JP2022076073A - Mold monitoring system - Google Patents

Abstract

To provide a mold monitoring system in which a judgement can be performed with appropriate accuracy when performing determination of a mounting quality of an insert part and determination of a presence or absence of a foreign matter on a processing surface based on a captured image of a processing surface of a mold.SOLUTION: An insert determination unit 32 determines a mounting quality of an insert component in mounting unit area images 71 to 73, and a foreign matter determination unit 33 determines a presence or absence of a foreign matter in a processing surface residual image 75 from which the mounting unit area images 71 to 73 are cut off. The insert determination unit 32 has a neural network for insert determination to which mounting unit area images 71 to 73 are input to an input layer, the mounting quality of the insert component is output from an output layer. The foreign matter determination unit 33 has a neural network for determining a foreign matter to which the processing surface residual image 75 is input to the input layer, and the presence or absence of the foreign matter is output from the output layer, in which in the configuration, a number of hidden layers of the neural network for determining the foreign matter is smaller than that of the neural network for insert determination.SELECTED DRAWING: Figure 2

Description

This specification discloses a mold monitoring system for monitoring a mold for insert molding.

Conventionally, a monitoring system for monitoring a mold used for resin molding has been known. For example, in Patent Document 1, the presence or absence of an abnormality is determined by capturing an image of a mold with a camera. Regarding the determination of the presence or absence of an abnormality, for example, in Patent Document 2, the reference image of the mold imaged in advance and the captured image are compared, and the presence or absence of the abnormality is determined.

Japanese Patent No. 6584472 Japanese Unexamined Patent Publication No. 2012-19354

By the way, in mold monitoring, the machined surface of the mold for insert molding may be monitored. The machined surface refers to the surface of the mold on which at least one of the cavity and the core and the mounting portion for mounting the insert component are formed. When the computer is made to monitor the machined surface, the captured image of the machined surface and the reference image which is a normal image are compared.

For example, as monitoring items of the machined surface, there are a determination of whether or not the insert component is mounted and a determination of the presence or absence of foreign matter. In the former case, for example, a missing item in which the insert component is not mounted on the mounting portion or insufficient insertion (floating) of the insert component into the mounting portion is mentioned as an example of mounting failure. Regarding the floating of the insert component, it is necessary to recognize the deviation of the relative position between the mounting portion and the insert component, and the required determination accuracy is relatively fine.

On the other hand, in the determination of the presence or absence of foreign matter, it is determined whether or not foreign matter such as an insert part has fallen on the machined surface. In this determination, it suffices if a foreign substance can be recognized, and a coarser determination accuracy than the determination of whether or not the insert component is mounted is sufficient.

In addition, so-called gas burning occurs on the machined surface of the mold. In gas burning, the gas inside the mold or the gas generated from the resin material is compressed during mold clamping and burns at a high temperature, and the soot adheres to the machined surface and the machined surface becomes Refers to the phenomenon of blackening. If the effect of gas burning is minor, such as a color change on the machined surface due to blackening or a slight blackening area, the quality of the molded product will not be affected, so the molding process will be stopped immediately. There is no need.

However, if the presence or absence of foreign matter is judged with the judgment accuracy required for the mounting quality of the insert component, a slight gas burn on the machined surface may be detected, and based on this, it may be erroneously judged that there is foreign matter. ..

Therefore, in the present specification, a mold monitoring system capable of making a judgment with appropriate accuracy in determining whether or not the insert component is mounted or not and whether or not there is a foreign substance on the processed surface based on the captured image of the machined surface of the mold. Will be disclosed.

This specification discloses a mold monitoring system for monitoring a mold for insert molding. The system includes an imager, an image extraction unit, an insert determination unit, and a foreign matter determination unit. The imager captures the machined surface of the mold in which at least one of the cavity and the core is formed and the mounting portion on which the insert component is mounted is formed. The image extraction unit cuts out a mounting area area image from the processed surface image captured by the imager. The insert determination unit determines whether or not the insert component is mounted in the mounting portion area image. The foreign matter determination unit determines the presence or absence of foreign matter in the processed surface residual image from which the mounting portion region image is cut off. The insert determination unit includes an insert determination neural network in which an image of the mounting area is input to the input layer and the mounting quality of the insert component is output from the output layer. The foreign matter determination unit includes a foreign matter determination neural network in which the residual image of the machined surface is input to the input layer and the presence or absence of foreign matter is output from the output layer. In the above configuration, the number of hidden layers of the foreign matter determination neural network is smaller than that of the insert determination neural network.

Generally, in a neural network, it is known that the determination accuracy increases as the number of hidden layers (intermediate layers) increases. According to the above configuration, by reducing the number of layers of the hidden layer of the foreign matter determination neural network to be smaller than the number of layers of the hidden layer of the insert determination neural network, a relatively rough determination becomes possible, and a relatively rough determination can be made on the machined surface. It is possible to suppress the detection of a slight gas burn and erroneous determination as a foreign substance.

Further, in the above configuration, the processed surface residual image is divided into a plurality of images according to the illuminance, and a foreign matter determination neural network may be provided for each of the divided images.

In the captured image of the processed surface, the change in the processed surface due to gas burning appears relatively large (clearly) in the image region with relatively high illuminance (bright), and in the image region with relatively low illuminance (dark). , The change in the processed surface due to gas burning appears relatively small. By dividing the processed surface residual image according to the illuminance, the low illuminance region, which is an image region with low illuminance, is separated from the high illuminance region, which is an image region with high illuminance. As a result, the low-light region is not affected by changes in the machined surface due to gas burning in the high-light image region.

Further, in the above configuration, the mold monitoring system may include a light source that illuminates the machined surface. In this case, the light source may be a red light source.

When the machined surface of the mold is irradiated with a light source, the irradiation light is reflected and halation occurs in the machined surface image, which may make it difficult to grasp the machined surface. Here, by utilizing the characteristic that the total reflection of the red light source with respect to the metal mirror surface is lower than that of other color light sources, the occurrence of halation in the processed surface image can be suppressed by using the red light source.

Further, in the above configuration, the learning teacher data may be updated in the foreign body determination neural network. In this case, the machined surface image in the latest predetermined period may be used as the teacher data.

By re-learning the neural network for determining foreign matter using the teacher data using the processed surface image in the latest predetermined period, it is possible to suppress erroneous determination especially in the high illuminance region where the change in gas burning on the processed surface is large. ..

According to the work management system disclosed in the present specification, when the insertion quality of the insert component is determined and the presence or absence of foreign matter on the processed surface is determined based on the captured image of the machined surface of the mold, the determination is made with appropriate accuracy. Is possible.

It is a figure which illustrates the hardware configuration of the mold monitoring system which concerns on this embodiment. It is a figure which illustrates the functional block of the mold monitoring system which concerns on this embodiment. It is a figure which illustrates the machined surface image of a mold. It is a figure which illustrates the machined surface image of a mold when an insert part is correctly mounted. It is a figure which illustrates the machined surface image of a mold when an insert part is erroneously mounted. It is a figure which illustrates the image of the machined surface of a mold when a foreign substance falls on a machined surface. It is a figure explaining the cutout of the mounting part area image. It is a figure explaining the process of dividing a machined surface residual image into a plurality of images. It is a figure which illustrates the determination unit of the insert determination part. It is a figure which illustrates the neural network (neural network for insert determination) in the determination unit of the insert determination unit. It is a figure which illustrates the determination unit of the foreign matter determination part. It is a figure which illustrates the neural network (neural network for foreign matter determination) in the determination unit of the foreign matter determination part. It is a figure which illustrates the mold monitoring flow in the mold monitoring system which concerns on this embodiment. It is a figure which shows another example of the mold monitoring flow in the mold monitoring system which concerns on this embodiment.

<Overall configuration>
FIG. 1 illustrates a mold monitoring system according to this embodiment. As will be described later, the system monitors the mold for insert molding. Then, the system determines whether or not the insert component is correctly mounted on the mold and whether or not foreign matter has fallen on the machined surface of the mold at the time of insert molding. This mold monitoring system includes a mold monitoring device 10, an imager 40, and a light source 42.

The mold monitoring device 10 is composed of, for example, a computer. The mold monitoring device 10 includes a CPU 11 of an arithmetic unit, a system memory 12 as storage means, and a storage 13. The storage 13 may be a non-transient storage device such as a hard disk drive (HDD) or a solid state drive (SSD). Further, the mold monitoring device 10 includes an input / output controller 15 that manages input / output of information between an input unit 14 such as a keyboard and a mouse and an external device such as an imager 40.

Further, the mold monitoring device 10 has a GPU 16 (Graphics Processing Unit), a frame memory 17, a RAMDAC 18 (Random Access Memory Digital-to-Analog Converter), and a display control unit as means for processing the captured image captured by the imager 40. 19 is provided. In addition, the mold monitoring device 10 includes a display unit 20 that displays a processed image. The mold monitoring device 10 may include a touch panel display in which the input unit 14 and the display unit 20 are integrated.

The GPU 16 is an arithmetic unit for image processing, and is mainly operated when performing insert determination and foreign matter determination, which will be described later. The frame memory 17 is a storage device that stores an image captured by the imager 40 and arithmetically processed by the GPU 16. The RAMDAC 18 converts the image data stored in the frame memory 17 into an analog signal for the display unit 20 which is an analog display.

The display control unit 19 causes the display unit 20 to display the image processed through the GPU 16, the frame memory 17, and the RAMDAC 18. For example, the display control unit 19 superimposes and displays the image captured by the imager 40 and the highlighting (described later) that emphasizes the portion that is the basis of the abnormality determination by the insert determination or the foreign matter determination.

Further, as will be described later, when the insert determination unit 32 (see FIG. 2) determines that the mounting is defective, or the foreign matter determination unit 33 determines that there is a foreign matter, the display control unit 19 receives this and displays the display. Display a warning message in unit 20. Further, when the insert determination unit 32 determines that the mounting is normal and the foreign matter determination unit 33 determines that there is no foreign matter, the display control unit 19 causes the display unit 20 to display a pass message.

In FIG. 1, the mold monitoring device 10 includes a display unit 20, but for example, the mold monitoring device 10 (computer) including a configuration other than the input unit 14 and the display unit 20 is separated from the assembly site. It may be installed in the server room. Further, the display unit 20 may be installed at the work site so that the insert molding operator can view it.

Further, a touch panel display in which the display unit 20 and the input unit 14 are integrated may be installed at the work site. As will be described later, when at least one of mounting failure and foreign matter detection occurs, the molding operation is temporarily stopped. A confirmation button is displayed on the touch panel display so that the operator can operate the button so that the molding work can be restarted after the improper mounting of the insert component is corrected and the foreign matter is removed.

FIG. 2 illustrates the functional block of the mold monitoring device 10 in a mixed form with the hardware block shown in FIG. This functional block diagram is configured by the CPU 11 executing a program stored in a storage 13 or stored in a non-transient storage medium readable by a computer such as a DVD.

The mold monitoring device 10 includes an image extraction unit 31, an insert determination unit 32, a foreign matter determination unit 33, and a molding device control unit 34 as functional blocks. As will be described later, the image extraction unit 31 cuts out the mounting unit region images 71, 72, 73 (see FIG. 7) from the processed surface image 50 (see FIG. 3) captured by the imager 40. The insert determination unit 32 determines whether or not the insert parts (first insert screw 61, second insert screw 62, cap 63 in FIG. 4) in the mounting unit area images 71, 72, 73 are mounted. The foreign matter determination unit 33 determines the presence or absence of foreign matter in the machined surface residual image 75 (see FIG. 8) from which the mounting portion region images 71, 72, and 73 are cut off.

The molding device control unit 34 can output control commands (stop command and restart command) to the insert molding device. As will be described later, when the mounting of the insert component on the mold of the insert molding apparatus is completed, the mold clamping button is pressed by the worker in order to align (mold) the fixed mold and the movable mold. At this time, when the mold monitoring device 10 detects a mounting defect of the insert component or a foreign substance on the machined surface, the molding device control unit 34 transmits a stop command to the insert molding device.

Further, as will be described later, when the worker corrects the improper mounting of the insert part and the foreign matter is removed from the machined surface, the worker means that the abnormality is resolved from the input unit 14 (confirmation button) to the molding device control unit 34. A confirmation signal is sent. In response to this, the molding apparatus control unit 34 transmits a restart command for restarting the stopped process to the insert molding apparatus.

The imager 40 is, for example, a camera device installed at a work site of insert molding, and can capture an insert molding work with a still image and a moving image. The imager 40 includes an image pickup device such as a CMOS or a CCD. Further, since the imager 40 performs fixed imaging, imaging conditions such as position / posture, angle of view, and magnification are fixed.

In particular, the imager 40 takes an image of the machined surface of the molding die provided in the insert molding apparatus. The machined surface of the mold indicates a surface on which at least one of a cavity (female mold) and a core (male mold) is formed and a mounting portion on which an insert component is mounted is formed. .. For example, when the molds of the movable plates are closed according to the molds of the fixed plates, the facing surfaces (matching surfaces) of the molds are the machined surfaces.

For example, the imager 40 performs an image pickup by an operator's operation at a predetermined timing. For example, after the worker mounts the insert component on the mounting portion provided on the machined surface of the mold in the open mold state, the mold clamping button is pressed by the worker to proceed to the next process, for example, the mold closing process. Is done. When the mold clamping button is pressed, the imager 40 images the machined surface of the mold.

With reference to FIG. 2, the light source 42 irradiates the machined surface of the mold, which is the image pickup target of the imager 40, with light. The light source 42 may be, for example, a red light source, and the machined surface of the mold is illuminated in red by the light source. It is generally known that the total internal reflection of a red light source with respect to a metal mirror surface is lower than that of other color light sources. In the mold monitoring system according to the present embodiment, by utilizing this characteristic and using a red light source as the light source 42, the occurrence of halation in the machined surface image is suppressed.

<Example of captured image of mold>
FIG. 3 illustrates a machined surface image 50, which is an image of a machined surface captured by the imager 40. For example, this machined surface may be a machined surface of a mold attached to a fixing plate of an insert molding apparatus.

Further, FIG. 4 shows an example when the insert component is correctly mounted on the mounting portion of the machined surface. In the example of FIG. 3, the mounting portion includes a first pin 56, a second pin 57, and a cap receiver 58. Further, in the example of FIG. 4, the insert component includes a first insert screw 61, a second insert screw 62, and a cap 63. However, in the mold monitoring system according to the present embodiment, the mounting portion and the insert component are not limited to the above example.

As illustrated in FIG. 3, a parting surface 55, which is a main surface and indicates a parting line, is formed on the processed surface image. Further, the cavity 51 is formed so as to be depressed from the parting surface 55. The cavity 51 faces the core provided in the mold of the movable plate when the mold is closed, and the gap formed between the two becomes a molding space for the molded product.

A cylindrical protrusion 52 protruding from the bottom surface of the cavity 51 in the height direction is formed in the cavity 51. A cylindrical protrusion 53 is formed in the central opening of the cylindrical protrusion 52. The cylindrical protrusion 53 is concentric with the cylindrical protrusion 52, and its outer peripheral surface is separated from the inner peripheral surface of the cylindrical protrusion 52.

A cap receiver 58 slightly protruding from the upper end surface is formed on the upper end surface of the cylindrical protrusion 53. The cap receiver 58 is a mounting portion on which the cap 63 (see FIG. 4), which is an insert component, is mounted (fitted). A groove 58A for alignment is carved in the central portion of the cap receiver 58, and a key (not shown) provided on the facing surface of the cap 63 with the cap receiver 58 is inserted into the groove 58A. As a result, the cap 63 is normally attached to the cap receiver 58.

A rod-shaped protrusion 54 extends from the cylindrical protrusion 52 along the radial direction thereof. A first pin 56 and a second pin 57, which are mounting portions protruding from the upper end surface, are formed on the upper end surface of the rod-shaped protrusion 54. Both the first pin 56 and the second pin 57 have a cylindrical shape, and as illustrated in FIG. 4, a first insert screw 61 and a second insert screw 62, which are cylindrical insert parts, are inserted.

FIG. 5 shows an example of improper mounting of the insert component. In this example, the lower end of the first insert screw 61 does not reach the lower end of the first pin 56, that is, a so-called “float” occurs. In order to determine such an abnormality on the image, it is necessary to be able to recognize at least that the relative positions of the first insert screw 61 and the first pin 56 are displaced.

FIG. 6 shows an example of a foreign object falling. In the mold monitoring system according to the present embodiment, it is mainly assumed that the insert component falls on the machined surface as a foreign substance. Therefore, in determining the presence or absence of foreign matter, the insert component may be set as the minimum foreign matter to be determined. When the insert component is composed of a plurality of types of members having different sizes, the smallest of them may be regarded as the smallest foreign matter, and the presence or absence of the foreign matter may be determined.

In the example of FIG. 6, the insert screw 65 has fallen into the cavity 51. In order to determine such an abnormality on the image, it is sufficient that the insert screw 65 is visible at least, and the required determination accuracy is low as compared with the improper mounting of FIG.

Further, in insert molding, it is known that the resin as a molding material is burnt on a mold, resulting in "gas burning" in which the processed surface is blackened. If such gas burning occurs excessively on the processed surface, there is a possibility that a defect may occur in the molded product, but a certain degree of gas burning does not affect the quality of the molded product. That is, if it is a slight gas burn, the molding work may be continued. As described above, in the determination of foreign matter on the machined surface, the coarse determination accuracy, which does not detect such a slight gas burn as an abnormality, is the appropriate determination accuracy.

As will be described later, in the mold monitoring system according to the present embodiment, an insert determination neural network for determining whether or not the insert component is mounted and a foreign matter determination neural network for determining the presence or absence of foreign matter are provided. In view of the above difference in determination accuracy, in the mold monitoring system according to the present embodiment, each network is configured so that the number of hidden layers of the foreign matter determination neural network is smaller than that of the insert determination neural network. Will be done. Since the number of hidden layers is relatively small, it is possible to determine the coarseness as described above, and it is possible to suppress false detection of slight gas burning as an abnormality.

<Image extraction unit>
7 and 8 show an example of image extraction by the image extraction unit 31 (see FIG. 2). The image extraction unit 31 cuts out a mounting area area image from the processed surface image 50. In FIG. 7, the first pin image 71, the second pin image 72, and the cap receiving image 73 are cut out from the machined surface image 50 as the mounting portion region image.

Since the image of the mounting portion region is used for determining whether or not the insert component is mounted on the mounting portion, the mounting portion region image includes each mounting portion and includes the peripheral region thereof. The area of the mounting portion area image in the machined surface image 50 is set in advance by, for example, the administrator of the mold monitoring device 10.

With reference to FIG. 8, the processed surface residual image 75, which is a residual image from which the mounting portion region image is cut off, is divided into a plurality of images. This division is performed according to the illuminance (brightness) of each pixel in the processed surface residual image 75.

The machined surface of the mold has three-dimensional irregularities, and on the captured image, there are areas that appear dark and areas that appear bright. For example, a portion including a shadow due to the cylindrical protrusion 52, the cylindrical protrusion 53, and the rod-shaped protrusion 54, or a portion away from the light source 42 is a low illuminance region with relatively low illuminance. On the other hand, the upper end surface of the cylindrical protrusion 52, the cylindrical protrusion 53, and the rod-shaped protrusion 54, and the portion of the parting surface 55 near the light source 42 are high illuminance regions with relatively high illuminance.

As described above, in the high illuminance region, the surface change (blackening) of the processed surface due to gas burning clearly appears as compared with the low illuminance region. For example, in the originally dark low-light region, the blackening of the processed surface due to gas burning is limited to a small change in the image, but in the high-light region brightly illuminated by the light source 42, the change (blackening) of the processed surface due to gas burning is clear. appear. The machined surface residual image 75 is divided into a plurality of images based on the difference in the degree of change in gas burning.

For example, as illustrated in FIG. 8, the machined surface residual image 75 is divided into an A region 75A, a B region 75B, a C region 75C, a D region 75D, and an E region 75E with a thick broken line as a boundary line. As will be described later, in the foreign matter determination, a neural network for determining a foreign matter is constructed for each of these divided region images, and the presence or absence of a foreign matter is determined for each region.

By dividing the image according to the illuminance in this way and constructing a neural network for determining foreign matter for each image, for example, in the low illuminance region, it is affected by the change (blackening) of the machined surface in the high illuminance region. Foreign matter can be determined without any need. Further, in the high illuminance region, as will be described later, by frequently re-learning, that is, by setting a short period from the previous learning to the period before the next re-learning, it is possible to determine the foreign matter following the change in the machined surface. ..

<Insert judgment unit>
With reference to FIG. 2, the mounting portion area image (first pin image 71, second pin image 72, cap receiving image 73 in the example of FIG. 7) extracted by the image extraction unit 31 is sent to the insert determination unit 32. Be done. FIG. 9 illustrates the functional block of the insert determination unit 32. The insert determination unit 32 includes a determination unit for each of a plurality of types of mounting unit area images transmitted from the image extraction unit 31. For example, in the example of FIG. 9, the first pin determination unit 32A, the second pin determination unit 32B, and the cap reception determination unit 32C are provided in the insert determination unit 32 according to the mounting unit area image extracted in FIG. 7.

Each of these determination units includes an insert determination neural network, which is an independent neural network. For example, these neural networks are composed of convolutional neural networks (CNN) from the viewpoint of performing image recognition.

Learning by teacher data is performed in advance in these neural networks. The teacher data includes a mounting area image as input data, and includes an assembly pass / fail class corresponding to the image as output data. The assembly pass / fail class indicates, in short, the result of the mounting necessity determination, and is roughly classified into a normal class and a defective class. Defects are further subdivided into classes such as "floating", "missing parts", and "mistakes in parts". The selection of the individual mounting area image and the assignment of the class corresponding to each image are performed by, for example, the administrator of the mold monitoring system.

For example, FIG. 10 illustrates a neural network of the first pin determination unit 32A. The teacher data of the first pin determination unit 32A includes the first pin image 71 as input data, and includes classes such as "normal", "floating", "missing item", and "part error" as output data corresponding to each image. ..

The number of nodes in the input layer of the neural network of each determination unit provided in the insert determination unit 32, including the first pin determination unit 32A, is the same as the number of pixels of the mounting area image transmitted from the image extraction unit 31. May be. For example, the number of nodes in the input layer of the neural network of the first pin determination unit 32A may be the same as the number of pixels of the first pin image 71. Further, the number of nodes in the output layer of the neural network of each determination unit may be the same as the number of the above-mentioned assembly pass / fail classes.

Further, the number of layers of the hidden layer (intermediate layer) of the neural network of each determination unit of the insert determination unit 32 is configured to be larger than the number of layers of the hidden layer of the neural network in the foreign matter determination unit 33 described later. For example, each of the neural networks of each determination unit in the insert determination unit 32 is set so that the number of layers of the hidden layer is larger than that of the neural network of any of the determination units in the foreign matter determination unit 33.

Generally, in a neural network, it is known that the number of hidden layers increases, the higher the determination accuracy. Each of the neural networks of each determination unit in the insert determination unit 32 is set so that the number of layers of the hidden layer is larger than that of the neural network of any of the determination units in the foreign matter determination unit 33, so that the foreign matter is formed. A higher determination accuracy can be obtained in the insert determination unit 32 as compared with the determination unit 33.

Further, in the image of the mounting portion region, most of the image region is occupied by the insert component. The insert component is attached to the mold when the mold is open, and is imaged by the imager 40 before the mold is closed. Since gas burning occurs on the machined surface of the mold after the mold is closed, the surface of the insert component when the image is taken by the imager 40 before the mold is closed is not affected by the gas burning. That is, since the mounting area image is basically narrowed down to the area that is not easily affected by gas burning, the possibility of erroneous judgment due to gas burning is low as in the case of foreign matter judgment. As described above, in the mounting quality determination, even if the determination accuracy is higher than that of the foreign matter determination unit 33, erroneous determination due to gas burning is unlikely to occur.

For example, as illustrated in FIG. 10, the first pin image is input to the input layer of the first pin determination unit 32A. In the example of this figure, an image in which the first insert screw 61 is floating with respect to the first pin 56 is input. In response to this input, the output layer of the first pin determination unit 32A outputs whether or not the insert component needs to be mounted. Specifically, the certainty of each class is obtained by the neural network for insert determination, and the output is obtained so that the certainty of floating, which is one aspect of improper mounting, is the highest. The first pin determination unit 32A outputs "float" having the highest certainty as the final determination result.

<Foreign matter judgment unit>
With reference to FIG. 2, the processed surface residual image 75 extracted by the image extraction unit 31 is divided according to the illuminance (5 images of regions 75A to 75E in the example of FIG. 8) and is subjected to the insert determination unit 32. Will be sent.

FIG. 11 illustrates a functional block of the foreign matter determination unit 33. The foreign matter determination unit 33 includes a determination unit for each of the divided regions 75A to 75E of the processed surface residual image 75 transmitted from the image extraction unit 31. For example, in the example of FIG. 11, the A area determination unit 33A, the B area determination unit 33B, the C area determination unit 33C, the D area determination unit 33D, and the E area determination unit 33E are provided in the foreign matter determination unit 33.

Each of these determination units includes a foreign matter determination neural network, which is an independent neural network. For example, these neural networks are composed of a convolutional neural network (CNN), similarly to the determination unit of the insert determination unit 32. A machined surface residual image is input to the input layer of these foreign matter determination neural networks, and the presence or absence of foreign matter is output from the output layer.

The neural network of each determination unit of the foreign body determination unit 33 is previously learned by the teacher data. The teacher data includes each divided region image as input data, and includes a class of foreign matter presence / absence corresponding to the image as output data. For example, for any of the determination units 33A to 33E, two types of classes, "with foreign matter" and "without foreign matter", are provided as output layers. The division area image in the teacher data and the corresponding class are assigned by, for example, the administrator of the mold monitoring system.

The number of nodes in the input layer of the neural network of each of the determination units 33A to 33E provided in the foreign matter determination unit 33 may be the same as the number of pixels of the divided region image transmitted from the image extraction unit 31. For example, the number of nodes in the input layer of the neural network of the C region determination unit 33C exemplified in FIG. 12 may be the same as the number of pixels in the C region 75C of the machined surface residual image 75. Further, the number of nodes in the output layer of the neural network of each determination unit may be two, "with foreign matter" and "without foreign matter".

Further, the number of layers of the hidden layer (intermediate layer) of the neural network of each of the determination units 33A to 33E of the foreign matter determination unit 33 is configured to be smaller than the number of layers of the hidden layer of the neural network in the insert determination unit 32. For example, each of the neural networks of the determination units 33A to 33E in the foreign matter determination unit 33 is set so that the number of hidden layers is smaller than that of the neural networks of any of the determination units 32A to 32C in the insert determination unit 32. Will be done.

For example, FIG. 12 illustrates a neural network of the C region determination unit 33C. The number of layers n_SFC_C of the hidden layer of the C region determination unit 33C includes, for example, the number of layers n_IST_PIN1 (see FIG. 10) of the hidden layer of the first pin determination unit 32A of the insert determination unit 32, the second pin determination unit 32B, and the cap receiver. The neural network is constructed so as to be smaller than the number of layers of any of the hidden layers of the determination unit 32C.

Each of the neural networks of the determination units 33A to 33E in the foreign matter determination unit 33 is set to have a smaller number of hidden layers than the neural network of any of the determination units 32A to 32C in the insert determination unit 32. As a result, the determination accuracy of the foreign matter determination unit 33 becomes coarser as compared with the insert determination unit 32. By coarsening the determination accuracy of the foreign matter determination unit 33, in other words, by making it blunted against slight changes on the machined surface, slight gas burning generated on the machined surface is ignored, and this is erroneously detected as a foreign matter. Is suppressed.

Further, each of the neural networks of the determination units 33A to 33E in the foreign matter determination unit 33 has a smaller number of nodes in the hidden layer than the neural networks of any of the determination units 32A to 32C in the insert determination unit 32. May be set to. Even with such a method, it is possible to lower the determination accuracy of any of the neural networks of the determination units 33A to 33E in the foreign matter determination unit 33 as compared with any of the determination units 32A to 32C in the insert determination unit 32. Become.

The determination units 33A to 33E of the foreign body determination unit 33 may have a shorter update period of the teacher data for learning than the determination units 32A to 32C of the insert determination unit 32. Further, in response to this, the determination units 33A to 33E of the foreign matter determination unit 33 have a shorter learning update period as compared with the determination units 32A to 32C of the insert determination unit 32. At this time, as the teacher data to be updated, the machined surface image 50 in the latest predetermined period may be used.

By re-learning the neural network for determining foreign matter by using the teacher data using the machined surface image 50 in the latest predetermined period (for example, the period from the time of re-learning to one week before), gas burning, especially on the machined surface, is performed. It is possible to suppress erroneous determination in a high illuminance region where a large change in the above appears.

<Mold monitoring flow>
FIG. 13 illustrates a mold monitoring flow in the mold monitoring system according to the present embodiment. This flow is started, for example, when the operator completes the mounting of the insert component on the mold and presses the mold clamping button to proceed to the mold closing process. As the mold clamping button is pressed down, the imager 40 takes an image of the machined surface of the mold (S10). The machined surface image 50 (see FIG. 4) thus obtained is sent to the image extraction unit 31 (see FIG. 2).

The image extraction unit 31 divides the processed surface image 50 into a mounting portion region image and a processed surface residual image 75 (S12). At this time, as for the processed surface image, the mounting portion area image is cut out from the processed surface image 50 for each mounting portion. For example, referring to FIG. 7, the image extraction unit 31 obtains three types of mounting portion region images, a first pin image 71, a second pin image 72, and a cap receiving image 73, from the processed surface image 50 as the mounting portion region image. cut out. The cut-out mounting area area image is sent to the insert determination unit 32.

Further, the image extraction unit 31 divides the processed surface residual image 75 into a plurality of regions according to the illuminance (S14). The divided image is sent to the foreign matter determination unit 33.

The insert determination unit 32 receives the determination results of each determination unit 32A to 32C, and at least one of the determination units outputs the determination result of improper mounting as the final value, that is, as the class with the highest certainty. It is determined whether or not it is present (S16). That is, the insert determination unit 32 confirms whether or not there are determination units 32A to 32C that output results other than "normal" with reference to FIG.

When at least one of the determination units 32A to 32C of the insert determination unit 32 outputs the determination result of improper mounting as the final value, the insert determination unit 32 outputs the determination result of improper installation. A signal to that effect (hereinafter, appropriately referred to as a mounting failure signal) is transmitted to the molding apparatus control unit 34. Upon receiving this signal, the molding apparatus control unit 34 stops the operation of the molding apparatus and stops the transition to the mold clamping step (S18).

Further, the mounting failure signal is also transmitted from the insert determination unit 32 to the display control unit 19. The display control unit 19 causes the display unit 20 (for example, a touch panel display) to display a warning message (S24). For example, the display control unit 19 causes the display unit 20 to display the characters "NG" beside the captured photograph of the mold. In addition, the display control unit 19 may perform image processing on the captured image of the mold to highlight the image area (for example, the first pin image 71) in which the determination result of improper mounting is output. Further, the display control unit 19 causes the display unit 20 to display a confirmation button in order to restart the mold clamping process after eliminating the mounting defect.

On the other hand, if it is determined in step S16 that no mounting abnormality is output from any of the determination units 32A to 32C, that is, it is determined that all the mounting conditions are normal, the determination result to that effect is displayed in the display control unit 19. Send to. At the same time, the presence or absence of foreign matter is determined by the foreign matter determination unit 33 (S20).

The foreign matter determination unit 33 confirms the determination result of the A region determination unit 33A to the E region determination unit 33E, that is, the class having the highest certainty. When the determination results of the A area determination unit 33A to the E area determination unit 33E are all free of foreign matter, the foreign matter determination unit 33 sends the determination result to that effect to the display control unit 19.

The insert determination unit 32 receives the determination result that all the determination units 32A to 32C are normally mounted, and the foreign matter determination unit 33 receives the determination result that all the determination units 33A to 33E have no foreign matter, and the display control is performed. The unit 19 causes the display unit 20 to display a pass message (S22). For example, the display control unit 19 causes the display unit 20 to display the characters "OK" beside the captured photograph of the mold.

On the other hand, in step S20, when at least one of the A area determination unit 33A to the E area determination unit 33E outputs the determination result that there is a foreign substance as the final value, the mounting failure signal is output to the molding apparatus control unit. It is transmitted to 34, and the flow proceeds to step S18. In the subsequent step S24, the display control unit 19 may perform image processing on the captured photograph of the mold to highlight the image area in which the determination result of the presence of foreign matter is output.

When the warning message is displayed on the display unit 20, the insert molding operator reattaches the insert component determined to be improperly attached, or removes the foreign matter that has fallen on the machined surface. The molding apparatus control unit 34 determines whether or not the confirmation button for restarting the molding process has been pressed by the operator (S26).

If the confirmation button is not pressed by the operator, the process returns to step S26 and the flow is in the standby state. On the other hand, when the confirmation button is pressed by the operator, the molding apparatus control unit 34 restarts the molding apparatus that has been temporarily stopped, and shifts the molding process to the molding process (S28).

<Another example of mold monitoring flow>
FIG. 14 shows another example of the mold monitoring flow by the mold monitoring system according to the present embodiment. In this example, steps S20 and S22 in FIG. 13 are deleted, and step S30 is provided instead of step S16.

In this step, when an abnormality is detected in at least one of the mounting portion region image and the machined surface residual image (S30), the molding apparatus control unit 34 transmits a stop command to the molding apparatus. Such a flow may be used, for example, at the time of inspection for confirming whether or not the abnormality detection by the mold monitoring system according to the present embodiment is operating normally. For example, in this case, as the captured image, an image in which a foreign substance is intentionally dropped on the machined surface or an image in which the insert component is intentionally forgotten to be attached to the mounting portion is used. Even in such a flow, it is possible to determine whether the mounting is good or not and whether or not there is a foreign substance.

10 Mold monitoring device, 20 Display unit, 31 Image extraction unit, 32 Insert determination unit, 32A 1st pin determination unit, 32B 2nd pin determination unit, 32C cap reception determination unit, 33 Foreign matter determination unit, 33A A area determination unit , 33BB area judgment unit, 33CC area judgment unit, 33DD area judgment unit, 33EE area judgment unit, 34 molding device control unit, 40 imager, 42 light source, 50 machined surface image, 51 cavity, 52 cylindrical protrusion , 53 Cylindrical protrusion, 54 Rod-shaped protrusion, 55 Parting surface, 56 First pin, 57 Second pin, 58 Cap holder, 61 First insert screw, 62 Second insert screw, 63 cap, 71 First pin image , 72 Second pin image, 73 Cap receiving image, 75 Machined surface residual image, 75A to 75E Divided area of machined surface residual image.

Claims (4)

It is a mold monitoring system that monitors the mold for insert molding.
An imager that captures the machined surface of the mold, in which at least one of the cavity and the core is formed and the mounting portion on which the insert component is mounted is formed.
An image extraction unit that cuts out a mounting area image from the processed surface image captured by the imager, and an image extraction unit.
An insert determination unit that determines whether or not the insert component is installed in the image of the mounting unit area,
A foreign matter determination unit that determines the presence or absence of foreign matter in the machined surface residual image from which the mounting portion region image is cut out,
Equipped with
The insert determination unit includes an insert determination neural network in which the mounting unit area image is input to the input layer and the mounting quality of the insert component is output from the output layer.
The foreign matter determination unit includes a neural network for determining foreign matter, in which the residual image of the machined surface is input to the input layer and the presence or absence of foreign matter is output from the output layer.
The number of hidden layers of the foreign matter determination neural network is smaller than that of the insert determination neural network.
Mold monitoring system. The mold monitoring system according to claim 1.
The processed surface residual image is divided into a plurality of images according to the illuminance, and the foreign matter determination neural network is provided for each of the divided images.
Mold monitoring system. The mold monitoring system according to claim 1 or 2.
A light source for irradiating the machined surface is provided.
The light source is a red light source,
Mold monitoring system. The mold monitoring system according to any one of claims 1 to 3.
In the foreign body determination neural network, the teacher data for learning is updated.
As the teacher data, the machined surface image in the latest predetermined period is used.
Mold monitoring system.

Images