slug well."

FIG. 5 is a cross-sectional view showing the mold part 24 taken along the cutting line indicated by the arrow V in FIG.
The gates 245 correspond one-to-one to the runners 244. The inner diameter of the gates 245 decreases as the gates approach the cavity C1. Hereinafter, when the gates 245 are to be particularly distinguished from one another, a specific gate 245 (e.g., the gate 245 located at the top of FIG. 5) will be referred to as gate GT1, and the remaining three gates 245 will be referred to as gates GT2, GT3, and GT4, respectively, in the clockwise direction from the gate GT1.

If the position of gate GT1 in mold 24 is 0 degrees, gate GT2 is at 90 degrees, gate GT3 is at 180 degrees, and gate GT4 is at 270 degrees. Gate GT3 is located on the opposite side of gate GT1, and gate GT4 is located on the opposite side of gate GT2.

The multiple sensors 25 are provided at positions facing the cavity C1 of the mold 241 and the mold 242. The multiple sensors 25 each detect the pressure of the molding material in the cavity C1. The sensor 25 may also be a sensor that detects temperature. The sensor 25 outputs a detection signal related to the pressure (or temperature) to the control unit 271. In this embodiment, as an example, the multiple sensors 25 are provided in the mold 241, but the sensor 25 may also be provided in the mold 242.

The multiple sensors 25 are provided at positions corresponding to the multiple gates 245, respectively. In this embodiment, the four sensors 25 correspond one-to-one to the four gates 245. Hereinafter, when making a particular distinction, the sensor 25 corresponding to gate GT1 among the multiple sensors 25 will be referred to as sensor SN1. Similarly, the sensors 25 corresponding to gates GT2, GT3, and GT4, respectively, will be referred to as sensors SN2, SN3, and SN4.

Sensor SN1 (corresponding to the "first sensor" in this disclosure) is provided in the region of the multiple gates 245 that is closest to gate GT1 (corresponding to the "first gate" in this disclosure). In other words, the distance from sensor SN1 to gate GT1 is shorter than the distance from sensor SN1 to each of the other gates GT2, GT3, and GT4. In other words, the sensor 25 closest to gate GT1 is sensor SN1.

Similarly, sensors SN2, SN3, and SN4 (all of which correspond to the "second sensor" in this disclosure) are provided in the area of the multiple gates 245 that is closest to gates GT2, GT3, and GT4 (all of which correspond to the "second gate" in this disclosure).

The first distance d1 that the molding material travels from gate GT1 to sensor SN1 is equal to the second distance d2 that the molding material travels from gate GT2 to sensor SN2 (d1 = d2). The first distance d1 is also equal to the third distance d3 and the fourth distance d4 that the molding material travels from gate GT3 to sensor SN3 and from gate GT4 to sensor SN4, respectively (d1 = d3 = d4). In other words, the distances from the multiple sensors 25 to the nearest gate 245 are all equal.

Refer to FIG. 4. The temperature sensor 26 is built into the mold 242 and detects the temperature of the mold 242. The temperature sensor 26 outputs a detection signal related to the temperature to the control unit 271. The temperature sensor 26 may be installed in the mold 241. The temperature sensor 26 may also be installed in the cylinder 222. In other words, it is sufficient for the temperature sensor 26 to be able to directly or indirectly detect the temperature of the molding material supplied into the cavity C1.

Refer to FIG. 2. The injection section 22 has a hopper 221, a cylinder 222, a screw 223, a ball screw 225, a motor 226, a pressure sensor 227, a movement amount sensor 228, and a heater 229. The hopper 221 is connected to the cylinder 222 and supplies molding material into the cylinder 222. The cylinder 222 is a hollow cylindrical member extending in the axial direction. The diameter of one end of the cylinder 222 on one axial side narrows as it approaches the extreme end on one radial side, and a nozzle 224 is provided at the extreme end on one axial side of the cylinder 222. The nozzle 224 is connected to a spool 243 of a mold 242.

The screw 223 is inserted into the cylinder 222 from the other axial end of the cylinder 222. A ball screw 225 is connected to the other axial end of the screw 223, and a motor 226 is connected to the other axial end of the ball screw 225. When the motor 226 rotates in response to an operation command from the control unit 271, the ball screw 225 moves in the axial direction. Accordingly, the screw 223 also moves in the axial direction. At this time, the screw 223 rotates in the circumferential direction with the axial direction as the central axis.

The pressure sensor 227 detects the pressure applied from the ball screw 225 to the motor 226 (the reaction force of the pushing force of the screw 223). In other words, the pressure sensor 227 detects the pressure that the screw 223 receives from the molding material. The pressure sensor 227 outputs a detection signal related to the pressure to the control unit 271. Note that the pressure sensor 227 may be installed in another position as long as it is a position where the pushing force of the screw 223 can be detected.

The movement amount sensor 228 detects the amount of axial movement of the ball screw 225. The movement amount sensor 228 outputs a detection signal related to the amount of movement to the control unit 271. Note that the movement amount sensor 228 may be installed in another position as long as it is a position where the amount of axial movement of the ball screw 225 can be detected.

The heater 229 is, for example, a resistance heater formed by winding a resistance wire into a coil. The heater 229 heats the inside of the cylinder 222 by resistance heat when a current flows through the resistance wire in response to an operation command from the control unit 271.

<Manufacturing method using a molding device>
2 and 3, a method for manufacturing a molded product using the molding device 20 will be described. The method for manufacturing a molded product using the molding device 20 includes a pre-process ST1, a mold clamping process ST2, a filling process ST3, a pressure holding process ST4, a pressure releasing process ST5, and a demolding process ST6, which are performed in this order. In this embodiment, the molded product is a resin cage used for a rolling bearing. However, this is only one example of a molded product, and the molded product molded by the molding device according to the present invention may be a molded product of other shapes and applications.

Refer to FIG. 2. First, the pre-process ST1 is carried out. In the pre-process ST1, the motor 226 rotates the screw 223, and while the inside of the cylinder 222 is heated by the heater 229, pellets of the molding material are supplied from the hopper 221 into the cylinder 222. The pellets of the molding material melt in the cylinder 222 due to frictional heat caused by the rotation of the screw 223 and heating by the heater 229, becoming molten molding material (melting process).

Next, the screw 223 rotates and moves to the other axial side, so that a predetermined amount of molding material is stored in the cylinder 222 on one axial side of the screw 223 (measurement process). This completes the pre-process ST1.

Next, when the mold clamping process ST2 is started, in the molding device 20 in the state shown in FIG. 2, the ball screw 234 moves to the other axial side in response to an operation command from the control unit 271, and the mold 241 is brought into contact with the mold 242 as shown in FIG. 3. With the mold 241 and the mold 242 combined in this manner, the ball screw 234 further presses the mold 241 against the mold 242 in the other axial direction with a predetermined mold clamping force. In other words, the multiple molds 241, 242 are clamped together. As a result, a circular cavity C1 is formed between the multiple molds 241, 242. This completes the mold clamping process ST2.

Here, the mold clamping force is one of the molding conditions and is determined according to other molding conditions such as the shapes of the molds 241 and 242. The mold clamping force is detected by the mold clamping force sensor 236.

Next, when the filling step ST3 is started, the ball screw 225 moves to one side in the axial direction while maintaining the above-mentioned mold clamping force. This causes the screw 223 to push the molding material to one side in the axial direction, and the molten molding material is injected from the nozzle 224 of the cylinder 222 through the spool 243, the multiple runners 244, and the multiple gates 245 into the cavity C1 (filling operation).

When the molding material is filled into the cavity C1, the filling step ST3 is completed. In the filling step ST3, the molten molding material is supplied to the cavity C1 while gradually solidifying from near the surfaces of the molds 241 and 242.

Next, when the pressure holding process ST4 is started, the screw 223 pushes the molding material further to one side in the axial direction, and the molding material is further injected from the nozzle 224 of the cylinder 222 into the cavity C1. This applies a predetermined pressure (e.g., several tens to several hundred MPa) to the molding material filled in the cavity C1. The screw 223 then maintains this state for a predetermined time, thereby continuing to apply the predetermined pressure to the molding material for a predetermined time (e.g., several seconds) (pressure holding operation). The pressure (pressure) with which the screw 223 pushes the molding material into the cavity C1 is detected by the pressure sensor 227.

Next, when the pressure release step ST5 is started, the screw 223 moves to the other axial side and releases the pressure of the molding material (pressure release operation). After a predetermined time has elapsed after the pressure release operation, the pressure release step ST5 ends when the pressure of the molding material in cavity C1 falls below a predetermined value. Then, when the demolding step ST6 is started, the mold section 24 is cooled, causing the molding material in cavity C1 to solidify and form a molded product. Then, the ball screw 234 moves to one axial side, and the mold 241 separates from the mold 242, allowing the molded product to be removed. Note that cooling of the mold section 24 may be started simultaneously with the pressure release step ST5.

<About gate clogging>
Fig. 6 is a diagram for explaining clogging of the gate 245 detected in this embodiment. Fig. 6 shows a state in which the gate GT1 is clogged with a foreign object Fm1 as an example of clogging. Fig. 6(a) is a diagram showing a state in the middle of the filling step ST3, and Fig. 6(b) is a diagram showing a state in the middle of the pressure holding step ST4. Figs. 6(a) and (b) both show the same cross section as Fig. 5.

As shown in FIG. 6(a), when the filling step ST3 is performed, the molding material L1 flows into the cavity C1 via the spool 243, the runner 244, and the gate 245. Here, because the gate GT1 is clogged with a foreign matter Fm1, the molding material L1 does not flow into the cavity C1 from the gate GT1, but flows into the cavity C1 from the other gates GT2, GT3, and GT4.

Conventionally, the mold section 24 was not provided with a sensor 25, and the pressure of the molding material L1 was monitored by a pressure sensor 227 provided in the injection section 22. For example, when all gates 245 are clogged, the pressure rise of the pressure sensor 227 in the filling process ST3 becomes clearly faster than normal (when there is no clogging), so it is possible to detect the clogging of the gate 245 even with the conventional configuration.

However, if some of the gates 245 (particularly less than half of the multiple gates 245) are clogged, as shown in FIG. 6(a), molding material L1 flows into cavity C1 from the other unclogged gates GT2, GT3, and GT4, so the pressure detected by pressure sensor 227 is almost the same as the normal pressure. For this reason, it is not possible to detect the clogging of some of the gates 245 based on the pressure of pressure sensor 227.

Also, as shown in FIG. 6(b), the molding material L1 flowing in from the other gates GT2, GT3, and GT4 eventually flows into the cavity C1 near the gate GT1, so that the cavity C1 is filled with the molding material L1. As a result, even if an abnormality occurs in which some of the gates 245 are clogged, a molded product is still produced.

However, because molding material L1 does not flow into cavity C1 near clogged gate GT1 as usual, there is a risk that the dimensions (e.g., circularity), weight, and quality (e.g., strength) of the molded product will deviate from the design range. For example, a weld area may form in an unintended area (an area other than the designed area), reducing the strength of the molded product. A molded product that does not fall within the design range will be defective. In addition, because the density of molding material L1 is low near gate GT1, there is a risk that abnormalities such as sink marks and voids will occur in the molded product.

Conventionally, such abnormalities in molded products have been detected by measuring the dimensions, appearance, weight, etc. of the molded product. In the molding device 20, molding of molded products is performed continuously, so with conventional detection methods, abnormal molded products continue to be wastefully molded after molding until the molded product is measured outside the molding device 20 and the abnormality is detected. For this reason, it is important to have technology that can sense blockages in some of the gates 245 in the molding device 20 and determine abnormalities immediately after molding without having to measure the molded product.

The inventors therefore came up with the idea of providing multiple sensors 25 at positions corresponding to multiple gates 245 in cavity C1, and detecting blockage of some of the gates 245 based on the difference in time series data of the pressure detected by these multiple sensors 25. Below, we will explain the difference in time series data caused by blockage of some of the gates 245.

<Discrepancies in time series data due to clogging of some gates>
Fig. 7 is an example of a graph showing time series data of pressure detected by a plurality of sensors 25. In Fig. 7, the vertical axis represents pressure, and the horizontal axis represents time.
In this embodiment, the origin of time (t=0) is the time when the control unit 271 issues an operation command to the motor 226 to move the screw 223 to the other axial side. In other words, it is the time when the injection unit 22 starts to supply the molding material to the mold unit 24 in the filling process ST3. Note that the origin of time (t=0) is not limited to this, and may be the time when the pressurization sensor 227 detects a predetermined pressure, or the time when the movement amount sensor 228 detects a predetermined movement amount.

Graph line F11 in FIG. 7 shows time series data of pressure detected by sensor SN1 (corresponding to the "first time series data" in this disclosure). Similarly, graph line F12 shows time series data of pressure detected by sensor SN2, graph line F13 shows time series data of pressure detected by sensor SN3, and graph line F14 shows time series data of pressure detected by sensor SN4 (all of which correspond to the "second time series data" in this disclosure). Graph line AvF1 shows time series data of the average pressure of the four graph lines F11 to F14. Graph lines F11 to F14 may be plots of the pressures detected by sensors SN1 to SN4 at each predetermined time, or may be plots of moving averages (e.g., five-point moving averages) of the pressures detected at each predetermined time.

Although the positions at which the graph lines rise and the peak values are different, the overall trends of graph lines F11 to F14 are consistent. When the filling process ST3 starts, the pressure of graph lines F11 to F14 rises and then increases monotonically. This is because molding material L1 is continuously supplied from the injection section 22, and molding material L1 presses against sensors SN1 to SN4 increasingly strongly.

The monotonically increasing graph lines F11 to F14 continues until halfway through the pressure holding process ST4. The pressure of graph lines F11 to F14 reaches a maximum value halfway through the pressure holding process ST4, and then monotonically decreases until the end of the pressure holding release process ST5. The monotonically increasing pressure until halfway through the pressure holding process ST4 is due to additional pressure being applied from the injection section 22 to the molding material L1 filled in the cavity C1. The monotonically decreasing pressure that begins halfway through the pressure holding process ST4 is due to the gate 245 being sealed by the cooled and solidified molding material, the application of pressure from the injection section 22 to the cavity C1 ceasing, and the molding material L1 in the cavity C1 cooling and shrinking.

Next, we will look at the individual trends of graph lines F11 to F14. First, graph line F11 of sensor SN1, which corresponds to the clogged gate GT1, rises later than the other graph lines F12 to F14. This is because, as shown in Figure 6, molding material L1 wraps around to the position of sensor SN1 from gates GT2 to GT4, so the point at which molding material L1 reaches sensor SN1 is later than the points at which it reaches the other sensors SN2 to SN4.

For the same reason, the peak time Xt1 at which the pressure of graph line F11 reaches its maximum value Pt1 is later than the peak times Xt2 to Xt4 of the other graph lines F12 to F14 (Figure 7 shows, as a representative example, the peak time Xt3 at which the pressure of graph line F13 reaches its maximum value Pt3).

The time when the pressure rises can be defined as appropriate, but in this embodiment, the time when a predetermined pressure Ps1 is reached is taken as the rise time. The predetermined pressure Ps1 is, for example, a value determined based on the average value AvPt of the maximum pressures of graph lines F11 to F14, and more specifically, a value that is a predetermined percentage (for example, 10 to 20%) of the average value AvPt. Note that the time when the slope of the pressure reaches a predetermined value for each of graph lines F11 to F14 may also be taken as the respective rise time.

As shown in FIG. 7, the rising point Xs1 of the graph line F11 is later than the rising points Xs2 to Xs4 of the other graph lines F12 to F14 (FIG. 7 shows the rising point Xs3 of the graph line F13 as a representative example). Also, the rising point Xs1 is later than the rising point AXs of the graph line AvF1.

The maximum pressure value Pt1 of the graph line F11 is lower than the maximum pressure values Pt2 to Pt4 of the other graph lines F12 to F14 (in FIG. 7, the maximum pressure value Pt3 of the graph line F13 is representatively shown) and the maximum pressure value AvPt of the graph line AvF1. This is because the molding material L1 that flows into the cavity C1 contracts as it is cooled by the mold section 24. Because the gate GT1 is clogged, the time it takes for the molding material L1 to reach the sensor SN1 is longer than the time it takes for the molding material L1 to reach the other sensors SN2 to SN4, and the molding material L1 contracts by that time. Due to this contraction, the force with which the molding material L1 presses the sensor SN1 is weaker than the force with which the molding material L1 presses the other sensors SN2 to SN4, so the maximum pressure value Pt1 of the graph line F11 is lower than the maximum values of the other graph lines F12 to F14.

Based on the above, if any of the following tendencies (1) to (3) is observed, there is a high possibility that the gate GT1 is clogged.
(1) The rising point Xs1 of the graph line F11 is later than the rising points Xs2 to Xs4 of the other graph lines F12 to F14. (2) The peak point Xt1 of the graph line F11 is later than the peak points Xt2 to Xt4 of the other graph lines F12 to F14. (3) The maximum pressure value Pt1 of the graph line F11 is lower than the maximum pressure values Pt2 to Pt4 of the other graph lines F12 to F14.

In this way, by focusing on the differences in the multiple time series data detected by the multiple sensors 25 corresponding to the multiple gates 245, it is possible to detect that some of the gates 245 are clogged. Although an example in which gate GT1 is clogged is described in Figures 6 and 7, a similar tendency is observed when any of the other gates GT2 to GT4 are clogged. For example, when only gate GT2 is clogged, the rising time point Xs2 and peak time point Xt2 of graph line F12 become relatively slow, and the maximum pressure value Pt2 of graph line F12 becomes relatively low.

Here, due to noise or the like, the pressure of graph line F11 may suddenly exceed the specified pressure Ps1 at a time earlier than the rising points Xs2 to Xs4. In this case, the rising point Xs1 of graph line F11 will be earlier than the rising points Xs2 to Xs4. Similarly, due to the influence of noise or the like, the maximum value Pt1 may suddenly exceed the other maximum values Pt2 to Pt4. In this case, if a comparison is made based only on the rising points or maximum values, blockage of some of the gates 245 cannot be correctly detected.

In order to reduce the effects of noise, etc., the time integrals of graph lines F11 to F14 may be compared instead of (or in addition to) comparing the rising time and maximum value. As the time integral, for example, integral value IV1 is calculated, which is the time integral from rising time Xs1 to peak time Xt1 of the pressure of graph line F11. Similarly, integral values IV2 to IV4 are calculated for graph lines F12 to F14, which are the time integral from rising time Xs2 to Xs4 to peak time Xt2 to Xt4. When gate GT1 is clogged, integral value IV1 is smaller than integral values IV2 to IV4. Furthermore, integral value IV1 is smaller than average value AIV of integral values IV1 to IV4.

<Difference in time series data when a sensor detects temperature>
Fig. 8 is an example of a graph showing time series data of temperatures detected by a plurality of sensors 25. In Fig. 8, the vertical axis represents temperature and the horizontal axis represents time.

Here, Figure 7 explains the difference in time series data when multiple sensors 25 detect pressure. However, each of the multiple sensors 25 may detect temperature. When detecting pressure, the surface of the sensor 25 (or an indirect member that presses the surface of the sensor 25) needs to face the cavity C1 and be in contact with the molding material L1. For this reason, if there is no space to expose the sensor 25 within the cavity C1, it is difficult to provide a sensor 25 that detects pressure.

In contrast, if the sensor 25 detects temperature, even if the sensor 25 is not exposed in the cavity C1 (i.e., even if it is buried in the mold 24 like the temperature sensor 26), it can detect the temperature of the molding material L1 by thermal conduction. And even if the sensor 25 detects temperature, it can detect clogging of some of the gates 245 based on differences in the time series data.

Graph line F21 in FIG. 8 shows time series data of temperature detected by sensor SN1 (corresponding to the "first time series data" in this disclosure). Similarly, graph line F22 shows time series data of temperature detected by sensor SN2, graph line F23 shows time series data of temperature detected by sensor SN3, and graph line F24 shows time series data of temperature detected by sensor SN4 (all of which correspond to the "second time series data" in this disclosure). Graph line AvF2 shows time series data of the average temperature of the four graph lines F21 to F24. Graph lines F21 to F24 may be plots of the temperatures detected by sensors SN1 to SN4 at each predetermined time, or may be plots of moving averages (e.g., five-point moving averages) of the temperatures detected at each predetermined time.

Although the positions at which the graph lines rise and the peak values are different, the overall trends of graph lines F21 to F24 are consistent. First, graph lines F21 to F24 show a rise in temperature, which then increases monotonically. This is because the molding material L1 supplied to cavity C1 heats sensors SN1 to SN4. After reaching their respective maximum values, graph lines F21 to F24 decrease monotonically. This is because the supply of molding material L1 from injection section 22 to cavity C1 stops and the molding material L1 in cavity C1 cools down.

Next, we will look at the individual trends of graph lines F21 to F24. First, the rising point Xs1 (the point at which the specified temperature Ts1 is reached) and peak point Xt1 of graph line F21 are later than the rising points Xs2 to Xs4 and peak points Xt2 to Xt4 of the other graph lines F22 to F24. In addition, the rising point Xs1 is later than the rising point AXs of graph line AvF2. This is because, similar to the reason why the rising of graph line F11 in Figure 7 is delayed, gate GT1 is clogged, so the point at which molding material L1 reaches sensor SN1 is later than the points at which it reaches the other sensors SN2 to SN4.

From the above, when the difference in the time series data of temperature is detected, there is a high possibility that the gate GT1 is clogged when the following tendency of (1) or (2) is observed.
(1) The rising point Xs1 of the graph line F21 is later than the rising points Xs2 to Xs4 of the other graph lines F22 to F24. (2) The peak point Xt1 of the graph line F21 is later than the peak points Xt2 to Xt4 of the other graph lines F22 to F24.

In this way, even when the sensor 25 detects temperature, by focusing on the differences in the multiple time series data detected by the multiple sensors 25 corresponding to the multiple gates 245, it is possible to detect that some of the gates 245 are clogged.

As explained above, the inventors have discovered through extensive research that when pressure or temperature is detected using multiple sensors 25 corresponding to multiple gates 245, if one of the gates 245 becomes clogged, the time series data detected by the sensor 25 (e.g., sensor SN1) corresponding to the clogged gate 245 (e.g., gate GT1) differs from the time series data detected by the other sensors 25 (e.g., sensors SN2 to SN4).

The inventors also discovered that the differences in the time series data are particularly evident in the pressure or temperature rise times Xs1 to Xs4, the peak times Xt1 to Xt4 when the pressure or temperature reaches its maximum value, the maximum pressure values Pt1 to Pt4, and the integral values IV1 to IV4 obtained by integrating the time from the pressure rise time to the peak time. For this reason, the inventors came up with an invention that quantifies these values as evaluation values and detects blockages in some of the gates 245 based on the evaluation values.

The evaluation value is obtained, for example, for each of the multiple sensors 25. In this embodiment, four evaluation values R1 to R4 are calculated based on the time series data detected by each of the four sensors 25. Evaluation value R1 (corresponding to the "first evaluation value" in this disclosure) is a value calculated based on the time series data (graph line F11 or F21) of pressure or temperature detected by sensor SN1, and is, for example, at least one value of the rising time point Xs1, the peak time point Xt1, the maximum value Pt1, and the integral value IV1.

Evaluation values R2 to R4 (each of which corresponds to the "second evaluation value" in this disclosure) are values calculated based on time series data of pressure or temperature detected by sensors SN2 to SN4. Evaluation values R1 to R4 are corresponding values. For example, if evaluation value R1 includes the rising point, evaluation values R2 to R4 also include the rising point. If evaluation value R1 includes the peak point, evaluation values R2 to R4 also include the peak point.

In the anomaly detection system 10 according to this embodiment, the learning device 30 generates a trained model Tm1 that learns the correlation between multiple evaluation values R1 to R4 and blockage of the gate 245, and the anomaly detection device 40 acquires detection information for detecting blockage of the gate 245 based on the trained model Tm1. The learning device 30 and the anomaly detection device 40 will be described below.

<Explanation of the learning device>
9 is a block diagram showing the functional configuration of a learning device 30 according to this embodiment. The learning device 30 has a training data acquisition unit 31, a learning calculation unit 32, a molding information storage unit 33, and a learned model storage unit 34. Each of these units is realized by a computer device having a calculation unit such as a CPU and a storage unit such as a HDD.

The molding information storage unit 33 stores various molding information. The molding information is, for example, information in a table format that associates various first information with second information. For example, if the first information is the type of mold, the second information includes various dimensions of the mold and the volume of the cavity C1. If the first information is the type of molding material or the lot number, the second information includes the physical properties of the molding material (viscosity, moisture content, etc.).

The training data acquisition unit 31 acquires information about training data from each part of the anomaly detection system 10. The training data includes gate information, molding information corresponding to the gate information, and multiple evaluation values R1 to R4. The training data is acquired based on data detected by each part of the anomaly detection system 10 (e.g., sensor 25) when, for example, a molded product to be learned is molded.

The gate information is a value related to the blockage of the gate 245, and more specifically, is a value that quantifies the degree of blockage of the gate 245. For example, a state in which all gates 245 are not clogged is quantified as "0", a state in which only gate GT1 is clogged is quantified as "1", a state in which only gate GT2 is clogged is quantified as "2", a state in which only gate GT3 is clogged is quantified as "3", and a state in which only gate GT4 is clogged is quantified as "4".

The training data acquisition unit 31 acquires gate information through input by an operator. For example, in order to train the learning device 30, the operator artificially clogs a specific gate 245 and inputs to the input unit 50 how the gate 245 is clogged. In this way, the training data acquisition unit 31 acquires the gate information.

The training data acquisition unit 31 also acquires molding information based on the information input by the worker to the input unit 50 and the molding information storage unit 33. For example, when molding a molded product to be learned, the worker inputs the lot number of the molding material related to the molded product. The training data acquisition unit 31 acquires molding information related to the physical properties (e.g., viscosity) of the molding material corresponding to the lot number from the molding information storage unit 33.

The molding device 20 molds molded products multiple times using both an unclogged gate 245 and a gate 245 that has been artificially clogged to a specific extent. The training data acquisition unit 31 then acquires multiple evaluation values R1 to R4 to which gate information has been assigned as teaching data, based on the time series data of the pressure or temperature detected by the sensor 25. For example, the multiple evaluation values R1 to R4 acquired when a molded product is molded using an unclogged gate 245 are assigned "0" as gate information.

The training data acquisition unit 31 further acquires multiple environmental values as training data. The environmental values are, for example, values related to the temperature detected by the temperature sensor 26 when the molded product to be learned is molded. The environmental values may further include values related to the environment around and inside the molding device 20 detected by the pressure sensor 227, the mold clamping force sensor 236, and other sensors (e.g., humidity sensors) not shown when the molded product to be learned is molded.

The training data acquisition unit 31 acquires one set of training data (a set of gate information, molding information, multiple evaluation values R1 to R4, and environmental values acquired when molding the molded product) for each molding of a molded product to be learned, for example. The training data acquisition unit 31 acquires multiple sets of training data for a predetermined number of moldings of the molded product to be learned.

The learning calculation unit 32 performs supervised machine learning calculations based on multiple sets of training data to generate a learned model Tm1 that models the correlation between the shaping information, multiple evaluation values R1 to R4, and environmental values, and the gate information. In this embodiment, a support vector machine (SVM) is used as the machine learning model, but other models may also be used. For example, a convolutional neural network (CCN) may be used, or a regression tree model, which is a model related to grouping of data, may be used.

Specifically, when generating the trained model Tm1, the forming information, the multiple evaluation values R1 to R4, and the environmental value (collectively referred to as "input information") are used as explanatory variables, and the gate information is used as the objective variable, thereby modeling the correlation between the input information and the gate information. In other words, the learning calculation unit 32 generates the trained model Tm1 that constitutes an intermediate layer for outputting gate information corresponding to the input information from the output layer when the input information is input.

Here, we will explain the tendency of the trained model Tm1. Consider a case where multiple evaluation values R1 to R4 are the pressure or temperature rising time points Xs1 to Xs4, respectively. In this case, if the evaluation value R1 input to the input layer is a time point later than the other evaluation values R2 to R4 (evaluation value R1 is a value greater than the other evaluation values R2 to R4), the output layer will have a high probability of gate information indicating that gate GT1 is clogged (in the above example, "1"). Also, for example, if the evaluation value R4 input to the input layer is a time point later than the other evaluation values R1 to R3, the output layer will have a high probability of gate information indicating that gate GT4 is clogged (in the above example, "4").

In addition, when multiple evaluation values R1 to R4 are each the maximum pressure, if the evaluation value R1 input to the input layer is smaller than the other evaluation values R2 to R4, the output layer will have a higher probability of the gate information indicating that gate GT1 is clogged (in the above example, "1").

When generating the trained model Tm1, the input information only needs to include multiple evaluation values R1 to R4, and does not need to include molding information and environmental values. Here, the more moisture content (molding information) in the molding material L1, or the higher the temperature (environmental value) detected by the temperature sensor 26, the lower the viscosity of the molding material L1 and the easier it is to flow. When the molding material L1 becomes easier to flow, the molding material L1 fills the entire cavity C1 more quickly, and the difference between the rising time Xs1 (evaluation value R1) and the rising time Xs3 (evaluation value R3) shown in FIG. 7 becomes smaller.

In this way, the multiple evaluation values R1 to R4 also change depending on the molding information and environmental values. For this reason, when there is a large variation in the environmental values and molding information, or when the variation has a large effect on the multiple evaluation values R1 to R4, it is preferable to include the environmental values and molding information in the input information when generating the trained model Tm1 in order to more accurately predict the gate information corresponding to the input information.

The learned model Tm1 generated by the learning calculation unit 32 is stored in the learned model storage unit 34. When new information is input to the learning device 30 and new training data is acquired by the training data acquisition unit 31, the learned model Tm1 stored in the learned model storage unit 34 is updated appropriately according to the contents of the training data. In addition, the learned model Tm1 is transmitted from the learning device 30 to the anomaly detection device 40 described below, and is also stored in the learned model storage unit 45 of the anomaly detection device 40.

<How to generate a trained model>
Next, a method for generating the trained model Tm1 by the learning device 30 will be described. The method for generating the trained model Tm1 includes a training data acquisition step and a learning calculation step. These steps are realized by a computer device constituting the learning device 30 executing a predetermined program.

First, when the training data acquisition process is started, the training data acquisition unit 31 acquires multiple sets of training data. For example, an operator artificially clogs a specific gate 245 and inputs into the input unit 50 how the gate 245 is clogged, so that the training data acquisition unit 31 acquires gate information. Then, a molded product is molded using the gate 245, and multiple evaluation values R1 to R4 (for example, rise times Xs1 to Xs4) corresponding to the gate information are acquired based on the time series data of the pressure or temperature detected by the sensor 25 during molding. This completes the training data acquisition process.

Next, when the learning calculation process is started, the learning calculation unit 32 learns the correspondence between the input information and the gate information based on multiple sets of training data, and generates a learned model Tm1. The learned model Tm1 is stored in the learned model storage units 34 and 45. This completes the learning calculation process.

<Explanation of the abnormality detection device>
10 is a block diagram showing the functional configuration of an abnormality detection device 40 according to this embodiment. The abnormality detection device 40 has a data acquisition unit 41, an abnormality detection unit 42, an output unit 43, a molding information storage unit 44, and a trained model storage unit 45. Each of these units is realized by a computer device having a calculation unit such as a CPU and a storage unit such as a HDD. The calculation unit executes data acquisition processing and abnormality detection processing, which will be described later, based on a program stored in the storage unit.

The molding information storage unit 44 stores molding information in a table format that associates various types of first information with second information, similar to the molding information storage unit 33. The trained model storage unit 45 stores the trained model Tm1 generated by the learning device 30.

The molding information storage unit 44 and the trained model storage unit 45 may be realized in the same storage area as the molding information storage unit 33 and the trained model storage unit 34 of the learning device 30 in the computer device, or may be realized in a different storage area. That is, the learning device 30 and the anomaly detection device 40 may be configured to share the same molding information storage unit 33 and trained model storage unit 34, or the learning device 30 and the anomaly detection device 40 may be configured to have independent molding information storage units 33, 44 and trained model storage units 34, 45.

The data acquisition unit 41 executes a data acquisition process to acquire information for performing anomaly detection from each part of the anomaly detection system 10. The information for performing anomaly detection is, for example, a set of environmental values, molding information, and multiple evaluation values R1 to R4 (hereinafter, this information is collectively referred to as the "detection data set") acquired by the molding device 20 when molding the molded product to be detected.

The anomaly detection unit 42 inputs the detection data set acquired by the data acquisition unit 41 to the trained model Tm1. Based on these inputs, the trained model Tm1 outputs detection information D1 for detecting blockage of the gate 245.

The detection information D1 is information that includes the accuracy of each piece of gate information. Specifically, the detection information D1 includes first to fourth accuracy, which are the accuracy that a blockage has occurred in each of the gates GT1 to GT4, and a fifth accuracy, which is the accuracy that no blockage has occurred in any of the gates 245. As an example, when a detection dataset is input to the trained model Tm1, the trained model Tm1 outputs detection information D1 with a first accuracy of 60%, a second accuracy of 10%, a third accuracy of 10%, a fourth accuracy of 10%, and a fifth accuracy of 10%.

The output unit 43 determines the manner in which the gate 245 is clogged (i.e., the gate 245 that is most likely to be clogged among the multiple gates 245) based on the detection information D1 acquired by the abnormality detection unit 42. The output unit 43 acquires, for example, the most probable manner of clogging from the detection information D1 as the determination result of the manner in which the gate 245 is clogged. For example, if the first probability is the highest from the above first to fifth probabilities, the output unit acquires "gate GT1 is clogged" as the determination result.

The output unit 43 outputs the judgment result to the display unit 60 and the control unit 271. The judgment result is displayed on the display unit 60. If it is judged that any of the multiple gates 245 is clogged, the judgment result may be displayed in a highlighted color such as red on the display of the display unit 60, and an alert may be issued via the speaker.

In this case, the molding device 20 including the gate 245 determined to be clogged may be stopped with the mold section 24 open by an operational command from the control section 271. In this case, the worker inspects the mold section 24 based on an alert from the display section 60, and cleans or replaces the mold section 24 as necessary.

In this embodiment, the output unit 43 may not be provided, and the detection information D1 obtained by the abnormality detection unit 42 may be displayed directly on the display unit 60. In this case, the operator may determine whether or not there is a blockage in the gate 245 based on the detection information D1 displayed on the display unit 60.

<Abnormality detection method by abnormality detection device>
Next, a description will be given of an anomaly detection method performed by the anomaly detection device 40. The anomaly detection method includes a data acquisition step and an anomaly detection step. These steps are realized by a computer device constituting the anomaly detection device 40 executing a predetermined program.

When the data acquisition process is started, the data acquisition unit 41 acquires a detection dataset that is acquired when the molded product to be detected is molded. This ends the data acquisition process. Next, when the anomaly detection process is started, the anomaly detection unit 42 acquires detection information D1 by inputting the detection dataset into the trained model Tm1. Next, the output unit 43 acquires a judgment result based on the detection information D1. Finally, the output unit 43 outputs the judgment result to the display unit 60 and the control unit 271. This ends the anomaly detection process.

<Actions and Effects of the Anomaly Detection System>
The abnormality detection system 10 of this embodiment obtains multiple evaluation values R1 to R4 based on time series data of pressure detected by the sensor 25, and inputs the multiple evaluation values R1 to R4 into a learned model Tm1 to obtain detection information D1 for detecting blockage of some of the gates 245.

More specifically, the abnormality detection system 10 includes a molding device 20 and an abnormality detection device 40. The molding device 20 includes a mold section 24 having a plurality of gates 245 and a cavity C1 connected to the plurality of gates 245 formed therein, an injection section 22 that fills the cavity C1 with a molten molding material L1 via the plurality of gates 245, a first sensor (e.g., sensor SN1) that is provided in an area closest to a first gate (e.g., gate GT1) among the plurality of gates 245 and detects the pressure or temperature of the molding material L1 in the cavity C1, and a second sensor (e.g., at least one of sensors SN2 to SN4) that is provided in an area closest to a second gate (e.g., at least one of gates GT2 to GT4) different from the first gate among the plurality of gates 245 and detects the pressure or temperature of the molding material L1 in the cavity C1.

The abnormality detection device 40 includes a data acquisition unit 41 that acquires a first evaluation value (e.g., evaluation value R1) calculated based on first time series data (e.g., graph line F11 or F21) of pressure or temperature detected by a first sensor, and a second evaluation value (e.g., at least one of evaluation values R2 to R4) calculated based on second time series data (e.g., at least one of graph lines F12 to F14 or F22 to F24) of pressure or temperature detected by a second sensor, and an abnormality detection unit 42 that detects blockage of the first gate or the second gate based on the first evaluation value and the second evaluation value.

When the first gate or the second gate is clogged, a difference occurs between the first time series data and the second time series data. To make it easier to compare this difference, the first time series data is quantified as a first evaluation value, and the second time series data is quantified as a second evaluation value, and a clog in the first gate or the second gate is detected based on these first evaluation value and second evaluation value. This makes it possible to detect clogs in some gates 245 that could not be detected in the past.

In this embodiment, the first distance (e.g., first distance d1) that the molding material L1 travels from the first gate to the first sensor is equal to the second distance (e.g., at least one of the second to fourth distances d2 to d4) that the molding material L1 travels from the second gate to the second sensor. By configuring in this manner, when the gate 245 is "not" clogged, the time at which the molding material L1 reaches the first sensor and the second sensor (i.e., the rising time and the peak time) can be aligned. This makes it possible to ensure that a delay in the rising time or the peak time occurs only when some of the gates 245 are clogged. As a result, the accuracy of clog detection can be improved.

In this embodiment, the first evaluation value includes the first rising time point of the first time series data (e.g., time point Xs1) or the first peak time point at which the pressure or temperature of the first time series data is at its maximum value (e.g., time point Xt1), and the second evaluation value includes the second rising time point of the second time series data (e.g., at least one of times Xs2 to Xs4) or the second peak time point at which the pressure or temperature of the second time series data is at its maximum value (e.g., at least one of times Xt2 to Xt4), and the abnormality detection unit 42 detects a blockage in the first gate when the first rising time point is later than the second rising time point or when the first peak time point is later than the second peak time point.

In addition, in this embodiment, the first sensor and the second sensor detect the pressure of the molding material L1, the first evaluation value includes a first peak value (e.g., maximum value Pt1) that is the maximum value of the pressure in the first time series data, and the second evaluation value includes a second peak value (e.g., at least one of maximum values Pt2 to Pt4) that is the maximum value of the pressure in the second time series data, and the abnormality detection unit 42 detects a blockage in the first gate when the first peak value is lower than the second peak value.

In this embodiment, the first sensor and the second sensor detect the pressure of the molding material L1, the first evaluation value includes a first integral value (e.g., integral value IV1) which is a time integral from a first rising point of the pressure in the first time series data to a first peak point at which the pressure in the first time series data is at its maximum value, and the second evaluation value includes a second integral value (e.g., at least one of integral values IV2 to IV4) which is a time integral from a second rising point of the pressure in the second time series data to a second peak point at which the pressure in the second time series data is at its maximum value, and the abnormality detection unit 42 detects a blockage in the first gate when the first integral value is smaller than the second integral value.

As described above, in this embodiment, the first evaluation value includes at least one of the first rising time, the first peak time, the first peak value, and the first integral value, and the second evaluation value includes at least one of the second rising time, the second peak time, the second peak value, and the second integral value corresponding to the first evaluation value. By comparing the first rising time and the second rising time, the first peak time and the second peak time, the first peak value and the second peak value, or the first integral value and the second integral value, it is possible to identify that the first gate is clogged out of the first and second gates. In other words, the abnormality detection system 10 not only detects clogging of some gates 245 that could not be detected in the past, but also identifies which gate 245 is clogged.

The abnormality detection unit 42 also detects blockage of at least one of the first gate and the second gate by inputting the first evaluation value and the second evaluation value to a trained model Tm1 that has been machine-learned to determine the correlation between the first evaluation value and the second evaluation value and the blockage of the multiple gates 245.

With this configuration, once the trained model Tm1 is generated, it becomes possible to detect blockages in some of the gates 245 from the first evaluation value and the second evaluation value. By using the trained model Tm1, it is possible to more accurately detect blockages in some of the gates 245 even when there is variation in the molding conditions.

Second Embodiment
The anomaly detection system 10 according to the first embodiment has been described above. However, the implementation of the present invention is not limited to this, and various modifications can be made. Below, an anomaly detection system 11 according to a second embodiment of the present invention will be described. In the following description, parts that are unchanged from the first embodiment are given the same reference numerals, and descriptions thereof will be omitted.

Figure 11 is a block diagram that shows a schematic diagram of an abnormality detection system 11 according to the second embodiment. The abnormality detection system 11 includes a plurality of molding devices 20, an abnormality detection device 40a, an input unit 50, and a display unit 60.

In this embodiment, the anomaly detection device 40a of the anomaly detection system 11 detects that the gate GT1 is clogged when the absolute value of the difference between the average value AvR of the multiple evaluation values R1 to R4 and the evaluation value R1 (i.e., the absolute value DIV1 of the deviation of the evaluation value R1) exceeds a predetermined reference value REF1. In other words, the anomaly detection system 11 differs from the anomaly detection system 10 according to the first embodiment in that it detects an anomaly by comparing the absolute value DIV1 of the deviation with the reference value REF1 without using the trained model Tm1.

Figure 12 is a block diagram showing the functional configuration of an abnormality detection device 40a according to this embodiment. The abnormality detection device 40a has a data acquisition unit 41, an abnormality detection unit 42a, an output unit 43a, a molding information storage unit 44, and a reference value storage unit 46. Each of these units is realized by a computer device having a calculation unit such as a CPU and a storage unit such as a HDD. The data acquisition unit 41 acquires multiple evaluation values R1 to R4, similar to the data acquisition unit 41 according to the first embodiment.

The reference value memory unit 46 stores a reference value REF1. The reference value REF1 is, for example, a value obtained by adding a predetermined margin to the standard deviation of the multiple evaluation values R1 to R4 obtained when none of the multiple gates 245 are clogged.

The abnormality detection unit 42a calculates the absolute value DIV1 of the deviation of the evaluation value R1 based on the multiple evaluation values R1 to R4 acquired by the data acquisition unit 41. Then, by comparing the absolute value DIV1 of the deviation with a reference value REF1 stored in the reference value storage unit 46, it acquires detection information D2 for detecting blockage of the gate GT1. The detection information D2 is, for example, the difference between the absolute value DIV1 of the deviation and the reference value REF1 (DIV1-REF1).

The output unit 43a judges whether or not the gate GT1 is clogged based on the detection information D2 acquired by the abnormality detection unit 42a. For example, when the detection information D2 is the above-mentioned difference (DIV1-REF1), the output unit 43a judges that the gate GT1 is clogged when the detection information D2 is a positive value (i.e., when the absolute value of the deviation DIV1 exceeds the reference value REF1). The output unit 43a outputs the judgment result to the display unit 60 and the control unit 271.

When determining whether gate GT2 is clogged, the abnormality detection unit 42a calculates the absolute value DIV2 of the deviation of evaluation value R2 based on the multiple evaluation values R1 to R4 acquired by the data acquisition unit 41, and compares the absolute value DIV2 of the deviation with a reference value REF1 to acquire detection information D3 for detecting a clog in gate GT2. In this way, by comparing the absolute value of the deviation of evaluation values R1 to R4 corresponding to gates GT1 to GT4 for which a clog is to be detected with the reference value REF1, detection information for detecting a clog in each of gates GT1 to GT4 can be acquired.

In this embodiment, the output unit 43a may not be provided, and the detection information obtained by the abnormality detection unit 42a may be displayed directly on the display unit 60. In this case, the operator may determine whether the gate 245 is clogged based on the detection information displayed on the display unit 60.

The abnormality detection system 11 according to this embodiment can easily detect blockages in some of the gates 245 by comparing multiple evaluation values R1 to R4 with a reference value REF1.

＜Other＞
The embodiments disclosed above are illustrative in all respects and are not restrictive. In other words, the anomaly detection system of the present invention is not limited to the embodiments shown in the drawings, and may have other embodiments within the scope of the present invention.

10 Abnormality detection system 11 Abnormality detection system 20 Molding device 21 Bed 22 Injection section 221 Hopper 222 Cylinder 223 Screw 224 Nozzle 225 Ball screw 226 Motor 227 Pressurization sensor 228 Movement amount sensor 229 Heater 23 Mold clamping section 231 Fixed platen 231a Through hole 232 Movable platen 232a Through hole 233 Tie bar 234 Ball screw 235 Support platen 236 Mold clamping force sensor 237 Motor 24 Mold section 241 Mold 242 Mold 243 Spool 243a End 243b End 244 Runner 245 Gate 25 Sensor 26 Temperature sensor 27 Control panel 271 Control unit 272 Communication unit 30 Learning device 31 Training data acquisition unit 32 Learning calculation unit 33 Molding information storage unit 34 Learned model storage unit 40 Anomaly detection device 40a Anomaly detection device 41 Data acquisition unit 42 Anomaly detection unit 42a Anomaly detection unit 43 Output unit 43a Output unit 44 Molding information storage unit 45 Learned model storage unit 46 Reference value storage unit 50 Input unit 60 Display unit C1 Cavity L1 Molding material Fm1 Foreign object GT1 to GT4 Gates SN1 to SN4 Sensors d1 First distance d2 Second distance d3 Third distance d4 Fourth distance R1 to R4 Evaluation value Tm1 Learned model D1 Detection information REF1 Reference value DIV1 Absolute value of deviation

Claims (9)

An abnormality detection system including a molding device that molds a molded product and an abnormality detection device that detects an abnormality in the molding device,
The molding device includes:
a mold part having a plurality of gates and a cavity connected to the plurality of gates formed therein;
an injection section that fills the cavity with a molten molding material through the plurality of gates;
a first sensor provided in a region of the plurality of gates closest to a first gate and configured to detect a pressure or a temperature of the molding material in the cavity;
a second sensor provided in a region closest to a second gate different from the first gate among the plurality of gates, the second sensor detecting a pressure or a temperature of the molding material in the cavity;
having
The abnormality detection device includes:
a data acquisition unit that acquires a first evaluation value calculated based on first time-series data of the pressure or the temperature detected by the first sensor, and a second evaluation value calculated based on second time-series data of the pressure or the temperature detected by the second sensor;
an abnormality detection unit that detects a blockage of the first gate or the second gate based on the first evaluation value and the second evaluation value;
having
The anomaly detection system, wherein the anomaly detection unit detects blockage of at least one of the first gate and the second gate by inputting the first evaluation value and the second evaluation value into a trained model that has been machine-learned to determine the correlation between the first evaluation value and the second evaluation value and the blockage of the multiple gates. The anomaly detection system of claim 1, wherein a first distance traveled by the molding material from the first gate to the first sensor is equal to a second distance traveled by the molding material from the second gate to the second sensor. The first evaluation value is
A first rising point of the first time series data, or
A first peak time point at which the pressure or temperature of the first time series data is a maximum value is included,
The second evaluation value is
A second rising point of the second time series data, or
A second peak time point at which the pressure or temperature of the second time series data is a maximum value is included,
3. The abnormality detection system according to claim 2, wherein the abnormality detection unit detects a blockage of the first gate when the first rising time point is later than the second rising time point or when the first peak time point is later than the second peak time point. the first sensor and the second sensor detect a pressure of the molding material;
the first evaluation value includes a first peak value that is a maximum value of pressure in the first time-series data,
the second evaluation value includes a second peak value that is a maximum value of the pressure of the second time-series data,
4. The abnormality detection system according to claim 1, wherein the abnormality detection unit detects a blockage of the first gate when the first peak value is lower than the second peak value. the first sensor and the second sensor detect a pressure of the molding material;
the first evaluation value includes a first integral value which is a time integral from a first rising point in time of the pressure of the first time series data to a first peak point in time at which the pressure of the first time series data becomes a maximum value,
the second evaluation value includes a second integral value which is a time integral from a second rising point in the pressure of the second time series data to a second peak point in the pressure of the second time series data at which the pressure is a maximum value,
5. The abnormality detection system according to claim 1, wherein the abnormality detection unit detects a blockage of the first gate when the first integral value is smaller than the second integral value. An abnormality detection device for detecting an abnormality in a molding device that molds a molded product,
The molding device includes:
a mold part having a plurality of gates and a cavity connected to the plurality of gates formed therein;
an injection section that fills the cavity with a molten molding material through the plurality of gates;
a first sensor provided in a region of the plurality of gates closest to a first gate and configured to detect a pressure or a temperature of the molding material in the cavity;
a second sensor provided in a region closest to a second gate different from the first gate among the plurality of gates, the second sensor detecting a pressure or a temperature of the molding material in the cavity;
having
The abnormality detection device includes:
a data acquisition unit that acquires a first evaluation value calculated based on first time-series data of the pressure or the temperature detected by the first sensor, and a second evaluation value calculated based on second time-series data of the pressure or the temperature detected by the second sensor;
an abnormality detection unit that detects a blockage of at least one of the first gate and the second gate based on the first evaluation value and the second evaluation value;
having
The anomaly detection unit detects blockage of at least one of the first gate and the second gate by inputting the first evaluation value and the second evaluation value into a trained model that has been machine-learned to determine the correlation between the first evaluation value and the second evaluation value and the blockage of the multiple gates. 1. An abnormality detection method for detecting an abnormality in a molding device including a mold section having a plurality of gates and a cavity connected to the plurality of gates formed therein, and an injection section that fills the cavity with a molten molding material via the plurality of gates, comprising:
a data acquisition process for acquiring a first evaluation value calculated based on first time-series data of the pressure or temperature of the molding material in the cavity detected by a first sensor provided in an area closest to a first gate among the plurality of gates, and a second evaluation value calculated based on second time-series data of the pressure or temperature of the molding material in the cavity detected by a second sensor provided in an area closest to a second gate different from the first gate among the plurality of gates;
an abnormality detection process for detecting clogging of at least one of the first gate and the second gate by inputting the first evaluation value and the second evaluation value into a trained model that has been machine-learned to learn correlations between the first evaluation value and the second evaluation value and clogging of the plurality of gates;
The anomaly detection method comprises: A program for detecting an abnormality in a molding device including a mold section having a plurality of gates and a cavity connected to the plurality of gates formed therein, and an injection section that fills the cavity with a molten molding material via the plurality of gates,
a data acquisition process for acquiring a first evaluation value calculated based on first time-series data of the pressure or temperature of the molding material in the cavity detected by a first sensor provided in an area closest to a first gate among the plurality of gates, and a second evaluation value calculated based on second time-series data of the pressure or temperature of the molding material in the cavity detected by a second sensor provided in an area closest to a second gate different from the first gate among the plurality of gates;
an abnormality detection process for detecting clogging of at least one of the first gate and the second gate by inputting the first evaluation value and the second evaluation value into a trained model that has been machine-learned to learn correlations between the first evaluation value and the second evaluation value and clogging of the plurality of gates;
A program that causes a computer device to execute the above. A trained model for detecting an abnormality in a molding device including a mold section having a plurality of gates and a cavity connected to the plurality of gates formed therein, and an injection section that fills the cavity with a molten molding material via the plurality of gates,
a first evaluation value calculated based on first time-series data of a pressure or a temperature of the molding material in the cavity detected by a first sensor provided in an area closest to a first gate among the plurality of gates;
a second evaluation value calculated based on second time-series data of the pressure or temperature of the molding material in the cavity detected by a second sensor provided in an area closest to a second gate different from the first gate among the plurality of gates; and
Explanatory variables, including
A trained model for causing a computer to output a response variable including a value relating to a clogged gate among the plurality of gates.

Images