Computer program, abnormality detection method, abnormality detection device, molding machine system, and learning model generation method

Abstract

To provide an abnormality detection device for providing a computer program which is capable of converting time series data of a physical quantity related to a movable part of an industrial machine into a time-series data image expressed by an image, and determining an abnormality of the industrial machine based on the time-series data image.SOLUTION: Provided is a computer program for causing a computer to execute processing of detecting an abnormality of an industrial machine having a movable section, the processing executed by the computer including: acquiring physical amount data, of a time series, outputted from a sensor that detects a physical amount relating to movement of the movable section; converting the acquired physical amount data of the time series to a time series data image expressed by an image; inputting the converted time series data image into a learning model that has been trained by using features of the time series data image relating to a normal movable section, and thereby calculating a feature amount of the time series data image; and determining presence or absence of an abnormality in the industrial machine on the basis of the calculated feature amount.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a computer program for detecting an abnormality in an industrial machine having a moving part, an abnormality detection method, an abnormality detection device, a molding machine system, and a learning model generation method. [Background technology]

[0002] Patent Document 1 discloses a monitoring method for monitoring the vibration of each moving part of an injection molding machine by an acceleration sensor and detecting the state of the molding process and the occurrence of an abnormality in the moving parts. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 5-50480 Summary of the Invention [Problem to be solved by the invention]

[0004] However, Patent Document 1 does not disclose a specific method for analyzing time-series data such as acceleration data or a method for detecting an abnormality. The vibration waveform of the moving part is complex, and it is not necessarily easy to determine an abnormality in the injection molding machine.

[0005] An object of the present disclosure is to provide a computer program, an anomaly detection method, an anomaly detection device, a molding machine system, and a learning model generation method that can convert time series data of physical quantities related to moving parts of industrial machinery into a time series data image represented by an image, and determine abnormalities in the industrial machinery based on the time series data image. [Means for solving the problem]

[0006] A computer program according to the present disclosure is a computer program for causing a computer to execute a process for detecting an abnormality in industrial machinery having a moving part, which acquires time-series physical quantity data output from a sensor that detects physical quantities related to the movement of the moving part, converts the acquired time-series physical quantity data into a time-series data image represented by an image, and inputs the converted time-series data image into a learning model that has learned the characteristics of the time-series data image related to a normal moving part, thereby calculating feature quantities of the time-series data image, and causes the computer to execute a process for determining whether or not there is an abnormality in the industrial machinery based on the calculated feature quantities.

[0007] The anomaly detection method disclosed herein is an anomaly detection method for detecting an abnormality in industrial machinery having a moving part, which involves acquiring time-series physical quantity data output from a sensor that detects physical quantities related to the movement of the moving part, converting the acquired time-series physical quantity data into a time-series data image represented by an image, inputting the converted time-series data image into a learning model that has learned the features of the time-series data image related to normal moving parts, calculating feature quantities of the time-series data image, and determining the presence or absence of an abnormality in the industrial machinery based on the calculated feature quantities.

[0008] The abnormality detection device according to the present disclosure includes the abnormality detection device and a molding machine, and the abnormality detection device is configured to detect an abnormality in the molding machine.

[0009] The anomaly detection device according to the present disclosure is an anomaly detection device that detects anomalies in industrial machinery having moving parts, and includes a sensor that detects physical quantities related to the movement of the moving part, an acquisition unit that acquires time-series physical quantity data output from the sensor, a conversion unit that converts the time-series physical quantity data acquired by the acquisition unit into a time-series data image represented by an image, a calculation unit that calculates feature quantities of the time-series data image by inputting the converted time-series data image into a learning model that has learned the features of the time-series data image related to the normal moving part, and a judgment unit that judges the presence or absence of an abnormality in the industrial machinery based on the feature quantities calculated by the calculation unit.

[0010] The learning model generation method according to the present disclosure is a method for generating a learning model for detecting an abnormality in a twin-screw kneading extruder having a first screw and a second screw, and includes generating a first image depicting displacements in the first axial direction and the second axial direction intersecting the rotation central axis of the first screw based on time series data showing displacements in the first axial direction and the second axial direction intersecting the rotation central axis of the first screw of the twin-screw kneading extruder during normal operation and time series data showing displacements in the third axial direction and the fourth axial direction intersecting the rotation central axis of the second screw of the twin-screw kneading extruder during normal operation; A computer executes a process of generating a plurality of time series data images including a first image and a second image depicting displacements in a third axial direction and a fourth axial direction intersecting the central axis of rotation of the second screw in a single image, and generating a learning model that outputs features corresponding to normal operation and abnormal operation of the twin-screw kneading extruder when a time series data image including a first image and a second image depicting displacements in a third axial direction and a fourth axial direction intersecting the central axis of rotation of the second screw is input based on a learning dataset including the generated plurality of time series data images and a plurality of reference images having any feature. Effect of the Invention

[0011] According to the present disclosure, time-series data of physical quantities related to moving parts of industrial machinery can be converted into a time-series data image represented by an image, and an abnormality in the industrial machinery can be determined based on the time-series data image. [Brief description of the drawings]

[0012] [Figure 1] 1 is a block diagram showing an example of the configuration of a molding machine system according to an embodiment of the present invention; [Diagram 2] FIG. 1 is a schematic diagram showing a configuration example of a twin-screw kneading extruder according to an embodiment of the present invention. [Diagram 3] FIG. 4 is a schematic diagram showing the mounting position of a displacement sensor. [Figure 4] 1 is a block diagram showing an example of the configuration of an anomaly detection device according to an embodiment of the present invention; [Diagram 5] 5 is a flowchart showing an abnormality detection process procedure according to the embodiment. [Figure 6] 11 is a first time-series data image showing the displacement of the first screw. [Figure 7] 4 is a time-series data image showing the displacement of the first and second screws. [Figure 8] 1 is a conceptual diagram showing feature amounts and anomaly detection results of a time-series data image. [Figure 9] FIG. 1 is a conceptual diagram showing an overview of a method for generating a learning model. [Figure 10] FIG. 1 is a conceptual diagram showing a learning model in a learning phase. [Figure 11] FIG. 1 is a conceptual diagram showing a learning model in a test phase. [Figure 12] 11 is a scatter plot showing feature amounts of a time-series data image. FIG. [Figure 13] 13 is a chart showing evaluation results of anomaly detection using a learning model. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] Specific examples of a computer program, an anomaly detection method, an anomaly detection device, a molding machine system, and a learning model generation method according to embodiments of the present invention will be described below with reference to the drawings. Note that the present invention is not limited to these examples, but is intended to include all modifications within the meaning and scope of the claims. In addition, at least some of the embodiments and modifications described below may be combined in any combination.

[0014] 1 is a block diagram showing an example of the configuration of a molding machine system according to this embodiment. The molding machine system includes a twin-screw kneading extruder 1, a displacement sensor 2, and an abnormality detection device 3.

[0015] <Twin-screw kneading extruder 1> 2 is a schematic diagram showing a configuration example of a twin-screw kneading extruder 1 according to this embodiment. The twin-screw kneading extruder 1 includes a cylinder 10 having a hopper 10a into which a resin raw material is introduced, a first screw 11, and a second screw 12. The first and second screws 11 and 12 are disposed substantially parallel to each other while meshing with each other and rotatably inserted into holes in the cylinder 10, and convey the resin raw material introduced into the hopper 10a in the extrusion direction (to the right in FIG. 2), melting and kneading the resin raw material. The first screw 11 is configured as a single screw by combining and integrating multiple types of screw pieces. For example, the first screw 11 is configured by combining forward flight pieces of a flight screw shape that transport the resin raw material in the forward direction, reverse flight pieces that transport the resin raw material in the reverse direction, kneading pieces that knead the resin raw material, etc., in an order and position according to the characteristics of the resin raw material. The second screw 12 has the same configuration as the first screw 11.

[0016] The twin-screw kneading extruder 1 also includes a motor 13 that outputs a driving force for rotating the first and second screws 11 and 12, and a reducer 14 that reduces and transmits the driving force of the motor 13. The drive shafts 11a and 12a of the first and second screws 11 and 12 are connected to an output shaft of the reducer 14. The first and second screws 11 and 12 are rotated by the driving force of the motor 13 that is reduced and transmitted by the reducer 14.

[0017] <Displacement sensor 2> The displacement sensor 2 is a sensor that detects the displacement of the rotation central axis of the rotating first and second screws 11, 12. The displacement sensor 2 is preferably disposed at an appropriate location between the reducer 14 and the cylinder 10, specifically, near the drive shafts 11a, 12a of the first and second screws 11, 12. The first and second screws 11 and 12 are an example of a movable part of an industrial machine. More specifically, the first and second screws 11 and 12 are an example of a rotating shaft of a twin-screw kneading extruder 1, which is an example of an industrial machine.

[0018] 3 is a schematic diagram showing the mounting position of the displacement sensor 2. The displacement sensor 2 includes a first displacement sensor 21 that detects displacement in the X-axis direction (first direction) perpendicular to the central axis of rotation of the first screw 11, and a second displacement sensor 22 that detects displacement in the Y-direction (second axis direction) perpendicular to the central axis of rotation of the first screw 11 and the X-axis. The first displacement sensor 21 and the second displacement sensor 22 are disposed to face the drive shaft 11a of the first screw 11. The X-axis and Y-axis directions are, for example, horizontal and vertical directions perpendicular to the central axis of rotation of the first screw 11. The displacement sensor 2 also includes a third displacement sensor 23 that detects displacement in the X-axis direction (third direction) perpendicular to the rotation axis of the second screw 12, and a fourth displacement sensor 24 that detects displacement in the Y-direction (fourth axis direction) perpendicular to the rotation axis of the second screw 12 and the X-axis. The third displacement sensor 23 and the fourth displacement sensor 24 are disposed to face the drive shaft 12a of the second screw 12. The X-axis and Y-axis directions are, for example, horizontal and vertical directions perpendicular to the rotation axis of the second screw 12. The first to fourth displacement sensors 21, 22, 23, 24 constantly detect the displacement of the rotation central axes of the rotating first and second screws 11, 12, and output time-series displacement data (physical quantity data) indicating the detected displacement to the abnormality detection device 3. Hereinafter, the first to fourth displacement sensors 21, 22, 23, 24 will be collectively referred to as the displacement sensor 2 as appropriate. The displacement data output by the first and second displacement sensors 21 and 22 corresponds to time series data indicating a first physical quantity according to the present disclosure, and the displacement data is a two-dimensional physical quantity consisting of two numerical values. The displacement data output by the third and fourth displacement sensors 23 and 24 corresponds to time series data indicating a second physical quantity according to the present disclosure, and the displacement data is a two-dimensional physical quantity consisting of two numerical values.

[0019] <Anomaly detection device 3> 4 is a block diagram showing an example of the configuration of the anomaly detection device 3 according to this embodiment. The anomaly detection device 3 is a computer, and includes a processing unit 31, a storage unit 32, an input interface 33 (input I / F), and an output interface 34 (output I / F). The storage unit 32, the input interface 33, and the output interface 34 are connected to the processing unit 31.

[0020] The processing unit 31 includes a central processing unit (CPU), a multi-core CPU, a graphics processing unit (GPU), a general-purpose computing on graphics processing units (GPGPU), a tensor processing unit (TPU), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), a neural processing unit (NPU), and other arithmetic processing circuits, a read only memory (ROM), a random access memory (RAM), and other internal storage devices, an I / O terminal, and the like. The processing unit 31 functions as the anomaly detection device 3 according to this embodiment by executing a computer program P stored in a storage unit 32, which will be described later. Each functional unit of the anomaly detection device 3 may be realized by software, or a part or all of the functional units may be realized by hardware.

[0021] The storage unit 32 includes a hard disk, an EEPROM (Electrically Erasable Programmable Read Only Memory), The storage unit 32 is a non-volatile memory such as a ROM or a flash memory. The storage unit 32 stores a computer program P for causing a computer to execute the anomaly detection method according to this embodiment, and a learning model 35. The learning model 35 is a convolutional neural network model that, when a time-series data image, which is described later and represents time-series data obtained by detecting the displacement of the first and second screws 11 and 12, is input, extracts features of the time-series data image and outputs the extracted feature amount. The learning model 35 is a model having a feature extraction layer of a CNN (Convolutional Neural Network), for example, a one-class classification model. The method of generating and the configuration of the learning model 35 will be described in detail later.

[0022] The computer program P and the learning model 35 according to the present embodiment may be recorded in a computer-readable manner on the recording medium 4. The storage unit 32 stores the computer program P and the learning model 35 read from the recording medium 4 by a reading device (not shown). The recording medium 4 is a semiconductor memory such as a flash memory. The recording medium 4 may also be an optical disk such as a CD (Compact Disc)-ROM, a DVD (Digital Versatile Disc)-ROM, or a BD (Blu-ray (registered trademark) Disc). The recording medium 4 may also be a magnetic disk such as a flexible disk or a hard disk, a magnetic optical disk, or the like. Furthermore, the computer program P and the learning model 35 according to the present embodiment may be downloaded from an external server (not shown) connected to a communication network (not shown) and stored in the storage unit 32.

[0023] The input interface 33 is connected to the first to fourth displacement sensors 21, 22, 23, 24, and receives the displacement data, which is time-series data output from the first to fourth displacement sensors 21, 22, 23, 24, as input.

[0024] The display unit 3a is connected to the output interface 34. The display unit 3a displays whether or not there is an abnormality in the twin-screw kneading extruder 1, and the like.

[0025] <Anomaly detection processing> 5 is a flowchart showing the procedure of the abnormality detection process according to this embodiment. The processing unit 31 acquires time-series displacement data output from the displacement sensor 2 (step S11). The processing unit 31 preferably acquires the displacement data over a period of time during which the first and second screws 11, 12 rotate multiple times.

[0026] Next, the processing unit 31 converts the acquired displacement data into a first time series data image (first image) and a second time series data image (second image) that respectively represent the displacements of the rotation central axes of the first and second screws 11, 12 (step S12).

[0027] Fig. 6 is a first time series data image showing the displacement of the first screw 11. The first time series data image is a substantially square image, and the X-axis and Y-axis of the image correspond to the X-axis direction (first axis direction) and the Y-axis direction (second axis direction) shown in Fig. 4, respectively, and the displacement amounts in the X-axis direction and the Y-axis direction are plotted as coordinate values ​​of the X-axis and the Y-axis. The first time series data image is a plot of the displacement for three rotations.

[0028] The second time series data image showing the displacement of the second screw 12 is also a substantially square image, like the first time series data image, and the X-axis and Y-axis of the image correspond to the X-axis direction (third axis direction) and the Y-axis direction (fourth axis direction), respectively, as in FIG. 4, and the displacement amounts in the X-axis and Y-axis directions are plotted as coordinate values ​​on the X-axis and Y-axis.

[0029] Next, the processing unit 31 combines the first time series data image and the second time series data image into one time series data image (step S13).

[0030] FIG. 7 is a time series data image showing the displacement of the first and second screws 11, 12. The synthesized time series data image is a substantially square image having twice the size (four times the area) of the first time series data image. The time series data image includes the information contained in the original image as it is, without changing the aspect ratio of the first and second time series data images. Specifically, the time series data image has the first and second time series data images arranged side by side on the upper left and right in FIG. 7, and a blank image arranged on the lower side to make the time series data image substantially square. The arrangement method of the first and second time series data images and the blank image is not particularly limited, and the arrangement method is not particularly limited as long as the first and second time series data images are included as they are.

[0031] Next, the processing unit 31 calculates the feature amount of the time-series data image by inputting the time-series data image synthesized in step S13 to the learning model 35 (step S14). The feature amount is expressed, for example, as a multi-dimensional vector amount, for example, a numerical value of about several hundred to several thousand.

[0032] Then, the processing unit 31 calculates an outlier score for the feature of the time series data image calculated in step S14 (step S15). The outlier score is a numerical value obtained by evaluating the degree of outlier of the feature of the time series data image calculated in step S14 with respect to the feature of the time series data image obtained during normal operation of the twin-screw kneading extruder 1 (hereinafter referred to as sample feature). The outlier score is, for example, LOF (Local Outlier Factor). LOF is the local density ld(P) of the feature to be detected as an anomaly and the local density average of the neighborhood group of the feature (the k nearest sample features to the feature).<ld(Q)> Compared to<ld(Q)> / ld(P). The greater the LOF is, the greater the likelihood of it being an outlier. Note that P represents the feature that is the target of anomaly detection, and Q represents the sample feature.

[0033] Then, the processing unit 31 judges whether the calculated outlier score is equal to or greater than a predetermined threshold value, thereby judging whether the twin-screw kneading extruder 1 or the first and second screws 11, 12 are normal or not (step S16). When the outlier score is the above-mentioned LOF, the threshold value is a value equal to or greater than 1. When the LOF is equal to or greater than the threshold value, the processing unit 31 judges that there is an abnormality, and when the LOF is less than the threshold value, it judges that there is a normality.

[0034] The abnormality determination using the LOF described above is one example, and it may be determined whether or not the twin-screw kneading extruder 1 is abnormal by the k-nearest neighbor method or SVM. Also, it may be configured to determine whether or not the twin-screw kneading extruder 1 is abnormal by determining whether or not the feature amount of the abnormality detection target is far from the sample feature amount in normal state by the Hotelling method.

[0035] FIG. 8 is a conceptual diagram showing feature quantities and anomaly detection results of a time-series data image. FIG. 8 shows high-dimensional feature quantities of a time-series data image that have been reduced in dimension to two-dimensional feature quantities and plotted on a two-dimensional plane. The horizontal and vertical axes of FIG. 8 indicate the first and second feature quantities of the reduced-dimensional time-series data image. The dimensionality of the feature quantities can be reduced by a dimension reduction algorithm such as t-SNE. As shown in FIG. 8, the learning model 35 is trained so that the local density of the feature quantity in a normal state output from the learning model 35 is high. The dashed circle shows an image of the threshold value. When the statistical distance between the feature amount of the time series data image that is the target of abnormality detection (e.g., a star-shaped hexagonal plot) and the feature amount group of the time series data image in a normal state is short (the local density is relatively high) and the LOF value is small, the twin-screw kneading extruder 1 is estimated to be in a normal state. When the statistical distance between the feature amount of the time series data image that is the target of abnormality detection (e.g., an X-mark plot) and the feature amount group of the time series data image in a normal state is long (the local density is relatively low) and the LOF value is large, the twin-screw kneading extruder 1 is estimated to be in an abnormal state.

[0036] When the twin-screw kneading extruder 1 is determined to be normal (step S16: YES), the processing unit 31 ends the processing. When the twin-screw kneading extruder 1 is determined to be abnormal (step S16: NO), the processing unit 31 outputs a warning that the twin-screw kneading extruder 1 is abnormal (step S17), and ends the processing. For example, the processing unit 31 displays a warning screen for notifying that the twin-screw kneading extruder 1 is abnormal on the display unit 3a. In addition, as shown in FIG. 8, the processing unit 31 may be configured to display an image representing the feature amount of the current time-series data image to be evaluated and the feature amount of the time-series data in normal state on the display unit 3a. Specifically, the processing unit 31 reduces each high-dimensional feature amount to two dimensions by any dimension reduction algorithm such as t-SNE, and plots each feature amount expressed in two dimensions on a two-dimensional graph, thereby generating an image representing the feature amount. Then, the processing unit 31 outputs the image data to the display unit 3a, thereby causing the display unit 3a to display an image showing the features of the current time series data image to be evaluated and the features of the time series data under normal conditions.

[0037] <Learning model generation method> 9 is a conceptual diagram showing an overview of a method for generating the learning model 35. An example in which the processing unit 31 of the anomaly detection device 3 trains the learning model 35 by machine learning will be described below.

[0038] First, as a learning data set required to generate a one-class classification model, a plurality of time-series data images detected and obtained when the twin-screw kneading extruder 1 is operating normally, and a plurality of arbitrary reference images R0, R1, and R2 unrelated to the time-series data images are prepared. The reference images R0, R1, and R2 are image data of a plurality of classes selected from a learning data set such as Image Net. In the example shown in FIG. 9, reference images R0, R1, and R2 with labels 0, 1, and 2 are prepared.

[0039] Then, the learning model 35 having a feature extraction layer consisting of multiple convolution layers and pooling layers that extract image features is machine-learned on the time-series data image during normal operation and the reference images R0, R1, and R2. Specifically, the processing unit 31 trains the learning model 35 so as to output features that can distinguish the features of each image, that is, features that have a high local density of the features of the time-series data image and are highly distinguishable from the features of the reference images R0, R1, and R2, as shown in the right diagram of FIG. The method for generating a one-class classification model will be described in detail below.

[0040] 10 is a conceptual diagram showing a learning model 35 in the learning phase. First, a first neural network (reference network) 35a and a second neural network (secondary network) 35a are prepared.

[0041] The first neural network 35a is a CNN having an input layer, a feature extraction layer, and a classification layer. The feature extraction layer has a repeated structure of multiple convolution layers and pooling layers. The classification layer has, for example, one or more fully connected layers. The first neural network 35a is pre-trained using a training dataset such as Image Net. The second neural network 35a has the same network configuration as the first neural network 35a, and the various parameters (weighting coefficients) that characterize the feature extraction layer and classification layer are also the same. Hereinafter, in the learning phase of the first and second neural networks 35a, 35a, the various parameters of the feature extraction layer and classification layer share the same values.

[0042] Then, the processing unit 31 inputs a reference image R0 to the first neural network 35a, and calculates a descriptive loss, which is a loss function. Meanwhile, the processing unit 31 inputs a normal time-series data image to the second neural network 35a, and calculates a compactness loss, which is a loss function. The descriptive loss is a loss function that is generally used in learning a classifier, and is, for example, a cross-entropy error. The compact loss is expressed by the following formulas (1), (2), and (3). The compact loss is a value that corresponds to the variance of the output of the second neural network 35a within the batch, that is, in the learning data set.

[0043]

number

[0044] Then, the processing unit 31 calculates a total loss based on the descriptive loss and the compact loss. The total loss is expressed by, for example, the following formula (4).

[0045]

number

[0046] The processing unit 31 optimizes various parameters of the first and second neural networks 35a, 35a using a backpropagation method or the like so as to reduce the total loss represented by the above formula (4), thereby training the first and second neural networks 35a, 35a. Note that, when training the networks by machine learning, it is advisable to fix various parameters in the front stages and adjust parameters in the rear stages.

[0047] The features output from the feature extraction layers of the first and second neural networks 35a, 35a trained in this manner have a high local density with the features of the time-series data images obtained when the twin-screw kneading extruder 1 is normal, and a low local density with the features of other arbitrary images, as shown in FIG.

[0048] Fig. 11 is a conceptual diagram showing the learning model 35 in the test phase. As shown in Fig. 11, the learning model 35 according to this embodiment can be configured using the input layer and feature extraction layer of the second neural network 35a trained as described above. Therefore, the classification layer shown in Fig. 10 is not an essential component of the learning model 35 according to this embodiment. However, it is also possible to use the second neural network 35a shown in Fig. 10 as the learning model 35 as it is, and use the feature amount output from the feature extraction layer. Note that, although two learning models 35 are illustrated in FIG. 11, this is a conceptual illustration of a state in which a time series data image to be detected and a normal time series data image as a sample are input, and does not indicate the existence of two models.

[0049] In the test phase, the processing unit 31 inputs a time series data image obtained under normal conditions as sample data to the learning model 35, thereby outputting the feature quantities of the time series data image under normal conditions. On the other hand, the processing unit 31 inputs a time series data image of an abnormality detection target to the learning model 35, thereby outputting the feature quantities of the time series data image. Then, the processing unit 31 calculates an outlier score of the feature quantity of the abnormality detection target with respect to the feature quantities of the sample data, and compares the calculated outlier score with a threshold value to determine whether or not the time series data image of the abnormality detection target is abnormal, that is, whether or not the twin-screw kneading extruder 1 is abnormal.

[0050] Note that FIG. 11 shows an example in which sample data is input into the learning model 35 and the features of the sample data are calculated. However, the features of a plurality of sample data items calculated in advance may be stored in the memory unit 32, and in the anomaly detection process, the outlier score may be calculated using the features of the sample data stored in the memory unit 32.

[0051] FIG. 12 is a scatter diagram showing the feature quantities of the time series data images. Like FIG. 8, FIG. 12 is a scatter diagram in which the high-dimensional feature quantities of the time series data images are reduced to two-dimensional feature quantities and plotted on a two-dimensional plane. In the scatter diagram, a feature quantity A of the time series data images obtained during normal operation of the twin-screw kneading extruder 1, which is the learning data, a feature quantity B of the time series data images during normal operation, which is the test data, and a feature quantity C of the time series data images during abnormal operation, which is the test data, are plotted. As can be seen from FIG. 12, the feature quantity B of the test data (normal) is in the vicinity of the feature quantity A of the learning data, and the feature quantity C of the test data (abnormal) is distinguishably separated from the feature quantity A of the learning data.

[0052] FIG. 13 is a chart showing the evaluation results of anomaly detection using the learning model 35. In the table of FIG. 13, "simulated abnormality" indicates the type of abnormality. Wear indicates an abnormal state in which the first and second screws 11, 12 are worn. In this experiment, a wear state was simulated by using a FF (forward lead full flight) piece before the kneading section and an FK (forward shift kneading disk) at the beginning of the kneading section with a smaller outer diameter, and the displacement of the first and second screws 11, 12 was detected to create a time series data image. Specifically, when a screw outer diameter of 68.0 mm is normal, the wear state was simulated by using the above pieces with an outer diameter of 63.0 mm. Screw incompatibility indicates an abnormal state in which the twin-screw kneading extruder 1 was operated using a resin raw material of a different type from the expected resin raw material. In this experiment, an abnormal state was indicated in which nylon (PA) was used as the resin raw material using the first and second screws 11, 12 for polypropylene (PP). "Normal data" shows time series data images obtained when there was no wear on the first and second screws 11, 12, an appropriate resin raw material (PP) was used, and the first and second screws 11, 12 were rotated at rotation speeds of 100 rpm, 200 rpm, and 400 rpm. "Abnormal data" shows time series data images obtained when the first and second screws 11, 12 were rotated at speeds of 100 rpm, 200 rpm, and 400 rpm using an appropriate resin raw material (PP) in the above-mentioned wear abnormality state for "wear". "Screw unsuitability" shows time series data images obtained when the first and second screws 11, 12 were rotated at speeds of 100 rpm, 200 rpm, and 400 rpm using the above-mentioned inappropriate resin raw material (PA) without wear on the first and second screws 11, 12.

[0053] Then, a part of the normal data in the table of Fig. 13 was used as a learning data set to train the learning model 35, and the discrimination ability between the remaining normal data and the abnormal data was evaluated using the AUC (Area Under the Curve). The AUCs for the normal data and the abnormal data were 0.96, 1.00, 0.99, 1.00, 1.00, and 0.94, confirming high discrimination ability.

[0054] As described above, according to the abnormality detection device 3 of this embodiment, time series data of the displacement of the first and second screws 11, 12 of the twin-screw kneading extruder 1 can be converted into a time series data image represented by an image, and an abnormality in the twin-screw kneading extruder 1 can be determined based on the time series data image.

[0055] Furthermore, by configuring the learning model 35 as a one-class classification model, the learning model 35 can be machine-trained using time-series data images obtained when the twin-screw kneading extruder 1 is operating normally, without using time-series data images of the twin-screw kneading extruder 1 during abnormal times.

[0056] Furthermore, by expressing the time series changes in the displacements in the X-axis direction and the Y-axis direction of each of the first and second screws 11, 12 in a single time series data image and calculating the feature amount of the time series data image using the learning model 35, it is possible to easily obtain feature amounts that accurately represent the states of the first and second screws 11, 12 of the twin-screw kneading extruder 1. By using the feature amounts obtained in this manner, abnormalities in the twin-screw kneading extruder 1 can be detected with high accuracy.

[0057] Furthermore, by using a time series data image that includes the first and second time series data images representing the displacement of each of the first and second screws 11 and 12 without any distortion, abnormalities in the twin-screw kneading extruder 1 can be accurately detected. It has been confirmed that if images obtained by stretching the first and second time-series data images vertically or horizontally are used, the characteristics of the first and second screws 11, 12 are affected, and the accuracy of abnormality detection decreases.

[0058] In this embodiment, an example has been described in which the displacement sensor 2 is provided on the drive shafts 11a, 12a of the first and second screws 11, 12, but the displacement sensor 2 may be provided in other locations as long as it can detect abnormalities in the first and second screws 11, 12. For example, the displacement sensor 2 may be provided on the output shaft of the motor 13. The output shaft is an example of a rotating shaft of the molding machine.

[0059] In addition, in this embodiment, an example has been described in which a displacement sensor 2 is provided on each of the first and second screws 11, 12. However, the displacement sensor 2 may be provided on only one of the first and second screws 11, 12 to detect abnormalities.

[0060] Furthermore, in the present embodiment, an example of detecting the displacement of the rotation central axis of the first and second screws 11, 12 has been described, but the twin-screw kneading extruder 1 may be detected by detecting physical quantities such as torque, rotation speed, or rotation acceleration applied to the first and second screws 11, 12 and using time series data of the physical quantities. In addition, the twin-screw kneading extruder 1 may be configured to detect an abnormality using a time series data image including a plurality of images depicting different types of physical quantities. For example, the twin-screw kneading extruder 1 may be configured to detect an abnormality using a time series data image including a first time series data image depicting the displacement of the rotation central axis of the first and second screws 11, 12 and a time series data image depicting the torque or rotation acceleration applied to the first and second screws 11, 12. Similarly, an arbitrary physical quantity in a moving part may be detected to detect an abnormality in the industrial machine.

[0061] Furthermore, in the present embodiment, an example has been described in which a time series data image including a two-dimensional first time series data image (first image) showing the displacement of the first screw 11 with two numerical values ​​and a two-dimensional second time series data image (second image) showing the displacement of the second screw 12 with two numerical values ​​is used, but a time series data image obtained by combining four images depicting four types of two-dimensional physical quantity data may be used to detect an abnormality in the twin-screw kneading extruder 1. Also, a time series data image including five or more images depicting time series data may be used to detect an abnormality in the twin-screw kneading extruder 1.

[0062] Furthermore, in the present embodiment, an example has been described in which a time series data image is used in which physical quantities that change over time are directly depicted. However, the detected time series physical quantities may be frequency converted, and a time series data image in which the frequency spectrum of the physical quantities is depicted may be used to detect abnormalities in the twin-screw kneading extruder 1 or the industrial machinery.

[0063] Furthermore, in this embodiment, an example of detecting an abnormality in the first and second screws 11, 12 of the twin-screw kneading extruder 1 has been described, but the abnormality detection device 3 may be configured to detect an abnormality in other moving parts, such as a single-screw extruder, an injection molding machine, or a moving part of other industrial machines. In this case, the learning model 35 learns time-series data images of the moving parts when they are normal. In the anomaly detection device 3 for detecting anomalies in a moving part of an industrial machine, the processing unit 31, like the present embodiment, acquires time series data indicating a two-dimensional first physical quantity consisting of two numerical values ​​and time series data indicating a two-dimensional second physical quantity consisting of two numerical values, converts the acquired first time series data into a first image depicting the first physical quantity, converts the acquired second time series data into a second image depicting the second physical quantity, and synthesizes a time series data image including the converted first image and second image in one image. Then, the processing unit 31 inputs the synthesized time series data image into a learning model to calculate the feature amount of the time series data image, and calculates an outlier score for the feature amount of the time series data image in a normal state. The processing unit 31 can detect an anomaly in the industrial machine by comparing the outlier score with a predetermined threshold value. Incidentally, a configuration may be adopted in which an abnormality in industrial machinery is detected using a time-series data image including three or more images depicting three or more types of physical quantities. The method of generating the learning model 35 for detecting anomalies in moving parts of industrial machinery is the same as in the embodiment. A plurality of time-series data images generated during normal operation of the industrial machinery and a plurality of reference images are used as a learning data set, and a one-class classification model learning model is machine-learned by the method shown in FIG.

[0064] Furthermore, in this embodiment, an example has been described in which the processing unit 31 of the anomaly detection device 3 trains the learning model 35 by machine learning, but the learning model 35 may be trained by machine learning using another external computer or server.

[0065] Furthermore, although a one-class classification model has been described as an example of the learning model 35, general neural networks such as CNN, U-Net, RNN (Recurrent Neural Network), other SVMs (Support Vector Machines), Bayesian networks, regression trees, etc. may also be used.

[0066] Furthermore, the anomaly detection device 3 may be configured to discriminate time-series image data at typical abnormal times from time-series image data at other abnormal times using a learning model 35 that is a plurality of one-class classification models. For example, the anomaly detection device 3 includes a first learning model that has been trained with time-series image data at normal times, as in the above embodiment. The anomaly detection device 3 also includes a second learning model 35 that has been trained with time-series image data at normal times and time-series image data at a first abnormal time. The anomaly detection device 3 can discriminate between time-series image data at normal times, time-series image data at a first abnormal time, and time-series image data at other abnormal times by using the first learning model 35 and the second learning model 35. Similarly, the anomaly detection device 3 can be configured to discriminate between time-series image data at two or more abnormal times by providing three or more learning models 35. The anomaly detection device 3 may also be configured to classify time-series image data at normal times and time-series image data at abnormal times using a machine-learned multi-class classification model.

[0067] Furthermore, although an example of machine learning mainly using neural networks such as CNN has been described, the configuration of the trained model is not limited to CNN, RNN, etc., and may be configured using neural networks other than CNN and RNN, SVM (Support Vector Machine), Bayesian network, regression tree, etc. The means of the present disclosure are appended. (Appendix 1) A computer program for causing a computer to execute a process for detecting an abnormality in an industrial machine having a moving part, acquiring time-series physical quantity data output from a sensor that detects a physical quantity related to the motion of the movable part; converting the acquired time-series physical quantity data into a time-series data image represented by an image; Calculating a feature amount of the time-series data image by inputting the converted time-series data image into a learning model that has learned features of the time-series data image related to the normal movable part; The presence or absence of an abnormality in the industrial machine is determined based on the calculated feature amount. A computer program for causing the computer to execute a process. (Appendix 2) The learning model is a model obtained by machine learning. 2. The computer program of claim 1. (Appendix 3) The learning model is a one-class classification model obtained by machine learning. 10. A computer program according to claim 1 or 2. (Appendix 4) the physical quantity data includes time series data indicating a first physical quantity and time series data indicating a second physical quantity; The time series data image is A first image depicting the first physical quantity and a second image depicting the second physical quantity are included in one image. 4. A computer program according to any one of claims 1 to 3. (Appendix 5) The first physical quantity and the second physical quantity are two-dimensional physical quantities each consisting of two numerical values. 5. A computer program according to claim 4. (Appendix 6) The movable part is a rotating shaft of a molding machine, and the physical quantity data includes time-series data indicating a displacement, a torque, a rotation speed, or a rotation acceleration of the rotating shaft of the molding machine. 6. A computer program according to any one of claims 1 to 5. (Appendix 7) The movable part is a screw of a molding machine, and the physical quantity data includes time-series data indicating a displacement, a torque, a rotation speed, or a rotation acceleration of the screw of the molding machine. 7. A computer program according to any one of claims 1 to 6. (Appendix 8) The movable parts are a first screw and a second screw of a twin-screw kneading extruder, and the physical quantity data includes time series data indicating displacements in a first axial direction and a second axial direction intersecting a rotation central axis of the first screw, and time series data indicating displacements in a third axial direction and a fourth axial direction intersecting a rotation central axis of the second screw, The time series data image is A first image in which the displacement of the first screw is depicted using the intersecting first and second axial directions as coordinate axes, and a second image in which the displacement of the second screw is depicted using the intersecting third and fourth axial directions as coordinate axes are included in one image. 4. A computer program according to any one of claims 1 to 3. (Appendix 9) The time series data image is a substantially square image, The first image and the second image, and a blank image that supplements the portion other than the first image and the second image. 9. The computer program of claim 8. (Appendix 10) An anomaly detection method for detecting an anomaly in an industrial machine having a moving part, comprising: acquiring time-series physical quantity data output from a sensor that detects a physical quantity related to the movement of the movable part; converting the acquired time-series physical quantity data into a time-series data image represented by an image; Calculating a feature amount of the time-series data image by inputting the converted time-series data image into a learning model that has learned features of the time-series data image related to the normal movable part; The presence or absence of an abnormality in the industrial machine is determined based on the calculated feature amount. Anomaly detection methods. (Appendix 11) An abnormality detection device for detecting an abnormality in an industrial machine having a moving part, A sensor for detecting a physical quantity related to the movement of the movable part; an acquisition unit that acquires time-series physical quantity data output from the sensor; a conversion unit that converts the time-series physical quantity data acquired by the acquisition unit into a time-series data image represented by an image; a calculation unit that calculates a feature amount of the time-series data image by inputting the converted time-series data image into a learning model that has learned features of the time-series data image related to the normal movable part; a determination unit that determines whether or not there is an abnormality in the industrial machine based on the feature amount calculated by the calculation unit; An anomaly detection device comprising: (Appendix 12) An anomaly detection device according to claim 11; Molding machine and Equipped with The abnormality detection device detects an abnormality in the molding machine. Molding machine system. (Appendix 13) A method for generating a learning model for detecting an abnormality in a twin-screw kneading extruder having a first screw and a second screw, Based on time series data showing displacements in a first axial direction and a second axial direction intersecting a rotation central axis of a first screw of the twin-screw kneading extruder during normal operation and time series data showing displacements in a third axial direction and a fourth axial direction intersecting a rotation central axis of a second screw of the twin-screw kneading extruder during normal operation, a plurality of time series data images are generated, the single image including a first image depicting displacements in the first axial direction and the second axial direction intersecting a rotation central axis of the first screw and a second image depicting displacements in the third axial direction and the fourth axial direction intersecting a rotation central axis of the second screw; A learning model is generated based on a learning data set including the generated plurality of time-series data images and a plurality of reference images having any feature, and when a time-series data image including a first image and a second image depicting the displacement of the rotation central axes of the first screw and the second screw is input, the learning model outputs a feature amount corresponding to normal operation and abnormal operation of the twin-screw kneading extruder. A method for generating a learning model in which processing is performed by a computer. [Explanation of symbols]

[0068] 1 Twin screw extruder 2 Displacement Sensor 3. Anomaly detection device 3a Display section 4 Recording media 10 Cylinders 10a Hopper 11 First screw 11a Drive shaft 12 Second screw 12a Drive shaft 13 Motor 14 Reducer 21 First displacement sensor 22 Second displacement sensor 23 3rd displacement sensor 24 4th displacement sensor 31 Processing section 32 Storage section 33 Input Interface 34 Output Interface 35 Learning Model 35a Neural Networks P Computer Program

Claims

1. A computer program for causing a computer to execute a process for detecting an abnormality in a molding machine having a rotation shaft, acquiring time-series physical quantity data output from a sensor that detects a physical quantity related to the motion of the rotating shaft, the physical quantity data including time-series data indicating time changes of at least two numerical values ​​including a torque and a displacement of the rotating shaft; converting the acquired time-series physical quantity data into a time-series data image represented by an image obtained by plotting the two numerical values ​​at each time point in the time series with intersecting X-axis and Y-axis directions as coordinate axes; Calculating a feature amount of the time-series data image by inputting the converted time-series data image into a learning model that has learned features of the time-series data image related to the normal rotation axis; The presence or absence of an abnormality in the molding machine is determined based on the calculated characteristic amount. A computer program for causing the computer to execute a process.

2. The learning model is a model obtained by machine learning.

2. The computer program product of claim 1.

3. The learning model is a one-class classification model obtained by machine learning.

3. A computer program according to claim 1 or 2.

4. the physical quantity data includes time series data indicating a first physical quantity and time series data indicating a second physical quantity; The time series data image is A first image depicting the first physical quantity and a second image depicting the second physical quantity are included in one image. A computer program according to any one of claims 1 to 3.

5. The first physical quantity and the second physical quantity are two-dimensional physical quantities each consisting of two numerical values.

5. A computer program according to claim 4.

6. The time series data image is A first image depicting the first physical quantity and a second image depicting the second physical quantity are included in one image without being modified. A computer program according to claim 4 or claim 5.

7. The physical quantity data includes time-series data indicating the rotation speed or the rotation acceleration of the rotation shaft of the molding machine. A computer program according to any one of claims 1 to 6.

8. The rotating shaft is a screw of the molding machine. A computer program according to any one of claims 1 to 7.

9. The rotating shafts are a first screw and a second screw of a twin-screw kneading extruder, and the physical quantity data includes time series data indicating displacements in a first axial direction and a second axial direction intersecting a rotation central axis of the first screw, and time series data indicating torques of the first screw and the second screw, The time series data image is A first image in which the displacement of the first screw is depicted using the intersecting first axial direction and second axial direction as coordinate axes, and a second image in which the torque of the first screw and the torque of the second screw are depicted using the intersecting third axial direction and fourth axial direction as coordinate axes, are included in one image. A computer program according to any one of claims 1 to 3.

10. An abnormality detection method for detecting an abnormality in a molding machine having a rotating shaft, comprising: acquiring time-series physical quantity data output from a sensor that detects a physical quantity related to the motion of the rotating shaft, the physical quantity data including time-series data indicating time changes of at least two numerical values ​​including a torque and a displacement of the rotating shaft; converting the acquired time-series physical quantity data into a time-series data image represented by an image obtained by plotting the two numerical values ​​at each time point in the time series with intersecting X-axis and Y-axis directions as coordinate axes; Calculating a feature amount of the time-series data image by inputting the converted time-series data image into a learning model that has learned features of the time-series data image related to the normal rotation axis; The presence or absence of an abnormality in the molding machine is determined based on the calculated characteristic amount. Anomaly detection methods.

11. An abnormality detection device for detecting an abnormality in a molding machine having a rotating shaft, a sensor for detecting a physical quantity related to the motion of the rotation shaft; an acquisition unit that acquires time-series physical quantity data output from the sensor, the physical quantity data including time-series data indicating time changes of at least two numerical values ​​including a torque and a displacement of the rotation shaft; a conversion unit that converts the time-series physical quantity data acquired by the acquisition unit into a time-series data image represented by an image in which the two numerical values ​​at each time point in the time series are plotted with intersecting X-axis and Y-axis directions as coordinate axes, based on the time-series physical quantity data acquired by the acquisition unit; a calculation unit that calculates a feature amount of the time-series data image by inputting the converted time-series data image into a learning model that has learned features of the time-series data image related to the normal rotation axis; a determination unit that determines whether or not there is an abnormality in the molding machine based on the feature amount calculated by the calculation unit; An anomaly detection device comprising:

12. The anomaly detection device according to claim 11 ; Molding machine and Equipped with The abnormality detection device detects an abnormality in the molding machine. Molding machine system.

Images