State monitoring apparatus and method

Abstract

In selecting a sensor set, when the selection is based on engineer's knowledge, it is not possible to select a sensor set that addresses a number of problems due to constraints on man-hours and knowledge. Alternatively, when a sensor set is selected on the basis of a correlation between sensor data and a problem, the regularity of the data when equipment is in an abnormal state is lost, and it could therefore be difficult to detect a correlation with the problem, and a large amount of prior data is required.　It is necessary to select a sensor set ad hoc that addresses a problem set by a user as an analysis target, on the basis of a knowledge database pertaining to the structure and failures of equipment, and to present the sensor data to the user.

Description

Condition Monitoring Device and Method

[0001] The present invention relates to a technique for monitoring the condition of an asset including equipment.

[0002] In many fields such as infrastructure, railways, industrial equipment, and medical equipment, it is necessary to monitor the condition of an asset after its introduction. Based on the condition of the target asset collected through this condition monitoring, a diagnosis for analyzing the presence or absence of abnormalities and problems is applied, and then an appropriate maintenance operation is applied. As a result, by continuously performing maintenance on the equipment, a predetermined performance can be maintained.

[0003] In asset diagnosis, based on the results of the investigation of the target asset by the maintenance worker and the condition of the equipment obtained by sensors, and by integrating the worker's knowledge of the target asset and general engineering knowledge, the failure location is gradually narrowed down, and finally, it is determined whether a failure has occurred and which equipment should be subjected to measures such as replacement, repair, and restoration. This is a common method.

[0004] In particular, due to the recent development of information technology, it has become easier to use IT technologies such as diagnostic technologies that collect the condition of an asset with sensors and grasp the current condition of the asset, and diagnostic support using such technologies is being studied.

[0005] To realize such a condition monitoring system, it is necessary to extract the appropriate sensors related to each problem that occurs and to display the data and provide analysis support to the user. For this purpose, for example, Patent Document 1 has been proposed. Patent Document 1 aims to "facilitate the generation of a model for monitoring equipment within a facility." To solve this problem, Patent Document 1 discloses a configuration that "includes the steps of acquiring measurement data from each of the multiple sensors that monitor each of the multiple pieces of equipment (S1815), detecting an abnormality in the equipment based on the measurement data (S1815), selecting the first measurement data with the highest correlation to the abnormality from among the measurement data as the objective variable (S1820), selecting multiple explanatory variables from among the measurement data excluding the first measurement data (S1825, S1830, S1835), and inputting the objective variable, the multiple explanatory variables, and other set values ​​into a learning engine (S1840)."

[0006] Japanese Patent Publication No. 2022-103931

[0007] To implement a condition monitoring system, it is necessary to build a set of sensors that can address potential problems and, when a problem occurs, to selectively present the user with data from the relevant sensors. However, the selection of sensors to address a problem is primarily done by skilled engineers based on their knowledge and experience of the equipment. This results in a large development time and limits the number of sensor sets that can be created in advance.

[0008] Furthermore, there are cases where we need to investigate problems that have never occurred before or problems where the sensor set has not been prepared in advance. However, in such cases, selecting the sensor set to be checked on an ad-hoc basis and conducting the investigation requires advanced equipment and assets, as well as knowledge of data analysis, and there are few engineers capable of doing so.

[0009] Furthermore, the appropriateness of sensor selection depends on the engineer's skill. Another approach is a data-driven method that specifies a set of sensors based on the correlation between the occurrence of a problem event and past data at the time of the problem. However, failure events are generally rare, and it may not be possible to collect enough data for sufficient statistical processing. Alternatively, the behavior of equipment after a failure may lose its regularity, making it impossible to detect a correlation between the problem and the data. Conversely, sensors that are not important and have no causal relationship may be extracted due to abnormal data after the failure.

[0010] Therefore, the objective of the present invention is to enable efficient problem analysis by allowing the identification of a more appropriate sensor set to address the problem.

[0011] To this end, the present invention uses knowledge data about the asset to select a sensor set corresponding to an operational problem in the target asset. More preferably, candidate sensor sets related to the operational problem are extracted, the relationship between the candidate sensor sets and the operational problem is evaluated, and a sensor set suitable for the operational problem is selected from the candidate sensor sets according to the evaluation result. Specifically, the present invention relates to a condition monitoring device for monitoring the state of a target asset, and includes an asset knowledge storage unit that stores asset knowledge data about the target asset, a sensor set extraction unit that extracts candidate sensor sets related to an analysis target which is an operational problem of the target asset, and a sensor set selection unit that evaluates the relationship between the analysis target and the extracted candidate sensor sets and selects a sensor set according to the evaluation result.

[0012] Furthermore, the present invention also includes a status monitoring device or a status monitoring system including a status monitoring device, a status monitoring program that enables the status monitoring device to function with a computer, and a storage medium that stores this status monitoring program, as well as a status monitoring method executed by the status monitoring device or a status monitoring system including a status monitoring device, and a status monitoring program that stores this status monitoring program.

[0013] By using this invention, it becomes possible to select a more appropriate sensor set, enabling efficient problem analysis. As a result, more appropriate maintenance operations can be carried out, improving the reliability and availability of equipment.

[0014] This is a functional block diagram showing the configuration of the condition monitoring system 100 in one embodiment of the present invention. This is a hardware configuration diagram showing an implementation example of the condition monitoring system 100 in one embodiment of the present invention. This is a flowchart showing the processing flow in one embodiment of the present invention. This is a flowchart showing the analysis target setting process in Example 1. This is a diagram showing an example of asset knowledge data in Example 1. This is a diagram showing an example of data showing the structure of the equipment in Example 1. This is a diagram showing an example of asset knowledge graph 3001 in Example 1. This is a diagram showing an example of sensor data 6001 in Example 1. This is a diagram showing a larger example of failure knowledge data which is asset knowledge graph 3001 in Example 1. This is a flowchart showing the sensor set extraction process in Example 1. This is a diagram showing an example of the display content of the sensor set selection results on the analysis dashboard 1000 in Example 1.

[0015] In this embodiment, a sensor set for monitoring the status of equipment is selected as the target asset. Therefore, in this embodiment, the sensor set is automatically or semi-automatically formulated based on a knowledge base constructed by aggregating the characteristics of the equipment. This reduces the man-hours required to construct the sensor set, improves coverage, and enables problem analysis of the equipment without relying on highly uncertain data analysis. The assets of this invention include not only equipment, but also parts, assemblies, modules, devices, equipment, systems, etc. Operational problems include asset failures, malfunctions, defects, and deterioration. Furthermore, the knowledge base regarding the asset, i.e., equipment, includes its structure (including configurations), functions, and problems. Additionally, the sensor set in this embodiment also includes individual sensors or single-type sensors.

[0016] Furthermore, this embodiment assists in identifying the location of a failure in the equipment. To this end, this embodiment uses an asset knowledge graph that records knowledge such as the structure and failures of the target asset, a sensor set extraction unit that extracts sensor sets related to failures and operational problems, a sensor set selection unit that evaluates the relationship between the problem and the extracted sensor sets to determine the sensor sets, an analysis dashboard that receives input of analysis targets from the user when a problem occurs and presents data of the selected sensor sets to the user, and a sensor DB that records sensor information. The specific configuration and processing flow of this embodiment will be described below with reference to Figures 1A to 1C.

[0017] Figure 1A is a functional block diagram showing the configuration of the status monitoring system 100 in this embodiment. In Figure 1A, the status monitoring system 100 consists of an analysis dashboard 1000, an analysis target setting unit 2000, an asset knowledge storage unit 3000, a sensor set extraction unit 4000, a sensor set selection unit 5000, a sensor data storage unit 6000, a sensor function supplementation unit 7000, and an asset knowledge update unit 8000. The status monitoring system 100 is used by a user 200. Each part will be described below. Although this embodiment is not limited to specific equipment, diagnostic technology, or analysis technology, the following explanation will use the diagnosis of a railway passenger door as an example.

[0018] First, the analysis dashboard 1000 receives input of analysis targets from the user 200 and presents data from the selected sensor set to the user 200. In this way, the analysis dashboard 1000 functions as an input / output unit. The analysis target setting unit 2000 sets analysis targets for problems such as equipment failures.

[0019] Furthermore, the asset knowledge storage unit 3000 stores asset knowledge data such as asset knowledge graphs that show knowledge about the structure and problems of the equipment. In addition, the sensor set extraction unit 4000 extracts sensor sets related to the analysis targets and problems set in the analysis target setting unit 2000.

[0020] Furthermore, the sensor set selection unit 5000 evaluates the relationship between the sensor sets extracted by the sensor set extraction unit 4000 (also referred to as sensor set candidates) and operational problems, and selects a sensor set from the sensor set candidates according to the evaluation results. In addition, the sensor data storage unit 6000 stores sensor data 6001 related to the sensors that constitute the sensor set.

[0021] Furthermore, the sensor function complementation unit 7000 complements the sensor's function using asset knowledge data such as asset knowledge graphs. The asset knowledge update unit 8000 updates asset knowledge data such as asset knowledge graphs in response to input from the user 200 to the analysis dashboard 1000. The above status monitoring system 100 can be implemented using a computer system. An example of its implementation is described below.

[0022] Figure 1B is a hardware configuration diagram showing one implementation example of the status monitoring system 100 in this embodiment. In Figure 1B, the status monitoring system 100 is configured such that the status monitoring device 10 and the terminal device 201 are connected via a network 500. The network 500 is further connected to sensor sets for each piece of equipment (equipment A300A, equipment B300B).

[0023] First, the terminal device 201 has a configuration corresponding to the analysis dashboard 1000 in Figure 1A, and is a computer operated by the user 200. The terminal device 201 can be implemented using a PC, tablet, or smartphone.

[0024] Furthermore, the status monitoring device 10 can be implemented using a computer such as a server. For this reason, the status monitoring device 10 includes a processing unit 101, a communication device 102, a main memory 103, and a sub-memory 104, which are connected to each other via a communication path.

[0025] First, the processing unit 101 can be implemented with a processor such as a CPU and performs calculations according to the state monitoring program 105 stored in the sub-memory 104, which will be described later. The state monitoring program 105 will be described later. The communication device 102 connects to the terminal device 201 and the sensor set (401A to 403A, 401B to 403B) via the network 500.

[0026] Furthermore, the main memory 103 is where the status monitoring program 105 stored in the secondary memory 104 and the information used for processing in the processing unit 101 are unpacked. The secondary memory 104 can be implemented as so-called storage. The secondary memory 104 stores the status monitoring program 105, the asset knowledge graph 3001, and the sensor data 6001. The asset knowledge graph 3001 is asset knowledge data, which is knowledge. In Figure 1B, only the asset knowledge graph 3001 is shown, but other data is also stored. In addition to being implemented as a storage device within the status monitoring device 10, the secondary memory 104 may be implemented as various storage media such as an external HDD (Hard Disk Drive), SSD (Solid State Drive), or memory card, or it may be implemented as a separate device from the status monitoring device 10, such as a database system or file server. The secondary memory 104 corresponds to the asset knowledge storage unit 3000 and the sensor data storage unit 6000 in Figure 1A.

[0027] Furthermore, the status monitoring program 105 consists of an analysis target setting module 106, an asset knowledge update module 107, a sensor function supplementation module 108, a sensor set extraction module 109, and a sensor set selection module 110. Note that each of these modules may be implemented as an individual program or as a combination of some of them.

[0028] Furthermore, the configuration shown in Figure 1A, which performs the same function as each module, is as follows: Analysis target setting module 106: Analysis target setting unit 2000 Asset knowledge update module 107: Asset knowledge update unit 8000 Sensor function supplementation module 108: Sensor function supplementation unit 7000 Sensor set extraction module 109: Sensor set extraction unit 4000 Sensor set selection module 110: Sensor set selection unit 5000 Therefore, the processing unit 101 will execute the processing of the analysis target setting unit 2000, the sensor set extraction unit 4000, the sensor set selection unit 5000, the sensor function supplementation unit 7000, and the asset knowledge update unit 8000 according to the status monitoring program 105.

[0029] Furthermore, in Figure 1B, multiple pieces of equipment, specifically equipment A300A and equipment B300B, are designated as assets to be monitored. Therefore, sensor sets 401A-403A and 401B-403B are provided for each of these. The condition monitoring device 10 then selects the appropriate sensor set from 401A-403A and 401B-403B depending on the problem with equipment A300A and equipment B300B. Note that in Figure 1A, descriptions corresponding to equipment A300A, equipment B300B, and sensor sets 401A-403A and 401B-403B are omitted, but the condition monitoring system 100 will have connection points to connect to these.

[0030] Next, the processing flow in this embodiment will be explained, mainly with reference to the configuration in Figure 1A. Figure 1C is a flowchart of the processing flow in this embodiment. In step S20, the analysis target setting unit 2000 executes the analysis target setting process. To do this, first, it reads the asset knowledge graph 3001 from the asset knowledge storage unit 3000. Then, the analysis target setting unit 2000 uses the read asset knowledge graph 3001 to set analysis targets for problems such as failures in the target equipment.

[0031] Furthermore, in step S40, the sensor set extraction unit 4000 performs a sensor set extraction process. That is, the sensor set extraction unit 4000 extracts candidate sensor sets that are related to the analysis target set in step S20. Then, in step S50, the sensor set selection unit 5000 performs a sensor set selection process. For this purpose, the sensor set selection unit 5000 evaluates the relationship between the candidate sensor sets extracted in step S40 and the operational problems of the target equipment. Then, the sensor set selection unit 5000 selects a sensor set from the candidate sensor sets according to the evaluation result. Here, the selected sensor set is also referred to as the selected sensor set.

[0032] Then, using the sensing data from the selected sensor set, it becomes possible to monitor the status of the target equipment. For this purpose, a status monitoring unit may be provided in the status monitoring system 100. This status monitoring unit may be implemented as a separate device from the status monitoring device 10 in Figure 1B. Furthermore, a status monitoring module may be provided in the status monitoring program 105 of the status monitoring device 10 in Figure 1B. This concludes the description of this embodiment, and below, a specific example of this embodiment, Example 1, will be described.

[0033] Example 1 describes one specific example of the processing flow in Figure 1C. Therefore, the configuration in Example 1 is the same as that in Figures 1A and 1C described in the Embodiments. Figure 2 is a flowchart of the analysis target setting process in Example 1. In other words, Figure 2 shows a specific example of step S20 in Figure 1C.

[0034] First, when the analysis dashboard 1000 receives an instruction from the user 200 to start the analysis process, in step S21, the analysis target setting unit 2000 reads the asset knowledge graph 3001 from the asset knowledge storage unit 3000.

[0035] Here, the asset knowledge graph 3001 records the structure of the target asset, the failure modes, the relationship between functional failures and failure modes, the relationship between functional failures and failure effects, and the relationship between phenomena observed during failure and failure modes. Furthermore, the asset knowledge graph 3001 does not need to record the failure knowledge of the entire target asset in a single relationship network; failure knowledge created for each structural or functional part of the target knowledge may also be stored. It is also assumed that the structural deployment data of the entire target asset is recorded. Now, we will conclude the explanation of Figure 2 and explain the data used in step S21.

[0036] Figure 3 shows an example of asset knowledge data stored in the asset knowledge storage unit 3000 in Embodiment 1. This data describes the target part, failure mode, failure cause, and failure effect. In addition, the asset knowledge graph 3001, which shows failure knowledge, also records data showing the structural development of the equipment, as shown in Figure 4. That is, it is a tree structure with the highest-level part or equipment itself, "P1 Door," at the top, and the components that make up the equipment, such as "P2 Door Rail" and "Door Panel," are shown below it. Furthermore, in Figure 4, the "Actuator," which is a lower-level component, is shown, and below that are components such as "Piston" (grandchild relationship from P1 Door).

[0037] Furthermore, each row of data shown in Figure 3 represents a combination of the target component, functional failure, failure mode, and failure effect, and in this invention, this is considered a small unit of failure knowledge. By adding the relationship between the failure cause and failure effect between each element in the data, it is possible to describe what factors cause the failure effect. For a failure effect occurring in a certain component, the relationship between each data ID describes which failure mode causes the failure effect, and further, which failure cause causes the failure mode.

[0038] Furthermore, even between different components, it is possible to describe a causal chain in which a failure in a component providing a lower-level function can lead to a failure in a higher-level function.

[0039] Here, Figure 5 shows an asset knowledge graph 3001 that graphically represents this relationship for a portion of the failure knowledge data from Figure 3. In the following explanation, this asset knowledge graph 3001, expressed in graphical form, will be used primarily as an example of asset knowledge data. It is also assumed that the equipment is equipped with sensors for control data and for detecting or measuring normal or abnormal operation.

[0040] Here, an example of sensor data 6001 in Example 1 is shown in Figure 6. In the asset knowledge graph 3001 of this embodiment, the sensors (401A to 403A, 401B to 403B) are managed in association with the components, functions, functional failures, and failure modes they monitor. Furthermore, similar knowledge can be constructed even if the sensors themselves are treated as components. In addition to sensors installed in the equipment, data from measuring instruments used by workers and inspection results may also be included.

[0041] Here, Figure 7 shows a larger example of failure knowledge data, which is an asset knowledge graph 3001. However, although the data shown in Figure 7 is created from engineers' knowledge and failure records, it is not always completely documented due to incomplete knowledge, unknown failures and causal relationships, and the amount of work required to create it, and may contain missing causal links or failure modes. For example, the relationship between a sensor node and a failure mode may, if properly recorded, be directly linked between the failure mode and the sensor node, or a functional failure node may be connected to the sensor for purposes such as detecting a malfunction. Therefore, it is expected that the failure mode linked to that functional failure is associated with a small number of links.

[0042] However, in reality, due to incomplete knowledge construction, sensors are only recorded as mere components, and there may be no link to functional failures or failure modes. Or, a link may only be established to failures directly related to the purpose of installing the sensor, and sensors that can actually be used for investigating other failures may not be linked. Also, something like a control signal, which is not directly recognized as a monitoring sensor, is particularly unlikely to be linked to a failure, but in a failure investigation, the control signal is often checked to confirm the operation of the equipment, and it is actually useful. With the above, the explanation of the data used in step S21 is completed. Subsequently, returning to FIG. 2, the processing after step S22 will be explained.

[0043] Next, in step S22, the analysis target setting unit 2000 sets an analysis target. For this purpose, first, the analysis target setting unit 2000 receives the analysis target setting from the user 200 through the analysis dashboard 1000 via the analysis dashboard 1000. The reception of the analysis target is performed by specifying it when there is a failure phenomenon, functional failure, or a component or failure mode that the user particularly wants to investigate occurring in the target equipment. This is implemented by specifying a single or multiple nodes in the asset knowledge graph 3001.

[0044] Also, as a method of specifying a node, a method can be considered in which the analysis dashboard 1000 displays a graph corresponding to FIG. 7 on the GUI for the user 200 to select. Alternatively, it can be realized by the user inputting the content to be investigated and the event of the failure in text or voice, and extracting the nodes with a high degree of coincidence calculated by the coincidence of the text description content and words or the combination of words and synonyms.

[0045] Further, the analysis target setting unit 2000 extracts only the nodes of the related failure modes from the specified node and sets this as the analysis target. This is because the analysis of the failure ultimately aims to identify the failure mode, take measures such as repair and component replacement to eliminate the failure mode, and eliminate the functional failure and failure impact. It can be said that the failure mode is the final analysis target. Also, the asset knowledge graph 3001 describes data in the form of a combination of components, failure modes, functional failures, and failure impacts. For this reason, no matter which phenomenon related to the failure is selected, it is possible to extract the related failure modes, and thus use this as the analysis target.

[0046] Further, in step S23, the analysis target setting unit 2000 delivers the list of the failure modes of the analysis target set in this way to the sensor set extraction unit 4000. Then, the analysis target setting process ends.

[0047] Next, the sensor set extraction unit 4000 performs a process of extracting a sensor set candidate, which is a sensor set related to the received analysis target. That is, the sensor set extraction process, which is step S40 in FIG. 1C, is executed. FIG. 8 is a flowchart showing the sensor set extraction process in the first embodiment. First, in step S41, the sensor set extraction unit 4000 receives the analysis target delivered in step S23 from the analysis target setting unit 2000. Also, in step S42, the sensor set extraction unit 4000 reads the asset knowledge graph 3001 related to the target equipment from the asset knowledge storage unit 3000.

[0048] Next, in steps S43 to S45, the sensor set extraction unit 4000 repeatedly performs a process of extracting a sensor set candidate, which is a related sensor set, from the read asset knowledge graph 3001 for each failure mode. Several methods can be considered for extracting the sensor set candidate.

[0049] (1) First, one can consider extracting sensor sets that are close in distance from the connection relationships on the graph. This can be achieved by searching the graph from the target failure mode and counting the distance to all sensor nodes with the minimum number of links, recording it as a distance score. With this method, it is possible to extract all sensors and obtain the relationship with the failure mode as distance information, but since the sensors and related events are not always properly linked, the distance may not be measured accurately. Here, "close in distance" means being within a predetermined threshold or being among the top (e.g., the highest) of a predetermined number of sensors in a sensor set that are close in distance.

[0050] (2) Next, a method of extraction using the co-occurrence relationship of the descriptions between nodes can be considered. This method infers the relationship between the failure mode node and the sensor node from the similarity of the descriptions between the failure mode node and the sensor node. For example, if the failure mode node contains a description of the details of the failure phenomenon or the phenomena associated with the anomaly, it is possible to record a description of the details of the detected phenomenon on the sensor node side. This can be achieved by comparing word matches and semantic comparisons between the node name and the description, or by comparing the similarity of the sentences. The obtained similarity between nodes is recorded as a node similarity score. In this way, sensor nodes that have a predetermined similarity to the failure mode node can be identified, and candidate sensor sets can be extracted according to this result. Furthermore, regarding the co-occurrence relationship, nodes may be grouped and the co-occurrence relationship of the group may be used. Furthermore, the position of the description in the description may be used. This can be done by using the proximity of the distance between the description positions, as in (1). Furthermore, the description can be a daily report, report, manual, etc., related to the equipment. Furthermore, these documents may be external information that can be obtained via the network, or they may be stored by the condition monitoring device 10. Furthermore, if the information stored by the status monitoring device 10 is insufficient, external information may be used to supplement it.

[0051] (3) Next, expanding on the above, it is conceivable to detect similarities by including failure modes and failure records that the sensor is already linked to. Even if there is no detailed description of the sensor itself, it is conceivable to extract the sensor by comprehensively scoring the co-occurrence of content between the node to which the sensor is connected and the target failure mode, thereby enabling a broader search for sensors to be extracted. The similarity in this case is defined as a peripheral similarity score.

[0052] Other possible approaches include utilizing a graph structure of links, or using a graph neural network.

[0053] An example of the sensor set candidates and their scores extracted in this way is shown in Figure 9. Then, in step S46, the sensor set extraction unit 4000 hands over these extracted sensor set candidates to the sensor set selection unit 5000 and terminates the process.

[0054] Next, the sensor set selection unit 5000 executes the sensor set selection process shown in step S50. Figure 10 is a flowchart showing the sensor set selection process in Embodiment 1.

[0055] First, in step S51, the sensor set selection unit 5000 receives the sensor set candidates delivered in step S46. Then, in steps S52 to S55, the sensor set selection unit 5000 repeatedly executes step S53 (calculation of sensor score) and step S55 (sensor selection) for each failure mode to be analyzed. At this time, the sensor set selection unit 5000 also performs a process to rank the extracted sensor set candidate data that is thought to be highly related to the failure mode that is actually the target of analysis. The details of these processes are explained below. First, after the sensor set selection unit 5000 receives the sensor set candidates from the sensor set extraction unit 4000, it performs the following process for each failure mode to be analyzed (step S52).

[0056] Furthermore, in step S53, the sensor set selection unit 5000 executes a sensor score calculation process. That is, the sensor set selection unit 5000 calculates an overall score using the sensor scores calculated by the sensor set extraction unit 4000. For example, in this embodiment, the overall score W can be calculated by taking a weighted sum of the distance D from the failure mode, the similarity S between nodes, and the similarity SN including surrounding nodes, as W = W1 × D + W2 × S + W3 × SN.

[0057] Furthermore, in step S54, the sensor selection process, the sensor set selection unit 5000 calculates and ranks the calculated overall score W for each failure mode and sensor. This ranking can be achieved by sorting in descending or ascending order of the overall score W.

[0058] Furthermore, at this stage, sensor sets are selected for each failure mode, which spans multiple failure modes extracted in relation to the analysis targets set by the user 200. For this reason, it is desirable to calculate the overall score W by summing the scores of each sensor for all failure modes. The sensor set selection unit 5000 then selects a sensor set from the candidate sensor set according to the ranking of the overall score W. This selection can be achieved by selecting sensor set candidates that rank high (including the highest rank) by a predetermined number, or sensor set candidates whose overall score is above (or below) a threshold. The selection process is completed by performing this selection process for each failure mode of the equipment to be analyzed (step S55).

[0059] Then, in step S56, the sensor set selection unit 5000 hands over the sensor set selected in this manner to the analysis dashboard 1000, and the processing of the sensor set selection unit is completed.

[0060] Using the results of the sensor sets configured for each failure mode, the analysis dashboard 1000 displays the selected sensor sets corresponding to the analysis objectives entered by the user 200. In this case, the analysis dashboard 1000 displays the selected sensor sets from the time-series data of each sensor stored in the sensor DB 6000. One possible display method is to display the sensors sequentially based on the overall score corresponding to the analysis objectives initially set by the user 200. It is also possible to stop displaying sensors at a certain score value, or to stop displaying sensors below a point where the score for the ranking drops significantly, indicating a low correlation. Alternatively, since the user 200's ultimate goal is often to identify the failure mode, it is also possible to display the sensors for each failure mode.

[0061] Here, Figure 11 shows an example of the display content of the sensor set selection results on the analysis dashboard 1000. In the display example shown in Figure 11, the rank, sensor, and integrated score are shown for each analysis objective. The analysis objective indicates a problem that occurred in the equipment. The rank indicates the rank in step S54. The sensor indicates the target sensor or sensor set. In the example in Figure 11, it is shown that the sensor set was selected in the order of door closing sensor, door opening / closing time sensor, and obstacle detection sensor. Thus, in this embodiment, the selected sensor set and its score are displayed in association. That is, in this embodiment, the sensor set selection unit 5000 of the condition monitoring device 10 outputs the selected sensor set, which is the evaluation result, and its score to the analysis dashboard 1000 in association with the score of that sensor set.

[0062] By viewing this dashboard display, user 200 can access appropriate sensor data corresponding to the problem they have defined, enabling efficient problem investigation. Furthermore, even if a sensor set corresponding to a problem has not been pre-configured, it is possible to view an appropriate sensor set for an ad-hoc specified problem. On the other hand, this ad-hoc process of setting sensor sets may result in the sensor set changing depending on how user 200 defines the problem, potentially preventing proper investigation. Therefore, it may be advisable to automatically select sensor sets for all failure modes and functional failures recorded in the asset knowledge graph 3001 during pre-processing, allowing the user to display sensor data simply by specifying a problem.

[0063] Alternatively, a skilled engineer with expertise in equipment and fault investigation could pre-design typical problems and select sensor sets using this system, creating a problem template that users performing analysis can then select. In this case, the skilled engineer would not need to select individual sensors and build sensor sets, thus reducing the effort required to build templates. Alternatively, it would become possible to prepare sensor sets for a wider range of problems.

[0064] Through the above process, a condition monitoring system is realized that can select and present to user 200 an appropriate sensor set corresponding to the equipment problem specified by user 200.

[0065] Furthermore, in this embodiment, it is possible to expand its usefulness by having the following functions and processing units.

[0066] In this embodiment, the relationship with the failure mode is inferred using the sensor name and the description of the sensor's function and role. However, the sensor name and description of its function and role are sometimes not sufficiently described. For example, in the case of a sensor used solely for monitoring purposes, the designer may not adequately define the functional requirements because the sensor itself is not related to the function or control of the equipment. Alternatively, even when the sensor data only records monitoring values ​​of control signals, the large number of data points and the fact that the data comes from intermediate processing rather than the equipment's function itself may result in insufficient explanation, or the names may be symbolic and lack meaningful interpretation.

[0067] Therefore, it is conceivable to infer and supplement the functions of the sensor using information from neighboring nodes connected to the sensor in the asset knowledge graph 3001. The processing unit, sensor function supplementation unit 7000, which performs this function, could, for example, summarize and generate descriptions of the components linked to the sensor, functional failures, and failure modes. For example, the sensor function supplementation unit 7000 could estimate the missing sensor data using the asset knowledge graph 3001.

[0068] Another possible extension is that when a user views sensor data using the analysis dashboard, the system can identify the sensors actually used in the analysis, extract sensors that are deemed to have a relationship with the problem, and use these in the score calculation of the sensor set selection unit 5000. In other words, the sensor set selection unit 5000 learns the sensors or sensor sets used by the user in the analysis, and updates the asset knowledge data stored in the asset knowledge storage unit 3000 accordingly. Then, the sensor set selection unit 5000 selects a sensor set based on the updated asset knowledge data. For example, the sensor set selection unit 5000 uses the updated asset knowledge data to calculate a score indicating the relationship between the analysis target and the sensor set, and makes a selection accordingly.

[0069] An example of a processing unit for this purpose, the asset knowledge update unit 8000, is to monitor the user's operation of the analysis dashboard 1000 and extract sensor data with long viewing times based on screen display time, mouse overlay time, etc. Alternatively, the user could simply be allowed to select sensors they deem relevant using checkboxes or similar. The asset knowledge update unit records this result as the relationship between the corresponding failure mode node and the sensor node. In this case, the sensor set selection unit reflects this by adding points when calculating the score for each failure mode if the failure mode and sensor are related.

[0070] Furthermore, the present invention also includes an embodiment in which the usage status of sensor data presented to the user during problem analysis is acquired, and the frequently used sensor data is associated with the problem and learned, thereby improving the presented items by reflecting the user's analytical insights when extracting the next sensor set, and enabling efficient problem investigation. For this reason, in the above embodiment, a learning unit is provided in the state monitoring system 100, and the above processing is performed with this unit.

[0071] As described above, the embodiments and Example 1 utilize the "relationship between the analysis target and the extracted candidate sensor sets," allowing for the selection of a more appropriate sensor set that better matches the problems of the target asset in the analysis target.

[0072] 100 Status Monitoring System 200 Users 1000 Analysis Dashboard 2000 Analysis Target Setting Unit 3000 Asset Knowledge Storage Unit 4000 Sensor Set Extraction Unit 5000 Sensor Set Selection Unit 6000 Sensor Data Storage Unit 7000 Sensor Function Complementary Unit 8000 Asset Knowledge Update Unit

Claims

1. A condition monitoring device for monitoring the status of a target asset, comprising: an asset knowledge storage unit for storing asset knowledge data relating to the target asset; a sensor set extraction unit for extracting candidate sensor sets related to an analysis target, which is an operational problem of the target asset; and a sensor set selection unit for evaluating the relationship between the analysis target and the extracted candidate sensor sets, and selecting a sensor set according to the evaluation result.

2. A status monitoring device according to claim 1, wherein the asset knowledge data is an asset knowledge graph in which knowledge data relating to an asset is shown in graph format.

3. A condition monitoring device according to claim 2, wherein the sensor set extraction unit extracts candidate sensor sets based on the degree of similarity between a set of failure modes related to the analysis target set by the user and the installation location of the sensor, the monitoring item, or the function description of the sensor, from among the information recorded in the asset knowledge graph.

4. A condition monitoring device according to claim 1 or 2, wherein the sensor set selection unit evaluates the sensor set based on a score indicating the relationship between the analysis target and the sensor set.

5. A condition monitoring device according to claim 4, wherein the sensor set selection unit outputs an evaluation result including a selected sensor set and a score associated with the sensor set.

6. A condition monitoring device according to claim 2, further comprising a sensor function complementation unit that estimates missing data from the sensor data of the sensor set using the asset knowledge graph.

7. A condition monitoring device according to claim 1, wherein the sensor set selection unit learns the sensor set used by the user for analysis, updates the asset knowledge data according to the results, and selects the sensor set based on the updated asset knowledge data.

8. A condition monitoring method for monitoring the state of a target asset, which is performed by a condition monitoring device, wherein an asset knowledge storage unit stores asset knowledge data relating to the target asset, a sensor set extraction unit extracts candidate sensor sets related to an analysis target which is an operational problem of the target asset, and a sensor set selection unit evaluates the relationship between the analysis target and the extracted candidate sensor sets, and selects a sensor set according to the evaluation result.

9. A state monitoring method according to claim 8, wherein the asset knowledge data is an asset knowledge graph in which knowledge data relating to an asset is shown in graph format.

10. A state monitoring method according to claim 9, wherein the sensor set extraction unit extracts candidate sensor sets based on the degree of similarity between a set of failure modes related to the analysis target set by the user and the installation location of the sensor, the monitoring item, or the function description of the sensor, from among the information recorded in the asset knowledge graph.

11. A state monitoring method according to claim 8 or 9, wherein the sensor set selection unit evaluates the sensor set based on a score indicating the relationship between the analysis target and the sensor set.

12. A state monitoring method according to claim 11, wherein the sensor set selection unit outputs an evaluation result including a selected sensor set and a score associated with the sensor set.

13. A state monitoring method according to claim 9, wherein the sensor function complementation unit further estimates missing data from the sensor data of the sensor set using the asset knowledge graph.

14. A state monitoring method according to claim 8, wherein the sensor set selection unit learns the sensor set used by the user for analysis, updates the asset knowledge data according to the results, and selects the sensor set based on the updated asset knowledge data.

Images