Method and apparatus for predicting structural abnormalities

Abstract

To predict a structure abnormality in a processing machine or device that has periodicity like a press device in advance with high accuracy.SOLUTION: An AE detection signal from an AE sensor provided in a part of a structure is converted into AE image information as visual information in accordance with the strength of the AE detection signal in each one cycle. The AE image information is arranged in an image map area having vertical and horizontal axes with one axis representing the number of cycles and the other axis representing a fixed time for one cycle. Cumulative mapping image information corresponding to the number of cycles is acquired, and abnormality prediction is performed on the basis of image determination with the mapping image information and either one or both of an image pattern based on prestored mapping image information for a normal state and prestored mapping image information for an abnormal sign.EFFECT: Occurrences of defective products can be almost eliminated because occurrence of an abnormality is not detected but a sign of an abnormality occurrence is detected.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a method and apparatus for accurately predicting structural abnormalities in periodic processing machines and devices in advance.

Background Art

[0002] Forging has high productivity and is suitable for mass production. However, on the other hand, if there are abnormalities in the mold, a large number of defective products will occur. Therefore, grasping the mold state is important for continuous stable production.

[0003] Sensor technology is used to detect deterioration and abnormalities in industrial machines and structures. Depending on what the sensor detects, a suitable one is adopted. For example, as an example of detecting and predicting abnormalities using an AE sensor (Acoustic Emission Sensor) adopted in the present invention, Patent Documents 1 to 4 exist as follows.

[0004] In Patent Document 1 (Japanese Patent Laid-Open No. 52-150090), when examining the fatigue strength of a material in material mechanics, in order to accurately grasp the occurrence time, progress, and behavior of cracks, a plurality of sets of preamplifiers and AE sensors are connected to an AE position calibration monitoring device with a zone selection function, one end of a pinpoint contact type wave guide is connected to each of the AE sensors, the wave guide is supported by a wave guide fixture of a sound wave insulating material, and the other end of the wave guide is acoustically connected to the test object, and a configuration is proposed.

[0005] Further, in Patent Document 2 (Japanese Patent Laid-Open No. 4-31-0857), in order to monitor the progress of minute cracks that do not affect the strength in a bridge structure, a plurality of AE sensors are arranged in the vicinity of the crack tip position of the bridge structural member and in the predicted crack progress direction in advance so as to form a spatial filter, and a limited monitoring area is provided. Only AE signals generated from within each monitoring area are extracted from the vibrations generated by the running wheels, and the progress of cracks is estimated from the change over time of the signals.

[0006] Furthermore, Patent Document 3 (Japanese Patent Publication No. Hei 8-159151) proposes that in order to accurately estimate the remaining life of a rolling bearing, vibration information, temperature information, and load information of the rolling bearing section during operation are simultaneously monitored, and the results are compared with pre-measured fluctuations in vibration information and temperature information associated with bearing damage corresponding to each load condition to predict the degree of bearing damage and the remaining life.

[0007] Furthermore, Patent Document 4 (Japanese Patent Publication No. 2004-170397) proposes a system in which one AE sensor is attached to each measurement point to detect and monitor the degree of damage to the entire structure. The AE sensors are attached in accordance with the predetermined predicted crack occurrence or propagation locations of the structure, and perform primary processing of the AE signal detected by the composite probe, store the primary processing data, and communicate with an external source. A data processing unit is connected to the AE sensor to request and receive output of primary processing data, and to perform secondary processing of the received primary processing data. Based on the secondary processing data, a safety evaluation of the structure is performed. The primary processing of the AE signal in the AE sensor involves counting the number of times the signal exceeds a pre-input threshold per unit time, and creating primary processing data that indicates an evaluation rank according to the count. The secondary processing in the data processing unit involves noise reduction by comparing the data with supplementary data that is pre-stored in the data processing unit or transmitted from an external source.

[0008] However, Patent Documents 1 and 2 focus on identifying cracks that have already "occurred" at a minute stage, Patent Document 3 focuses on estimating the remaining "lifespan," and Patent Document 4 involves attaching an AE sensor to a "predicted crack occurrence location" or a "location where a crack has already occurred" to determine the extent of damage. None of these methods detect cracks or abnormalities in advance.

[0009] For example, forging offers high productivity and allows for mass production, but if a problem occurs with the mold, a large number of defective products will be produced immediately. Moreover, since it is not possible to directly inspect the mold during production, the location and nature of the mold malfunction can only be clearly determined from the large number of defective products produced.

[0010] In other words, if Patent Documents 1 to 4 were to be adopted in a forging process, stopping operations after a mold malfunction occurs would be too late, the location of the malfunction is unpredictable, and even if the lifespan could be estimated, there are malfunctions that affect the product but not the mold lifespan due to high operating rates from that point onward, and ultimately none of them "predict malfunctions that will occur in the near future." [Prior art documents] [Patent Documents]

[0011] [Patent Document 1] Japanese Patent Application Publication No. 52-150090 [Patent Document 2] Japanese Patent Application Publication No. 4-310857 [Patent Document 3] Japanese Patent Application Publication No. 8-159151 [Patent Document 4] Japanese Patent Publication No. 2004-170397 [Overview of the Initiative] [Problems that the invention aims to solve]

[0012] The problem that this invention aims to solve is the inability to accurately predict mold abnormalities that will occur in the near future. [Means for solving the problem]

[0013] To solve the above problems, the structural abnormality prediction method of the present invention , Zhou Processing machines with a limited lifespan or device in Parts that move periodicallyA method for predicting structural abnormalities in advance, which converts an AE detection signal from an AE sensor provided in a part of a structure into AE image information as visual information according to the strength of the AE detection signal every cycle, arranges the AE image information in an image map area with one axis being the number of cycles and the other axis being a fixed time for one cycle, obtains cumulative mapping image information according to the number of cycles, and performs anomaly prediction by image determination of the mapping image information and either or both of the mapping image information stored in advance in a normal state and the mapping image information registered as an anomaly precursor.

Effect of the Invention

[0014] The present invention is applicable to periodic processing Machine or device Regarding the structural abnormality of a periodically movable part in a process, it is possible to predict that the abnormality will occur (become apparent) in the near future. Therefore, for example, in forging, measures can be taken before a die abnormality occurs, so that a large number of defective products can be prevented from occurring.

Brief Description of the Drawings

[0015] [Figure 1] In the structural abnormality prediction device of the present invention, (a) shows the configuration of a cold forging device as a processing machine, and (b) shows the overall configuration. [Figure 2] It is a diagram for explaining the effectiveness of the AE sensor. [Figure 3] It is a diagram showing the behavior of AE waves in one cycle for each cumulative shot number. [Figure 4] (a) to (d) are diagrams for explaining color graph generation. [Figure 5] (a) to (d) are diagrams for explaining anomaly prediction determination using an image pattern.

Embodiment for Carrying Out the Invention

[0016] The present invention is for accurately predicting die abnormalities that will occur in the near future in advance , ZhouPeriodic processing machine or device In Parts that move periodically A method for predicting structural abnormalities in advance, which converts AE detection signals from AE sensors provided in a part of the structure into AE image information as visual information every cycle according to the strength of the AE detection signals, and arranges the AE image information in an image map area with the vertical and horizontal axes where one axis is the number of cycles and the other axis is a fixed time for one cycle, obtains cumulative mapping image information according to the number of cycles, and performs anomaly prediction by image determination of the mapping image information and any one or both of the mapping image information of the normal state stored in advance and the mapping image information registered as an anomaly precursor.

[0017] In this example, an example of implementing the structural anomaly prediction device of the present invention adopting the method of the present invention in a cold forging device (bevel gear) as a periodic processing machine is shown. To implement the above-mentioned present invention, the structural anomaly prediction device 1 of the present invention, as shown in Fig. 1(b), includes an AE sensor S1 provided in a part of the cold forging device shown in Fig. 1(a), an image data processing unit 1A, a storage unit 1B, and an image determination unit 1C.

[0018] The image data processing unit 1A converts the AE detection signal from the AE sensor S1 into AE image information as visual information every cycle according to the strength of the AE detection signal. Further, the image data processing unit 1A arranges the above AE image information in an image map area with the vertical and horizontal axes where one axis is the number of cycles and the other axis is a fixed time for one cycle, and obtains cumulative mapping image information according to the number of cycles.

[0019] The storage unit 1B stores the mapping image information accumulated at any time. In this example, for example, an image pattern based on the mapping image information of the normal state (referred to as "normal image pattern") and an image pattern based on the mapping image information registered as an anomaly precursor (referred to as "abnormal image pattern") are stored. In this example, the storage unit 1B stores all the mapping image information accumulated at any time, but does not store it as an image pattern in a certain cycle unit.

[0020] The image determination unit 1C performs image determination between the image pattern that appears after accumulating the mapping image information for a certain number of cycles and the image data registered in the storage unit 1B as a normal image pattern or an indicator of an abnormality. Furthermore, when the image determination unit 1C predicts an abnormality, that is, when it determines that an abnormality will occur in the near future, it may activate the notification means described later or stop the operation of the cold forging machine. In this example, the machine is set to stop operation after providing notification.

[0021] Furthermore, the structural anomaly prediction device 1 includes notification means that generate a warning sound or the like to indicate that an anomaly has been predicted, and display means for displaying warnings and various other information. It also includes hardware components found in a typical personal computer, such as input means for inputting commands, storage means for storing the present invention method as software, RAM, CPU, etc. Note that the various sensors, including the AE sensor S1, and the image data processing unit 1A, storage unit 1B, and image determination unit 1C in the structural anomaly prediction device 1 shown in Figure 1(b) may be configured to be remote from each other, provided that signal transmission and reception are possible.

[0022] A cold forging machine, as a periodic processing machine, has the following basic components: slide 2, bolster 3, upper die 4, lower die 5, die height 6, inner punch 7, punch 8, die 9 (mold), die cushion 10, cushion plate 11, and knockout 12.

[0023] Furthermore, in this example, the structural abnormality prediction device 1 of the present invention has the following sensors. S1 is an AE sensor. The AE sensor S1 is installed on the cushion plate 11. S2 is a load sensor. The load sensor S2 is placed in the pressure-receiving section directly below the die 9 in order to measure the load applied to key components such as the die 9, knockout 12, and inner punch 7 during molding. S3 is a displacement sensor. Displacement sensor S3 measures the press stroke and the gap in the mold in order to clarify the positional relationship of the mold.

[0024] Furthermore, the structural anomaly prediction device 1 of the present invention may additionally employ sensors such as temperature, sound, vibration (acceleration), and various strain sensors in addition to the above-mentioned sensors S1 to S3, and may also additionally employ an image inspection method using a camera.

[0025] Furthermore, the structural abnormality prediction device 1 of the present invention can also perform evaluations in association with equipment and peripheral devices by collecting data such as signals for controlling equipment and peripheral devices such as press machines, actuator current, hydraulic and pneumatic pressure, and proximity sensor data.

[0026] In this example, the material is formed as follows. Before forming, the cushion plate 11 to which the die 9 is attached is positioned above the lower mold 5 by the die cushion 10. At the start of forming, the punch 8 and die 9 come into contact as the slide 2 descends. Subsequently, the die 9 is pushed by the punch 8, causing the cushion plate 11 to descend, and the material is pushed out into the forming section P by the fixed knockout 12, and forming is completed at the bottom dead center.

[0027] In forging, the plastic deformation of the material causes complex changes in the contact state with the die, and consequently, the stress state also changes. Furthermore, in actual production environments, there are many uncertain factors such as contact and sliding between other dies, surrounding equipment, and environmental disturbances. Until now, it has been extremely difficult to identify the relationship between these uncertain factors and die abnormalities, so the only way to reduce the number of defective products was to detect abnormalities as quickly as possible. In other words, it was not possible to "prevent the occurrence of abnormalities."

[0028] In this invention, the AE sensor is used as the primary sensor for "predicting the occurrence of anomalies." In this example, other sensors are also introduced to separate data caused by molding from data caused by the aforementioned uncertain factors (details omitted).

[0029] Here, we will explain why the AE sensor is used as the primary sensor in this invention. The AE sensor detects changes in elastic energy within a solid as waves associated with deformation and fracture, and the detection data from the AE sensor provides clues to understanding the dynamic structural changes within the material that will become the product in the mold.

[0030] What is important in this invention is not to observe large changes associated with an anomaly, nor to observe changes as the anomaly progresses, but to find "slight changes that appear before the anomaly occurs," and this is because it is ideal for capturing these slight (weak, minute) changes. The present invention enables anomaly prediction by reliably capturing these slight changes that are likely to be overlooked.

[0031] Next, the process of "predicting abnormalities" using the AE sensor as the primary sensor is described below. To confirm the effectiveness of the detection data from the AE sensor, Figure 2 compares the AE waves when (a) a mold (punch) was replaced with a new one and when (b) a mold was deemed unusable due to a crack in the mold (punch). In Figure 2, the horizontal axis represents time, and the vertical axis represents the AE wave (maximum value) on the left and the knockout load, which is the main load for molding, on the right.

[0032] In Figures 2(a) and 2(b), there was almost no change in the knockout load (right vertical axis), but AE waves were hardly present between 1.0 and 1.5 seconds when using a new mold, whereas they were significantly present when using an NG mold. This indicates that there is a significant change in AE waves before and after mold replacement, confirming the effectiveness of the AE sensor detection data.

[0033] Next, the detection signals from the AE sensor collected during press forming were analyzed. Figure 3 shows the relationship between stroke and AE wave, with the horizontal axis representing the distance to the bottom dead center of the die height (stroke), and the vertical axis representing the maximum value of the AE wave per unit time, both plotted logarithmically to easily capture small changes. Furthermore, to distinguish between pressurized and unloaded states, the stroke on the horizontal axis is shown with negative values ​​for distances before the bottom dead center and positive values ​​for distances after the bottom dead center.

[0034] In Figure 3, the AE wave changes significantly depending on the usage of the mold, as shown in Figure 2. Therefore, the AE waves at 100 shots, 31,000 shots, and 54,000 shots of the same mold are shown. As a result, it can be seen that there are significant changes in the AE wave in specific parts depending on the usage of the mold.

[0035] During production, a large amount of data is collected. When observing the changes in waveforms from this large amount of data, it is common to use features such as peak values ​​and mean values ​​of time-series data as trend data and focus only on the characteristic parts. However, focusing only on parts of the data, such as features, makes it difficult to capture small changes in the time series.

[0036] In other words, when operators visually compare changes in behavior from a large amount of data collected from periodic processing machines, they generally overlay or arrange data displays (e.g., graphs) of the same time unit. However, when overlaying, it is difficult to determine, for example, which cycle an anomaly occurred in, and when arranging vertically, there is a limit to the number of cycles that can be viewed side by side, and while waveform anomalies can be detected within the arranged range, it is difficult to detect waveform anomalies outside of that range.

[0037] Therefore, in this invention, in order to get an overview of the large amount of time-series detection data collected from the AE sensor, the data is visualized as shown in Figure 4. The visualized graph is hereinafter referred to as the color graph (mapping image information). The color graph is generated in the image data processing unit 1A as follows. Note that the image data within a certain range (a certain number of shots range) in the color graph is called the "image pattern".

[0038] The processing in the image data processing unit 1A will be explained with reference to Figure 4. The image data processing unit 1A obtains an AE detection signal, which is like a time-series graph in Figure 4(a) with time on the horizontal axis and the effective value on the vertical axis, for one shot of AE wave (AE detection signal). Next, in this example, the image data processing unit 1A transforms the graph to one with a logarithmic scale on the vertical axis, as shown in Figure 4(b), in order to make it easier to capture changes in the AE wave.

[0039] The image data processing unit 1A converts the AE wave (AE detection signal) for one shot, as shown in Figure 4(b), into a single color bar (AE image information). The color bar, as shown in Figure 4(b), represents the vertical axis, i.e., the height of the RMS value of the AE wave, using varying shades of color. Although shown in black and white in the illustration, it is actually in color.

[0040] Figure 4(c) shows a display with five color bars arranged vertically (on the axis) (five shots). In this way, the horizontal axis is fixed to the displacement of only one shot, and the vertical axis represents the number of shots and the strength of the AE sensor detection data, expressed in color (shade, etc.). In other words, in a molding operation where one shot is repeated, the AE detection signal is converted into a color bar, and this color bar is cumulatively placed in the map information and image data processing is performed on the mapping image information to obtain a color graph.

[0041] The image determination unit 1C determines the anomaly prediction as follows. In this example, the memory unit 1B stores both an image pattern based on mapping image information of a normal state as shown in Figure 5(a) (hereinafter referred to as the normal image pattern) and an image pattern based on mapping image information of an anomaly precursor as shown in Figure 5(b) (hereinafter referred to as the anomaly image pattern). Note that multiple anomaly image patterns are stored, corresponding to the number of types of anomalies that can be identified.

[0042] The image determination unit 1C receives a color graph in which a color bar is accumulated for each shot as it is being processed. Here, it compares the image pattern generated by accumulating the data from the currently input shot (hereinafter referred to as the real image pattern) with the normal image pattern and the abnormal image pattern. The image determination unit 1C uses not only the position of the AE image information in the map area but also the color (intensity) as information for determination.

[0043] An example of a judgment made by the image determination unit 1C will be explained. For example, the normal image pattern in Figure 5(a) means that within a set range of a certain number of shots, the color bar indicating the expected abnormality does not appear. Therefore, when the image determination unit 1C detects a real image pattern like those shown in Figures 5(c) and 5(d) within a set range of a certain number of shots, it determines that there is a mismatch with the image pattern other than the normal image pattern, i.e., the image pattern in Figure 5(a), and predicts the occurrence of an abnormality. In this example, it issues a warning signal (sound and display) and stops operation.

[0044] On the other hand, the abnormal image patterns registered in advance as precursors to abnormalities in Figure 5(b) refer to the appearance of color bars indicating any expected abnormalities, obtained empirically or through testing. Therefore, the image determination unit 1C determines the match or mismatch of image patterns in a brute-force manner for each shot. If a real image pattern like the one shown in Figure 5(c) appears within a set range of a certain number of shots, it does not yet match the abnormal image pattern in Figure 5(b), so the prediction of an abnormality is postponed. If a real image pattern like the one shown in Figure 5(d) appears, it determines that there is a mismatch with the abnormal image pattern in Figure 5(b), and in this example, it predicts an abnormality and issues a warning signal (sound and display) to stop operation.

[0045] In this example, we demonstrated a method that uses both a normal image pattern and an abnormal image pattern. Even if an abnormality is detected in the normal image pattern, if there is a margin of error in the abnormal image pattern, the abnormality detection is not performed until the color bar accumulates and the real image pattern completely matches the abnormal image pattern. This method determines an abnormality only when both the normal and abnormal image patterns are detected as abnormal. However, a detection method using only one of the two patterns may also be used depending on the processing machine or equipment.

[0046] Thus, the present invention makes it easier for humans to visually detect subtle signs before an anomaly occurs by fixing the period (cycle) time on one axis and accumulating color bar displays for each cycle on the other axis, thereby predicting periodic structural abnormalities in machinery and other devices that are difficult to find with normal graph displays on a continuous time axis. Furthermore, in terms of software processing, by accumulating color bar displays (mapping image information) within a certain range and determining image patterns, it is possible to easily capture intermittent, sudden, temporal, and periodic changes before an anomaly occurs, and thus reliably predict anomalies.

[0047] In short, the present invention is characterized by its ability to instantly detect even slight changes over a long period of time, rather than relying on the numerical values ​​and waveforms of each detection data from the AE sensor as they appear in the conventional method. This is because it determines anomaly prediction based on the image pattern that appears in a color graph where the color bars for each shot are accumulated.

[0048] For example, if the horizontal axis is a continuous and not fixed time axis, from a human visual perspective, previous data becomes out of view as time progresses, making comparison difficult. Also, from a software processing perspective, it is not possible to detect changes by comparing with all the data recorded up to that point. Instead, thresholds or waveforms are set in advance, and when these set values ​​are detected, it is judged as an anomaly, that is, an anomaly has occurred.

[0049] In this invention, from a human visual perspective, in this example, the display moves vertically (from the bottom to the top of the display surface) according to the number of shots rather than the passage of time, but the horizontal axis is fixed to the time axis for each shot, so it is possible to easily compare the changes in the detection data of the AE sensor for each shot cycle while constantly comparing them.

[0050] Furthermore, instead of detecting pre-set thresholds or waveforms as abnormalities, the software processing can detect them based on the matching or mismatch of image patterns that appear cumulatively on a shot-by-shot basis. This allows for the reliable detection of early signs of anomalies, making it possible to prevent anomalies from occurring before they happen. [Explanation of Symbols]

[0051] 1. Structural Anomaly Prediction Device 1A Image Data Processing Unit 1B Storage section 1C Image judgment section S1 AE sensor 9. Die (mold)

Claims

1. A method for predicting structural abnormalities in a periodically moving part of a periodic processing machine or apparatus, comprising: converting an AE detection signal from an AE sensor provided on a part of the structure into AE image information as visual information according to the strength of the AE detection signal for each cycle; arranging the AE image information in an image map region having vertical and horizontal axes with one axis representing the number of cycles and the other axis representing a fixed time for one cycle; obtaining cumulative mapping image information according to the number of cycles; and performing structural abnormality prediction by image judgment of the mapping image information and either or both of the image patterns of pre-stored normal state mapping image information and pre-stored abnormality precursor mapping image information.

2. A device for predicting structural abnormalities in a periodically moving part of a periodic processing machine or apparatus, comprising: an AE sensor provided on a part of the structure; an image data processing unit that converts an AE detection signal from the AE sensor into AE image information as visual information according to the strength of the AE detection signal for each cycle, arranges the AE image information in an image map region having vertical and horizontal axes with one axis representing the number of cycles and the other axis representing a fixed time for one cycle, and obtains cumulative mapping image information according to the number of cycles; a storage unit that stores the mapping image information and either or both of the image patterns from the normal state mapping image information and the image patterns from the mapping image information indicating an abnormality; and an image determination unit that performs image determination on either or both of the mapping image information and the image patterns from the normal state mapping information and the image patterns from the mapping information indicating an abnormality in the storage unit.

Images