JP6823576B2 - Anomaly detection system and anomaly detection method - Google Patents

Description

The present invention relates to a technique for diagnosing an operating condition, which is automatically implemented for a mechanical device or the like. In particular, it relates to an abnormality detection system and an abnormality detection method for devices and systems.

There is a technology that automatically diagnoses the operating status of various mechanical devices and detects abnormal conditions such as failures and signs of failures.

For example, Patent Document 1 collects diagnostic operation control means that performs diagnostic operation in a plurality of modes after the occurrence of an earthquake is detected, and sounds in the hoistway when the car is running in each mode of the diagnostic operation. Sound collecting means for sound collection, abnormal sound reference value storage means for storing a plurality of abnormal sound reference values set for each mode of diagnostic operation, and abnormal sound reference for each mode stored in the abnormal sound reference value storage means. Based on the value, the abnormal sound determining means for determining whether or not the sound collected by the sound collecting means is an abnormal sound, and the abnormal sound is generated in the hoistway when the car is driven multiple times in the same mode of the diagnostic operation. By providing an abnormal sound information determining means for determining whether or not the sound is detected at the same position, the abnormal sound suitable for each mode is detected and the accidental false detection of the sound is prevented. Is described.

JP-A-2007-223750

It is required to monitor the operating status of various machines, devices, systems, etc. (hereinafter referred to as "diagnosis targets") at low cost and with high accuracy. Therefore, it has been studied to automatically monitor the status of the diagnosis target by attaching various sensors to the diagnosis target and analyzing the data obtained from the sensors.

At this time, it is generally considered that the diagnosis target operates in various operation modes, but if the analysis is performed without distinguishing the operation modes, data in significantly different states are compared, and the analysis accuracy deteriorates.

For this reason, it is conceivable to perform diagnostic operation in each operation mode and perform analysis separately, but it is necessary to stop normal operation and perform diagnostic operation, which shortens the actual operating time. .. In addition, when diagnosing only the driving condition during diagnostic driving, it is not possible to respond to sudden abnormalities because monitoring during normal driving is not possible.

Therefore, it is desired to enable analysis of sensor data in consideration of the operation mode during normal operation and to perform highly accurate diagnosis.

One aspect of the present invention is an abnormality detection system that detects an abnormality of a diagnosis target based on a detection signal from a detection element provided in the diagnosis target, and is a diagnosis target based on a detection signal from the detection element during normal operation. In the combination of the abnormality determination unit that determines the presence or absence of an abnormality in the first operation mode and the second operation mode, and the presence or absence of an abnormality in the first operation mode and the second operation mode determined by the abnormality determination unit. Based on this, it is provided with an abnormality cause detection unit that determines and detects the cause of the abnormality to be diagnosed.

This is an abnormality detection method that detects an abnormality of a diagnosis target using an information processing device equipped with a storage device, an input device, a processing device, and an output device. The input device is a sensor from a detection element that detects the state of the diagnosis target. The data is acquired and the processing device performs a classification process for classifying the sensor data for each operation mode to be diagnosed and an abnormality detection process for detecting an abnormality for each classified operation mode.

It is possible to analyze sensor data in consideration of the operation mode during normal operation, and to perform highly accurate diagnosis.

A conceptual diagram showing a situation in which a sensor is mounted on a pneumatic machine. The graph which shows the example of the detection signal of a vibration sensor. The block diagram which shows the whole structure of the diagnostic system of the operation situation of an Example. A flow chart showing the flow of diagnostic processing performed by the edge calculation unit. The block diagram which shows the functional block in an abnormality determination part. The figure which shows the example of the abnormality cause estimation table. A flow diagram showing the operation of the learning phase. The conceptual diagram which shows the concept of the process which sets the threshold value for identifying a mode. A flow chart showing the operation of the operation phase. The conceptual diagram which shows the example of the operation schedule of the diagnostic system of an Example. FIG. 6 is a conceptual diagram showing a situation in which a sensor is mounted on a pneumatic machine in the second embodiment. The conceptual diagram which shows the concept of the abnormality determination data of Example 2. The figure which shows the abnormality cause estimation table example of Example 2. FIG. FIG. 6 is a conceptual diagram showing a situation in which a sensor is mounted on a pneumatic machine in the third embodiment. The figure which shows the abnormality cause estimation table example of Example 3. FIG. FIG. 6 is a conceptual diagram showing a situation in which a sensor is mounted on a cutting machine in the fourth embodiment. FIG. 6 is a conceptual diagram showing a situation in which a sensor is mounted on a semiconductor manufacturing apparatus in the fifth embodiment. A layout diagram showing an example of a factory monitoring status monitor displayed on a terminal. A layout diagram showing an example of a maintenance monitor displayed on a terminal.

The embodiment will be described in detail with reference to the drawings. However, the present invention is not construed as being limited to the description of the embodiments shown below. It is easily understood by those skilled in the art that a specific configuration thereof can be changed without departing from the idea or purpose of the present invention.

In the configuration of the invention described below, the same reference numerals may be used in common among different drawings for the same parts or parts having similar functions, and duplicate description may be omitted.

When there are a plurality of elements having the same or similar functions, they may be described by adding different subscripts to the same code. However, if it is not necessary to distinguish between a plurality of elements, the subscript may be omitted for explanation.

The position, size, shape, range, etc. of each configuration shown in the drawings and the like may not represent the actual position, size, shape, range, etc. in order to facilitate understanding of the invention. Therefore, the present invention is not necessarily limited to the position, size, shape, range, etc. disclosed in the drawings and the like.

In the first embodiment, a rotating machine constituting an air compressor (hereinafter referred to as “pneumatic compressor”) as a diagnosis target is taken as an example, and a system in which a vibration sensor is attached to the rotating machine to remotely monitor an abnormality will be described as an example. A specific example of a rotating machine is a motor, and a specific example of a vibration sensor is a MEMS (MEMS Micro Electro Mechanical System) sensor. However, as will be described in a later embodiment, various types of diagnostic objects and sensors can be applied.

Explaining with the example of the pneumatic machine, the pneumatic machine targeted in this example has three operation modes of stop, unload, and load. Stop is a state in which the rotating machine is stopped, unloading is a state in which the rotating machine is rotating but no load is applied, and load is a state in which the rotating machine is rotating and is loaded. .. The vibration states are significantly different in the three operation modes.

<1. Diagnosis target (pneumatic machine)>
FIG. 1 is a conceptual diagram showing a situation in which a sensor is mounted on a pneumatic machine. The pneumatic machine 100 includes a rotating machine 102, and the rotating machine 102 compresses the air in the air tank 106 via the air compression valve 104. The vibration sensor 108 is retrofitted to the pneumatic machine 100. As will be described in detail later, the mounting location is arbitrary and the number is arbitrary. As described above, the pneumatic machine can be operated in three operation modes depending on whether the rotating machine 102 is rotating and whether the rotating machine 102 is loaded with air compression.

<2. Sensor data>
FIG. 2 is a graph showing an example of sensor data of the vibration sensor 108 in each of the three operation modes. The pneumatic machine operates in three operation modes of stop, unload, and load, and a remarkable difference is observed in the acquired sensor data in the three operation modes. In FIG. 2, the horizontal axis shows time and the vertical axis shows acceleration. As can be judged from the graph of FIG. 2, the magnitude of acceleration differs by one digit or more between stop and load / unload. It can also be seen that the waveform patterns are also different. Therefore, if an abnormality judgment is made using the same judgment criteria without distinguishing between these data, it is difficult to make a highly accurate diagnosis.

<3. Diagnostic system configuration>
FIG. 3 is a block diagram showing a configuration of a diagnostic system for an operating condition of an embodiment. A vibration sensor 108 is attached to the pneumatic machine 100, and the sensor data is transmitted to the edge calculation unit 300 placed in the vicinity of the pneumatic machine 100.

The edge arithmetic unit 300 is a general information processing device including a storage device 301, an input device 302, a processing device (processor or CPU (Central Processing Unit) 303), an output device 304, and a bus (not shown) connecting them. For example, it can be configured by a microcomputer or a server. In this embodiment, functions such as calculation and control are realized by executing a program stored in the storage device 301 by the processing device 303 in cooperation with other hardware. A program executed by a computer or the like, its function, or a means for realizing the function may be referred to as a "function", a "means", a "part", a "unit", a "module", or the like. The processing device 303 includes an abnormality determination unit 305 and an abnormality cause detection unit 306 as functions, and transmits an abnormality determination result to the central calculation unit 320 via the network 310.

The central arithmetic unit 320 can also be composed of a storage device 321, an input device 322, a processing device 323, an output device 324, and a general microcomputer or server including a bus (not shown) connecting them. The central calculation unit 320 includes an operation screen display unit 325, and displays the operation screen on the terminal 330 via the output device 324.

The communication between the vibration sensor 108 and the edge calculation unit 300 may be either wireless communication or wired communication. In this embodiment, wired communication using an I2C (Inter-Integrated Circuit) serial bus is assumed. However, wireless communication such as WiFi, Bluetooth (registered trademark), ZigBee (registered trademark) may be used. The interface required for communication is such that the vibration sensor 108 and the input device 302 each have a known configuration.

In this embodiment, the abnormality cause detection unit 306 is included in the edge calculation unit 300, but may be included in the central calculation unit 320. Further, both the abnormality determination unit 305 and the abnormality cause detection unit 306 may be included in the central calculation unit 320. In that case, the edge calculation unit 300 transmits the sensor data to the central calculation unit 320 instead of the abnormality determination result. Further, the above configuration may be configured by a single computer attached to the diagnosis target, or another computer in which any part of the input device, output device, processing device, and storage device is connected by a network. It may be composed of.

In this embodiment, one vibration sensor 108 is assumed, but there may be a plurality of sensors or other types of sensors such as a temperature sensor. The plurality of sensors may transmit the sensor data to one edge calculation unit 300, or may transmit the sensor data to another edge calculation unit (not shown). That is, although one edge calculation unit 300 is assumed in this embodiment, there may be a plurality of edge calculation units.

In this embodiment, one terminal 330 is assumed, but there may be a plurality of terminals. Further, the communication between the central calculation unit 320 and the terminal 330 may be either wireless communication or wired communication. In this embodiment, the operation screen display unit 325 assumes an HTML server, and codes written in languages such as JavaScript (registered trademark), CSS (Cascading Style Sheets), PHP, Ruby, and Java (registered trademark) are used. Output. The terminal 330 is supposed to be a tablet device which is a small computer, and is supposed to display an operation screen on a web browser.

In the present embodiment, in the case of a storage device, an input device, a processing device, an output device, etc., a known general device can be appropriately applied. For example, a magnetic disk device or various semiconductor memories can be applied as the storage device. As an input device, a keyboard, a mouse, or various input interfaces can be applied. A display, a printer, or various output interfaces can be applied as the output device.

In the above description, functions equivalent to the functions configured by software can also be realized by hardware such as FPGA (Field Programmable Gate Array) and ASIC (Application Specific Integrated Circuit). Such aspects are also included in the scope of the present invention.

<4. Diagnostic process flow>
FIG. 4 is a flow chart showing the flow of diagnostic processing performed by the edge calculation unit 300.
In the process S401, the vibration is measured by the vibration sensor 108 attached to the rotating machine 102 constituting the pneumatic machine 100, and the sensor data is acquired via the input device 302. The acquired sensor data is stored in the storage device 301 as needed for later processing.

In the process S402, the abnormality determination unit 305 determines the operation mode of the sensor data. The compressed air generated by the pneumatic machine 100 is stored in the air tank 106. Compressed air is sent into equipment such as factories and used in other mechanical devices. The pneumatic machine 100 controls the discharge pressure so that the compressed air stored in the air tank 106 maintains a constant pressure level. For pneumatic machines, this operation time is the normal operation time. In general, normal operation refers to an operating state in which a device or equipment operates to achieve its original purpose, and is distinguished from other operating states that operate only for diagnosis, inspection, repair, or maintenance, for example. Will be done.

During normal operation, the pneumatic machine 100 shown in FIG. 1 has three operation modes. First, there is a stop mode in which the rotating machine 102 is stopped. Next, there is an unload mode in an idling state in which the rotating machine 102 is rotating but the air compression valve 104 is closed. Finally, there is a load mode in which the air compression valve 104 opens to feed compressed air into the air tank 106. These three operation modes are distinguished and collected as a sensor data database (DB) 3011 for each operation mode. At this time, it is preferable to use an independent database for each operation mode. A specific example of the algorithm for determining the operation mode in the process S402 will be described later. In the above explanation, a pneumatic machine using a load / unload method is assumed as the discharge pressure control method, but other methods such as an on / off method, a slide valve method, and a suction throttle method are also used. It is included in the scope of the present invention.

In the process S403, the abnormality determination unit 305 determines an abnormality for each operation mode. As a method for determining an abnormality based on sensor data, various known methods can be adopted. For example, it is envisioned to use deep learning, reinforcement learning, or Bayesian networks. The method of determining the abnormality is basically not limited, but an algorithm different from the algorithm for determining the operation mode of the process S402 may be selected. As the feature amount to be analyzed for the abnormality determination, a feature amount different from the operation mode determination may be used. In general, it is desirable that the algorithm for determining the operation mode has a small amount of processing, and the algorithm for determining an abnormality emphasizes accuracy rather than the amount of processing. It is also possible to use different algorithms or features for each operation mode.

FIG. 5 is a block diagram showing a functional block in the abnormality determination unit 305. The abnormality determination unit 305 includes a learning unit 3051 that learns a threshold value for determining an operation mode, a mode determination unit 3052 that determines a mode of sensor data using the learned threshold value, and an abnormality detection unit 3053. As for the algorithms and processing programs used in each of these parts, those stored in the storage device 301 shall be implemented in the processing device 303.

The mode determination unit 3052 performs the operation mode determination S402, and stores the sensor data divided for each mode as the sensor data databases 3011-1, 3011-2, and 3011-3 for each mode. The abnormality detection unit 3053 uses the sensor data divided for each mode and detects an abnormality for each mode data.

Returning to FIG. 4, in the process S404, the abnormality cause detection unit 306 estimates the abnormality cause with reference to the abnormality cause estimation table 3012. As a result of performing the abnormality determination processing for each of the three modes in the processing S403, the combination of each sensor data and the normal and abnormal becomes eight patterns. Using this combination, the cause of abnormality of the air compressor is estimated.

FIG. 6 is a table diagram showing an example of the abnormality cause estimation table 3012. FIG. 6A stores the probable cause of abnormality for each of the eight patterns. The abnormality cause of the abnormality cause estimation table 3012 shall be created and stored by the administrator based on experience, experiment, or simulation prior to the processing of FIG. In addition, the cause of the abnormality can be added or corrected by the administrator while the system is in operation.

For example, in pattern 2, since the abnormality occurs even though the pneumatic machine 100 is stopped, it can be estimated that the abnormality is caused by an external factor other than the pneumatic machine 100. In pattern 3, since the abnormality occurs without the influence of the air tank 106 which is a load when viewed from the rotating machine 102, it can be estimated that the abnormality is caused only by the rotating machine 102. For example, the deterioration of the bearing is deteriorated. Conceivable. In pattern 4, it can be assumed that the abnormality is caused by a load. For example, it is considered that the amount of compressed air used in the factory temporarily increases, resulting in an overload. In this way, by distinguishing the operation patterns and determining the abnormality, the state of the device can be monitored in detail.

In the process S405, the estimated cause of the abnormality is transmitted from the output device 304 to the central arithmetic unit 320 via the network 310. The central calculation unit 320 causes the terminal 330 to display the cause of the abnormality.

Conventionally, the sensor data is stored as one data, but as shown in FIG. 5, in this embodiment, the data is stored for each operation mode, and an abnormality is determined for each of them. Conventionally, it has been difficult to identify the cause of an abnormality because only two patterns, normal and abnormal, can be obtained as a combination of data at the time of abnormality. Since eight patterns can be obtained in this embodiment, it is possible to estimate the cause of the abnormality with high accuracy from the combination of the abnormalities.

Moreover, more information can be obtained by following the transition of patterns in time series as well as combinations. For example, in the case of transition from pattern 4 to pattern 6, it is possible to interpret such as "the motor is damaged as a result of continuously applying an overload". For this purpose, as shown in FIG. 6B, the abnormality cause corresponding to the pattern transition is registered in advance as a part of the abnormality cause estimation table 3012. This shall be created and stored by the administrator based on experience, experiment, or simulation prior to the processing of FIG. On the other hand, the pattern transition may be left as historical data in the storage device 301, and the data may be analyzed by collating with the abnormality cause estimation table 3012 at regular or arbitrary time points.

<5. Judgment of operation mode>
A method of determining the operation mode of the abnormality determination unit 305 in the process S402 will be described. A monitoring function may be installed in advance for a diagnosis target such as a pneumatic machine, but a sensor may be retrofitted. Therefore, a configuration that can identify the operation mode even with a retrofit sensor is desired.

The first configuration example is a configuration in which control information of the device is acquired from the device to be diagnosed. For example, the operation mode is determined by acquiring the control signal by using the interface for outputting the control signal provided in advance in the device to be diagnosed. This method is an ideal method that is accurate, does not require arithmetic processing, and has excellent real-time performance, if an interface is available. However, the legacy device may not output the control signal, and the feasibility depends on the device to be diagnosed.

Further, for a device whose control information is displayed on the control panel, it is conceivable to separately install a camera to acquire an image of the device to be diagnosed and recognize the image to determine the operation mode. If the control information is displayed on the control panel, it can be applied to legacy devices, but it requires the addition of hardware and software, which poses a cost issue.

The second configuration example is a configuration in which the sensor is made redundant. For example, in a pneumatic machine, a current sensor is arranged in an electric system, a barometric pressure sensor is arranged in an air system, and an operation mode is determined by a combination of sensor data from these. In order to determine the operation mode, it is necessary to perform arithmetic processing of the sensor data obtained from each sensor, and it is necessary to set a threshold value for that purpose. Therefore, it becomes a cost factor, but accurate operation mode determination can be expected if appropriate settings are made.

The third configuration example is a configuration in which the threshold value is determined using the effective value and the average value of the sensor data. When a sensor is retrofitted, no other sensor is required and additional hardware can be avoided, which is the most cost effective as compared with other configuration examples. However, it is necessary to set the threshold value by arithmetic processing and learning.

<5. Detailed explanation of operation mode judgment>
Hereinafter, a specific configuration example of the third configuration example will be described in detail. The configuration of the abnormality determination unit 305 shown in FIG. 5 corresponds to the third configuration example. In this example, the operation mode is determined using the data of the single sensor attached later. The operation mode is determined by the abnormality determination unit 305 in two phases, an operation phase and a learning phase. In the operation phase executed by the mode determination unit 3052, the mode is determined by the threshold determination with a small calculation cost. Therefore, first, the optimum threshold value is determined in the learning phase executed by the learning unit 3051.

FIG. 7 is a flow chart showing the operation of the learning phase. This function is implemented as a learning unit 3051. In the learning phase, a clustering judgment and a threshold value judgment are performed on a certain data, the clustering judgment is the correct answer, the results of both are compared, and the threshold value is set.

In the process S701, the sensor data acquired over a predetermined period is used as the data to be processed for learning. As the data to be processed, for example, a time-series sensor data for 1 minute is prepared, and this data is divided into 1 second × 60 pieces of data.

In the process S702, the expected number of modes is acquired. The number of modes may be a specified value if it is known in advance, but may be input by the user from the input device 302. In the case of a load / unload type pneumatic machine, the number of modes 3 is input.

In general clustering analysis, in order to improve the analysis accuracy, it is necessary to perform calculations with various cluster numbers for the same data, check the accuracy of cluster division, and adjust the optimum number of clusters. In this embodiment, an interface that allows the user to input "the number of clusters (= the number of operation modes)" as input data is introduced (described later). By making it possible to use OT (operational technology) of a user who understands the mechanical device to be diagnosed, the optimum number of modes can be selected and highly accurate diagnosis can be made possible.

In the process S703, clustering is executed on the data to be processed, and the mode is determined. At this time, processing such as standardization, principal component analysis, and dimension reduction is performed on the data to be processed as necessary. Since these are known as data analysis methods, details are omitted. As for the clustering method, various known methods can be applied. For data acquired by a vibration sensor, for example, features can be extracted from time waveforms and spectra, standardized and principal component analysis performed, and then clusters can be determined using the k-means method. As a premise of the clustering determination at this time, the number of modes obtained in the process S701 is used. In a simple example, an effective value and a specific frequency component can be assumed as corresponding to the main components.

As a result of the clustering determination, information on "which cluster the 60 points (data) belong to" can be obtained. In addition, information on "center points of three clusters" and "radius of three clusters" can be estimated. In this example, clustering by a known k-means method is assumed. In clustering by the k-means method, it is generally necessary to first determine the optimum number of clusters in order to improve the determination accuracy. That is, the process of inputting various numbers of clusters, recalculating, evaluating the judgment accuracy from the judgment results, and selecting the optimum number of clusters is performed. Since this process requires human judgment, it is difficult to automate and the cost to realize it is high. On the other hand, in this embodiment, as described above, the number of operation modes can be input in advance as the optimum number of clusters, so that the process of determining the optimum number of clusters becomes unnecessary. Therefore, it is possible to make a highly accurate determination while suppressing the cost for realization.

Next, a threshold value for identifying each mode is set. Since the threshold value is set for each mode, the process is performed n times for the number of modes (S704).

In the process S705, the initial value An of the threshold value is set for the mode. The value may be arbitrary, but for example, set a value that will include some correct answers, and gradually expand or contract it.

In the process S706, the input data is discriminated by using the threshold value set in the process S705.

In the process S707, for each data, a match / mismatch between the mode determined by clustering in the process S703 and the mode determined by the threshold value in the process S706 is determined, and the correct answer rate is calculated.

FIG. 8 shows the concept of the process of setting the threshold value for identifying the mode. For example, 60 data are plotted on a plane defined by two principal components. Each data displayed as a square, a circle, and a star is a mode of data determined as a result of the clustering determination. In the process S706, the input data is discriminated using the threshold value.

As an example, consider the case where the threshold value A2 is applied to the cluster 2. In the two-dimensional space of FIG. 8, the initial value A2 of the threshold value is set (process S705), the threshold value is determined (process S706), the correct answer rate calculation (process S707), and the correct answer rate evaluation (process S708) are performed, and the threshold value is gradually set. (Process S709).

FIG. 8A assumes a case where the threshold value A2 is an initial value. The initial value of the threshold value A2 is defined by the radius with the center coordinates as the geometric center of gravity of the data of the cluster 2. In FIG. 8A, since all the data in the threshold value A2 indicated by the dotted circle is cluster 2, the correct answer rate by the correct answer rate calculation S707 is 100%. In this case, the correct answer rate evaluation result by the correct answer rate evaluation S708 is regarded as NG, and the threshold value A2 is updated in the process S709. In this case, the threshold value A2 is increased by a predetermined number.

FIG. 8B shows the result of increasing the threshold value A2 several times by a predetermined number, and the correct answer rate is 100% (S707). This state is the optimum value of the threshold value, but in order to finally determine the optimum value, the correct answer rate evaluation result is NG (S708), and the threshold value A2 is further increased by a predetermined number (S709).

FIG. 8C shows the result of further increasing the threshold value A2 by a predetermined number, and since the data of the cluster A3 is included, the correct answer rate is less than 100%.

In the correct answer rate evaluation process S708 of the learning phase, the learning unit 3051 sets the correct answer rate evaluation to NG when the correct answer rate is 100%, and updates the threshold value An in the increasing direction. However, when the correct answer rate is less than 100%, the threshold value An is updated in the decreasing direction. Then, the threshold value An at the time when the correct answer rate reaches 100% is set as the optimum threshold value. As a result, the threshold value A2 shown in FIG. 8B becomes the optimum threshold value.

In the above description, the threshold value is transitioned from increase to decrease, but conversely, the optimum threshold value can be set in the same manner by the method of transitioning from decrease to increase. In the above example, the center coordinates are immovable and the radius of the threshold value is adjusted, but the center coordinates can be adjusted in the same manner.

As described above, the learning unit 3051 can set the threshold values (center coordinates and radius) for the three modes. The obtained threshold value shall be updated as appropriate to maintain the optimum value.

FIG. 9 is a flow chart showing the operation of the operation phase. This function is implemented as a mode determination unit 3052. In the operation phase, the data mode is determined using the threshold value determined in the learning phase.

In the process S901, the sensor data acquired over a predetermined period is set as the process target data. As the data to be processed, for example, a time-series sensor data for 1 minute is prepared, and this data is divided into 1 second × 60 pieces of data.

In the process S902, the data mode is determined using the threshold value determined in the learning phase. Although omitted in FIG. 9, the discrimination is performed in parallel or in series for each mode.

In the process S903, the mode of the data is determined. That is, as described in FIG. 8, the data existing within the threshold value indicated by the dotted circle is classified into, for example, three modes.

In the process S904, as shown in FIG. 5, the classified data is stored in the storage device 301 for each mode. At this time, the time information is stored in association with the data.

FIG. 10 is a conceptual diagram showing an example of an operation schedule of the diagnostic system of the embodiment. As shown in the operation time division 1001, in this embodiment, the pneumatic machine 100 is basically in a normal operation state. As described above, the pneumatic machine 100 has three operation modes 1002: stop, unload, and load, which can be switched at any timing.

Phase 1003 includes a learning phase shown in FIG. 7 and an operation phase shown in FIG. The learning phase shown in FIG. 7 can be performed at any timing, but it is basically premised that the learning phase is operating normally. From this point of view, for example, when the vibration sensor 108 is installed, it is conceivable to manually activate the learning unit 3051 after confirming the normal operation separately to enter the learning phase and learn the threshold value. As shown in FIG. 10, for learning, a learning time sufficient to cover all three operation modes is required.

The threshold value once set may change due to changes over time in the pneumatic machine 100. Therefore, it is conceivable to periodically perform the learning phase, calculate the threshold value, and update the threshold value. However, in the learning phase, if it is not guaranteed that the pneumatic machine 100 is operating normally, the threshold value is updated by reflecting a slight shift of the threshold value due to a change over time, but the fluctuation amount of the threshold value exceeds a predetermined value. In that case, it is conceivable that the configuration is not updated or an error signal is generated.

As shown in the processing operation 1004, the threshold value is calculated and the threshold value is updated in the learning phase. Further, in the operation phase, as shown in FIG. 9, the mode determination unit 3052 determines the operation mode, and the abnormality detection unit 3053 detects the abnormality.

Although a single vibration sensor 108 is used in the first embodiment, the same method can be applied even if a plurality of sensors are used. That is, a single or a plurality of compressor rotating mechanisms (hereinafter, rotating machines) included in one pneumatic machine are measured by different sensors, and data is stored in each sensor. Examples of the rotating machine include a screw compressor, a scroll compressor, and a reciprocating compressor.

FIG. 11 shows an example of a pneumatic machine 100 composed of a motor 102a, a one-stage side screw 102b, and a two-stage side screw 102c. As shown in FIG. 11, a sensor 108a is installed on the motor 102a, a sensor 108b is installed on the first-stage screw 102b, and a sensor 108c is installed on the second-stage screw 102c, each of which outputs sensor data.

As shown in FIG. 12, the mode of each sensor data is determined by the same method as described in the first embodiment, and the sensor data is stored in the storage device 301 as the sensor data database 3011.

As shown in FIG. 13, in this case, the abnormality data combination table stored in the abnormality cause estimation table 3012 has 8 ³ = 512 patterns, and the abnormality cause can be estimated in more detail.

Pneumatic machines are generally operated by controlling the number of multiple units. Therefore, even if the state of unit control is treated as an operation mode, the same concept as in the first embodiment can be applied.

FIG. 14 is a diagram showing a configuration example of a system operated by controlling the number of a plurality of units. For example, the load absorber 100-3 can be controlled by controlling the number of constant speed base load machines (100-1, 100-2) and two variable speed load absorbers (100-3, 100-4). When measuring with a sensor, if the number of base load machines (100-1, 100-2) and load absorbers 100-4 that are moving at the same time is input as the number of operation modes, there are 2 ⁴ = 16 combinations of abnormal data. It becomes.

FIG. 15 shows an abnormality data combination table stored in the abnormality cause estimation table 3012 in this case. It is possible to estimate the cause of the abnormality in more detail.

In this configuration, the magnitude of the load changes depending on the number of base load machines (100-1, 100-2) and load absorbers 100-4 that are moving at the same time. For example, when only Unit 3 100-3 is operating and when Unit 1 100-1 and Unit 3 100-3 are operating at the same time, the data of the sensors provided in Unit 3 100-3 are different. Conceivable. Further, in the case of a vibration sensor, it is conceivable that the vibration generated from Unit 1 100-1 is transmitted through the piping, the ground, etc. and affects the vibration sensor of Unit 3 100-3. Therefore, by distinguishing between these situations and performing the abnormality determination, it is possible to make a highly accurate determination.

In the above embodiment, the pneumatic machine has been described as an example, but it can also be applied to the state detection of other devices. An embodiment in the case of abnormality monitoring of a cutting machine will be described. A cutting machine is a device that cuts or grinds a workpiece using a cutting tool, and includes, for example, a drilling machine, a milling machine, a machining center, and an NC (Numerically Control) lathe.

FIG. 16 shows an example of a cutting machine equipped with a drill 161 as a cutting tool. The drill 161 is held by the drill holding portion 162. The drill holding portion 162 rotates the drill 161 and the drill holding portion 162 is lowered so that the drill 161 and the work 163 come into contact with each other to perform drilling. A vibration sensor 108v is attached to the drill holding portion 162 in order to monitor the abnormality of the cutting machine. Further, a vibration sensor 108i is attached to the work 163 in order to monitor the processing state of the drilling process.

This device has four main modes of operation. First, there is a stop mode in which the drill 161 is stopped. Next, there is an idling mode in which the drill 161 rotates but the work 163 is not cut. After that, the drill 161 is lowered, and there is a drilling cutting mode for drilling a hole in the work 163. Finally, there is a line cutting mode in which the drill holding portion 162 moves while cutting the work 163 to cut the work 163 linearly. By processing the sensor data obtained from the sensor 108 with the same system configuration as in the first embodiment, these operation modes can be distinguished and the state of the cutting machine can be accurately determined.

An embodiment of monitoring an abnormality of a semiconductor manufacturing apparatus will be described. Semiconductor manufacturing equipment is generally equipped with a vacuum chamber, for example, metal vapor deposition equipment, sputtering equipment, plasma CVD (Chemical Vapor Deposition) equipment, and ICP (Inductively Coupled Plasma) etching equipment. There are also RIE (Reactive Ion Etching) equipment, ashing equipment, electron beam drawing equipment, and FIB (Focused Ion Beam) equipment. In addition, some semiconductor inspection devices are also provided with a vacuum chamber, and for example, there are SEM (Scanning Electron Microscope) and TEM (Transmission Electron Microscope). Although there are differences in the basic configuration, it can be similarly applied as a semiconductor manufacturing apparatus.

FIG. 17 shows an example of the configuration of a semiconductor manufacturing apparatus provided with a roughing pump and a main pulling pump. The rough pulling scroll pump 171 and the main pulling turbo molecular pump 172 are connected to the vacuum chamber 175 through the rough pulling valve 173 and the main pulling valve 174, respectively, as shown in the figure. In order to reduce the back pressure of the turbo molecular pump 172, the scroll pump 171 is connected to the turbo molecular pump 172 through a back pressure valve 176. In order to monitor the abnormality of this device, a vibration sensor 108v and a current sensor 108i are attached to the scroll pump 171 and a pressure sensor 108p is attached to the vacuum chamber 175.

The device has five modes of operation. First, there is a stop mode in which the scroll pump 171 and the turbo molecular pump 172 are stopped. Next, there is a roughing mode in which the scroll pump 171 operates, the roughing valve 173 opens, and the pressure in the vacuum chamber 175 is lowered from atmospheric pressure to vacuum pressure. After that, there is an idling mode in which the rough pull valve 173 is closed, the back pressure valve 176 is opened, and the turbo molecular pump 172 is operated. Next, there is a main pull mode in which the main pull valve 174 is opened and the pressure in the vacuum chamber 175 is lowered from the vacuum pressure to the ultra-high vacuum pressure. Finally, there is a machining mode in which semiconductor machining is performed in the vacuum chamber 175. In the machining mode, elements of the device (not shown) operate. For example, in a RIE apparatus, a reactive gas is injected into the vacuum chamber 175. By processing the sensor data obtained from the sensor 108 with the same system configuration as in the first embodiment, these operation modes can be distinguished and the state of the semiconductor manufacturing apparatus can be accurately determined.

Examples 1 to 5 are used to show an example of an interface for displaying the monitoring status when a factory equipment observer (hereinafter referred to as a user) monitors the equipment. The user can grasp the status of the device and the like through the factory monitoring status monitor displayed on the terminal 330 shown in FIG. Specifically, the monitor is displayed on the display of a personal computer or the display of a tablet terminal. The displayed information is sent from the edge calculation unit 300 to the terminal 330 via the central calculation unit 320.

FIG. 18 is an example of a factory monitoring status monitor displayed on the terminal 330. The factory monitoring status monitor 181 includes an operation mode number input field 182 for setting the number of operation modes of each device, and the user can set the number of operation modes. The number of modes input here is acquired in the process S702 of FIG.

The monitor 181 includes one or a plurality of device status display windows 183 that display the state of the device for each operation mode. The abnormality determination in each operation mode is executed in the process S403 of FIG. 4, and the result is sent to the terminal 330. When an abnormality occurs in each operation mode, the abnormality lamp 184 of the corresponding mode lights up, so that the user can know the abnormality of the device.

In the maintenance proposal window 185, the optimum maintenance method is displayed according to the possible cause of the abnormality. The probable cause of abnormality is estimated by the process S404 of FIG. 4, and the estimation result is sent to the terminal 330. When the maintenance button 186 is pressed, an estimate request email is sent to the maintenance company via an arbitrary network.

An example of an interface in which an equipment observer (hereinafter referred to as a vendor) of a maintenance company monitors the equipment of each factory and displays the maintenance status will be shown using Examples 1 to 5. Vendors monitor one or more factories at the same time and can provide users with timely repairs (or maintenance) depending on equipment status. The hardware and software configurations are the same as in the sixth embodiment.

FIG. 19 is an example of a maintenance management monitor displayed on the terminal 330. The vendor can grasp the device status and the maintenance status through the maintenance management monitor 191 displayed on the terminal 330 shown in FIG. The device status window 192 displays the device error status and the probable cause. The probable cause of abnormality is estimated by the process S404 of FIG. 4, and the estimation result is sent to the terminal 330.

The repair proposal status is displayed in the maintenance status window 193. When the vendor presses an email, phone, fax, or mail button, each contact method is used to send a repair proposal to the user. In addition, the name of the maintenance staff in charge and the schedule of the maintenance staff in charge are displayed in the maintenance status window. The vendor can check the maintenance staff's schedule and contact the user with a repair proposal, which enables a quick decision on the repair schedule. In addition, the number of repair parts in stock is displayed in the maintenance status window. When the vendor presses the order button 194, the order information is transmitted to each member manufacturer. Since the vendor can manage the inventory of parts required for repair after checking the equipment status, the efficiency of inventory management is improved. For example, when the occurrence of a device failure is detected, the delivery time of repair can be shortened by securing the inventory of necessary members in advance. In addition, it is possible to reduce the assets owned by the maintenance company by keeping a small inventory of the parts necessary for failures that occur infrequently (for example, failures that occur once every 10 years). , ROE (return on equity) will be improved.

As described in detail above, when the analysis is performed without distinguishing the operation modes, the analysis accuracy is deteriorated because the data in different states are compared. For example, when principal component analysis and cluster analysis are used, the resolution of each component is lowered because the data that are significantly different are standardized together. In addition, there was a concern that abnormal data would be erroneously determined as normal data in another cluster.

By using the configuration of this embodiment and performing abnormality determination by distinguishing the operation modes, the analysis accuracy is improved. In particular, since the diagnosis can be performed using the sensor data during normal operation, the diagnosis can be performed without stopping the operation. That is, since it is not necessary to provide a diagnostic operation period in particular, it is possible to prevent a decrease in the operating rate of the equipment. In addition, it is possible to precisely estimate the cause of the abnormality from the combination of data for each operation mode. Further, when both the operation mode is specified and the abnormality is determined from the sensor data, it is not necessary to introduce an additional sensor for specifying the operation mode. Therefore, the introduction cost is reduced. Further, on the input / output interface screen such as the monitoring screen, the amount of information that the user can obtain increases because there is an indicator that displays the normal / abnormal state for each operation mode.

Further, as described in the embodiment, an operation mode number input unit for inputting the number of operation modes of the mechanical device when detecting an abnormality of the mechanical device based on a signal from a detection element provided in the mechanical device to be diagnosed. The operation mode threshold calculation unit that calculates the threshold for specifying the operation mode based on the signal of the detection element, and the operation mode specification that specifies the operation mode of the mechanical device using the threshold based on the signal of the detection element. Use a system that includes a unit. In this configuration, the operation mode can be determined by light processing by comparison with the threshold value. In addition, when executing the algorithm that performs the classification that serves as a reference when learning the threshold value, the number of operation modes (example: three types of stop, unload, and load) can be input, and the operation is performed by utilizing OT. The mode discrimination accuracy can be improved.

The present invention is not limited to the above-described embodiment, and includes various modifications. For example, it is possible to replace a part of the configuration of one embodiment with the configuration of another embodiment, and it is possible to add the configuration of another embodiment to the configuration of one embodiment. In addition, it is possible to add / delete / replace the configurations of other examples with respect to a part of the configurations of each embodiment.

305: Abnormality determination unit 306: Abnormality cause detection unit 3011: Sensor data database 3012: Abnormality cause estimation table 3051: Learning unit 3052: Mode determination unit 3053: Abnormality detection unit

Claims (13)

An abnormality detection system that detects an abnormality in the diagnosis target based on a detection signal from a detection element provided in the diagnosis target.
An abnormality determination unit that determines the presence or absence of an abnormality in the first operation mode and the second operation mode of the diagnosis target based on the detection signal from the detection element during normal operation.
An abnormality cause detection unit that determines and detects the cause of the abnormality to be diagnosed based on the combination of the presence or absence of the abnormality in the first operation mode and the second operation mode determined by the abnormality determination unit.
Equipped with a,
The abnormality determination unit
A learning unit that learns a threshold value for determining an operation mode based on the detection signal,
A mode determination unit that determines the operation mode of the diagnosis target using the threshold value based on the detection signal and classifies the detection signal for each operation mode.
Abnormality detection system for each classified the detection signal, the abnormality detecting unit for detecting an abnormality of the diagnostic object, wherein Rukoto equipped with. The abnormality detection system according to claim 1.
The diagnosis target is composed of a first element device and a second element device.
An abnormality detection system, wherein the first operation mode and the second operation mode are modes that depend on a combination of operating states of the first element device and the second element device. The abnormality detection system according to claim 1.
The detection signal from the detection element changes due to an external factor generated regardless of the operation mode of the diagnosis target.
The abnormality cause detection unit determines whether the abnormality is the diagnosis target or the abnormality caused by the external factor, based on the combination of the presence or absence of the abnormality in the first operation mode and the second operation mode. Anomaly detection system featuring. The abnormality detection system according to claim 1.
A plurality of the detection elements are provided.
The abnormality determination unit determines the presence or absence of an abnormality based on the detection signal detected in each operation mode for each detection element.
The abnormality detection system is characterized in that the abnormality cause detection unit determines and detects the cause of the abnormality to be diagnosed based on the combination of the presence or absence of the abnormality detected from each detection element in each operation mode. The abnormality detection system according to claim 1.
A storage device that stores an abnormality cause estimation table that stores a possible abnormality cause for a combination of the presence or absence of an abnormality in the first operation mode and the second operation mode is provided.
The abnormality detection system is characterized in that the abnormality cause detection unit determines and detects the cause of an abnormality by referring to the abnormality cause estimation table. The abnormality detection system according to claim 1 .
The learning unit
The operation mode is determined based on the detection signal using the first algorithm, and the threshold value is learned using the determination result.
The abnormality detection unit
An abnormality detection system, characterized in that an abnormality to be diagnosed is detected by using a second algorithm different from the first algorithm. The abnormality detection system according to claim 6 .
The first algorithm is clustering by the k-means method.
It is characterized by including an input device for inputting the number of operation modes to be diagnosed from the outside.
Anomaly detection system. It is an abnormality detection method that detects an abnormality of a diagnosis target by using an information processing device equipped with a storage device, an input device, a processing device, and an output device.
The input device acquires sensor data from a detection element that detects the state of the diagnosis target, and obtains sensor data.
The processing device, a classification process for classifying the sensor data for each operating mode of the diagnostic object, rows that have the abnormality detection process performed by the abnormality detection to the operation mode for each classified,
In the classification process,
A learning process in which the sensor data is classified for each operation mode to be diagnosed using the first algorithm and a threshold value is learned for each operation mode based on the classification result.
A mode determination process for determining the sensor data for each operation mode based on the learned threshold value is performed.
Anomaly detection method. In the learning process,
The k-means method is used as the first algorithm, and when the k-means method is executed, the number of operation modes can be input from the input device, and the sensor data is clustered into the number of operation modes.
The abnormality detection method according to claim 8 . The storage device stores an abnormality cause estimation table that stores possible abnormality causes for the combination of presence / absence of abnormality for each operation mode.
In the abnormality detection process,
Anomaly detection is performed for each operation mode, the result of the abnormality detection is applied to the abnormality cause estimation table, and the corresponding abnormality cause is output from the output device.
The abnormality detection method according to claim 8 . The abnormality cause estimation table further stores the assumed abnormality cause for the transition pattern of the combination of the presence or absence of the abnormality for each operation mode.
The abnormality detection method according to claim 10 . In the abnormality detection process, an abnormality is detected for each operation mode by using a second algorithm different from the first algorithm.
The abnormality detection method according to claim 8 . In the abnormality detection process, abnormality detection is performed using a different algorithm or feature amount for each operation mode.
The abnormality detection method according to claim 12 .

Images