JP2022113410A - Learning device and program, and mold abnormality predictor - Google Patents

Abstract

To accurately detect the abnormality of a mold mounted on a press machine.SOLUTION: A learning device (100) includes: data obtaining means (101) that obtains measurement data (DT1) including load data applied on a workpiece during press operation by a press machine (11) and vibration data of a flame of the press machine, for each press cycle; feature amount calculating means (102) for calculating plural kinds of future amounts including load feature amount and vibration feature amount for each press cycle, on the basis of the obtained measurement data; generating means (104) for mechanically learning correlation between the calculated plural kinds of feature amounts and a mold state, and generating a prediction model for predicting the presence/absence of abnormality of a mold for each press cycle; and model memory means (114) that stores the generated prediction models (M1, M2,...).SELECTED DRAWING: Figure 5

Description

TECHNICAL FIELD The present invention relates to a learning device, a program, and a die abnormality prediction device, and in particular, a learning device and a program for generating a prediction model for predicting the presence or absence of an abnormality in a die mounted on a press machine by machine learning, and , relates to a mold abnormality prediction device that predicts the presence or absence of a mold abnormality based on a prediction model generated by machine learning.

A forging press device presses a workpiece by rotationally driving a crankshaft to move a slide up and down. In such a forging press apparatus, as described in Japanese Patent Application Laid-Open No. 2019-13947 (Patent Document 1), the load applied to the workpiece during processing is monitored, and an abnormality is detected based on the peak value of the load. This is traditionally done. This is because it is known that the peak value of the load has a high correlation with the quality of the product.

In addition, as described in Japanese Patent Application Laid-Open No. 2019-13976 (Patent Document 2), there is a conventional technology for predicting failure of a press by using a plurality of sensors that detect the operating conditions of the press. Proposed.

JP 2019-13947 A JP 2019-13976 A

In Patent Literature 1, defective products can be detected based on the peak value of the load, but the cause (abnormality factor) cannot be identified.

In Patent Document 2, changes from the initial normal state of the measurement results obtained from a plurality of sensors are confirmed by a monitoring PC, and past empirical rules are also compared to predict failure of the press machine. However, it is difficult to accurately predict failures.

In particular, the die of a press machine has a higher failure frequency than other parts, so there has been a demand for a technology that can detect die abnormalities (failure, breakage, etc.) without relying on the experience of the operator.

The present invention has been made to solve the above problems, and its object is to provide a learning device and a program for accurately detecting an abnormality in a mold mounted on a press, and , to provide a mold abnormality prediction device.

A learning device according to one aspect of the present invention is a learning device that generates a prediction model for predicting the presence or absence of an abnormality in a die mounted on a press machine by machine learning, and comprises data acquisition means and feature amount calculation means. , generating means, and model storage means. The data acquisition means acquires measurement data including load data applied to the workpiece during press operation by the press machine and vibration data of the frame of the press machine for each press cycle. The feature amount calculation means calculates a plurality of types of feature amounts including the load feature amount and the vibration feature amount for each press cycle based on the measurement data acquired by the data acquisition means. The generation means machine-learns the correlation between the plurality of types of feature quantities calculated by the feature quantity calculation means and the state of the mold to generate a prediction model for predicting the presence or absence of an abnormality in the mold for each press cycle. . The model storage means stores the prediction model generated by the generation means.

Preferably, the generation means machine-learns the correlation between the plurality of types of feature quantities and the mold state for each mold pattern, and the model storage means associates the mold pattern with mold identification information for identifying the mold pattern, and predicts remember the model.

If the press machine is a multi-stage press machine equipped with a plurality of dies, the generating means receives a plurality of types of feature values as input and generates a prediction model that outputs the presence or absence of anomalies for individual dies. is desirable. In a broad sense, an individual mold means a mold for each process without distinguishing between upper mold and lower mold, and in a narrow sense means an upper mold and a lower mold for each process.

In a mode in which the frame has first and second pillars respectively provided upstream and downstream of a plurality of molds, the feature amount calculation means is configured to calculate vibration from a vibration meter attached to the first pillar It is desirable to calculate the first vibration feature amount based on the vibration data of , and calculate the second vibration feature amount based on the vibration data from the vibration meter attached to the second support. Each vibration feature quantity includes at least one of a wave height vector, a time vector, and a frequency vector, and preferably includes at least a frequency vector. More preferably, it contains a frequency vector and a wave height vector or a time vector.

In addition to the load feature amount and the vibration feature amount, the feature amount calculation means further calculates at least one of a die contact time, a die height value, a die temperature, a die lubricant amount, and a die shot number. is desirable. In addition, it is preferable that the types of the feature amount include the mold temperature and the amount of the mold lubricating fluid. By adding these to the explanatory variables of the prediction model, the magnitude of the load on the mold can be detected with high accuracy. If the press is a multi-stage press with multiple dies, the die temperature, die lubrication volume, and die shot count should be measured for each die (at least for each run) It is desirable that

A mold abnormality prediction device according to another aspect of the present invention is a mold abnormality prediction device for predicting the presence or absence of an abnormality in a mold mounted on a press, and is generated by any one of the learning devices described above. and a storage means for storing the predicted model. Data acquisition means for acquiring measurement data including load data applied to the workpiece during press operation by the press machine and vibration data of the frame of the press machine for each press cycle, and based on the measurement data acquired by the data acquisition means feature amount calculating means for calculating a plurality of types of feature amounts including the load feature amount and the vibration feature amount for each press cycle; and storing the plurality of types of feature amounts calculated by the feature amount calculating means in a storage means. Prediction means for predicting the presence or absence of an abnormality in the mold for each press cycle by inputting the data into the predicted model, and output means for outputting the prediction result of the prediction means.

Preferably, the storage means stores a plurality of prediction models in association with mold identification information for identifying mold patterns. In this case, the die abnormality prediction device selects a prediction model to be used for die abnormality prediction based on the die pattern corresponding to the current press operation from among a plurality of prediction models before starting press operation by the press machine. It is desirable to further include a model discriminating means for discriminating.

A learning program according to still another aspect of the present invention is a learning program for generating a prediction model for predicting the presence or absence of an abnormality in a die mounted on a press by machine learning. a step of calculating a plurality of types of feature quantities including load feature quantity and vibration feature quantity for each press cycle based on measurement data including load waveform data and vibration waveform data of the frame of the press machine; a step of machine learning the correlation between the multiple types of feature values obtained and the state of the mold to generate a prediction model for predicting the presence or absence of abnormalities in the mold for each press cycle; and a step of storing in a storage means.

According to the present invention, it is possible to accurately detect an abnormality in a mold using a predictive model.

BRIEF DESCRIPTION OF THE DRAWINGS It is a figure which shows schematic structure of the forging press apparatus which concerns on embodiment of this invention. BRIEF DESCRIPTION OF THE DRAWINGS It is a figure which shows typically the press and sensors with which the forging press apparatus which concerns on embodiment of this invention is provided. FIG. 3 is a diagram schematically showing the peripheral structure of the press and the operating state information detected by sensors in the embodiment of the present invention; It is a figure which shows typically the example of arrangement|positioning of the metal mold|die with which the press in embodiment of this invention is provided. 1 is a block diagram showing the functional configuration of a learning device according to an embodiment of the present invention; FIG. 4 is a flow chart showing a prediction model generation method by the learning device according to the embodiment of the present invention. (A) to (C) are graphs for explaining a method of calculating a feature amount of frame vibration. (A) to (D) are graphs for explaining a method of calculating a feature amount of frame vibration. 1 is a block diagram showing the functional configuration of a mold abnormality prediction device according to an embodiment of the present invention; FIG. 4 is a flow chart showing a mold abnormality prediction method by the mold abnormality prediction device according to the embodiment of the present invention.

An embodiment of the present invention will be described in detail with reference to the drawings. The same or corresponding parts in the drawings are denoted by the same reference numerals, and the description thereof will not be repeated.

In the present embodiment, the forging press machine will be described as having a function of detecting an abnormality in the die based on the learning result of the learning device.

<Schematic configuration of forging press>
First, referring to FIGS. 1 to 3, a schematic configuration of a forging press device 10 will be described. As shown in FIG. 1, the forging press apparatus 10 includes a press machine 11 that presses a workpiece, and a PLC (Programmable Logic Controller) 40 that controls the operation of the press machine 11 .

As shown in FIG. 2, the press machine 11 has a slide 1 and a bolster 2 which are provided in a frame 3 so as to face each other vertically. The frame 3 includes a pair of struts 3 a and 3 b arranged on the front side of the press machine 11 and a pair of struts 3 c and 3 d arranged on the rear side of the press machine 11 . The press machine 11 is configured such that the slide 1 connected to the connecting rod 5 moves up and down by rotationally driving the crankshaft 4 .

A flywheel 8 rotatably driven by a belt 7 is connected via a clutch 9 to one end of the crankshaft 4 . A brake device 12 is attached to the other end of the crankshaft 4 to stop the slide 1 by stopping the rotation of the crankshaft 4 . The press machine 11 rotates the flywheel 8 at a constant speed with the main motor 6, engages the clutch 9, and transmits the rotation of the flywheel 8 to the crankshaft 4, thereby elevating the slide 1 and moving it on the bolster 2. Forging the work placed on. The workpiece is processed into a size and shape according to the pattern of the mold including the lower mold (not shown) installed on the bolster 2 and the upper mold (not shown) installed at the lower end of the slide 1. , the final product. "Die" means a die used for forging. The mold is typically made of special steel.

A forging press device 10 in the present embodiment is a device for processing a work by hot forging. The forging press machine 10 is provided with a plurality of sensors as shown below in order to detect the operating conditions of the press machine 11. As shown in FIG. As will be described later, the press machine 11 may be a multi-stage press machine that processes a workpiece through a plurality of processes. That is, the press machine 11 may be equipped with a plurality of dies.
Load sensor 20: Detects the load applied to the work. For example, it is composed of a strain gauge that detects strain of the frame 3 .
Angle sensor 21: detects the position of the slide 1 and the speed (stroke) of the slide 1 by detecting the rotation angle (press angle) of the crankshaft 4;
· Brake slack pressure sensor 22 : detects the slack pressure of the brake device 12 .
· Clutch pressure sensor 23 : detects the pressure of the clutch 9 .
· Clutch tank pressure sensor 24 : detects the pressure of the hydraulic oil that drives the clutch 9 .
· Brake tank pressure sensor 25 : detects the pressure of the cooling water of the brake device 12 .
Cooling water flow rate sensor 26 : Detects the flow rate of cooling water for the brake device 12 .
BKO position displacement sensor 27: detects the position of the lower knockout device of the press machine 11 (BKO position).
SKO position displacement sensor 28: detects the position of the upper knockout device (SKO position).
· Temperature measuring resistor 29 : Detects the temperature of each part of the press machine 11 .
- Lubricant flow rate sensor 30: detects the flow rate of the lubricating liquid that lubricates the mold.
· Air blow pressure sensor 31 : detects the pressure of the high pressure air supplied to the press machine 11 .
- Material temperature sensor 32: detects the temperature of the material supplied to the press machine 11;
Die height displacement sensor 33: detects the die height of the press machine 11 (the distance between the bolster 2 and the bottom dead center position of the slide 1).
- Mold temperature sensor 34: detects the temperature of the mold.

As shown in FIG. 3 , the forging press device 10 further includes a carrier device 13 and a heater 14 . The conveying device 13 is composed of, for example, a belt conveyor, and conveys the work W from the previous process to the press machine 11 (position of the bolster 2). The heater 14 is arranged on the upstream side of the press machine 11 and is provided to prevent hardening of the work W being transported by the transport device 13 . Heater 14 is configured by, for example, a tunnel furnace.

The material temperature sensor 32 described above detects the temperature of the workpiece W immediately after passing through the heater 14 (the workpiece W near the outlet of the heater 14). A mold temperature sensor 34 detects the temperature of the mold mounted on the press machine 11 .

In addition to the above-described sensors, the forging press machine 10 includes a timer (not shown) for detecting the time required for the workpiece W to reach the press position from the exit of the heater 14 (heater exit - press time). may further include Further, a work presence/absence sensor (not shown) for detecting the presence/absence of the work W may be further provided near the outlet of the heater 14 .

The press machine 11 in the present embodiment is a multi-stage press machine, and forms the workpiece W step by step using a plurality of dies. Fig. 4 shows an example of mold arrangement. In the example of FIG. 4, the press machine 11 has three molds 51 to 53 arranged along the work W conveying direction A1. In this case, the workpiece W is molded step by step through a first process (process I) by the mold 51, a second process (process II) by the mold 52, and a third process (process III) by the mold 53 in this order. be done.

Upper molds 51a, 52a and 53a of molds 51 to 53 are fixed to the lower end of slide 1 in frame 3 and move up and down at the same time. Lower molds 51 b , 52 b , 53 b of molds 51 to 53 are fixed to the upper end of bolster 2 in frame 3 . In the following description, when there is no need to distinguish between the upper and lower three pairs of molds 51a, 51b, 52a, 52b, 53a, 53b, they are referred to as "molds 50".

The dies 51 to 53 of the press machine 11 are each lubricated with lubricating liquid supplied from a die lubricating means 60 (indicated by phantom lines). In this case, the lubricant flow rate sensor 30 and mold temperature sensor 34 are typically provided for each mold 50 .

The load sensor 20 described above is attached to one of the columns (for example, column 3c) of the frame 3. As shown in FIG. Vibration meters 35 are attached to the left and right struts 3c and 3d of the frame 3, respectively. One of the columns 3c and 3d (column 3c) is positioned upstream in the conveying direction of the mold 51 in step I, and the other of the columns 3c and 3d (column 3d) is located in the conveying direction of the mold 53 in step III. located downstream. Each vibrometer 35 is composed of an acceleration sensor, for example, and detects vibration of the frame 3 . Specifically, each vibrometer 35 detects the vibration of the supports 3c and 3d to which it is attached in a substantially horizontal direction (back and forth direction or left and right direction).

Referring to FIG. 1 again, the forging press machine 10 includes a mold abnormality prediction device (hereinafter abbreviated as "abnormality prediction device") 200 for predicting the presence or absence of abnormality in the dies 51-53. The abnormality prediction device 200 predicts whether or not each type 50 has an abnormality based on the prediction model generated by the learning device 100 . The learning device 100 and the abnormality prediction device 200 constitute a mold abnormality prediction system SYS.

Learning device 100 and abnormality prediction device 200 acquire detection signals from a plurality of sensors via PLC 40 or without PLC 40 . The PLC 40 and these devices 100 and 200 are connected by wire or wirelessly. Learning device 100 and anomaly prediction device 200 are implemented by a computer including a processor such as a CPU (Central Processing Unit) and a memory. In the present embodiment, an example in which the PLC 40 and the abnormality prediction device 200 are provided separately is shown, but this is not limiting, and the PLC 40 has the function of the abnormality prediction device 200 as described later. may

<Learning device>
(About functional configuration)
FIG. 5 is a block diagram showing the functional configuration of learning device 100 according to the present embodiment. The learning device 100 mainly includes a data acquisition unit 101 , a format conversion unit 102 , a result acquisition unit 103 , a learning unit 104 and a model storage unit 114 . The model storage unit 114 stores prediction models M1, M2, . . . for each mold pattern.

In the present embodiment, the “mold pattern” is specified by a product number (model number) corresponding to the shape of the molds 51 to 53 and a workpiece pattern indicating the arrangement pattern of the workpieces W on the molds 51 to 53. . When the works W are continuously pressed by the forging press machine 10, the part numbers of these works W are common. The arrangement patterns of the workpieces W on the molds 51 to 53 include a pattern in which the workpieces W are arranged in all the molds 51 to 53, a pattern in which the workpieces W are arranged only in the first mold 51, and a pattern in which the workpieces W are arranged only in the first mold 51. A pattern in which the workpiece W is arranged only at 53, and the like are included. Although the arrangement mode of the work W corresponding to the part number is determined on a one-to-one basis, since the works W are conveyed sequentially, the patterns differ from the arrangement mode of the work W at the start and end of the press operation. , the workpiece W is placed only on part of the molds 51-53. The arrangement mode of the workpiece W corresponding to the part number includes a pattern in which the workpieces W are arranged sequentially on all of the molds 51 to 53, a pattern in which the workpieces W are arranged in the molds 51 to 53 by skipping one, and the like. be The information for specifying the mold pattern may further include the projected area of the workpiece W.

The data acquisition unit 101 acquires measurement data from sensors included in the forging press device 10 during operation of the press machine 11 (during press operation by the forging press device 10). In this embodiment, the data acquisition unit 101 acquires measurement data (hereinafter referred to as “cycle measurement data”) and die pattern data for each press cycle.

Cycle measurement data is primary data and includes at least load data obtained from load sensor 20 and vibration data obtained from vibrometer 35 . The cycle measurement data desirably includes temperature data for each die 50 obtained from the die temperature sensor 34, lubricating fluid data for each die 50 obtained from the lubricating fluid flow sensor 30, and each die 50 counted for each press cycle. further includes at least one shot number data of

The cycle measurement data acquired by the data acquisition unit 101 is stored in the primary data storage unit 111 in time series with a unique identification number (hereinafter referred to as “cycle No.”) as an index. Each cycle measurement data DT1 is stored in the primary data storage unit 111 together with the corresponding mold pattern data. The mold pattern data includes the product number and work pattern of the target work W, as described above. The product number of the workpiece W is input by a user (operator or administrator) via an input unit (not shown) before the forging press device 10 starts the press operation, for example. The work pattern is detected by work pattern detection means (not shown) provided in the forging press device 10 .

The format conversion unit 102 converts the format of primary data into secondary data in a format suitable for machine learning. Specifically, the format conversion unit 102 functions as feature amount calculation means, and calculates a plurality of types of feature amounts as explanatory variables of the prediction model based on the cycle measurement data DT1. More specifically, a plurality of types of feature quantities including at least the load feature quantity and the vibration feature quantity are calculated for each press cycle.

Details of the types of feature amounts (secondary data) calculated by the format conversion unit 102 are shown in Table 1 below. Table 1 lists the types of feature values and the data items for each type. Also, the type of each data item (scalar or vector) is written together.

The "load feature amount" is an index that can be calculated based on the load data obtained from the load sensor 20, and includes a load peak value or a load increase slope. The load increase slope is calculated as the slope of a linear expression passing through two points, the rising position and the peak position of the forming load curve. In the present embodiment, the load feature amount includes both the load peak value and the load increase slope. Therefore, it is considered possible to accurately predict the presence or absence of an abnormality in each die 50, which is directly related to the quality of the product.

The “mold contact time” is also an index that can be calculated based on the load data obtained from the load sensor 20, and the time between two points between the rising position and the peak position of the forming load curve is calculated as the mold contact time. be done. In this way, the mold contact time is calculated not for each of the molds 51 to 53 but as one feature quantity.

The feature amount of the “die height value” is calculated from the output value (die height data) from the die height displacement sensor 33 . By considering not only the load feature amount but also the die height value, it becomes possible to predict the presence or absence of wear of the dies 51 to 53 with higher accuracy.

The “mold temperature” feature amount is calculated based on temperature data obtained from the mold temperature sensor 34 attached to each mold 50 . For example, the mold temperature sensors 34 may be provided only on the lower molds 51b, 52b, and 53b in each of the steps I to III, and only the feature values of the temperatures of the lower molds 51b, 52b, and 53b may be calculated. The feature value of the temperature of each die 50 (at least the lower dies 51b, 52b, 53b) is calculated, for example, as the average temperature of the specified area immediately before pressing (while the workpiece is being conveyed). In addition to/instead of the average temperature, the maximum temperature, the minimum temperature, and the like may be adopted as the feature amount of the mold temperature.

The characteristic value of the “mold lubricating fluid amount” is calculated based on the lubricating fluid amount data obtained from the lubricating fluid flow rate sensor 30 provided in the lubricating fluid passage to each die 50 . The feature amount of the amount of lubricating fluid for each die 50 is, for example, the amount (total amount) of lubricating fluid for each die 50 immediately before pressing (while the workpiece is being conveyed).

The feature amount of “frame vibration” is calculated based on vibration data obtained from vibration meters 35 provided on the left and right struts 3c and 3d of the frame 3, respectively. The feature quantity of each of the left and right frame vibrations includes, for example, at least one of a wave height vector, a time vector, and a frequency vector in a specific section of one cycle. As shown in Table 1, the feature amount of frame vibration desirably includes two items, wave height vector and frequency vector, but may include two items, time vector and frequency vector. Thus, it is desirable to include at least the frequency vector. The details of a specific method for calculating the feature amount of frame vibration will be described later.

The feature amount of the “number of mold shots” is a value common to the primary data, and includes the number of shots of each mold 50 (total).

Cycle feature data DT2 including a plurality of types of feature amounts calculated by format conversion section 102 is stored in secondary data storage section 112 as secondary data. Each cycle characteristic data DT2 has, for example, the same cycle No. as the original cycle measurement data DT1. and is stored in the secondary data storage unit 112 in association with the mold pattern data.

The result acquisition unit 103 acquires mold abnormality information input by the user via input means such as the operation unit 120, for example. The mold abnormality information is information indicating, for example, in which mold 50 an abnormality occurred in which press cycle. The result acquisition unit 103 generates result data (hereinafter referred to as “cycle result data”) indicating the die state for each press cycle based on the acquired die abnormality information. The cycle result data DT3 generated by the result acquisition unit 103 is, for example, cycle number. , and stored in the result data storage unit 113 . “Mold state” represents the state of each mold 50 (abnormal/abnormal). It should be noted that each cycle result data DT3 only needs to be able to identify which cycle measurement data DT1 or which cycle characteristic data DT2 the cycle result data DT3 corresponds to. is not limited to the form stored in association with.

The learning unit 104 uses the large number of cycle feature data DT2 stored in the secondary data storage unit 112 and the large number of cycle result data DT3 stored in the result data storage unit 113 as teacher data to obtain a plurality of types of data for each press cycle. machine-learning the correlation between the feature quantity and the mold state. That is, a prediction model is generated for predicting the presence or absence of an abnormality in the mold for each press cycle. A known algorithm such as isolation forest is used as a machine learning algorithm for the prediction model. Isolation forest is a general-purpose machine learning algorithm used for anomaly detection.

In this embodiment, the learning unit 104 machine-learns the correlation between a plurality of types of feature quantities and the mold state for each “mold pattern”, number of combinations), predictive models M1, M2, . . . are generated. A plurality of prediction models M1, M2, . . . generated by learning unit 104 are stored in model storage unit 114. FIG. That is, the model storage unit 114 stores each prediction model in association with the mold identification information that identifies the mold pattern.

The functions of the data acquisition unit 101, the format conversion unit 102, the result acquisition unit 103, and the learning unit 104 described above are realized by the processor executing software. Operation unit 120 is configured by a user interface such as a keyboard, mouse, and touch panel, for example. The primary data storage unit 111, the secondary data storage unit 112, the result data storage unit 113, and the model storage unit 114 are typically configured by non-volatile storage devices included in the computer.

(About operation)
FIG. 6 is a flowchart showing a prediction model generation method by learning device 100 according to the present embodiment. The model generation process shown in FIG. 6 is implemented by the processor of learning device 100 executing a learning program stored in advance in the memory. Note that this processing is started in response to the user inputting an instruction to start learning via the operation unit 120 . In the present embodiment, a plurality of cycle measurement data DT1 are stored in the primary data storage unit 111 before this process is started. and mold pattern data are stored in chronological order. Further, in the result data storage unit 113, a plurality of cycle result data DT3 are stored as cycle numbers. are stored in association with .

Referring to FIG. 6, learning device 100 identifies one mold pattern (in a predetermined order) from a combination of the product number of workpiece W and the workpiece pattern (step S2), and in primary data storage unit 111, Cycle measurement data DT1 associated with the mold pattern is retrieved (step S4).

After that, the format conversion unit 102 reads all of the searched (target) cycle measurement data DT1 from the primary data storage unit 111 (step S6), and converts each read cycle measurement data DT1 into cycle feature data DT2. (step S8). That is, the format conversion unit 102 calculates a plurality of types of feature amounts shown in Table 1 above based on each data included in the cycle measurement data DT1, and records them in the secondary data storage unit 112 as the cycle feature data DT2. . Here, a calculation example of the feature amount of "frame vibration" will be described with reference to FIGS. 7 and 8. FIG.

FIG. 7A is a graph showing strokes of slide 1 along the time axis. 7B and 7C are graphs showing changes in the magnitude of vibration (pulse) along the same time axis as in FIG. 7A. ² ”. FIG. 8(B) is a graph showing the stroke of slide 1, like FIG. 7(A). FIG. 8A is a graph showing the brake slack signal along the same time axis. FIG. 8(C) is a graph showing changes in the magnitude of vibration (pulse), like FIGS. 7(B) and (C). FIG. 8(D) is a graph showing the transition of the power spectral density with the frequency as the horizontal axis (unit: Hz). Here, it is assumed that the vibration detected by the vibrometer 35 attached to the post 3c on the rear left side of the frame 3, for example, is shown.

A method of calculating the wave height vector will be described with reference to FIG. 7(B). The format conversion unit 102 extracts a pulse in a section TS exceeding the baseline (predetermined value) BL, and divides the extracted section TS into d strips along the time axis. Then, the pulse data is processed into a _{d -} dimensional vector (p ₌ [h1,h2,..., _hj ,...,hd]) to calculate the wave height vector. However, “h _j ” is the average value within each strip.

A method of calculating the time vector will be described with reference to FIG. The format conversion unit 102 divides the pulse in the section TS into _dw pieces in the wave height direction, and obtains _2dw pieces of time that intersect the division line (horizontal line). Then, the time data is processed into a 2d _w -dimensional vector (q=[w ₁ , w ₂ , . . . , w _j , . . . , w _2d ]) to calculate the time vector. However, for 'w _j ', select the two outer intersections with respect to the pulse apex position. For example, looking at the dividing line Lx, the outer intersection times t1 and t4 are selected and the times t2 and t3 are not selected. Time t1 corresponds to the time when the upper molds of the molds 51 to 53 start to come into contact with the work W, and time t4 corresponds to the time when the upper molds of the molds 51 to 53 start to separate from the work W.

A method of calculating the frequency vector will be described with reference to FIG. The format conversion unit 102 cuts out three types of analysis intervals, that is, a “pre-stroke interval,” a “pre-mold contact interval,” and a “during-mold contact interval,” from the vibration waveform shown in FIG. 8(C). The "pre-stroke section" is a section before time t11 when the brake loosening signal shown in FIG. 8A is turned ON. The “pre-mold contact section” is a section from time t11 to time t12 when the stroke position shown in FIG. 8B becomes the mold contact position. The "during mold contact section" is a section from time t12 to time t13 when the brake loosening signal shown in FIG. 8A is turned OFF.

The format conversion unit 102 Fourier-transforms the vibration signal for each analysis interval and calculates the power spectral density PSD. FIG. 8(D) shows a power spectrum density waveform with a frequency of 0 to fs/2 Hz. "fs" is the sampling frequency, fs/2=50 in this example.

Format conversion section 102 sets, for example, lower limit frequency f _min =0 and upper limit frequency f _max =fs/2, divides this section FS into d strips (for example, 20 sections), and converts the power spectrum in section FS to Let the density PSD be a _d -dimensional vector (p ₌ [h ₁ , h ₂ , . . . , h _j , . value). Thereby, a frequency vector for each analysis interval is obtained.

Of the three types of analysis intervals, only the frequency vector of the “mold contacting interval” that has a strong relationship with the dies 51 to 53 may be calculated. That is, the frequency vectors of the "pre-stroke interval" and the "pre-mold contact interval" need not be included in the cycle feature data DT2.

Also, the analysis interval used for calculating the frequency vector may be used for calculating the above-described wave height vector. That is, as schematically shown in FIG. 8C, each analysis section is divided into d strips (for example, 20 pieces), and the pulse data is processed into a d-dimensional vector for each analysis section, and the wave height A vector may be calculated. Also in this case, the format conversion unit 102 may calculate only the wave height vector of the “during mold contact period”.

The cycle feature data DT2 including at least one of the vibration wave height vector, vibration time vector, and vibration frequency vector calculated by the above method has the same cycle number as the target cycle measurement data DT1. is recorded in the secondary data storage unit 112 in association with . As shown in FIG. 5, the cycle feature data DT2 may also be stored in association with the target mold pattern data.

Referring to FIG. 6 again, next, learning unit 104 reads all cycle result data DT3 corresponding to cycle measurement data DT1 read in step S6 from result data storage unit 113 (step S10). Specifically, the same cycle No. as the target cycle measurement data DT1. read out the cycle result data DT3 associated with . As a result, result data corresponding to each cycle feature data DT2 obtained in step S8 (that is, data indicating the presence or absence of abnormality in each die 50) is extracted.

The learning unit 104 machine-learns the correlation between the cycle feature data DT2 obtained in step S8 and the cycle result data DT3 read out in step S10 (step S12). That is, the learning unit 104 receives a plurality of types of feature amounts as input and generates a prediction model that outputs the presence or absence of an abnormality for each mold 50 (individual mold), that is, the mold state. After generating the prediction model, the learning unit 104 stores the model storage unit 114 in association with the mold identification information that identifies the mold pattern (product number, work pattern) identified in step S2 (step S14).

If there is an unprocessed die pattern among the combinations of the product number of the work W and the work pattern (YES in step S16), the process returns to step S2 and repeats the above process. When the processing for all mold patterns is completed (NO in step S16), the series of model generation processing ends. As a result, the model storage unit 114 stores prediction models M1, M2, . In the example of FIG. 5, the prediction model M1 is a learned model for product number "30" and work pattern No. "15", and the prediction model M2 is a learned model for product number "80" and work pattern No. "15". shown to be a model.

As described above, according to the present embodiment, in addition to the load feature amount including the load peak value or the load increase slope, at least the vibration feature amount of the left (upstream) strut 3c and the right (downstream) are used as explanatory variables of the prediction model. When an abnormality occurs in one of the plurality of molds 51 to 53, depending on the position (process), there is a difference between the left and right vibration feature amounts, or even if there is almost no difference, both vibration feature amounts are abnormal values. Therefore, it is possible to predict in which of the molds 51 to 53 an abnormality has occurred by including the left and right vibration feature amounts in explanatory variables of the prediction model. In addition, in the present embodiment, by adding at least one of the temperature data, lubricating fluid amount data, and shot number data for each mold 50 to the explanatory variables of the prediction model, the upper and lower molds of the molds 51 to 53 It is possible to accurately predict in which of the molds an abnormality has occurred.

<Abnormality prediction device>
(About functional configuration)
FIG. 9 is a block diagram showing the functional configuration of abnormality prediction device 200 according to the present embodiment. The abnormality prediction device 200 mainly includes a model storage unit 213 , a data acquisition unit 202 , a format conversion unit 203 , a prediction unit 204 and an output unit 230 .

The model storage unit 213 pre-stores a plurality of prediction models M1, M2, . . . generated by the learning device 100 . Each prediction model is also stored in the model storage unit 213 in association with the mold identification information. Learning device 100 and anomaly prediction device 200 are connected, for example, via a network. , and stores the received prediction models M1, M2, . . . in the model storage unit 213 . Alternatively, prediction models M1, M2, . . . stored in model storage unit 114 of learning device 100 may be stored in model storage unit 213 using a removable recording medium.

After the operation of the press machine 11 is started, the data acquisition unit 202 acquires the same type of measurement data (cycle measurement data) DT11 as the cycle measurement data DT1 in the learning device 100 together with the die pattern data for each press cycle.

The format conversion unit 203 functions as feature amount calculation means, like the format conversion unit 102 of the learning device 100 . That is, based on the cycle measurement data DT11, a plurality of types of feature quantities shown in Table 1 are calculated, and cycle feature data DT12 including these feature quantities are generated.

First, the prediction unit 204 determines a prediction model to be used for predicting abnormality of the dies 51 to 53 for each cycle based on the acquired die pattern data. Specifically, among the plurality of prediction models M1, M2, . . . Then, the presence or absence of abnormality in the molds 51 to 53 is predicted by inputting the plurality of types of feature amounts included in the cycle feature data DT12 to the discriminated prediction model. In other words, the "mold state" output by the prediction model is the prediction result.

The output unit 230 outputs the result of prediction by the prediction unit 204 . The output unit 230 outputs whether or not there is an abnormality in each die 50 for each press cycle. Moreover, the importance of the data items listed in Table 1 (the degree of influence on the prediction result) and the certainty of the prediction result may be further output.

The functions of the data acquisition unit 202, the format conversion unit 203, and the prediction unit 204 described above are realized by the processor executing software. The model storage unit 213 is typically composed of a non-volatile storage device included in the computer. The output unit 230 is typically configured by a display unit that displays prediction results. In addition, the output part 230 may be comprised by the communication part etc. which output a prediction result to monitoring PC42 shown in FIG.

(About operation)
FIG. 10 is a flow chart showing an abnormality prediction method for the dies 51 to 53 by the abnormality prediction device 200 according to the present embodiment. The abnormality prediction process shown in FIG. 10 is implemented by executing an abnormality prediction program pre-stored in the memory by the processor of the abnormality prediction device 200 .

Referring to FIG. 10, when the operation of press machine 11 is started, data acquiring unit 202 converts cycle measurement data DT11 including output data from predetermined sensors included in forging press machine 10 into die pattern data. (Step S26).

Subsequently, the format conversion unit 203 converts the cycle measurement data DT11 acquired by the data acquisition unit 202 in step S26 into cycle feature data DT12 (step S28). That is, the format conversion unit 203 calculates a plurality of types of feature amounts shown in Table 1 above based on each data included in the cycle measurement data DT11.

Also, the prediction unit 204 selects the model to be used this time among the plurality of prediction models M1, M2, . . . A prediction model is discriminated (step S29). The prediction model discrimination process may be performed in parallel with the data format conversion process.

The prediction unit 204 inputs the cycle feature data DT12 obtained in step S28 (that is, a plurality of types of feature amounts for one cycle) to the prediction model determined in step S29, and performs each die 50 for each press cycle. Presence or absence of abnormality is predicted (step S30).

The output unit 230 outputs the prediction result, which is the output of the prediction model, that is, the presence or absence of abnormality in each mold 50 (step S32). As an example, as shown in FIG. 9, the upper mold of mold 51 (upper mold I), the lower mold of mold 51 (lower mold I), the upper mold of mold 52 (upper mold II), . . , the presence or absence of abnormality is displayed for each of the lower molds of the mold 53 (lower mold III). This makes it possible to easily identify the type 50 that may be abnormal. The abnormality prediction results for the molds 51 to 53 may be output without distinguishing between the upper mold and the lower mold.

In addition, the output unit 230 outputs the degree of influence of the data items listed in Table 1 on the prediction result (output of the prediction model) in step S30, for example, as a graph, and outputs the degree of influence of the prediction result in step S30. degree is output as a numerical value (%) (step S33). A certainty factor may be output for prediction results (abnormal/no abnormal) for each type 50 .

In this way, in the present embodiment, i) the result of prediction of abnormality of the molds 51 to 53, ii) the degree of importance of each data item with respect to the result of prediction, and iii) the degree of certainty of the result of prediction are output. Accordingly, it is possible to easily take measures for the prediction result.

The processes of steps S26 to S33 are repeatedly executed until the press operation ends (NO in step S34).

As described above, the mold abnormality prediction system SYS according to this embodiment includes the learning device 100 . Therefore, by using the prediction model generated by the learning device 100, the abnormality prediction device 200 can detect (predict) the abnormality of the molds 51 to 53 without relying on the experience of the operator. Since the dies 51 to 53 have a higher failure frequency than other parts, the burden on the operator can be reduced according to this embodiment.

Moreover, in the present embodiment, since the presence or absence of an abnormality in the dies 51 to 53 can be predicted in real time for each press cycle, an abnormality in the dies 51 to 53 can be detected early. Therefore, mass production of defective products by the press machine 11 can be prevented or suppressed. It should be noted that the operation of the press machine 11 may be automatically stopped when any of the molds 50 is predicted to be abnormal by the prediction model.

(Modification)
In the present embodiment, the vibration meter 35 is attached to the rear struts 3c and 3d of the frame 3, and the vibration feature amount of the rear struts 3c and 3d is used for abnormality prediction. Alternatively, the vibration meter 35 may be attached to the front struts 3a and 3d, and the vibration feature amounts of the rear struts 3c and 3d may be used for abnormality prediction. Alternatively, the vibration meter 35 may be attached to all four pillars 3a to 3d, and the vibration feature amounts of the four pillars 3a to 3d may be used for abnormality prediction.

Further, in the present embodiment, the learning unit 104 generates a prediction model that outputs the presence or absence of an abnormality in each die 50. It is also possible to generate a predictive model whose output is the presence or absence of anomalies in the molds 51 to 53.

Further, in the present embodiment, an example in which the press machine 11 mounts the dies 51 to 53 has been described, but the number of dies is not limited to three. Also, the number of molds is not limited to a plurality, and may be one.

In addition, the prediction models M1, M2, . M1, M2, . . . may be updated. Alternatively, the abnormality prediction device 200 may have the function of the learning device 100 .

Further, in the present embodiment, the forging press apparatus 10 is provided with the abnormality prediction device 200, but the abnormality prediction device 200 may predict the presence or absence of an abnormality in the dies 51 to 53 off-line. good.

The learning method executed by the learning device 100 can also be provided as a program. Similarly, an abnormality prediction method for the dies 51 to 53 executed by the abnormality prediction device 200 can be provided as a program. Such a program can be provided by being recorded on an optical medium such as a CD-ROM (Compact Disc-ROM) or a computer-readable non-transitory recording medium such as a memory card. The program can also be provided by downloading via a network.

The program according to the present invention may call necessary modules out of program modules provided as part of an operating system (OS) of a computer in a predetermined sequence at a predetermined timing to execute processing. . In that case, the program itself does not include the above module, and the process is executed in cooperation with the OS. Programs that do not include such modules may also be included in the programs according to the present invention.

Also, the program according to the present invention may be provided by being incorporated into a part of another program. Even in that case, the program itself does not include the modules included in the other program, and the processing is executed in cooperation with the other program. A program incorporated in such other programs may also be included in the program according to the present invention.

It should be considered that the embodiments disclosed this time are illustrative in all respects and not restrictive. The scope of the present invention is indicated by the scope of the claims rather than the above description, and is intended to include all modifications within the meaning and range of equivalents of the scope of the claims.

3 frames, 3a, 3b, 3c, 3d support, 10 forging press device, 11 press machine, 20 load sensor, 35 vibration meter, 51, 52, 53 mold, 100 learning device, 101, 202 data acquisition unit, 102, 203 format conversion unit 103 result acquisition unit 104 learning unit 111 primary data storage unit 112 secondary data storage unit 113 result data storage unit 114, 213 model storage unit 120 operation unit 200 abnormality prediction device 204 Prediction section 230 Output section DT1, DT11 Cycle measurement data DT2, DT12 Cycle characteristic data DT3 Cycle result data M1, M2 Prediction model SYS Mold abnormality prediction system W Work.

Claims (8)

A learning device that uses machine learning to generate a prediction model for predicting the presence or absence of an abnormality in a mold mounted on a press,
data acquisition means for acquiring measurement data including load data applied to a workpiece during press operation by the press machine and vibration data of the frame of the press machine for each press cycle;
feature amount calculation means for calculating a plurality of types of feature amounts including a load feature amount and a vibration feature amount for each press cycle based on the measurement data acquired by the data acquisition means;
generating means for generating a prediction model for predicting the presence or absence of an abnormality in the mold for each press cycle by machine-learning the correlation between the plurality of types of feature quantities calculated by the feature quantity calculating means and the mold state; ,
A learning device comprising model storage means for storing the prediction model generated by the generation means. The generating means machine-learns a correlation between a plurality of types of feature quantities and mold states for each mold pattern,
2. The learning device according to claim 1, wherein said model storage means stores a prediction model in association with mold identification information for identifying a mold pattern. The press is a multistage press equipped with a plurality of dies,
3. The learning device according to claim 1, wherein said generating means receives a plurality of types of feature quantities as input and generates a prediction model outputting whether or not there is an abnormality in each mold. the frame has first and second struts provided upstream and downstream of the plurality of molds, respectively;
The feature amount calculation means calculates a first vibration feature amount based on vibration data from the vibration meter attached to the first support, and calculates vibration data from the vibration meter attached to the second support. 4. The learning device according to claim 3, wherein the second vibration feature amount is calculated based on. In addition to the load feature amount and the vibration feature amount, the feature amount calculation means further calculates at least one of a die contact time, a die height value, a die temperature, a die lubricant amount, and a die shot number. The learning device according to any one of claims 1 to 4, which calculates. A mold abnormality prediction device for predicting the presence or absence of an abnormality in a mold mounted on a press,
a storage means for storing a prediction model generated by the learning device according to any one of claims 1 to 5;
data acquisition means for acquiring measurement data including load data applied to a workpiece during press operation by the press machine and vibration data of the frame of the press machine for each press cycle;
feature amount calculation means for calculating a plurality of types of feature amounts including a load feature amount and a vibration feature amount for each press cycle based on the measurement data acquired by the data acquisition means;
prediction means for inputting the plurality of types of feature amounts calculated by the feature amount calculation means into the prediction model stored in the storage means and predicting the presence or absence of an abnormality in the mold for each press cycle;
and output means for outputting a result of prediction by the prediction means. The storage means stores a plurality of prediction models in association with mold identification information for identifying mold patterns,
Further comprising a model discriminating means for discriminating a prediction model to be used for predicting an abnormality of the die based on the die pattern corresponding to the current press operation, among the plurality of prediction models, before the press operation is started by the press machine. 7. The mold abnormality prediction device according to claim 6. A learning program for generating a prediction model for predicting the presence or absence of an abnormality in a mold mounted on a press machine by machine learning,
Based on the measurement data including the load waveform data applied to the work during the press operation by the press machine and the vibration waveform data of the frame of the press machine, a plurality of types of feature amounts including the load feature amount and the vibration feature amount are calculated. a calculating step for each cycle;
generating a prediction model for predicting the presence or absence of an abnormality in the mold for each press cycle by performing machine learning on the correlation between the calculated multiple types of feature values and the mold state;
A learning program that causes a computer to execute a step of storing the generated prediction model in a model storage means.

Images