Failure detection device, failure detection system, failure detection method, and program

The fault detection system addresses the challenge of adapting to diverse on-site conditions by using a trained model and terminal device interaction to reduce workload and enhance failure detection adaptability.

JP2026182409APending Publication Date: 2026-11-19TOPPAN HOLDINGS INC
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
JP2025078636
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2025-05-09
Publication Date
2026-11-19

AI Technical Summary

Technical Problem

Existing failure detection systems face challenges in adapting to diverse on-site conditions due to the difficulty in performing pre-learning for all possible failures, leading to a heavy workload when re-learning is required at manufacturing sites.

Method used

A fault detection system comprising a sensor data acquisition unit, classification unit, and output unit, utilizing a trained fault detection model to classify device states and adapt to on-site conditions by allowing terminal devices to transmit labels for re-learning unknown faults.

Benefits of technology

The system reduces workload by enabling the fault detection model to adapt to on-site situations while detecting new faults and facilitating re-learning, thereby improving its responsiveness to diverse failure scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2026182409000001_ABST
    Figure 2026182409000001_ABST
Patent Text Reader

Abstract

The goal is to adapt the fault detection model to on-site conditions while minimizing the workload. [Solution] The system comprises a sensor data acquisition unit that acquires sensor data indicating the state of a target device, a classification unit that classifies the state of the target device using the sensor data acquired by the sensor data acquisition unit and a fault detection model, and an output unit that outputs the classification result from the classification unit. The fault detection model is a trained model that learns the correspondence between the sensor data and the state of the device by using training data in which labels indicating the state of the device are associated with the training sensor data, and is trained to classify and output the state of the device based on the sensor data input to the model. The classification unit determines from the sensor data classified by the fault detection model as not being in a faulty state whether the target device is normal or whether a new fault that the fault detection model has not yet learned has been detected in the target device.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a failure detection device, a failure detection system, a failure detection method, and a program.

Background Art

[0002] There is a technology for detecting failures of devices and the like at manufacturing sites and the like. For example, Patent Document 1 discloses a technology for accurately predicting failures of factory equipment by performing learning and inference based on state variables and the like obtained from output data of sensors.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of Invention

Problems to be Solved by the Invention

[0004] However, at manufacturing sites and the like, various apparatuses and devices that are targets of failure detection are installed in various environments, and various failures occur. For this reason, in failure classification using supervised learning, it is difficult to perform learning in advance for all of the various failures. As a countermeasure, it is conceivable to operate a failure detection model at a manufacturing site or the like and perform re-learning on failures occurring at the site to adapt to the site. However, the work in which a person in charge at the site records the actual equipment state while checking (inspecting) the actual operation state each time, or accumulates sensor data and adds labels when a new failure occurs, imposes a heavy load and is not realistic.

[0005] The present invention has been made in view of such a situation, and an object thereof is to provide a failure detection device, a failure detection system, a failure detection method, and a program that can make a failure detection model adapt to the on-site situation while suppressing the work load.

Means for Solving the Problems

[0006] To solve the above-mentioned problems, one aspect of the present invention is a fault detection device comprising: a sensor data acquisition unit that acquires sensor data indicating the state of a target device that is to be monitored, acquired by a sensor; a classification unit that classifies the state of the target device using the sensor data acquired by the sensor data acquisition unit and a fault detection model; and an output unit that outputs the classification result from the classification unit. The fault detection model is a trained model that has been trained to classify and output the state of a device based on the sensor data input to the model by learning the correspondence between the sensor data and the state of the device using training data in which labels indicating the state of the device are associated with the training sensor data. The classification unit determines whether the target device is normal or whether a new fault that the fault detection model has not yet learned has been detected in the target device, based on the sensor data that the fault detection model has classified as not being in a faulty state.

[0007] Furthermore, one aspect of the present invention is a fault detection system comprising: a fault detection device described above; and a terminal device that is communicably connected to the fault detection device and transmits a label to the fault detection device to the fault detection device in response to the detection of a new fault that the fault detection model has not yet learned.

[0008] Furthermore, one aspect of the present invention is a fault detection method performed by a computer used in a fault detection device, wherein a sensor data acquisition unit acquires sensor data indicating the state of a target device that is to be monitored, acquired by a sensor; a classification unit classifies the state of the target device using the sensor data acquired by the sensor data acquisition unit and a fault detection model; an output unit outputs the classification result by the classification unit; the fault detection model is a trained model that has been trained to classify and output the state of the device based on the sensor data input to the model by learning the correspondence relationship between the sensor data and the state of the device using training data in which labels indicating the state of the device are associated with the training sensor data; and the classification unit determines from the sensor data classified by the fault detection model as not being in a faulty state whether the target device is normal or whether a new fault that the fault detection model has not yet learned has been detected in the target device.

[0009] Furthermore, one aspect of the present invention is a program that causes a computer used in a fault detection device to acquire sensor data indicating the state of a target device that is being monitored, acquired by a sensor, to classify the state of the target device using the acquired sensor data and a fault detection model, and to output the classification result based on the classification. The fault detection model is a trained model that has been trained to classify and output the state of a device based on the sensor data input to the model by learning the correspondence between the sensor data and the state of the device using training data in which labels indicating the state of the device are associated with the training sensor data. In the classification, the program causes the computer to determine whether the target device is normal or whether a new fault that the fault detection model has not yet learned has been detected in the target device, based on the sensor data that the fault detection model has classified as not being in a faulty state. [Effects of the Invention]

[0010] According to the present invention, the fault detection model can be adapted to on-site conditions while reducing the workload. [Brief explanation of the drawing]

[0011] [Figure 1] This block diagram shows an example configuration of the fault detection system 1 according to an embodiment. [Figure 2] This is a block diagram showing an example configuration of the server device 10 according to the embodiment. [Figure 3] This is a diagram illustrating the processing performed by the server device 10 according to the embodiment. [Figure 4] This is a diagram illustrating the processing performed by the server device 10 according to the embodiment. [Figure 5] This is a diagram illustrating the processing performed by the server device 10 according to the embodiment. [Figure 6] This is a diagram illustrating the processing performed by the server device 10 according to the embodiment. [Figure 7] This is a diagram illustrating the processing performed by the server device 10 according to the embodiment. [Figure 8] This is a sequence diagram showing the processing flow performed by the fault detection system 1 of the embodiment. [Modes for carrying out the invention]

[0012] Embodiments of the present invention will be described below with reference to the drawings.

[0013] (Regarding fault detection system 1) First, the configuration of the fault detection system 1 will be explained using Figure 1. Figure 1 is a block diagram showing an example configuration of the fault detection system 1 according to an embodiment. The fault detection system 1 is a system that detects abnormalities in equipment installed in a manufacturing site or the like. The fault detection system 1 includes, for example, a server device 10, a plurality of sensors 20 (sensors 20-1, 20-2, ..., 20-3), a gateway device 30, and a terminal device 40.

[0014] Sensor 20 is a sensor that measures the status of equipment installed in a manufacturing site or similar location. Examples of sensors 20 include an acceleration sensor that measures the magnitude of vibration of equipment, a microphone that measures the magnitude of sound emitted from equipment, an ammeter that measures the magnitude of current flowing through circuits installed in equipment, and a temperature sensor that measures the temperature of equipment. Sensor 20 may be installed one per piece of equipment, or multiple sensors may be installed. Sensor 20 periodically or irregularly transmits sensor data as measurement results of the equipment status to the gateway device 30.

[0015] The sensor 20 comprises, for example, a sensor unit, a sensor control unit, a communication pre-processing unit, and a communication unit. The sensor unit measures the state of the device according to the control of the sensor control unit and outputs sensor data as a measurement result to the sensor control unit. The sensor control unit controls the sensor unit and acquires sensor data from the sensor unit. If the sensor control unit performs pre-processing before transmitting the data to the gateway device 30, it outputs the sensor data to the communication pre-processing unit. Pre-processing here refers to processing the sensor data so that it can be easily processed in a later process (in this case, the process of transmitting the sensor data to the gateway device 30). For example, the communication pre-processing unit processes the sensor data according to the communication format with the gateway device 30. The communication pre-processing unit outputs the pre-processed sensor data to the communication unit. The communication unit acquires the sensor data pre-processed by the communication pre-processing unit and transmits the acquired sensor data to the gateway device 30. The sensor 20 may be configured to be powered by a power supply or battery. In this case, when the sensor 20 is powered while connected to a battery, it may be configured to operate intermittently in acquiring sensor data and communicating with the gateway device. This avoids the situation where the battery power is consumed by continuous operation, and thus reduces power consumption.

[0016] The gateway device 30 is a relay device provided between the sensor 20 and the server device 10, and transmits sensor data measured by the sensor 20 to the server device 10. The gateway device 30 performs intermittent communication according to a wireless communication protocol such as LTE (Long Term Evolution), for example. In intermittent communication, communication and pause are repeated at predetermined communication intervals. This avoids maintaining a continuous always-on connection and reduces power consumption. The gateway device 30 includes, for example, a communication unit and a control unit. The communication unit communicates with the sensor 20 and the server device 10 under the control of the control unit. The control unit performs protocol conversion and the like on received data (including sensor data) received from the sensor 20 as necessary, and controls the relay of sensor data by the communication unit. Note that the gateway device 30 may be communicably connected to the server device 10 via wired communication, for example, by connection using an Ethernet cable or the like. In this case, the gateway device 30 transmits the sensor data to the server device 10 via the Ethernet cable or the like.

[0017] The server device 10 is a computer that performs failure detection. As the server device 10, for example, a cloud server, an on-premises server, a PC (personal computer), or the like can be applied. The server device 10 is communicably connected to the gateway device 30 and the terminal device 40. The server device 10 receives sensor data from the gateway device 30. The server device 10 performs failure detection based on the sensor data received from the gateway device 30 using a failure detection model that is an AI model for performing failure detection. The server device 10 stores the detection result of the performed failure detection in a web page provided by the server device 10. This web page is provided to monitor the state of the equipment that is the target of failure detection, and is, for example, a page on which the result of failure detection is displayed for each piece of equipment or for each set of sensor data.

[0018] The terminal device 40 is a computer operated by a user who is an administrator or a person in charge at a manufacturing site or the like. As the terminal device 40, for example, a PC, a smartphone, a tablet terminal, a mobile phone, or the like can be applied. The terminal device 40 allows the user to view the failure detection results displayed on the web page, for example, by accessing the web page provided by the server device 10 via a web browser.

[0019] Further, the terminal device 40 is communicably connected to the server device 10. For example, the terminal device 40 receives a notification related to a failure from the server device 10. When the type of the failure notified from the server device 10 is unknown, that is, the failure is an unlearned failure for the failure detection model, the terminal device 40 transmits a notification corresponding to the unlearned failure to the server device 10. The notification corresponding to the unlearned failure is, for example, a notification that notifies the server device 10 of a label assigned by the user to the unlearned failure, and requests the failure detection model to perform relearning so as to learn the unlearned failure.

[0020] (Configuration of Server Device 10) The configuration of the server device 10 will be described with reference to Figure 2. Figure 2 is a block diagram illustrating a configuration example of the server device 10 according to the embodiment. As shown in Figure 2, the server device 10 includes, for example, a communication unit 11, a storage unit 12, and a control unit 13.

[0021] The communication unit 11 communicates with various external devices connected via the communication network NW. The storage unit 12 is configured by a storage medium, for example, an HDD (Hard Disk Drive), a flash memory, an EEPROM (Electrically Erasable Programmable Read Only Memory), a RAM (Random Access read / write Memory), a ROM (Read Only Memory), or any combination of these storage media. The storage unit 12 stores programs for executing various processes of the server device 10, and temporary data used when performing the various processes. The storage unit 12 includes a sensor data storage unit 121. The sensor data storage unit 121 stores sensor data.

[0022] The control unit 13 is implemented by the CPU (Central Processing Unit) of the server device 10 executing a program pre-stored in the memory unit 12. Alternatively, the control unit 13 may be implemented as an integrated circuit such as an ASIC (Application Specific Integrated Circuit).

[0023] The control unit 13 includes a sensor data acquisition unit 131, a preprocessing unit 132, a classification unit 133, a display control unit 134, a notification unit 135, and an AI model update unit 136.

[0024] The sensor data acquisition unit 131 acquires sensor data. For example, the communication unit 11 of the server device 10 receives sensor data transmitted from the gateway device 30 and outputs the received sensor data to the sensor data acquisition unit 131. As a result, the sensor data acquisition unit 131 acquires sensor data. The sensor data acquisition unit 131 stores the acquired sensor data in the storage unit 12 as the sensor data storage unit 121.

[0025] The pre-processing unit 132 performs pre-processing. This pre-processing involves processing the sensor data to make it easier to process in subsequent processes (in this case, the process of outputting the sensor data to the classification unit 133). The pre-processing unit 132 processes the sensor data according to the input of the fault detection model used by the classification unit 133 for fault detection.

[0026] For example, to make it easier for a fault detection model to detect faults, it may be configured to perform a Fast Fourier Transform (FFT) on sensor data acquired over time (time-series data) and then input the resulting frequency-domain sensor data (frequency-domain data) into the fault detection model. Performing such a transformation makes it easier for the fault detection model to extract fault features. Alternatively, preprocessing such as passing the sensor data through a low-pass filter, high-pass filter, or Kalman filter may be performed to remove noise. Alternatively, the sensor data may be preprocessed by performing normalization, which involves adjusting the gain of the sensor data values ​​so that the maximum value of the sensor data becomes a reference value (for example, 1.0).

[0027] In this way, when sensor data is converted and input to the fault detection model, the preprocessing unit 132 performs a preprocessing step to convert the sensor data into a format that can be input to the fault detection model. For example, the preprocessing unit 132 performs a preprocessing step to convert time-series sensor data into frequency domain data. Alternatively, the preprocessing unit 132 may perform a preprocessing step to remove the effect of temperature changes caused by ambient temperature by subtracting an offset value corresponding to the ambient temperature at the time the sensor data was acquired from the sensor data indicating temperature. The preprocessing unit 132 outputs the preprocessed sensor data to the classification unit 133. Furthermore, if the sensor data notified from the gateway device 30 is in the format for input to the fault detection model, the preprocessing by the preprocessing unit 132 may be omitted.

[0028] The classification unit 133 performs fault detection using sensor data (or pre-processed sensor data if pre-processing has been performed). As part of fault detection, the classification unit 133 performs fault classification to determine which of the learned faults to classify the fault into, and detects unlearned faults.

[0029] The method by which the classification unit 133 performs fault detection will be explained using Figure 3. Figure 3 is a diagram illustrating the processing performed by the server device 10 according to this embodiment. As shown in this diagram, the classification unit 133 performs fault detection using two AI models (fault detection models), the first model and the second model.

[0030] The first model is a pre-trained model that learns the correspondence between sensor data and device states by using training data in which labels indicating the device state are associated with training sensor data, thereby classifying and outputting the device state based on the input sensor data.

[0031] The second model is an unsupervised learning model, specifically one that has learned only from sensor data in a normal, non-faulty state. It has learned linear or nonlinear separation boundaries composed of sets of sensor data in a normal state. The second model determines whether the input sensor data is an outlier located outside the distribution of normal sensor data (e.g., a normal distribution), and classifies sensor data that is not an outlier as normal and sensor data that is an outlier as abnormal.

[0032] The first model receives sensor data (or pre-processed sensor data). The first model classifies the input sensor data (or pre-processed data) as either trained fault data or other data. In this diagram, trained fault data refers to any of the trained fault A, trained fault B, or trained fault C. In this case, the first model is a trained model that has learned training data in which training sensor data is associated with one of the labels "Fault A," "Fault B," "Fault C," or "Other." Here, the sensor data labeled "Other" includes sensor data in a normal state (normal data) that is not faulty, and sensor data that indicates a fault state but has not been trained in the first model.

[0033] The second model receives sensor data classified as "other" by the first model. The second model classifies the input sensor data as either normal or an unknown fault. The second model classifies the input sensor data as "normal" if it is not an outlier indicating a normal state. If the second model classifies the input sensor data as an outlier indicating a normal state, it classifies it as an unknown fault, i.e., fault data that the first model has not yet learned.

[0034] Returning to the explanation of Figure 2, the display control unit 134 displays the fault detection results on the web page. The display control unit 134 updates the web page by displaying the latest detection result each time a detection result is output from the classification unit 133. Furthermore, if the classification unit 133 detects an unknown fault, the display control unit 134 notifies the terminal device 40 of this fact and causes the terminal device 40 to display a screen for assigning a label to the fault data corresponding to the unknown fault (see Figures 4-5).

[0035] The notification unit 135 notifies the terminal device 40 of the fault detection result. The notification unit 135 notifies the fault detection result to a pre-configured destination, such as an email address, SNS account, or telephone number. The notification unit 135 may also be configured to notify the pre-configured destination of the detection result if an abnormal detection result is obtained during fault detection. This allows on-site personnel to be alerted that a fault has been detected.

[0036] The AI ​​model update unit 136 updates the AI ​​model (mainly the first model). The AI ​​model update unit 136 generates training data for retraining (additional training) by associating new labels with sensor data corresponding to unknown faults. These new labels are those assigned by the user to the unknown fault in response to the server device 10 notifying the terminal device 40 that an unknown fault has been detected. The AI ​​model update unit 136 updates the first model by having the first model learn the training data for retraining. By learning the training data for retraining, the first model can learn the correspondence between sensor data corresponding to unknown faults and the state of the equipment, and will be able to detect fault states that were previously unknown based on the input sensor data.

[0037] For example, whenever an unknown fault is detected, the AI ​​model update unit 136 generates training data for retraining corresponding to the detected unknown fault, and trains the first model with the generated training data. This enables the first model to detect unknown faults, expands the types of faults that can be classified according to the field conditions, and allows it to respond to the field conditions.

[0038] Here, the processing performed by the server device 10 will be explained using Figures 4 to 7. Figures 4 to 7 are diagrams illustrating the processing performed by the server device 10 according to this embodiment.

[0039] Figure 4 shows an example of a screen displayed on a webpage when an unknown fault is detected. In this figure, the webpage displays the status of each of the multiple devices (device A, device B, and device C in this example) that are the target of fault detection, under the heading "Device List". In this figure, the status of device A is "Normal", the status of device B is a Type 1 "Fault", and the status of device C is an unregistered type "Fault". Here, a Type 1 "Fault" is a fault that the first model has learned. An unregistered type "Fault" is a fault that the first model has not learned. As shown in this figure, when an unregistered type of "fault," that is, a fault that the first model has not yet learned, is detected by the second model, the display control unit 134 displays a different display method, such as highlighting that an unlearned fault has been detected. The display control unit 134 also displays operation button B1 (the operation button labeled "Status Registration" in this figure) for registering a label corresponding to the unregistered type of "fault."

[0040] Figure 5 shows an example of the screen displayed on the web page when operation button B1 is pressed in Figure 4. In this figure, the web page shows an input field N for registering an unregistered type of "fault" fault label detected by device C, labeled as "Status Registration". In this figure, an example is shown where the user has entered "Label 2" as the fault label for the unregistered type of "fault" in input field N. When the user enters a fault label to be assigned to the unregistered type of "fault" in input field N and operates operation button B2, the fault label entered in input field N is output to the server device 10.

[0041] The server device 10 receives labels set on the web page (labels assigned by the user to unregistered types) and outputs the received labels to the AI ​​model update unit 136. The AI ​​model update unit 136 generates training data for retraining by associating the labels notified from the terminal device 40 with the sensor data corresponding to the unregistered types.

[0042] Figure 6 shows an example of how fault detection is performed when the first model is updated, in a diagram corresponding to Figure 3. In this diagram, the first model, through additional training, can now classify sensor data corresponding to "Fault D" in addition to the previously learned fault data of "Fault A," "Fault B," and "Fault C." The first model classifies the input sensor data (or pre-processed data) into one of the following categories: "Fault A," "Fault B," "Fault C," "Fault D," or "Other." Similar to Figure 3, the second model receives sensor data classified as "Other" by the first model.

[0043] Figure 7 shows an example of the screen displayed on the web page when a fault detection occurs in the same situation as in Figure 4 after relearning. In this figure, the web page, as in Figure 4, shows the status of each of the multiple devices (device A, device B, and device C in this example) that are the target of fault detection as a "device list". As in Figure 4, the status of device A is shown as "normal", and the status of device B is shown as a type 1 "fault". Here, regarding the state of device C, Figure 4 showed it as an unregistered type "failure," but through relearning, it became possible to detect that this failure is a type 2 "failure," and therefore Figure 7 shows that the state of device C is a type 2 "failure."

[0044] Figure 8 is a sequence diagram showing the processing flow of the fault detection system 1 of the embodiment. Sensor 20 transmits sensor data, which is the measurement result of measuring the status of the equipment, to the gateway device 30 (step S1). The gateway device 30 transmits the sensor data received from sensor 20 to the server device 10 (step S2).

[0045] The sensor data acquisition unit 131 of the server device 10 acquires the sensor data notified via the gateway device 30 and stores the acquired sensor data in the sensor data storage unit 121 (step S3). The preprocessing unit 132 acquires sensor data by referring to the sensor data storage unit 121, performs preprocessing on the acquired sensor data, and generates preprocessed sensor data (preprocessed data) (step S4). The preprocessing unit 132 outputs the preprocessed data to the classification unit 133 (step S5). The classification unit 133 performs fault detection based on the pre-processed data and outputs the detection result to the display control unit 134 (step S6). The classification unit 133 inputs the pre-processed data acquired from the pre-processing unit 132 to the first model, obtains a detection result indicating either a fault detected by the first model, an unknown fault detected by the second model, or normal, and outputs the obtained detection result to the display control unit 134. The display control unit 134 updates the web page based on the detection results obtained from the classification unit 133. The display control unit 134 displays the detection results on the terminal device 40 as needed (step S7). Furthermore, the classification unit 133 outputs the detection result to the notification unit 135 (step S8). The notification unit 135 notifies the terminal device 40 of the detection result as necessary (step S9). If an unknown fault is detected, the terminal device 40 accesses a web page in response to user input and sets a label to be assigned to the unknown fault. Information indicating the label to be assigned to the unknown fault is transmitted to the server device 10 in response to this operation (step S10).

[0046] When the AI ​​model update unit 136 of the server device 10 obtains a label to be assigned to the unknown fault set in step S10, it refers to the sensor data storage unit 121 and obtains sensor data corresponding to the unknown fault (step S11). The AI ​​model update unit 136 outputs the sensor data obtained in step S11 to the preprocessing unit 132 (step S12). The preprocessing unit 132 performs preprocessing on the sensor data obtained in step S12 and outputs the preprocessed data to the AI ​​model update unit 136 (step S13). The AI ​​model update unit 136 generates training data for retraining by associating the preprocessed data acquired in step S13 with the labels acquired in step S10, and trains the first model with the generated training data for retraining. The AI ​​model update unit 136 outputs the retrained first model to the classification unit 133 as an AI model (fault detection model) to be used for fault detection (step S14).

[0047] As described above, the server device 10 of the embodiment includes a sensor data acquisition unit 131, a classification unit 133, and a display control unit 134 (an example of an output unit). The sensor data acquisition unit 131 acquires sensor data indicating the state of the target device, which is the object of monitoring, acquired by the sensor 20. The classification unit 133 classifies the state of the target device using the sensor data and a first model and a second model (an example of a fault detection model). The display control unit 134 outputs the classification result from the classification unit 133. The first model is a trained model that has been trained to classify and output the state of the device based on the sensor data input to the model by learning the correspondence between the sensor data and the state of the device using training data in which labels indicating the state of the device are associated with the sensor data for learning. The classification unit 133 uses the second model to determine whether the target device is normal or whether a new fault that the first model has not learned has been detected in the target device, based on the sensor data that the first model (an example of a fault detection model) has classified as not being in a faulty state. As a result, the server device 10 of this embodiment can detect faults that the first model has not yet learned, and the fault detection model can be adapted to the on-site conditions while reducing the workload. Furthermore, the server device 10 of this embodiment can detect unlearned failures, allowing it to narrow down the sensor data that needs to be retrained in the first model. This reduces the workload of generating training data for retraining and the operational burden of manually associating labels with sensor data.

[0048] Furthermore, in the server device 10 of this embodiment, when the classification unit 133 detects that a new fault has been detected, the display control unit 134 outputs information for setting a label to be assigned to the new fault. As a result, the server device 10 of this embodiment can assign a label to a new fault that has not yet been learned, thereby enabling the fault detection model to adapt to the actual situation on site while reducing the workload.

[0049] Furthermore, the server device 10 of this embodiment further includes an AI model update unit 136. When a new fault is detected by the classification unit 133, the AI ​​model update unit 136 generates training data for retraining. The training data for retraining is data in which sensor data corresponding to the new fault is associated with a label indicating that the state of the equipment is that new fault. The AI ​​model update unit 136 uses the training data for retraining to retrain the first model (fault detection model). As a result, the server device 10 of this embodiment can learn previously unlearned faults through retraining, and the fault detection model can be made to correspond to the situation on site.

[0050] Furthermore, in the server device 10 of this embodiment, when a new fault is detected by the classification unit 133, the display control unit 134 outputs information to the terminal device 40 for setting a label to be assigned to the new fault. The AI ​​model update unit 136 generates training data for retraining in response to the label to be assigned to the new fault being notified by the user operating the terminal device 40. The training data for retraining is training data in which the label notified by the terminal device is associated with the sensor data corresponding to the new fault. The AI ​​model update unit 136 uses the training data for retraining to retrain the first model (fault detection model). As a result, in the server device 10 of this embodiment, previously unlearned faults can be retrained according to the labels set by the user, and the fault detection model can be made to correspond to the situation on site.

[0051] As described above, the fault detection system 1 of this embodiment comprises a server device 10 and a terminal device 40. The terminal device 40 is connected to the server device 10 (fault detection device) in a communicative manner. When the terminal device 40 detects a new fault that has not yet been learned by the first model (fault detection model) from the server device 10, it transmits a label to be assigned to the new fault to the server device 10. This allows the server device 10 of this embodiment to adapt the fault detection model to the actual site conditions while reducing the workload.

[0052] The server device 10 and fault detection system 1 in the above-described embodiment may be implemented in whole or in part by a computer. In that case, the program for implementing this function may be recorded on a computer-readable recording medium, and the program recorded on this recording medium may be loaded into the computer system and executed. The term "computer system" here includes hardware such as the OS and peripheral devices. The term "computer-readable recording medium" refers to portable media such as flexible disks, magneto-optical disks, ROMs, CD-ROMs, and storage devices such as hard disks built into the computer system. Furthermore, the term "computer-readable recording medium" may also include those that dynamically hold programs for a short period of time, such as communication lines used when transmitting programs via networks such as the Internet or communication lines such as telephone lines, and those that hold programs for a certain period of time, such as volatile memory inside the computer system that acts as a server or client in that case. The program may be for implementing a part of the functions described above, or it may be a program that can implement the functions described above in combination with a program already recorded in the computer system, or it may be implemented using a programmable logic device such as an FPGA.

[0053] While embodiments of this invention have been described in detail above with reference to the drawings, the specific configuration is not limited to these embodiments and includes designs and the like that do not depart from the spirit of this invention.

[0054] In the embodiment described above, an example was shown in which, when operation button B1 in Figure 4 is operated, the user transitions to a screen for registering labels as shown in Figure 5. However, the embodiment is not limited to this example. Images to assist the user in setting labels may be displayed along with Figure 4, or in response to the operation of operation button B1 in Figure 4.

[0055] Images can be used to assist users in setting labels, for example, by displaying attribute information of sensor data corresponding to an unknown fault, more specifically, time-series data including sensor data in which an unknown fault was detected, normal sensor data acquired in the time period before and after the detection, and pre-processed data. Images showing the installation location of sensor 20 in the equipment and the time when sensor data corresponding to the unknown fault was acquired may also be used to assist users in setting labels. By displaying such images, users can easily determine whether the fault detected as an unknown fault is indeed an unlearned fault, is included in existing faults, or is irregular data caused by another factor, such as maintenance, even though the equipment is in a normal state.

[0056] Furthermore, in the screen shown in Figure 5, a dropdown menu or similar method may be used to display options for labels to be assigned to new faults. This prevents erroneous registration, such as assigning a label already assigned to another fault to a newly detected fault. [Explanation of Symbols]

[0057] 1…Fault detection system 10…Server equipment (fault detection device) 11… Communications Department 12...Storage section 13…Control Unit 131...Sensor data acquisition unit 132... Pre-processing section 133...Classification section 134...Display control unit (output unit) 135...Notification unit (output unit) 136... AI Model Update Section 20...Sensor 30…Gateway device 40…Terminal device

Claims

1. A sensor data acquisition unit acquires sensor data indicating the status of the target device, which is the object of monitoring, obtained by the sensor. A classification unit classifies the state of the target device using the sensor data acquired by the sensor data acquisition unit and a fault detection model. An output unit that outputs the classification results from the classification unit, Equipped with, The fault detection model is a trained model that learns the correspondence between the sensor data and the state of the equipment by using training data in which labels indicating the state of the equipment are associated with the training sensor data, and is trained to classify and output the state of the equipment based on the sensor data input to the model. The classification unit determines, from the sensor data classified as not being in a faulty state by the fault detection model, whether the target device is functioning normally or whether a new fault that the fault detection model has not yet learned has been detected in the target device. Fault detection device.

2. The output unit outputs information for setting a label to be assigned to a new fault when the classification unit detects a new fault. The fault detection device according to claim 1.

3. If the classification unit detects a new fault, the system further includes an AI model update unit that retrains the fault detection model using training data for retraining, in which the sensor data corresponding to the new fault is associated with a label indicating that the equipment state is that new fault. The fault detection device according to claim 1.

4. When the classification unit detects a new fault, the output unit outputs information to the terminal device for setting a label to be assigned to the new fault. The system further includes an AI model update unit that, upon receiving notification from a user operating the terminal device of a label to be assigned to the new fault, retrains the fault detection model using training data for retraining, in which the labels notified by the terminal device are associated with the sensor data corresponding to the new fault. The fault detection device according to claim 1.

5. The fault detection device according to claim 1, A terminal device that is communicatively connected to the fault detection device and transmits a label to the fault detection device to the fault detection device in response to the detection of a new fault that the fault detection model has not yet learned, A fault detection system equipped with the following features.

6. A fault detection method performed by a computer used in a fault detection device, The sensor data acquisition unit acquires sensor data indicating the status of the target device, which is the object of monitoring, obtained by the sensor. The classification unit classifies the state of the target device using the sensor data acquired by the sensor data acquisition unit and the fault detection model. The output unit outputs the classification result from the classification unit. The fault detection model is a trained model that learns the correspondence between the sensor data and the state of the equipment by using training data in which labels indicating the state of the equipment are associated with the training sensor data, and is trained to classify and output the state of the equipment based on the sensor data input to the model. The classification unit determines, from the sensor data classified as not being in a faulty state by the fault detection model, whether the target device is functioning normally or whether a new fault that the fault detection model has not yet learned has been detected in the target device. Fault detection method.

7. In the computer used in fault detection devices, The system acquires sensor data that indicates the status of the target device being monitored, obtained by the sensor. Using the acquired sensor data and fault detection model, the state of the target device is classified. Output the classification results based on the above classification. The fault detection model is a trained model that learns the correspondence between the sensor data and the state of the equipment by using training data in which labels indicating the state of the equipment are associated with the training sensor data, and is trained to classify and output the state of the equipment based on the sensor data input to the model. In the above classification, the sensor data classified as not being in a faulty state by the fault detection model is used to determine whether the target device is functioning normally or whether a new fault that the fault detection model has not yet learned has been detected in the target device. program.

Citation Information

Patent Citations

  • Failure prediction system, failure prediction device, learning device, failure prediction method, and program

    JP2023167883A