Sign detection system, sign detection method, and sign detection program

Abstract

The present invention improves abnormality detection performance in a production facility, or identifies a cause of an abnormality to cope with the abnormality at an early stage. This sign detection system comprises one or more computers that execute: acquiring, from a production facility including an inspection target, sensing data that represents at least one among light, sound, temperature, vibration, odor, and the generation of a specific substance in a non-contact state; calculating the degree of abnormality pertaining to the acquired sensing data; detecting a sign and a cause of the abnormality on the basis of causal relationship information representing a correspondence relationship between the cause of the abnormality and a feature of the degree of abnormality, and the degree of abnormality corresponding to the acquired sensing data; and outputting information indicating the sign and the cause of the detected abnormality.

Description

Symptom detection system, symptom detection method, and symptom detection program

[0001] The present disclosure relates to a sign detection system, a sign detection method, and a sign detection program.

[0002] Conventionally, an anomaly detection device has been proposed that includes a data acquisition unit that reads image data and process data from a storage device that stores image data continuously output by an imaging device equipped in production equipment and process data continuously output by a sensor attached to the production equipment; an anomaly determination unit that calculates a degree of image abnormality using a model that has learned the tendency of feature quantities of image data in a normal state and the feature quantities of the image data acquired by the data acquisition unit, and calculates a sensor value abnormality degree that represents the degree of modulation of the process data read by the data acquisition unit; and a cause diagnosis unit that outputs a cause corresponding to the sensor value abnormality degree and image abnormality degree calculated by the anomaly determination unit, based on causal relationship information that defines the correspondence between the cause and the combination of the sensor value abnormality degree and the image abnormality degree (Patent Document 1).

[0003] In addition, a facility status monitoring system has been proposed that has a sensor that outputs data indicating the status of the facility used for status monitoring as sensor data, a communication unit that transmits the sensor data, and a power supply unit that supplies power to the sensor and communication unit, and that has a sensor node common to multiple facilities (Patent Document 2).

[0004] A remotely installed equipment status diagnostic device has also been proposed (Patent Document 3).

[0005] International Publication No. 2023 / 127748 Japanese Patent Application Laid-Open No. 2023-7350 Japanese Patent Application Laid-Open No. 07-49713

[0006] In general, it is desirable to predict anomalies in production facilities and minimize their impact on safety, stability, product quality, costs, the environment, and so on. However, when changes appear in process data in a plant, for example, it is not uncommon for a problem to have progressed to some extent. Therefore, this technology aims to at least one of improving the ability to detect anomalies in production facilities and identifying the cause of the anomaly and taking early action.

[0007] Examples of anomaly detection devices according to the present disclosure are as follows: (Aspect 1) An anomaly detection system including one or more computers that execute the following steps: acquire sensing data representing at least one of light, sound, temperature, vibration, odor, and generation of a specific substance from production equipment including an inspection target in a non-contact manner; calculate an anomaly level for the acquired sensing data; detect a sign of the anomaly and its cause based on causal relationship information representing a correspondence between the cause of the anomaly and characteristics of the anomaly level, and the anomaly level corresponding to the acquired sensing data; and output information indicating the detected sign of the anomaly and its cause. (Aspect 2) In Aspect 1, two or more types of acquired sensing data may be present, the anomaly level may be calculated for the two or more types of sensing data, and the causal relationship information may represent a correspondence between the cause of the anomaly and characteristics of the anomaly level based on each of the two or more types of sensing data. (Aspect 3) In Aspects 1 or 2, the production equipment is a chemical plant, and the sensing data may be data different from process data measured on a processing target of the chemical plant. (Aspect 4) In Aspect 3, the one or more computers may further acquire the process data, calculate an abnormality level for the process data, and detect a sign of the abnormality and its cause based on a combination of the abnormality levels of two or more types of sensing data for inspection objects of the chemical plant and the abnormality level of the process data. (Aspect 5) In any one of Aspects 1 to 4, a plurality of the inspection objects may be present in a production facility, and the system may further include a moving means equipped with a sensor for acquiring the sensing data and capable of moving within the area of ​​the production facility in accordance with a preset program, and the sensor may acquire the sensing data from each of the plurality of inspection objects. (Aspect 6) In Aspect 5, the moving means may include a holding mechanism for holding the sensor so that its orientation or position can be changed.(Aspect 7) In Aspect 5 or 6, an identification mark indicating that the production equipment is subject to inspection may be provided on or near the production equipment, and the transportation means may start acquiring the sensing data by detecting the identification mark using image recognition. (Aspect 8) In any one of Aspects 5 to 7, if the one or more computers are unable to acquire the sensing data from the sensor, they may request the transportation means to recreate the sensing data or notify a user terminal. (Aspect 9) In any one of Aspects 1 to 8, the one or more computers may acquire two or more types of sensing data and detect multiple combinations of signs of anomaly and their causes based on the anomaly level corresponding to each of the acquired sensing data. (Aspect 10) In Aspect 9, the one or more computers may output the multiple detected combinations of signs of anomaly and their causes in order of likelihood based on the degree of agreement between the anomaly level characteristics represented by the causal relationship information and the anomaly levels corresponding to each of the acquired sensing data. (Aspect 11) In Aspect 9 or 10, the one or more computers may store information representing a repair history of the production equipment, and for a combination of multiple detected signs of abnormality and their causes, it may be determined that the more recently a part was repaired, the less likely the defect is caused by that part. (Aspect 12) A sign detection method performed by one or more computers includes: acquiring, in a non-contact state, sensing data from production equipment including an inspection target, the sensing data including at least one of light, sound, temperature, vibration, odor, and generation of a specific substance; calculating an abnormality degree for the acquired sensing data; detecting the sign of abnormality and its cause based on causal relationship information representing a correspondence between the cause of the abnormality and characteristics of the abnormality degree, and the abnormality degree corresponding to the acquired sensing data; and outputting information indicating the detected sign of abnormality and its cause.(Mode 13) A symptom detection program that causes one or more computers to execute the following steps: acquiring, in a non-contact manner, sensing data representing at least one of light, sound, temperature, vibration, odor, and generation of a specific substance from production equipment including an inspection target; calculating an abnormality level for the acquired sensing data; detecting a symptom of the abnormality and its cause based on causal relationship information representing a correspondence between the cause of the abnormality and characteristics of the abnormality level, and the abnormality level corresponding to the acquired sensing data; and outputting information indicating the detected symptom of the abnormality and its cause. (Aspect 14) A sign detection method comprising: a sensing data acquisition step of acquiring, in a non-contact state, sensing data indicating at least one of light, sound, temperature, vibration, odor, and generation of a specific substance from production equipment including an inspection target; a calculation step of calculating an abnormality level for the acquired sensing data; a detection step of detecting a sign of the abnormality and its cause based on causal relationship information indicating a correspondence between the cause of the abnormality and a plurality of abnormality level characteristics, and the abnormality level corresponding to the acquired sensing data; an output step of outputting information indicating the detected sign of the abnormality and its cause; a process data acquisition step of further acquiring process data different from the sensing data, which is measured on a processing object of the production equipment; and a second calculation step of calculating an abnormality level for the process data, wherein the output step is performed while the number of items of the process data indicating an abnormality is fewer than the number of items of the sensing data indicating an abnormality.

[0008] The contents of the means for solving the problem can be combined as much as possible without departing from the problem and technical idea of ​​the present disclosure. The contents of the means for solving the problem can be provided as a device such as a computer or a system including multiple devices, a method executed by a computer, or a program executed by a computer. A recording medium storing the program may also be provided.

[0009] According to the disclosed technology, it is possible to improve the performance of detecting abnormalities in production equipment, or to identify the cause of the abnormality and take action early.

[0010] FIG. 1 is a diagram illustrating an example of a system according to this embodiment. FIG. 2 is a diagram illustrating acquisition of sensing data. FIG. 3 is a schematic perspective view illustrating an example of a holding structure for a five-sense sensor. FIG. 4 is a diagram illustrating an example of the flow of data and processing in the entire system. FIG. 5 is a process flow diagram illustrating an example of a sign detection process executed by the system. FIG. 6 is a diagram illustrating an anomaly detection using an autoencoder. FIG. 7 is a diagram illustrating an example of a determination logic. FIG. 8 is a diagram illustrating another example of the determination logic. FIG. 9 is a diagram illustrating another example of the determination logic. FIG. 10 is a diagram illustrating another example of the determination logic. FIG. 11 is a diagram illustrating another example of the determination logic.

[0011] Hereinafter, an embodiment of the symptom detection device will be described with reference to the drawings.

[0012] <Embodiment> Fig. 1 is a diagram illustrating an example of a system according to this embodiment. The system 100 includes a plant 1, a control station 2, a warning detection device 3, and a user terminal 4. These components are communicably connected via a predetermined control network. The system 100 is, for example, a distributed control system (DCS) and includes multiple control stations 2. That is, the control system of the plant 1 is divided into multiple sections, and each control section is controlled in a distributed manner by the control station 2.

[0013] The plant 1 is, for example, a process plant. A process plant is a plant that primarily handles materials with variable shapes, such as fluids, gases, and powders, and manufactures products through processes such as reactions and synthesis. Process plants include, but are not limited to, plants that handle technologies such as chemical products, petroleum, metals, paper and pulp, pharmaceuticals, glass and cement, and steel. Preferably, the plant 1 is, for example, a chemical process plant, a production facility that achieves a certain purpose through a series of chemical processes. The plant 1 is equipped with sensors (process instrumentation 111, five senses sensor 124), and plant data obtained from signals output from the sensors is transmitted to the predictive detection device 3 via the control station 2. The system 100 transmits and receives plant data in accordance with a communication standard, such as OPC (OLE for Process Control). The plant data is information for monitoring the operating status of the plant 1, and this plant data includes process data and sensing data.

[0014] Process data is data representing the state of a processing object being processed by, for example, the production equipment 11 included in the plant 1. The processing object may include raw materials, intermediates, products, catalysts, utilities (heat transfer medium, etc.), waste liquids, wastewater, etc. The production equipment 11 may also include reactors, distillation columns, heat exchangers, pumps, compressors, tanks, piping, etc. In particular, process data may be signals or data representing the state of a processing object obtained directly or indirectly by process instrumentation 111, such as sensors or field devices, attached to the production equipment 11, etc., to detect the state of the processing object or to understand the operating state (operating parameters) of the production equipment 11 with respect to the processing object. For example, the process data may be direct signals or data from the process instrumentation 111 attached to the production equipment 11, or data obtained by processing the acquired signals or data in advance. The process data may also include information from temperature sensors, pressure sensors, flow meters, level meters, pH measuring instruments, concentration meters, density meters, viscometers, etc. In this embodiment, the process data is data obtained without going through the patrol device 12 described later, and is data transmitted from the plant 1 to the sign detection device 3 without going through the patrol device 12.

[0015] On the other hand, the sensing data is data obtained by the patrol device 12 (described later) and transmitted from the plant 1 to the sign detection device 3 via the patrol device 12 (described later). For example, the sensing data represents the status of each piece of production equipment to be inspected. The sensing data is generated based on information sensed by at least one of the five senses (human senses) for the production equipment 11 to be inspected in the plant 1. Furthermore, the sensing data is also information that has not yet been digitized and is then digitized to enable understanding of the status of the production equipment to be inspected. In other words, the sensing data refers to signals or data that are acquired non-contact with the inspection target via a five-sense sensor 124 that replaces human senses (five senses) and output in digital form (sampled and quantized), or data that is transmitted and received non-contact, such as by wireless communication, to acquire signals or data from the inspection target. Furthermore, unlike the process data acquisition method described above, the sensing data can be acquired non-contactly via wireless communication or by accessing the five-sense sensor 124 from a remote location relative to the inspection target. The objects of the sensing include light, sound, temperature, vibration, odor, and the occurrence or presence of specific substances at the inspection target. Specifically, the sensing data includes at least one of visible light, infrared light, and ultraviolet light obtained from the inspection target, sound emitted from the inspection target, the temperature and vibration of the inspection target itself, odor around the inspection target, and the occurrence or presence of specific substances. The data is acquired by measuring these without physically contacting the inspection target with the five senses sensor 124. The orientation and position of the five senses sensor 124 may be changed for each inspection target so that measurements can be taken in close proximity to the inspection target even without contact. The sensing data may also be the signal output by a sensor or data obtained by processing the signal. The five senses sensor may include, for example, at least one of an imaging device (image sensor) that acquires visual information (light) such as an image or video, a microphone (acoustic sensor) that acquires auditory information such as sound, an odor sensor that acquires olfactory information by converting the detection result of an odorant (the occurrence of a specific substance) into an electrical signal, and a tactile sensor that acquires tactile information.The tactile sensor may be an inertial sensor (acceleration sensor, angular velocity sensor, IMU (Inertial Measurement Unit)), force sensor, slip sensor, vibration sensor, temperature sensor, thermal camera, etc. In other words, the sensing data is data obtained through various five senses sensors that are primarily in a state of non-contact, either physically or electrically, with the inspection target. Therefore, the five senses sensor 124 that obtains the sensing data is positioned at a physical and spatial distance from the equipment to be inspected, and monitors the status of the production equipment 11.

[0016] Alternatively, when various five-sense sensors are installed in the production equipment 11, the sensing data may be data transmitted and received by the sign detection device 3 via a patrol device 12 (described later) in a contactless manner, such as wireless communication, when the signals and data from the sensors are acquired by the sign detection device 3. In this case, the various five-sense sensors 124 that acquire the sensing data may be attached to the production equipment 11, and the data and signals emitted therefrom may be transmitted from a physical and spatial distance away from the production equipment 11.

[0017] In particular, in this embodiment, the sensing data may be acquired by acquiring five-sensory data and signals as the five-sensory sensors 124 themselves approach each piece of production equipment 11 (inspection target) fixed within the plant 1. Alternatively, data and signals from the five-sensory sensors 124 attached to each piece of production equipment 11 may be acquired by a mobile receiving device when the sensor approaches the production equipment 11. In such cases, the five-sensory sensors 124 themselves or the receiving device themselves are mobile, and, for example, the five-sensory sensors 124 and receiving device are fixed to the patrol device 12 described below. Data acquired in this manner can also be considered sensing data in this embodiment. In this case, the patrol device 12 patrols a path independent of the path of the processed object described above. The processed object and production equipment 11 are then monitored externally to collect sensing data. In this way, the influence of the processed object during measurement can be eliminated, and measurement flexibility can be increased, such as by stopping or reversing the patrol device 12 for remeasurement.

[0018] The plant 1 in FIG. 1 includes production equipment 11, a patrol device 12, and an access point (AP) 13. The production equipment 11 includes tanks, reactors, piping, and the like for producing chemicals, and corresponds to the inspection target in this embodiment. The production equipment 11 is equipped with general process instrumentation 111 for measuring process data. The patrol device 12 is an example of a mobile means including a processor 121, a storage device 122, a communication interface (IF) 123, a five-sensory sensor 124, and a drive device 125. The patrol device 12 will be described in detail later; the five-sensory sensor 124 acquires the sensing data described above. The AP 13 is a relay device that connects a computer to a network within the system and enables communication. In this embodiment, the patrol device 12 is wirelessly connected to the AP 13 via the communication IF 123 and can communicate with the warning detection device 3.

[0019] Each sensor signal from the process instrumentation 111 fixed to the production equipment 11 of the plant is converted into process data and transmitted to the control station 2. The control station 2 receives the process data from the plant 1 and outputs control signals to the plant 1. Actuators such as valves and other equipment provided in the plant 1 are controlled based on the control signals. The control station 2 may also output the process data obtained from the plant 1 to the symptom detection device 3 or the user terminal 4. The control station 2 may also function as an annunciator that outputs an alarm or the like to the user terminal 4 when, for example, the symptom detection device 3 detects an abnormality or its precursor. At this time, information regarding an operation to resolve the abnormality or its precursor may be output to the user terminal 4 depending on the cause of the detected abnormality or its precursor.

[0020] The sign detection device 3 directly acquires sensing data output by the plant 1 and also acquires process data output by the plant 1 via the control station 2. In this embodiment, the sign detection device 3 primarily uses sensing data among plant data to detect abnormalities or signs of abnormalities in the plant 1 and identify their causes. The sign detection device 3 is equipped with a sign detection model based on a knowledge base that stores correspondences between expected causes and, for example, effects that appear as abnormalities. The sign detection model is a model for identifying signs of abnormalities and their causes by detecting deviations from normal states in changes in sensing data. The sign detection device 3 may extract, for example, actions to suppress the occurrence of an abnormality based on a table that stores information indicating the causes of the abnormality or its signs and actions to take to address them, and the identified causes, and present the extracted actions to a user. Note that a deviation from an expected (desired) operating state or normal operating state in the plant 1 is also referred to as a "modulation."

[0021] The following describes a system that uses sensing data to detect signs of abnormality and identify their causes. FIG. 1 shows a block diagram illustrating an example of the configuration of an abnormality detection device 3. The abnormality detection device 3 is a computer and includes a processor 31, a storage device 32, and a communication interface (I / F) 33. The processor 31 is an arithmetic processing device such as a central processing unit (CPU), a microcontroller unit (MCU), a microprocessing unit (MPU), a digital signal processor (DSP), a field programmable gate array (FPGA), an application specific IC (ASIC), or an application specific standard product (ASSP). The processor 31 performs the processes described in this embodiment by, for example, executing a program. The storage device 32 is a main storage device such as a random access memory (RAM) or a read-only memory (ROM), and an auxiliary storage device (secondary storage device) such as a hard-disk drive (HDD), a solid-state drive (SSD), or a flash memory. The main storage device temporarily stores programs read by the processor 31 and reserves a work area for the processor 31. The auxiliary storage device stores programs executed by the processor 31 and other sensing data. The communication IF 33 is a network module for connecting to the control network of the system 100 and performs communication based on a predetermined protocol.

[0022] The user terminal 4 is a computer installed in an operation room and used by users such as operators (board operators) who operate the plant 1. The user terminal 4 also includes a processor, a storage device, and a communication IF. These components are similar to the processor 31, storage device 32, and communication IF 33 of the sign detection device 3. The user terminal 4 also includes a user interface (UI) such as a display panel, a touch panel stacked thereon, a keyboard, and a pointing device, and outputs information to the user and receives operations from the user via the UI. The user terminal 4 may display sensing data acquired from the control station 2. When the sign detection device 3 detects an abnormality or its precursor, information to that effect is notified to the user terminal 4, and the user terminal 4 outputs a warning or the like as appropriate. The user terminal 4 also transmits a control signal to the control station 2 for controlling the operation of the plant 1 based on the user's operation.

[0023] FIG. 2 is a diagram illustrating the acquisition of sensing data. The plant 1 shown in FIG. 2 includes a production device 11 (corresponding to an inspection target), a patrol device 12 equipped with five-sensory sensors, a patrol route 14, and an identification display 15. The processor 121 of the patrol device 12 is a processing unit such as a CPU, and executes programs to perform the processes described in this embodiment. The processor 121 also controls the drive unit 125 to move the patrol device 12 and, at a predetermined position, stores sensing data converted based on signals acquired by the five-sensory sensors 124 in the storage device 122. The storage device 122 includes, for example, a main storage device and an auxiliary storage device. The communication IF 123 is a communication module for transmitting and receiving data. The communication IF 123 transmits the sensing data to the sign detection device 3, for example, via an AP 13 installed at a predetermined position within the plant 1.

[0024] As described above, the sensing data is data representing the state of the production equipment 11, obtained using various sensors that are not in physical contact with the production equipment 11 (i.e., spatially separated). Such sensing data can be measured by the five-sense sensor 124 mounted on the patrol device 12. Alternatively, the sensing data may further include data received by a receiving device that is not in physical or spatial contact with a sensor arranged in contact with the production equipment 11 (the sensor and the receiving device communicate signals wirelessly). Such sensing data can be collected by a receiving device mounted on the patrol device 12 or a receiving device installed within the plant 1. The processor 121, storage device 122, communication IF 123, five-sense sensor 124, and drive device 125 are fixed to the chassis (main body) of the patrol device 12. This avoids the effects of movement of the patrol device 12 and external vibrations, eliminating noise during data collection. In particular, the fact that the five senses sensor 124 is fixed to the chassis of the patrol device 12 prevents the five senses sensor 124 from shaking due to the influence of wind outdoors, and does not reduce the accuracy of measurement.

[0025] The five-sense sensor 124 may be attached to the chassis of the patrol device 12 in an adjustable orientation and position. For example, if the five-sense sensor 124 is a camera, the lens orientation, or if the sensor is a microphone, the sound collection section, can be adjusted toward the production equipment 11 to be inspected while moving the patrol device 12, or the height position can be adjusted. This allows accurate data to be obtained without being obstructed by the shadow of the equipment. This allows for more accurate detection of abnormalities in the production equipment 11. Furthermore, since only one sensor, such as a camera or microphone, can be used to change its orientation for measurement, the patrol device 12 can be made lighter. Furthermore, since multiple production equipment 11 in the plant 1 can be measured with a minimal number of sensors, there is no need to attach multiple sensors to the plant 1 or the patrol device 12, resulting in a simple system. The orientation of each sensor installed in the patrol device 12 can be changed or adjusted using a pre-programmed program, or it may be adjustable by remote control (manual).

[0026] The five senses sensor 124 is a variety of sensors corresponding to items detected by the five human senses, such as an imaging device, an infrared sensor, a microphone, an odor sensor, a gas detector, and a tactile sensor, and includes at least one of these. In the example of FIG. 2 , the five senses sensor is provided on the mobile patrol device 12, but it may also be fixed within the plant 1. Depending on the type of sensor, some may be mounted on the patrol device 12 and others installed within the plant 1. For example, a tactile sensor such as an acceleration sensor for detecting vibration may be fixed to the production equipment 11 to be inspected, and an imaging device or microphone may be mounted on the patrol device 12. Alternatively, the patrol device 12 may be stopped for each production equipment 11 to be inspected, and a sensor attached to the tip of an extendable arm from the patrol device 12 may be brought into contact with the production equipment 11 for measurement. Furthermore, if the five senses sensor 124 is fixed to the production equipment 11, sensing data may be transmitted to the patrol device 12 or the AP 13 via wireless communication.

[0027] The patrol device 12 may hold the five-sensory sensor 124 so that at least one of its orientation and position can be changed. FIG. 3 is a schematic perspective view showing an example of a holding structure for the five-sensory sensor 124. The five-sensory sensor 124 in FIG. 3 is an imaging device (camera) that acquires image data. For example, the five-sensory sensor 124 is suspended by a holder 127 from a rotating platform 126 rotatably mounted on the lower part of the chassis of the patrol device 12. The five-sensory sensor 124 can be rotated around a vertical axis of rotation by, for example, rotating the rotating platform 126 with a motor. This allows the five-sensory sensor 124 to change its orientation (pan angle) horizontally, for example, 360 degrees, allowing various production equipment 11 present around the patrol route 14 to be measured. The holder 127 may further include a motor 128 for rotating the orientation (tilt angle) of the five-sensory sensor 124 up and down around the horizontal axis of rotation. In this case, it is preferable that the depression angle of the five-sensory sensor 124 be at least 90 degrees (directly above) and the elevation angle be at least 90 degrees (directly below). Furthermore, if the five-sensory sensor 124 is an imaging device, a so-called fisheye lens may be used to capture a wide area, or images captured using two or more lenses may be stitched together. Furthermore, the holding unit 127 extending vertically from the chassis may be equipped with a linear actuator that expands and contracts in the direction of extension. This allows the five-sensory sensor 124 to be brought closer to the production equipment 11 to be inspected or to measure from an angle looking up from below. By holding various five-sensory sensors 124 in a manner that allows at least one of their orientation and height positions to be changed, it becomes possible to measure production equipment 11 located in various directions around the patrol route 14. Furthermore, a composite sensor equipped with multiple types of sensors may share the rotating table 126, holding unit 127, and motor 128.

[0028] The drive device 125 is, for example, a driving motor that controls the rotation of the drive wheels, and causes the patrol device 12 to travel or stop based on the control of the processor 121. The drive device 125 may also include a steering motor for changing the steering angle of the wheels. Note that when the five senses sensor 124 is fixed to the production equipment 11, the five senses sensor 124 does not have a drive device 125.

[0029] The patrol device 12 may have an explosion-proof housing for housing at least some of the above-mentioned components. The explosion-proof housing may also include all of the processor 121, storage device 122, communication IF, five senses sensor 124, and drive device 125. In this way, even if there is a possibility of a flammable substance leaking from the plant 1, the equipment and circuits included in the patrol device 12 will not be affected. The housing structure of the patrol device 12 can be changed as appropriate depending on the type of production equipment 11 included in the plant 1, for example.

[0030] The patrol path 14 is, for example, a rail laid within the plant 1, and the patrol device 12 travels along the patrol path 14. The patrol path 14 is established in advance within the plant 1. Note that in the example of FIG. 2 , the patrol path 14 is an erected rail, and the patrol device 12 is a suspended running body, but this is not limited to this example. For example, the patrol path 14 may be a line drawn on the floor of the plant 1, and the patrol device 12 may be equipped with a sensor for tracing the line, or may be a robot that moves autonomously within the facility while sensing surrounding obstacles. Furthermore, the patrol path 14 may be a predetermined flight path, and the patrol device 12 may be a drone such as a multicopter that flies autonomously. Furthermore, the patrol path 14 may be a unit including a processor, a memory device, communication equipment, and five senses sensors so that it can be worn by a person and used to patrol the plant 1.

[0031] The identification mark 15 is, for example, identification information displayed on the production equipment 11 to be inspected or in its vicinity, and may be displayed on a sign or the like. The identification information is information for identifying the production equipment to be inspected. The identification information may be a predetermined mark, symbol, two-dimensional code, barcode, or the like. The processor 121 of the patrol device 12 detects the presence of the inspection target by image-recognizing the identification information using an imaging device (five-senses sensor 124), and performs measurement using the five-senses sensor 124, for example, by stopping the patrol device 12, or while the patrol device 12 is moving, or by slowing down the speed. For example, the patrol device 12 may be configured to detect the identification mark 15 only at the desired measurement point by adjusting the accuracy of the image recognition or the placement or orientation of the identification mark 15.

[0032] The patrol device 12 may determine that it has reached the desired measurement point based on location information or the like, rather than using the identification display 15. Alternatively, images of the desired measurement point may be machine-learned in advance using images acquired by an imaging device while traveling along the patrol route 14, so that the processor 121 can recognize the measurement point based on the images acquired from the imaging device while the patrol device 12 is traveling. Furthermore, the processor 121 may recognize the measurement point based on the three-dimensional shape of the surrounding area acquired by LiDAR or other means, rather than the imaging device, based on the three-dimensional shape of the production equipment 11 to be inspected, the distance to the production equipment 11, the elapsed traveling time, and the like. Furthermore, the processor 121 may recognize the measurement point by receiving a wireless signal (command) transmitted from a transmitter installed near the production equipment 11 using a receiver mounted on the patrol device 12. The wireless signal method is not particularly limited and may be a beacon signal, short-range wireless communication, or the like. Travel of the patrol device 12 and measurement by the five senses sensor 124 may be performed manually remotely while viewing information acquired from the imaging device. Two or more of the above methods for detecting measurement points can also be implemented in combination.

[0033] FIG. 4 illustrates an example of the flow of data and processing throughout the entire system. The diagram illustrates the patrol device 12, the process data 21 stored in the storage device of the control station 2, the sign detection device 3, the user terminal 4, and the user 5. The processor 31 of the sign detection device 3 executes a program according to the embodiment to function as an acquisition unit 311, an anomaly level calculation unit 312, a sign detection unit 313, a cause diagnosis unit 314, and an output control unit 315. The storage device 32 of the sign detection device 3 also stores causal relationship information 321. The causal relationship information 321 represents the correspondence between the cause of an anomaly and the characteristics of the anomaly level, and will be described in detail later. The example in FIG. 4 illustrates a case in which process data is used in addition to data obtained from the five senses sensors 124 to comprehensively identify signs of anomalies and their causes in the production equipment. The sign detection device 3 is located on a network constituting the system, facilitating the exchange of data from the five senses sensors 124 and process data.

[0034] The processing of signals from the five senses sensor 124 will now be described. The patrol device 12, which patrols the production equipment along the patrol route 14, is equipped with the five senses sensor 124. The five senses sensor 124 monitors the status of the production equipment. The sensing data output by the five senses sensor 124 is transmitted to the sign detection device 3 via the communication IF 123, and in FIG. 4 , an arrow is schematically connected from the five senses sensor 124 to the acquisition unit 311 of the sign detection device 3. Similarly, the acquisition unit 311 also acquires process data 21 from the control station 2.

[0035] Using the acquired sensing data and process data, the anomaly degree calculation unit 312 performs a predetermined calculation according to the production equipment to be measured and the type of data to calculate the anomaly degree. The anomaly degree calculation may be performed using a trained model that has been machine-learned using past normal data and data related to abnormal modulation. For specific phenomena observed in the production equipment, such as temperature rise or abnormal noise, the anomaly degree is calculated based on the sensing data.

[0036] The sign detection unit 313 determines whether there is a trend pattern of the degree of abnormality that matches or is similar to the trend pattern of the degree of abnormality calculated by the abnormality degree calculation unit 312, based on a combination of assumed causes and malfunction events (tendency patterns of the degree of abnormality) that are linked to and accumulated in the causal relationship information 321. Furthermore, the cause diagnosis unit 314 identifies, from the trend pattern of the degree of abnormality that is determined to match or be similar, the assumed cause linked to this as the cause of the malfunction.

[0037] For example, the anomaly degree calculation unit 312 may determine the presence or absence of an anomaly based on the calculated anomaly degrees using a predetermined threshold as a reference, and the sign detection unit 313 may determine whether there is a trend pattern of the anomaly degrees that matches or is similar to the combination of the determination results. Furthermore, if the combination of the calculated anomaly degrees matches or is similar to the trend pattern of the anomaly degrees corresponding to a plurality of assumed causes, the possibility that each of the candidates is the cause of the sign may be calculated and estimated.

[0038] 4, process data is also used to calculate the degree of abnormality and identify the cause of the malfunction. The process data is also input as a signal to the acquisition unit 311, and after undergoing the same processing as the sensing data, is processed by the sign detection unit 313. The sign detection unit 313 comprehensively judges the data from the five senses sensor 124 and the process data 21, and the cause diagnosis unit 314 identifies the occurrence of an abnormality and its cause.

[0039] The output control unit 315 drives the alarm system to issue a warning or notification if the series of processes determines that an abnormality or a sign of an abnormality has occurred in the production equipment. An example of the alarm system is a user terminal 4. The user terminal 4 may be, for example, a screen or terminal device that displays text data, an indicator that displays light, or an audio (speaker) circuit device that speaks specific information. The displayed or notified content is transmitted to the operator, user 5, who then takes action at the production equipment (on-site), such as replacing parts. Furthermore, if new knowledge is gained as a result of the response regarding the malfunction and its cause (causal relationship information 321), this can be further reflected in the causal relationship information 321, thereby broadening the scope of malfunction events and enabling immediate response to any abnormalities that occur. This also improves the accuracy of identifying malfunction events and their causes.

[0040] Note that signs of abnormality in the production equipment and their causes may be identified using only the sensing data obtained from the signals of the five senses sensors 124 (without using the process data 21). In this case, the process data 21 in FIG. 4 is unnecessary. Alternatively, the processor 121 in the patrol device 12 may be equipped with the functions of the sign detection device 3, the storage device 122 may store the causal relationship information 321, and the processor 121 may perform the processing of the acquisition unit 311, the abnormality degree calculation unit 312, the sign detection unit 313, the cause diagnosis unit 314, and the output control unit 315. The processing in each unit is the same as that described above, and therefore a description thereof will be omitted.

[0041] <Precursor Detection Process> FIG. 5 is a process flow diagram illustrating an example of the precursor detection process executed by the system. During operation of the plant 1, the precursor detection device 3 repeats the process illustrated in FIG. 5. In the plant 1, the processor 121 of the patrol device 12 controls the drive device 125 to move the patrol device 12 along a predetermined patrol route 14, acquires sensing data generated by signal conversion from the five senses sensors 124, and stores the data in the storage device 122. The sensing data may be generated in a format such as an image file, a video file, an audio file, information indicating odor intensity, odor quality (presence or absence of a specific odor), presence or absence of a substance such as a specific gas, infrared light intensity, acceleration, angular velocity, temperature, etc. The sensing data is generated, for example, by stopping the patrol device 12 at a predetermined position or after the patrol is completed. The sensing data is recorded in association with information indicating the date and time of generation and the production equipment 11 to be inspected. The sensing data may also be associated with information indicating the location where the patrol device 12 was generated (the location where the patrol device 12 was stopped) or coordinates. Then, the processor 121 transmits the sensing data stored in the storage device 122 to the sign detection device 3 at a predetermined timing.

[0042] The processor 31 of the symptom detection device 3 acquires sensing data via the communication IF 33 (FIG. 5: S1). In this step, the processor 31 reads from the storage device 32 at least one of information indicating, for example, an image file, a video file, an audio file, the intensity of an odor, the presence or absence of a specific odor, acceleration, angular velocity, temperature, etc.

[0043] After S1, the processor 31 calculates a predetermined abnormality level using the sensing data (FIG. 5: S2). In this step, a predetermined calculation is performed depending on the type of sensing data and the location where it was created to calculate the abnormality level.

[0044] When the sensing data is an image file, a video file, a heat map, or the like (i.e., a sensing signal representing light such as visible light or infrared light), a machine learning model (anomaly detection model) using an autoencoder may be used to learn the characteristics of normal image data and calculate the degree of anomaly. FIG. 6 is a diagram illustrating anomaly detection using an autoencoder. In this method, anomaly detection is performed using each pixel constituting an image file (or one frame of a video file) as input data. Specifically, unsupervised machine learning is performed using a neural network, for example, using normal images as training data, to create a model that can compress (encode) and restore (decode) the input data. In a neural network, for example, the number of nodes in the input layer and output layer corresponds to the number of pixels, and the number of nodes in the intermediate layer is smaller than the number of sensors. Information input to the input layer is compressed in the intermediate layer and restored in the output layer. Note that multiple intermediate layers may be present, and the connection structure between layers is not limited to full connection. Then, a learning process is performed using normal image data as training data, and a model is created in which parameters are adjusted so that the difference between the values ​​in the input layer and the values ​​in the output layer is small (the original image can be restored). In the example of FIG. 6 , a model is created in the learning process to obtain an output image that is substantially identical to the input image. Furthermore, in the anomaly determination process, image data, which is sensing data, is input, and the degree of anomaly is calculated based on the difference between the values ​​in the input layer and the values ​​in the output layer. The degree of anomaly can be calculated using L1 distance, L2 distance, SSIM (Structural SIMilarity), or the like. In the example of FIG. 6 , image data showing an oil leak or water leakage from a mechanical seal or gland packing is input in the anomaly determination process. If image data different from normal image data is input, the information compressed in the intermediate layer cannot be properly restored in the output layer. Therefore, the difference between the values ​​in the input layer and the output layer becomes large, and the degree of anomaly is calculated based on this difference. Note that instead of an autoencoder, a one-class SVM may be used to determine the degree of anomaly.Furthermore, the degree of abnormality, which is a one-dimensional index value, may be calculated based on some algorithm that evaluates the magnitude of the difference in hue, saturation, or brightness from the image data in a normal state that serves as a reference.

[0045] When the sensing data is an audio file, for example, data decomposed into frequency components by FFT (Fast Fourier Transform) analysis can be used as input data, and the degree of abnormality can be calculated based on the above-mentioned autoencoder, One Class SVM, or other algorithms. The normal sound of an audio file also differs depending on the location (inspection target) where it was created, and the presence or absence of an abnormality can be determined based on the degree of deviation from the normal sound at each location (inspection target) (i.e., the degree of abnormality).

[0046] Even when the sensing data is information representing acceleration, angular velocity, etc. for vibration detection, the degree of abnormality can be calculated based on the above-mentioned autoencoder, One Class SVM, or other algorithms after, for example, performing predetermined preprocessing. Information representing acceleration, etc., also has different normal characteristics depending on the location (inspection target) where it was created, and the presence or absence of an abnormality can be determined based on the degree of deviation from the normal characteristics at each location (inspection target) (i.e., the degree of abnormality).

[0047] If the sensing data is information representing odor intensity or odor quality (in other words, the occurrence of a specific substance) and its strength, for example, the features may be learned by unsupervised learning using principal component analysis using normal data as training data. In this case, the degree of anomaly of the sensing data can be evaluated based on its distance from the distribution of the features of the training data. If the five senses sensor 124 is an odor sensor that can detect odor quality in addition to odor intensity, the distribution of the features of the training data can be learned using information representing odor quality as a variable. Furthermore, the degree of anomaly may be calculated based on the above-mentioned autoencoder, One Class SVM, or other algorithms, or the odor intensity or temperature itself may be used as the degree of anomaly. The normal characteristics of information representing odor intensity, quality, etc. also vary depending on the location (inspection target) where it was created, and the presence or absence of an anomaly can be determined based on the degree of deviation from the normal characteristics (i.e., the degree of anomaly) at each location (inspection target).

[0048] After S2, the processor 31 determines whether the combination of the above-mentioned abnormality levels corresponds to a predetermined pattern that indicates an abnormality or a sign of an abnormality (FIG. 5: S3). In this step, the processor 31 determines whether each piece of sensing data indicates an abnormality, for example, based on the abnormality level calculated in S2 and a predetermined threshold value for each piece of sensing data. The processor 31 also determines whether the determination result or the combination thereof for one or more pieces of sensing data matches a predetermined pattern.

[0049] FIG. 7 is a diagram illustrating an example of a judgment logic. The table in FIG. 7 includes attributes for "assumed cause" and "judgment logic," and is causal relationship information that represents the correspondence between the cause of an abnormality and the characteristics of the abnormality level. The "assumed cause" field stores information indicating the cause of an abnormality or its precursor. In the example of FIG. 7, information indicating causes such as "deterioration of pump internal components," "deterioration of blower internal components," "oil leak," "mechanical seal or gland packing malfunction," "pump motor deterioration," etc. are registered. The "judgment logic" also includes attributes such as "leak," "abnormal noise," "odor," "vibration," "temperature," etc. Each item in the "judgment logic" corresponds to a type of sensing data. Each field stores information indicating the presence or absence of an abnormality. An abnormality is determined to exist when the above-described abnormality level exceeds a threshold value determined for each type of sensing data and the location where the sensing data was created. Furthermore, based on the combination patterns of the presence or absence of an abnormality registered in the judgment logic, an abnormality or its precursor caused by the assumed cause of the corresponding record can be detected. The storage device 32 may further store information indicating the action that the operator should take to resolve the problem, in association with the assumed cause.

[0050] The "Leak" field stores information indicating the presence or absence of a liquid leak, determined based on visual information such as an image file. The "Abnormal Sound" field stores information indicating the presence or absence of an abnormal sound, determined based on auditory information such as an audio file. Note that information such as "pump" and "motor" shown in parentheses indicates the location where the audio file was created (the object of inspection). The "Odor" field stores information indicating the presence or absence of an odor, determined based on olfactory information such as odor intensity. Note that data having predetermined characteristics related to the quality of the odor, such as "process fluid" shown in parentheses, may be extracted, and a determination may be made based on the extracted data as to whether an odor of a strength equal to or greater than a threshold has been detected. The "Vibration" field stores information indicating the presence or absence of vibration. For example, if the degree of abnormality based on acceleration information exceeds a predetermined threshold, it is determined that vibration exists. The "Temperature" field stores information indicating the presence or absence of a temperature abnormality. For example, if the degree of abnormality based on temperature information exceeds a predetermined threshold, it is determined that there is a temperature abnormality.

[0051] The decision logic (causal relationship information) shown in FIG. 7 is an example of a knowledge base that defines the combination of causal relationships between causal events and the resulting effects, implemented in a data structure suitable for this embodiment. Such a knowledge base can be created, for example, based on HAZOP (Hazard and Operability Study). HAZOP can comprehensively correlate and list, for example, check items on a field operator's patrol route, the control range (tolerance range) for the check items, a list of assumed causes of deviations from the control range, decision logic for determining which assumed cause caused the deviation, the impact of the deviation, and countermeasures to be taken when the deviation occurs. The knowledge base can also be created based on methods other than HAZOP, such as Fault Tree Analysis (FTA), Failure Mode and Effect Analysis (FMEA), Event Tree Analysis (ETA), or methods based on these or similar methods, information extracted from interviews with operators, or information extracted from work standards or technical standards. Creating the decision logic in this way eliminates personal and experience-based differences in inspection work, enabling objective judgments. It also improves the accuracy of cause estimation based on a combination of factors. If new knowledge about causal relationships is obtained during the operation of this system, the knowledge base and the decision logic created based on it may be updated as appropriate.

[0052] If it is determined in S3 that the combination of events indicating modulation and abnormality levels matches a predetermined pattern (S3: YES), the processor 31 outputs a warning (FIG. 5: S4). In this step, the processor 31 causes an annunciator on the monitoring system to be issued from the control station 2 to the user terminal 4, for example. The warning preferably also includes information indicating the probable cause. This helps the user identify the cause, allowing the user to quickly take appropriate action according to the cause.

[0053] After S4, or if it is determined in S3 that the predetermined pattern does not match (S3: NO), the process in Fig. 5 is terminated. Note that the process in Fig. 5 is executed repeatedly while the plant 1 is in operation.

[0054] <Effects> If anomaly detection is performed based on process data, for example, when an abnormality appears in the flow rate of a liquid being pumped through a pipe, damage to the equipment is often already occurring. For example, the flow rate may be reduced due to damage to the bearings of the pump (motor), causing the motor's rotation speed to drop. Furthermore, if the motor shaft is about to seize, it may not be possible to restore the equipment by simply replacing the part, and the motor itself may have to be replaced. This increases the time required to restore the equipment, including the time required to arrange for a motor, as well as the repair costs and labor.

[0055] Using the sensing data obtained by the five senses sensor 124, for example, if a malfunction occurs in a pump that moves liquid through a pipe, abnormal noise and vibration due to malfunctioning motor bearings and a rise in temperature around the motor shaft can be detected first. In this case, the motor's rotation speed may not be significantly affected. That is, the malfunction and its cause can be detected early based on the combination of abnormalities such as sounds in unusual frequency bands and a tendency for the shaft to be too hot. Such modulation is difficult to detect using a flow meter (process data) attached to the pipe. Simply replacing the motor bearings can reduce the time, cost, and labor required for repair. Therefore, the ability to detect signs of abnormalities in production equipment (production devices 11) can be improved. As described above, according to the above-described embodiment, signs of motor malfunction can be detected before modulation appears in the process data, while the degree of modulation in the process data is small, or while the number of items indicating modulation in the process data is small. Note that while the determination logic shown in FIG. 7 includes a combination of multiple types of sensing data, an abnormality or its sign may be detected based on a single type of sensing data. However, by using multiple types of sensing data, it is possible to improve the performance of identifying the cause, for example.

[0056] <Modification> In S3 and S4 of Fig. 5, the processor 31 may extract and output a suspected cause that matches only a part of the pattern (in other words, that has a high degree of match), not just a complete match with the pattern shown in Fig. 7. In this case, if it is not possible to narrow down the suspected cause to one, multiple possible suspected causes may be output. In this way, the user can be alerted early.

[0057] Furthermore, if multiple suspected causes are extracted, information other than sensing data may also be used to narrow down the suspected causes. Furthermore, multiple suspected causes may be displayed in descending order of likelihood, for example, based on the degree of match of the patterns shown in FIG. 7 . For example, information representing the repair history of the production equipment 11 may be further stored in the storage device 32 of the symptom detection device 3, and the suspected causes may be ranked based on the timing of part replacement. In other words, the system may include a step of determining that the more recently replaced a part is considered to be the cause, the less likely that part is the cause of the malfunction.

[0058] The determination logic shown in FIG. 7 may identify the cause of an anomaly by combining not only the sensing data from the five senses sensor 124 but also the presence or absence of an anomaly in the process data. Furthermore, the image data described above may indirectly represent process data, such as an image of a level gauge display, such as an oil gauge. For level gauge images, an anomaly detection model can be created using supervised machine learning, for example, using images of an anomaly. FIG. 8 illustrates an example of determination logic including process data. The table in FIG. 8 also shows causal relationship information including the attributes "assumed cause" and "determination logic." Furthermore, the "determination logic" further includes the attributes "flow rate," "discharge pressure," "oil gauge," "leak," "abnormal noise," "odor," "vibration," "temperature," and so on. The items corresponding to those in FIG. 7 are not described here. "Flow rate" and "discharge pressure" correspond to process data. Furthermore, the "oil gauge" corresponds to the type of information inspected by a field operator during rounds of inspection. In the example shown in FIG. 7, the determination logic includes a confirmation that there is no change in the process data. This allows for distinguishing between the stage after an anomaly has occurred and the stage before an anomaly occurs. In other words, an alarm can be output based only on the degree of anomaly in the sensing data before an abnormality occurs in the process data. Note that an oil gauge may also be included in the process data. In this case, an alarm can be output while the number of process data items indicating an abnormality is smaller than the number of sensing data items indicating an abnormality. In either case, repairs or other measures can be taken while the impact on the production equipment 11 is still small.

[0059] FIG. 9 illustrates an example of logic for predicting a deterioration in heat exchanger performance. The heat exchanger is configured to transfer heat between a process fluid and a utility fluid. The heat exchanger is, for example, a shell-and-tube heat exchanger, in which the process fluid flows through the tubes and the utility fluid flows through the shell, with heat being transferred via the tube walls. In the example of FIG. 9 , the following potential causes are registered as predictive patterns: "heat exchanger fouling (process)," "heat exchanger fouling (utility)," "heat exchanger blockage (process)," "heat exchanger blockage (utility)," "heat exchanger internal leakage," "heat exchanger external leakage (process)," "heat exchanger external leakage (utility)," and "thermal insulation deterioration." A heat exchanger internal leakage is the leakage of utility fluid (e.g., cooling water or steam) into the process side, which refers to the mixing of the heat medium used in the heat exchanger, reactor jacket or coil, or other heat exchange equipment into the process fluid. If contamination occurs, it may cause various abnormalities such as abnormal reactions, heat generation, and clogging. However, according to this embodiment, it is possible to detect the signs early. In the example of FIG. 9, the cause of the abnormality can be narrowed down with high accuracy based on the presence or absence of a change in the "flow rate" corresponding to the process data, the presence or absence of a "leak" and the presence or absence of an "odor" determined based on the sensing data, etc. In other words, the signs of an abnormality are detected by further combining the sensing data with the degree of anomaly in the process data. Note that if it is not possible to narrow down the suspected cause to one, multiple candidate suspected causes may be output to the user in order of likelihood.

[0060] Figure 10 is a diagram showing an example of judgment logic for predicting a decline in the capacity of a distillation column. Note that a description of items corresponding to the judgment logic described above will be omitted. In the example of Figure 10, in addition to the example of Figure 9, a record for the assumed cause "leakage" has been added. The presence or absence of an abnormality in a leak can be determined based on the presence or absence of a leak detected by the five senses sensor 124 and the presence or absence of an odor (the quality of the odor).

[0061] FIG. 11 is a diagram illustrating an example of logic for determining a sign of a decline in ejector performance. Items corresponding to the above-described determination logic will not be described. In the example of FIG. 11, in addition to the example of FIG. 10, the attributes "thermometer (utility)" and "pressure gauge (process)" are further included. Records for the assumed causes "decrease in driving steam pressure" and "ejector contamination" are also added. The ejector is, for example, a steam ejector equipped with a Venturi tube nozzle and used to create a vacuum within the system. Signs of "decrease in driving steam pressure" and "ejector contamination" can be determined based on information similar to process data, such as a thermometer or pressure gauge. The thermometer and pressure gauge may be configured such that the scale or gauge is photographed by an imaging device, which is the five senses sensor 124, and input as image data.

[0062] Unlike patrols by field operators, inspections of production equipment 11 using the patrol device 12 can be performed frequently. For example, the system can acquire the above-mentioned data for each production equipment 11 in the morning (first time), afternoon (second time), and night (third time). At this time, the system can focus on the attribute data indicated by each production equipment 11 and comprehensively assess how the data changes depending on the number of measurements to predict abnormalities and identify their causes. For example, if the first and third measurements are within the normal range and a change is observed only in the second measurement, it can be determined that, even if there is some cause, this is a coincidental phenomenon and does not have a significant impact on the production equipment. Conversely, if such changes occur repeatedly, it is possible that some abnormality is occurring. Another feature of the system of the present invention is that it can easily comprehensively identify abnormalities and their causes by understanding changes in data acquired at different times for the same attribute. Furthermore, increasing the measurement frequency can improve the accuracy of the assessment.

[0063] The threshold for determining whether or not an abnormality exists based on the degree of abnormality may be changed depending on the season or weather. For example, if the normal temperature range of the production equipment 11 varies depending on the season, the threshold may be changed accordingly. Furthermore, for audio files acquired during rainy weather, the threshold may be changed depending on the amount of noise. The season and weather may be determined based on a calendar or clock provided in the warning sign detection device 3 or information obtained from other devices, or environmental information such as outside temperature and humidity may be acquired from a specified sensor and used to make the determination. The patrol route 14 may be a dedicated route traveled only by the patrol device 12. In this case, the patrol device 12 can be operated at any time. If a measurement error occurs, the patrol device 12 can be returned to the location of the error at any time and remeasured. The time required for remeasurement can also be set arbitrarily.

[0064] Furthermore, the threshold value for determining the presence or absence of an abnormality based on the degree of abnormality may be changed depending on the operating state of the plant 1, such as steady operation or non-steady operation. Non-steady operation includes, for example, the period from equipment startup until operation stabilizes, operation before equipment shutdown, and some kind of irregular operation. For example, when the production volume per hour in the plant 1 is high or when the system is starting up, the load on the pump is high, and even if the pump temperature or vibration is relatively high, it may still be normal. In such cases, an algorithm for calculating the degree of abnormality may be created for each operating state and used separately for each operating state. Furthermore, the operating state can be determined based on a flag by appropriately controlling a flag representing the operating state.

[0065] The processor 31 may monitor the accumulation status of sensing data obtained from the patrol device 12, and if data is missing for a certain time period or location (inspection target), it may instruct the patrol device 12 to recreate the data or notify the user terminal 4. When notifying the user terminal 4, for example, the operation of the patrol device 12 may be switched to manual operation, and the user terminal 4 may instruct the patrol device 12 to operate based on the operator's operation. The operator may stop the patrol device 12 at a desired position and execute one or more data acquisitions, and confirm that the data acquisition was successful based on, for example, a comparison with tolerances or past data. Furthermore, the processor 121 of the patrol device 12 may detect missing data and determine whether or not it needs to be recreated.

[0066] The patrol device 12 may record sensing data while moving, but by stopping to acquire sensing data, noise due to movement can be reduced and the sensing data acquisition position can be standardized to improve detection accuracy. The patrol device 12 may also be capable of autonomous travel, or may operate based on user operation via the user terminal 4, for example.

[0067] At least some of the functions of the sign detection device 3 may be distributed across multiple devices (in other words, multiple processors), or the same function may be provided by multiple devices (in other words, multiple processors) operating in parallel. Furthermore, one sign detection device 3 may be configured to include multiple processors. That is, a system may be provided that includes one or more computers (or processors) for executing the above-described processing. Furthermore, at least some of the functions of the sign detection device 3 may be provided on a so-called cloud. Furthermore, the patrol device 12 may calculate the degree of anomaly and send it to the sign detection device 3, or the patrol device 12 may even identify the cause of the sign and send the result to a user terminal 4, etc.

[0068] The present disclosure also includes a method and a computer program for executing the above-described process, and a computer-readable recording medium having the program recorded thereon. The recording medium having the program recorded thereon enables the above-described process by causing a computer to execute the program.

[0069] Here, a computer-readable recording medium refers to a recording medium that stores information such as data and programs electrically, magnetically, optically, mechanically, or chemically and can be read by a computer. Among such recording media, those that can be removed from a computer include flexible disks, magneto-optical disks, optical disks, magnetic tapes, memory cards, etc. Furthermore, recording media that are fixed to a computer include HDDs, SSDs, ROMs, etc.

[0070] The configurations and combinations thereof in each embodiment and modified example are merely examples, and additions, omissions, substitutions, and other modifications of the configurations are possible as appropriate without departing from the spirit of the present disclosure. The present disclosure is not limited by the embodiments, but only by the scope of the claims. Furthermore, each aspect disclosed in this specification can be combined with any other feature disclosed in this specification.

[0071] 100: System 1: Plant, 11: Production equipment, 12: Patrol device, 121: Processor, 122: Storage device, 123: Communication interface, 124: Five senses sensor, 125: Drive device, 13: Access point, 14: Patrol route 2: Control station 3: Prediction detection device, 31: Processor, 32: Storage device, 33: Communication interface 4: User terminal

Claims

1. A symptom detection system including one or more computers that execute the following: acquiring, in a non-contact manner, sensing data representing at least one of light, sound, temperature, vibration, odor, and the generation of a specific substance from production equipment including an inspection target; calculating the degree of abnormality for the acquired sensing data; detecting a sign of the abnormality and its cause based on causal relationship information representing the correspondence between the cause of the abnormality and the characteristics of the degree of abnormality, and the degree of abnormality corresponding to the acquired sensing data; and outputting information indicating the detected sign of the abnormality and its cause.

2. The precursor detection system of claim 1, wherein there are two or more types of the acquired sensing data, the degree of abnormality is calculated for the two or more types of the sensing data, and the causal relationship information represents the correspondence between the cause of the abnormality and the characteristics of the degree of abnormality based on each of the two or more types of the sensing data.

3. The symptom detection system according to claim 1 or 2, wherein the production facility is a chemical plant, and the sensing data is data different from process data measured on the object to be processed in the chemical plant.

4. The symptom detection system according to claim 3, wherein the one or more computers further acquire the process data, calculate an abnormality level for the process data, and detect the signs of the abnormality and their causes based on a combination of the anomaly levels of the process data and the anomaly levels of two or more types of the sensing data for the inspection target of the chemical plant.

5. The precursor detection system according to claim 1 or 2, wherein there are a plurality of inspection targets in the production equipment, and the system further comprises a means of transportation that is equipped with a sensor for acquiring the sensing data and that can move within the area of ​​the production equipment in accordance with a pre-set program, and the sensor acquires the sensing data from each of the plurality of inspection targets.

6. The symptom detection system according to claim 5, wherein the moving means comprises a holding mechanism that holds the sensor so that its orientation or position can be changed.

7. The precursor detection system according to claim 5, wherein an identification mark indicating that the production equipment is subject to inspection is provided on or around the production equipment, and the moving means starts acquiring the sensing data by detecting the identification mark using image recognition.

8. The symptom detection system according to claim 5, wherein, if the one or more computers are unable to acquire the sensing data from the sensor, they request the means of transportation to recreate the sensing data or notify the user's terminal.

9. The symptom detection system according to claim 1 or 2, wherein the one or more computers acquire two or more types of sensing data and detect multiple combinations of the signs of the abnormality and their causes based on the degree of abnormality corresponding to each of the acquired sensing data.

10. The symptom detection system according to claim 9, wherein the one or more computers output combinations of multiple detected symptoms of anomalies and their causes in order of likelihood based on the degree of agreement between the characteristics of the degree of anomaly represented by the causal relationship information and the degree of anomaly corresponding to each of the acquired sensing data.

11. The symptom detection system described in claim 9, wherein the one or more computers retain information representing the repair history of the production equipment, and when multiple detected signs of an abnormality are combined with their causes, the more recently a part was repaired, the less likely the defect is caused by that part.

12. A sign detection method in which one or more computers perform the following steps: acquire sensing data including at least one of light, sound, temperature, vibration, odor, and generation of a specific substance in a non-contact manner from production equipment including an inspection target; calculate the degree of abnormality for the acquired sensing data; detect signs of the abnormality and their causes based on causal relationship information that indicates the correspondence between the cause of the abnormality and the characteristics of the degree of abnormality, and the degree of abnormality corresponding to the acquired sensing data; and output information indicating the detected signs of the abnormality and their causes.

13. A symptom detection program that causes one or more computers to execute the following steps: acquiring sensing data representing at least one of light, sound, temperature, vibration, odor, and the generation of a specific substance in a non-contact manner from production equipment including an inspection target; calculating the degree of abnormality for the acquired sensing data; detecting signs of the abnormality and their causes based on causal relationship information representing the correspondence between the cause of the abnormality and the characteristics of the degree of abnormality, and the degree of abnormality corresponding to the acquired sensing data; and outputting information indicating the detected signs of the abnormality and their causes.

14. A sign detection method comprising: a sensing data acquisition step of acquiring, in a non-contact manner, sensing data indicating at least one of light, sound, temperature, vibration, odor, and the generation of a specific substance from production equipment including an inspection target; a calculation step of calculating an abnormality level for the acquired sensing data; a detection step of detecting a sign of the abnormality and its cause based on causal relationship information indicating the correspondence between the cause of the abnormality and multiple abnormality level characteristics, and the abnormality level corresponding to the acquired sensing data; an output step of outputting information indicating the detected sign of the abnormality and its cause; a process data acquisition step of further acquiring process data different from the sensing data, which is measured on an object to be processed in the production equipment; and a second calculation step of calculating an abnormality level for the process data, wherein the output step is performed while the number of items of the process data indicating an abnormality is fewer than the number of items of the sensing data indicating an abnormality.

Images