Maintenance assistance system and maintenance assistance method

Abstract

This maintenance assistance system includes: a maintenance record information storage unit that stores maintenance record information which relates to diagnoses on a diagnosis-target device performed in the past and which includes data on a plurality of data items and failure component information; an importance degree training unit that divides the maintenance record information, stored by the maintenance record information storage unit, in accordance with division conditions predetermined on the basis of the data items, trains the relationship between each of the data items and a failure component in the maintenance record information for each division, and generates the importance degree of the data item for each of the division conditions; a similar case extraction unit that divides diagnostic data for diagnosing the diagnosis-target device to be diagnosed, in accordance with the division conditions, and extracts a case similar to the diagnostic data from the maintenance record information storage unit on the basis of the importance degree of the data item corresponding to each of the division conditions; and a failure component inference unit that infers a failure component in the diagnosis-target device on the basis of the similar case extracted by the similar case extraction unit.

Description

Maintenance support system and maintenance support method

[0001] The present disclosure relates to a maintenance support system and a maintenance support method.

[0002] In recent years, a maintenance support system has become known that presents candidate replacement parts using accumulated data including past malfunctions and parts replaced when responding to malfunctions (see, for example, Patent Document 1).

[0003] Japanese Patent Application Laid-Open No. 2020-009068

[0004] However, in the conventional maintenance support system described above, replacement part candidates are selected without considering differences in failure trends for each data item, such as regionality, etc. As a result, the conventional maintenance support system ignores the failure trends for each data item, such as regionality, resulting in a problem of reduced accuracy in failure diagnosis.

[0005] The present disclosure has been made to solve the above problems, and its purpose is to provide a maintenance support system and a maintenance support method that can improve the accuracy of fault diagnosis.

[0006] In order to solve the above problems, one aspect of the present disclosure is a maintenance support system including: a maintenance record information storage unit that stores maintenance record information obtained by previously diagnosing a device to be diagnosed, the maintenance record information including data on a plurality of data items and faulty part information; an importance learning unit that divides the maintenance record information stored in the maintenance record information storage unit according to predetermined division conditions based on the data items, learns the relationship between the data items in the maintenance record information and the faulty parts in each division, and generates an importance of the data items for each division condition; a similar case extraction unit that classifies diagnostic data used to diagnose the device to be diagnosed according to the division conditions, and extracts similar cases that are similar to the diagnostic data from the maintenance record information storage unit based on the importance of the data items corresponding to the classification; and a faulty part estimation unit that estimates a faulty part of the device to be diagnosed based on the similar cases extracted by the similar case extraction unit.

[0007] Another aspect of the present disclosure is a maintenance support method for a maintenance support system including a maintenance record information storage unit that stores maintenance record information from a past diagnosis of a device to be diagnosed, the maintenance record information including data on a plurality of data items and faulty part information, in which an importance learning unit divides the maintenance record information stored in the maintenance record information storage unit according to predetermined division conditions based on the data items, learns the relationship between the data items in the maintenance record information and faulty parts in each division, and generates importance of the data items for each division condition, a similar case extraction unit classifies diagnostic data for diagnosing the device to be diagnosed according to the division conditions, and extracts similar cases that are similar to the diagnostic data from the maintenance record information storage unit based on the importance of the data items corresponding to the classification, and a faulty part estimation unit estimates a faulty part of the device to be diagnosed based on the similar cases extracted by the similar case extraction unit.

[0008] According to the present disclosure, the accuracy of fault diagnosis can be improved.

[0009] 1 is a functional block diagram showing an example of a maintenance support system according to the present embodiment; FIG. 2 is a diagram showing an example of data in a maintenance record information storage unit according to the present embodiment; FIG. 3 is a diagram showing an example of data in a weight storage unit according to the present embodiment; FIG. 4 is a diagram showing an example of data in an estimation result storage unit according to the present embodiment; FIG. 5 is a diagram showing an example of data in an output information storage unit according to the present embodiment; FIG. 6 is a flowchart showing an example of model learning processing of a model learning device according to the present embodiment; FIG. 7 is a flowchart showing an example of weight learning processing of a model learning device according to the present embodiment; FIG. 8 is a flowchart showing an example of diagnostic processing of a diagnostic device according to the present embodiment; FIG. 9 is a diagram illustrating the hardware configuration of a diagnostic device and a model learning device of the maintenance support system according to the present embodiment;

[0010] A maintenance support system and a maintenance support method according to an embodiment of the present disclosure will be described below with reference to the drawings.

[0011] 1 is a functional block diagram showing an example of a maintenance support system 1 according to this embodiment. As shown in FIG. 1, the maintenance support system 1 according to this embodiment includes a diagnostic device 10, a plurality of devices to be diagnosed 20, a plurality of maintenance terminals 30, and a model learning device 40.

[0012] In this embodiment, among the multiple diagnosis target devices 20, devices that have been diagnosed in the past or devices that are currently in normal operation will be described as diagnosis target devices 21, and devices that are currently being diagnosed will be described as diagnosis target devices 22. In addition, in the maintenance support system 1, when referring to any diagnosis target device or when no particular distinction is made, they will be described as diagnosis target devices 20.

[0013] In this embodiment, of the multiple maintenance terminals 30, the terminal that transmitted the maintenance record information, which is the past diagnosis result, will be described as the maintenance terminal 31, and the terminal currently performing the diagnosis will be described as the maintenance terminal 32. In addition, in the maintenance support system 1, when referring to any maintenance terminal or when no particular distinction is made, the terminals will be described as the maintenance terminal 30.

[0014] The diagnostic device 10, the plurality of devices to be diagnosed 21, the plurality of maintenance terminals 30 (31, 32), and the model learning device 40 can be connected to a network NW1 and can communicate with each other via the network NW1. The device to be diagnosed 22 and the maintenance terminal 32 can be connected to a network NW2 and can communicate with each other via the network NW2.

[0015] The network NW1 is, for example, a wide area network (WAN), and the network NW2 is, for example, a local area network (LAN) within the building in which the diagnosis target device 22 is installed.

[0016] The diagnosis target devices 20 (21, 22) are, for example, home appliances such as air conditioners. The diagnosis target devices 20 (21, 22) are devices that are the targets of fault diagnosis.

[0017] The maintenance terminals 30 (31, 32) are terminal devices for maintaining the diagnosis target device 20, and are, for example, smartphones, tablet terminals, mobile PCs (mobile personal computers), etc. The maintenance terminals 30 (31, 32) are devices used by a maintenance company to diagnose and maintain the diagnosis target device 20 on-site, or to diagnose the diagnosis target device 20 before heading to the site.

[0018] In addition, the maintenance terminal 32 is a terminal that diagnoses the diagnosis target device 22 that is the target of diagnosis (maintenance), and is equipped with a NW (network) communication unit 321, an input unit 322, a display unit 323, a terminal memory unit 324, and a terminal control unit 325.

[0019] The NW communication unit 321 is a functional unit realized by a communication device such as a network adapter. The NW communication unit 321 is connected to the network NW2 and is capable of communicating with the diagnosis target device 22. The NW communication unit 321 is also connected to the network NW1 and is capable of communicating with, for example, the diagnosis device 10.

[0020] The input unit 322 is an input device such as a keyboard, a touch screen, or buttons. The input unit 322 accepts various input information in response to operations by a user (maintenance contractor). The input unit 322 is used, for example, for the maintenance contractor to input diagnostic data. The diagnostic data includes, for example, the model name, installation year, installation area, and malfunction symptoms of the diagnostic target device 22.

[0021] The display unit 323 is a display device such as a liquid crystal display. The display unit 323 displays, for example, an input screen for inputting diagnostic data and output information received from the diagnostic device 10 described below. Here, the output information is, for example, the diagnosis results for the diagnostic data, such as candidates for faulty parts.

[0022] The terminal storage unit 324 stores various types of information used by the maintenance terminal 32. For example, the terminal storage unit 324 stores input information from the input unit 322, display information on the display unit 323, information transmitted to and received from the diagnostic device 10, etc.

[0023] The terminal control unit 325 is a functional unit realized by, for example, causing a processor including a CPU (Central Processing Unit) to execute a program. The terminal control unit 325, for example, transmits diagnostic data received via the input unit 322 to the diagnostic device 10 via the network NW1. The terminal control unit 325 also transmits operating data acquired from the diagnosis target device 22 via the network NW2 to the diagnostic device 10 via the network NW1. The terminal control unit 325 also displays output information received from the diagnostic device 10 via the network NW1 on the display unit 323.

[0024] The above-mentioned operating data includes detection data from various sensors (not shown) provided in the diagnosis target device 22, error code information, and the like.

[0025] The model learning device 40 is, for example, a server device connectable to the network NW1. The model learning device 40 executes weight learning processing and learning processing of a faulty component detection model. The model learning device 40 also includes a network communication unit 41, a learning storage unit 42, and a learning processing unit 43.

[0026] The NW communication unit 41 is a functional unit realized by a communication device such as a network adapter. The NW communication unit 41 is connected to the network NW1 and is capable of communicating with the diagnosis target device 21, the maintenance terminal 31, and the diagnosis device 10.

[0027] The learning storage unit 42 is a storage device such as a RAM, a flash memory, or a hard disk drive (HDD), and stores various information used by the model learning device 40. The learning storage unit 42 includes a maintenance record information storage unit 421, an operating data storage unit 422, a weight storage unit 423, and a model storage unit 424.

[0028] The maintenance record information storage unit 421 stores maintenance record information collected from multiple maintenance terminals 31. The maintenance record information is, for example, a maintenance work report created by a maintenance company. The maintenance record information storage unit 421 stores, for example, maintenance record information of past diagnoses of the diagnosis target device 21, including data on multiple data items and faulty part information. Here, an example of data stored in the maintenance record information storage unit 421 will be described with reference to FIG. 2.

[0029] 2 is a diagram showing an example of data stored in the maintenance record information storage unit 421. As shown in Fig. 2, the maintenance record information storage unit 421 stores maintenance record information that associates a NO (number), a model name, years since installation, a region, symptoms, a replacement part P1, and a replacement part P2.

[0030] In FIG. 2, NO is an example of individual identification information of the diagnosis target device 21 (20). The model name indicates the model name of the diagnosis target device 21 (20). The model name is an example of device identification information that identifies the diagnosis target device 21 (20). The years of installation and region indicate the years (period) and region in which the diagnosis target device 21 (20) has been installed. The symptoms indicate symptoms of a malfunction or failure that occurred when the diagnosis target device 21 (20) was diagnosed in the past. Replacement parts P1 and P2 indicate parts that were replaced during past maintenance work. The model name, years of installation, region, and symptoms correspond to data items.

[0031] For example, in the example shown in Figure 2, the maintenance record information with a No. of "1" indicates that the model name is "MSZXXX01S" and the number of years since installation is "5" (five years). It also indicates that the region is "Tokyo" and the symptom is "not cooling." It also indicates that replacement parts P1 and P2 are a "compressor" and an "expansion valve."

[0032] Returning to the explanation of FIG. 1 , the operating data storage unit 422 stores operating data collected from each diagnosis target device 21 (20). Here, the operating data includes detection data from various sensors (not shown) provided in each diagnosis target device 21 (20), error code information, etc. The operating data storage unit 422 stores, for example, the above-mentioned numbers and model names in association with the operating data.

[0033] The weight storage unit 423 (an example of an importance storage unit) divides the maintenance record information stored in the maintenance record information storage unit 421 according to predetermined division conditions based on data items, and stores the learning results obtained by learning the relationship between the data items in the maintenance record information and the faulty parts for each division. Note that the learning results indicate the weights (importance) of the data items for each division condition. Here, an example of data stored in the weight storage unit 423 will be described with reference to FIG. 3 .

[0034] 3 is a diagram showing an example of data stored in the weight storage unit 423 in this embodiment. As shown in FIG. 3, the weight storage unit 423 stores data items and weights in association with each other for each division. The data items include, for example, the model, region, capacity range, and age.

[0035] 3 shows an example in which the area of ​​a data item is divided into a coastal area division A and an inland area division B. The division condition here is that the area of ​​a data item is either a coastal area or an inland area.

[0036] In division A (coastal area), the weight of model is "0.12", the weight of region is "0.83", the weight of capacity range is "0.26", and the weight of age is "0.38". In division B (inland area), the weight of model is "0.34", the weight of region is "0.34", the weight of capacity range is "0.44", and the weight of age is "0.59".

[0037] It should be noted that the larger the value of each weight, the higher the importance, and the smaller the value, the lower the importance.

[0038] 1, the model storage unit 424 stores a faulty part detection model that is a learning result obtained by learning the maintenance record information and the operation data as learning data. The faulty part detection model is, for example, an estimation model that estimates a faulty part from similar cases of malfunctions (failures) of the diagnosis target device 22 (20).

[0039] The learning processing unit 43 is a functional unit realized by, for example, causing a processor including a CPU to execute a program. The learning processing unit 43 executes a learning process to learn the weights of data items for each division and the faulty part detection model. The learning processing unit 43 includes a maintenance record information collection unit 431, an operation data collection unit 432, a weight learning unit 433, and a model learning unit 434.

[0040] The maintenance record information collection unit 431 collects maintenance record information from the maintenance terminal 31 (30) and stores the collected maintenance record information in the maintenance record information storage unit 421. The maintenance record information collected by the maintenance record information collection unit 431 is used as weights for data items and as learning data for learning a faulty part detection model.

[0041] The driving data collection unit 432 collects driving data from the diagnosis target device 21 (20) and stores the collected driving data in the driving data storage unit 422. The driving data collected by the driving data collection unit 432 may be used as part of the weights of the data items and learning data for learning the faulty part detection model.

[0042] The weight learning unit 433 (an example of an importance learning unit) divides the maintenance record information stored in the maintenance record information storage unit 421 according to predetermined division conditions based on data items, learns the relationship between the data items in the maintenance record information and the faulty parts for each division, and generates weights (importance) of the data items for each division condition. The weight learning unit 433 uses the past maintenance record information stored in the maintenance record information storage unit 421 and the driving data stored in the driving data storage unit 422 as learning data, for example, using a machine learning method such as LightGBM to calculate the importance of each data item, to generate weights of the data items for each division condition. Note that the weight learning unit 433 may also generate weights of the data items for each division condition using the maintenance record information as learning data without using the driving data.

[0043] The division condition is, for example, whether the area is a coastal area or an inland area in the data items of the area as shown in Fig. 3. In this case, the weight learning unit 433 generates a weight for each data item in the coastal area (division A) and a weight for each data item in the inland area (division B).

[0044] That is, the weight learning unit 433 divides the learning data based on whether the area in which the device to be diagnosed 21 (20) is installed is a coastal area (division A) or an inland area (division B), and performs a learning process on each of them to generate a weight for each data item.

[0045] The division conditions, such as the above-mentioned regional data items, are determined in advance by the maintenance company in consideration of the maintenance record information and operation data, so as to have a large effect on the defective parts (replacement parts).

[0046] The weight learning unit 433 stores the generated weights of the data items for each division condition in the weight storage unit 423. Furthermore, the weight learning unit 433 transmits, for example, the weights of the data items for each division condition stored in the weight storage unit 423 to the diagnostic device 10 via the NW communication unit 41.

[0047] The model learning unit 434 learns the maintenance record information as learning data and generates a faulty part detection model that estimates faulty parts from similar cases. The model learning unit 434 generates the faulty part detection model using a machine learning method such as LightGBM or SVM (Support Vector Machine) using, for example, past maintenance record information stored in the maintenance record information storage unit 421 and driving data stored in the driving data storage unit 422 as learning data. Note that the model learning unit 434 may also generate the faulty part detection model using the maintenance record information as learning data without using the driving data.

[0048] The model learning unit 434 stores the generated faulty part detection model in the model storage unit 424. Furthermore, the model learning unit 434 transmits, for example, the faulty part detection model stored in the model storage unit 424 to the diagnostic device 10 via the NW communication unit 41.

[0049] The diagnostic device 10 is, for example, a server device connectable to the network NW1. The diagnostic device 10 estimates a faulty part by using a faulty part detection model generated by the model learning device 40 and weights (importance) for each data item. The diagnostic device 10 estimates a faulty part of the diagnosis target device 22 using diagnostic data and operating data acquired from a maintenance terminal 32 via the network NW1 as input data. The diagnostic device 10 also generates a display screen as output information based on the estimation result of the faulty part and transmits it to the maintenance terminal 32 via the network NW1. The diagnostic device 10 also includes a NW communication unit 11, a device storage unit 12, and a diagnostic processing unit 13.

[0050] The NW communication unit 11 is a functional unit realized by a communication device such as a network adapter. The NW communication unit 11 is connected to the network NW1 and is capable of communicating with the maintenance terminal 32 and the model learning device 40.

[0051] The device storage unit 12 is a storage device such as a RAM, a flash memory, or an HDD, and stores various information used by the diagnostic device 10. The device storage unit 12 includes a diagnostic data storage unit 121, a driving data storage unit 122, a weight storage unit 123, a model storage unit 124, a similar case storage unit 125, an estimation result storage unit 126, and an output information storage unit 127.

[0052] The diagnostic data storage unit 121 stores diagnostic data acquired from the maintenance terminal 32. The diagnostic data has data items similar to the input data, excluding the item of replacement parts in the maintenance record information used in the learning process by the model learning device 40. The diagnostic data storage unit 121 stores diagnostic data for data items such as model, region, capacity range, and age.

[0053] The operating data storage unit 122 stores operating data of the diagnosis target device 22 acquired from the maintenance terminal 32. The operating data has the same data items as the operating data used in the learning process by the above-described model learning device 40. The operating data has data items such as detection data from various sensors (indoor temperature, indoor humidity, outdoor temperature, outdoor humidity, etc.) and error codes.

[0054] The weight storage unit 123 stores the weight of the data item for each division condition acquired from the model learning device 40. The weight storage unit 123 stores, for example, the same information as the weight storage unit 423 shown in FIG.

[0055] The model storage unit 124 stores the faulty part detection model acquired from the model learning device 40. The model storage unit 124 stores the same information as that stored in the model storage unit 424 of the model learning device 40.

[0056] The similar case storage unit 125 stores information about past similar cases extracted by the similar case extraction unit 133 (described later). The similar case storage unit 125 stores a plurality of similar cases (for example, approximately 100 similar cases) extracted using the weights of data items acquired from the model learning device 40.

[0057] The estimation result storage unit 126 stores candidates for faulty parts in the diagnosis target device 22 estimated by the faulty part estimation unit 134 described below. For each similar case, the estimation result storage unit 126 stores information associating the candidates for faulty parts estimated by the faulty part estimation unit 134 with the failure probability (likelihood), and the average value of the failure probabilities for all similar cases, as estimation results estimated using the faulty part detection model. Here, an example of data stored in the estimation result storage unit 126 will be described with reference to FIG. 4 .

[0058] 4 is a diagram showing an example of data stored in the estimation result storage unit 126 in this embodiment. As shown in FIG. 4, the estimation result storage unit 126 stores, for each similar case, candidates for a faulty part and a failure probability in association with each other. The estimation result storage unit 126 also stores the average value of the failure probabilities for all similar cases.

[0059] 4, in case EX1 of the similar case, the candidates for failed parts are the compressor, four-way valve, coil, fan motor, and electronic board, and the respective failure probabilities are "52.00%, "3.70%, "24.00%, "9.20%, and "4.00%." ​​It also shows that the average failure probabilities of cases EX1 to EXN of the similar case are "58.20%, "2.70%, "23.10%, "11.60%, and "4.10%."

[0060] 1, the output information storage unit 127 stores output information from the output information generation unit 135. The output information storage unit 127 stores output information based on the estimation result as shown in FIG.

[0061] 5 is a diagram showing an example of data stored in the output information storage unit 127 according to this embodiment. As shown in FIG. 5, the output information storage unit 127 stores display screen information that associates the estimation results with the failure probabilities. Here, the estimation results indicate the top three components with the highest failure probabilities among the candidate components for failure.

[0062] For example, in the example shown in Figure 5, the parts with the highest average failure probabilities of similar cases EX1 to EXN stored in the estimation result storage unit 126 described above are the "compressor," "coil," and "fan motor," and the respective failure probabilities are "50.20%, "23.10%," and "11.60%."

[0063] 1 , the diagnostic processing unit 13 is a functional unit realized by causing a processor including a CPU to execute a program, and includes a diagnostic data acquisition unit 131, an operating data acquisition unit 132, a similar case extraction unit 133, a faulty part estimation unit 134, and an output information generation unit 135.

[0064] The diagnostic data acquisition unit 131 acquires diagnostic data from the maintenance terminal 32 via the NW communication unit 11. The diagnostic data acquisition unit 131 stores the acquired diagnostic data in the diagnostic data storage unit 121.

[0065] The operating data acquisition unit 132 acquires operating data of the diagnosis target device 22 from the maintenance terminal 32 via the NW communication unit 11. The operating data acquisition unit 132 stores the acquired operating data of the diagnosis target device 22 in the operating data storage unit 122.

[0066] The similar case extraction unit 133 classifies the diagnostic data for diagnosing the diagnostic target device 22 according to predetermined division conditions, and extracts similar cases that are similar to the diagnostic data from the maintenance record information storage unit 421 of the model learning device 40 based on the weights of the data items corresponding to the classification (division conditions).

[0067] The similar case extraction unit 133 classifies the acquired diagnostic data and operating data according to the region where the diagnosis target device 22 is installed, for example, whether the data corresponds to a coastal region or an inland region. For example, in coastal regions, there is a strong tendency for failures due to metal rust, and the weights (importance) of data items for failure diagnosis differ, so it is considered effective to classify the data into the above-mentioned divisions (coastal region or inland region).

[0068] The similar case extraction unit 133 acquires the weight of each data item corresponding to the classification (division condition) from the weight storage unit 123, and uses the weight of each data item corresponding to the classification (division condition) to calculate the similarity with the past cases stored in the maintenance record information storage unit 421. The similar case extraction unit 133 calculates the similarity (Sim ｎ ) is calculated.

[0069]

[0070] Here, Sim ｎ indicates the similarity of the nth past case, and α, β, ... indicate the weight of each data item. Furthermore, x~, y~, ... indicate diagnostic data for the input value of each data item, and x, y, ... indicate diagnostic data for past cases of each data item. Furthermore, function f is a function that outputs "1" when the past case and the data of the input item match, and outputs "0" when they do not match. Note that in this embodiment, a variable with a horizontal line over the letter x is represented as x~, and a variable with a horizontal line over the letter y is represented as y~.

[0071] The similar case extraction unit 133 calculates the n-th similarity (Sim ｎ ) and multiply the weight (α, β) of each data item by the degree of match of each data item (using "1" for a match and "0" for a mismatch) to calculate the sum of the degrees of match of all items.

[0072] The similar case extraction unit 133 calculates the similarity (Sim ｎ ) are extracted as similar cases, for example, 100 cases. In this way, the similar case extraction unit 133 extracts a plurality of similar cases. The similar case extraction unit 133 stores the extracted similar cases in the similar case storage unit 125.

[0073] The failed part estimation unit 134 estimates a failed part of the diagnosis target device 22 based on the similar cases extracted by the similar case extraction unit 133. The failed part estimation unit 134 estimates a failed part from the similar cases using the failed part detection model stored in the model storage unit 124. The failed part estimation unit 134 estimates failed part candidates and failure probabilities for each of a plurality of similar cases (e.g., 100 cases) stored in the similar case storage unit 125 using the failed part detection model, for example, as shown in FIG. 4 . The failed part estimation unit 134 also stores the estimation results (failed part candidates and failure probabilities) in the estimation result storage unit 126.

[0074] The output information generation unit 135 generates output information based on the estimation results estimated by the failed part estimation unit 134. The output information generation unit 135 calculates the average value of the failure probabilities for the failed parts in the multiple similar cases stored in the estimation result storage unit 126. For example, as shown in FIG. 4 , the output information generation unit 135 stores the calculated average value of the failure probability in the estimation result storage unit 126.

[0075] The output information generating unit 135 also selects a specific number (e.g., three) of candidate faulty parts in descending order of average failure probability, and generates output information including the selected candidate faulty parts. The output information generating unit 135 generates output information (display screen) such as that shown in FIG. 5 and stores it in the output information storage unit 127.

[0076] The output information generation unit 135 transmits the generated output information to the maintenance terminal 32 via the NW communication unit 11. In this way, the output information generation unit 135 generates output information including the selected candidates for the faulty part and the average value of the failure probability, and transmits the generated output information to the maintenance terminal 32.

[0077] Next, the operation of the maintenance support system 1 according to this embodiment will be described with reference to the drawings.

[0078] First, the model learning process of the model learning device 40 will be described with reference to Fig. 6. Fig. 6 is a flowchart showing an example of the model learning process of the model learning device 40 in this embodiment.

[0079] 6 , the model learning device 40 collects maintenance work reports and operating data via the network NW1 (step S101). The maintenance record information collection unit 431 of the model learning device 40 collects maintenance work reports as maintenance record information from the maintenance terminal 31 via the NW communication unit 41, and stores the collected maintenance record information (maintenance work reports) in the maintenance record information storage unit 421. In addition, the operating data collection unit 432 of the model learning device 40 collects operating data from the diagnosis target device 21 via the NW communication unit 41, and stores the collected operating data in the operating data storage unit 422.

[0080] Next, the model learning unit 434 of the model learning device 40 classifies input data and output data from the maintenance work report and the operation data (step S102). The model learning unit 434 classifies the maintenance record information (maintenance work report) stored in the maintenance record information storage unit 421 and the operation data stored in the operation data storage unit 422 as learning data into input data and output data. For example, in the case shown in FIG. 2, "model name," "years of installation," "area," and "symptoms" are classified as input data, and "replacement part P1" and "replacement part P2" are classified as output data.

[0081] Next, the model learning unit 434 learns the relationship between the input data and the output data and generates a faulty component detection model (step S103). The model learning unit 434 generates the faulty component detection model from the above-mentioned learning data using a machine learning method such as LightGBM or SVM. The model learning unit 434 stores the generated faulty component detection model in the model storage unit 424.

[0082] Next, the model learning unit 434 transmits the failed part detection model to the diagnostic device 10 (step S104). The model learning unit 434 transmits the failed part detection model stored in the model storage unit 424 to the diagnostic device 10 via the NW communication unit 41. The transmitted failed part detection model is stored in the model storage unit 124 of the diagnostic device 10. After the processing of step S104, the model learning unit 434 ends the model learning process.

[0083] Next, the weight learning process of the model learning device 40 will be described with reference to Fig. 7. Fig. 7 is a flowchart showing an example of the weight learning process of the model learning device 40 in this embodiment.

[0084] 7, model learning device 40 first extracts data items for which the failure tendency differs significantly depending on the division from the maintenance work report and the operation data (step S201). Weight learning unit 433 of model learning device 40 extracts, for example, "region" as a data item for which the failure tendency differs significantly.

[0085] Next, the weight learning unit 433 divides the data of the data items according to the division conditions of the specified data items (step S202). The weight learning unit 433 divides the above-mentioned learning data into, for example, coastal areas and inland areas according to the "region" of the specified data item.

[0086] Next, the weight learning unit 433 learns the relationship between the diagnostic data and replacement parts for each division and calculates a weight for each data item (step S203). For example, the weight learning unit 433 calculates the weight of each data item for the coastal region using a machine learning method such as LightGBM from the diagnostic data (learning data) for which the division condition is the coastal region. Furthermore, for example, the weight learning unit 433 calculates the weight of each data item for the inland region using a machine learning method such as LightGBM from the diagnostic data (learning data) for which the division condition is the inland region. The weight learning unit 433 stores the calculated weight of each data item for each division condition in the weight storage unit 423, as shown in FIG. 3 .

[0087] Next, weight learning unit 433 transmits the weight for each data item to diagnostic device 10 (step S204). Weight learning unit 433 transmits the weight for each data item for each division condition stored in weight storage unit 423 to diagnostic device 10 via NW communication unit 41. The transmitted weight for each data item for each division condition is stored in weight storage unit 123 of diagnostic device 10. After processing in step S204, weight learning unit 433 ends the weight learning process.

[0088] Next, the diagnostic processing of the diagnostic device 10 will be described with reference to Fig. 8. Fig. 8 is a flowchart showing an example of the diagnostic processing of the diagnostic device 10 in this embodiment.

[0089] 8 , the diagnostic device 10 first extracts specified data items from the diagnostic data in response to receiving the diagnostic data and the operating data (step S301). The diagnostic data acquisition unit 131 of the diagnostic device 10 acquires the diagnostic data from the maintenance terminal 32 via the NW communication unit 11, and the operating data acquisition unit 132 acquires the operating data of the diagnosis target device 22 from the maintenance terminal 32 via the NW communication unit 11. The similar case extraction unit 133 of the diagnostic device 10 extracts the specified data item (e.g., "area") in response to acquiring (receiving) the diagnostic data and the operating data.

[0090] Next, the similar case extraction unit 133 classifies the data of the specified data item according to the division conditions used in weight learning (step S302). The similar case extraction unit 133 classifies the diagnostic data and driving data into, for example, a coastal area or an inland area.

[0091] Next, the similar case extraction unit 133 extracts weights for the classification from the weight learning results (step S303). The similar case extraction unit 133 extracts weights for each data item corresponding to the classification (division condition) obtained by classifying the diagnostic data and the driving data from the weight storage unit 123. For example, if the classification is coastal area, the similar case extraction unit 133 obtains weights for each data item corresponding to coastal area from the weight storage unit 123. Also, for example, if the classification is inland area, the similar case extraction unit 133 obtains weights for each data item corresponding to inland area from the weight storage unit 123.

[0092] Next, the similar case extraction unit 133 uses the extracted weight to calculate the similarity with the past case (step S304). The similar case extraction unit 133 calculates the similarity (Sim) between the past case stored in the maintenance record information storage unit 421 and the diagnostic data and the driving data using the above-mentioned formula (1). ｎ ) is calculated.

[0093] Next, the similar case extraction unit 133 sorts the past cases in descending order of similarity and selects highly similar cases (e.g., 100 cases) (step S305).The similar case extraction unit 133 stores the selected highly similar past cases (e.g., 100 cases) in the similar case storage unit 125 as similar cases.

[0094] Next, the failed part estimation unit 134 of the diagnostic device 10 uses the failed part detection model to estimate a failed part for each of the selected past cases (step S306). The failed part estimation unit 134 estimates failed part candidates and failure probabilities for each of the similar cases stored in the similar case storage unit 125, using the failed part detection model stored in the model storage unit 124. The failed part estimation unit 134 stores the estimation results in the estimation result storage unit 126, for example, as case EX1 to case EXN shown in FIG. 4 .

[0095] Next, the output information generating unit 135 of the diagnostic device 10 tally up the faulty parts and failure probabilities estimated from each past case (step S307). The output information generating unit 135 calculates the average value of the failure probabilities for the faulty parts in multiple past cases (similar cases). For example, as shown in FIG. 4 , the output information generating unit 135 calculates the average value of the failure probabilities for each faulty part and stores it in the estimation result storage unit 126.

[0096] Next, the output information generation unit 135 generates output information from the aggregation result and transmits it to the maintenance terminal 32 (step S308). The output information generation unit 135 sorts the average failure probability values ​​for each failed component in descending order and determines the top three failed component candidates with the highest average failure probability values. The output information generation unit 135 generates output information such as that shown in FIG. 5 using the top three failed component candidates with the highest average failure probability values. The output information generation unit 135 stores the generated output information in the output information storage unit 127 and transmits the output information to the maintenance terminal 32 via the NW communication unit 11. After the processing of step S308, the output information generation unit 135 terminates the diagnostic processing of the diagnostic device 10.

[0097] As described above, the maintenance support system 1 according to this embodiment includes the maintenance record information storage unit 421, the weight learning unit 433 (importance learning unit), the similar case extraction unit 133, and the faulty part estimation unit 134. The maintenance record information storage unit 421 stores maintenance record information obtained by past diagnoses of the diagnosis target device 21, the maintenance record information including data on multiple data items and faulty part information. The weight learning unit 433 (importance learning unit) divides the maintenance record information stored in the maintenance record information storage unit 421 according to predetermined division conditions (e.g., coastal area or inland area) based on the data items, learns the relationship between the data items in the maintenance record information and the faulty parts for each division, and generates weights (importance) of the data items for each division condition. The similar case extraction unit 133 classifies diagnostic data for diagnosing the diagnosis target device 22 according to the division conditions, and extracts similar cases similar to the diagnostic data from the maintenance record information storage unit 421 based on the importance of the data items corresponding to the division conditions. The faulty part estimation unit 134 estimates the faulty part of the diagnosis target device 22 based on the similar cases extracted by the similar case extraction unit 133 .

[0098] As a result, the maintenance support system 1 according to this embodiment extracts similar cases based on the weights (importance) of data items for each division condition (e.g., coastal area or inland area) that are determined by dividing (classifying) the area according to predetermined division conditions, thereby enabling more accurate and appropriate extraction of similar cases. As a result, the maintenance support system 1 according to this embodiment can improve the accuracy of fault diagnosis and the quality of diagnosis by maintenance companies.

[0099] Furthermore, the maintenance support system 1 according to this embodiment can extract appropriate similar cases by taking into account differences in failure trends for each data item, such as regionality, and can select candidates for replacement parts taking regionality into account.

[0100] The maintenance support system 1 according to this embodiment also includes a model learning unit 434. The model learning unit 434 learns the maintenance record information as learning data and generates a faulty part detection model that infers faulty parts from similar cases. The faulty part estimation unit 134 uses the faulty part detection model to infer faulty parts from similar cases.

[0101] As a result, the maintenance support system 1 according to this embodiment can more appropriately estimate faulty parts by using a faulty part detection model to estimate faulty parts from similar cases, thereby improving the diagnostic quality of maintenance companies.

[0102] The maintenance support system 1 according to this embodiment also includes an output information generation unit 135. The output information generation unit 135 generates output information based on the estimation results obtained by the failed part estimation unit 134. The similar case extraction unit 133 extracts a plurality of similar cases. The failed part estimation unit 134 uses a failed part detection model to estimate a failed part and a failure probability for each of the plurality of similar cases. The output information generation unit 135 calculates the average value of the failure probability for the failed part in the plurality of similar cases, selects a specific number of failed part candidates (e.g., the top three) in descending order of the average failure probability, and generates output information including the selected failed part candidates.

[0103] As a result, the maintenance support system 1 according to this embodiment selects candidates for the faulty part using the average value of the failure probability for the faulty part in a plurality of similar cases, thereby enabling more accurate estimation of candidates for the faulty part. Furthermore, the maintenance support system 1 according to this embodiment outputs a specific number (for example, the top three) of candidates for the faulty part in descending order of the average failure probability, thereby providing the maintenance company with information for determining the faulty part and improving the quality of the diagnosis.

[0104] In this embodiment, the similar case extraction unit 133 extracts similar cases that are similar to the diagnostic data received from the maintenance terminal. The output information generation unit 135 generates output information including the selected candidates for the faulty part and the average value of the failure probability, and transmits the generated output information to the maintenance terminal 32.

[0105] As a result, the maintenance support system 1 according to this embodiment transmits output information including the selected candidate faulty parts and the average value of the failure probability to the maintenance terminal 32, thereby providing the maintenance company with information to determine the faulty parts.

[0106] In this embodiment, the weight learning unit 433 generates weights (importance) of data items for each division condition using learning data including operation data of the diagnosis target device 21 (e.g., sensor detection data, error codes, etc.) and maintenance record information previously collected from the diagnosis target device 21. The model learning unit 434 generates a faulty part detection model using learning data including the operation data of the diagnosis target device 21 and maintenance record information.

[0107] As a result, the maintenance support system 1 according to this embodiment generates weights (importance) of data items and a faulty part detection model for each division condition taking into account operating data (e.g., sensor detection data, error codes, etc.), thereby enabling even more accurate estimation of faulty parts.

[0108] In this embodiment, the weight learning unit 433 divides the maintenance record information by the region in which the diagnosis target device 21 is installed. The similar case extraction unit 133 classifies the diagnostic data by region and extracts similar cases based on the weights (importance) of the data items corresponding to the classified regions.

[0109] As a result, the maintenance support system 1 according to this embodiment can extract appropriate similar cases by taking into account differences in regional failure trends, and can appropriately select candidates for replacement parts taking regional characteristics into account.

[0110] Furthermore, a maintenance support method according to this embodiment is a maintenance support method for a maintenance support system 1 including a maintenance record information storage unit 421, and includes a weight learning step, a similar case extraction step, and a faulty part estimation step. The maintenance record information storage unit 421 stores maintenance record information obtained by past diagnoses of a diagnosis target device 21, the maintenance record information including data on multiple data items and faulty part information. In the weight learning step, the weight learning unit 433 divides the maintenance record information stored in the maintenance record information storage unit 421 according to predetermined division conditions based on the data items, learns the relationship between the data items in the maintenance record information and the faulty parts for each division, and generates weights (importance) of the data items for each division condition. In the similar case extraction step, the similar case extraction unit 133 classifies diagnostic data for diagnosing the diagnosis target device 22 according to the division conditions, and extracts similar cases similar to the diagnostic data from the maintenance record information storage unit 421 based on the weights of the data items corresponding to the classifications (division conditions). In the failed part inferring step, the failed part inferring unit 134 infers a failed part of the diagnosis target device 22 based on the similar cases extracted by the similar case extracting unit 133. As a result, the maintenance support method according to this embodiment has the same effects as the above-described maintenance support system 1, and is able to extract more accurate and appropriate similar cases, thereby improving the quality of diagnosis.

[0111] 9 is a diagram illustrating the hardware configuration of the diagnostic device 10 and the model learning device 40 of the maintenance support system 1 according to this embodiment. The devices shown in FIG. 9 illustrate the hardware configuration of each device (the diagnostic device 10 and the model learning device 40) of the maintenance support system 1.

[0112] As shown in FIG. 9, each device (the diagnosis device 10 and the model learning device 40) of the maintenance support system 1 includes a communication device H11, a memory H12, and a processor H13.

[0113] The communication device H11 is a communication device such as a LAN card that can be connected to the network NW1. The memory H12 is a storage device such as a RAM, a flash memory, or an HDD, and stores various information and programs used by each device (the diagnostic device 10 and the model learning device 40).

[0114] The processor H13 is a processing circuit including, for example, a CPU, etc. The processor H13 executes various processes of each device (the diagnostic device 10 and the model learning device 40) by executing programs stored in the memory H12.

[0115] The present disclosure is not limited to the above-described embodiment and may be modified within the scope of the present disclosure. For example, in the above-described embodiment, an example was explained in which the weight division condition is divided into coastal areas and inland areas based on the data item "region," but this is not limited to this, and other data items and division conditions may be used. For example, if "years of installation" is used as a data item, the data may be divided (classified) into, for example, 5 years or more and less than 5 years as a division condition.

[0116] Furthermore, in the above embodiment, an example has been described in which the maintenance support system 1 includes the diagnostic device 10 and the model learning device 40, but this is not limited to this, and the diagnostic device 10 may include the functions of the model learning device 40 and be realized by a single device.

[0117] In the above embodiment, some of the functions of the diagnostic device 10 may be provided in the model learning device 40, or some of the functions of the model learning device 40 may be provided in the diagnostic device 10. Furthermore, the diagnostic device 10 and the model learning device 40 may be realized by three or more devices.

[0118] In the above embodiment, the similar case extraction unit 133 extracts a specific number of similar cases (for example, 100 cases), but this is not limited to this, and one past case with the highest similarity may be extracted as the similar case.

[0119] In the above embodiment, the model learning device 40 generates the weights for each data item and the faulty part detection model using maintenance record information and driving data. However, the present invention is not limited to this example. The weights for each data item and the faulty part detection model may be generated without using driving data. In this case, the diagnostic device 10 extracts similar cases from the diagnostic data without using driving data.

[0120] In addition, in the above embodiment, an example was described in which communication between each device is achieved using two networks, network NW1 and network NW2, but this is not limited to this and may be achieved using one network or three or more networks.

[0121] In the above embodiment, an air conditioner is described as an example of the diagnosis target device 20, but the diagnosis target device 20 is not limited to this. The diagnosis target device 20 may be, for example, another home appliance, an IoT device, or the like.

[0122] Each component of the maintenance support system 1 described above has an internal computer system. A program for implementing the functions of each component of the maintenance support system 1 described above may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read into a computer system and executed to perform processing in each component of the maintenance support system 1 described above. Here, "reading a program recorded on a recording medium into a computer system and executing it" includes installing the program into a computer system. The term "computer system" here includes an OS and hardware such as peripheral devices.

[0123] Furthermore, a "computer system" may include multiple computer devices connected via a network, including communication lines such as the Internet, WAN, LAN, and dedicated lines. Furthermore, a "computer-readable recording medium" refers to portable media such as flexible disks, optical magnetic disks, ROMs, and CD-ROMs, as well as storage devices such as hard disks built into a computer system. Thus, the recording medium storing the program may be a non-transitory recording medium such as a CD-ROM.

[0124] The recording medium also includes internal or external recording media accessible from a distribution server for distributing the program. The program may be divided into multiple parts, downloaded at different times, and then combined by each component of the maintenance support system 1, or each divided program may be distributed by a different distribution server. Furthermore, the term "computer-readable recording medium" also includes a medium that stores a program for a certain period of time, such as volatile memory (RAM) within a computer system that serves as a server or client when a program is transmitted over a network. The program may also be a medium for implementing part of the above-described functions. Furthermore, the program may be a so-called differential file (differential program) that can realize the above-described functions in combination with a program already stored in the computer system.

[0125] 1... Maintenance support system, 10... Diagnostic device, 11, 41, 321... Network communication unit, 12... Device storage unit, 13... Diagnostic processing unit, 20, 21, 22... Device to be diagnosed, 30, 31, 32... Maintenance terminal, 40... Model learning device, 42... Learning storage unit, 43... Learning processing unit, 121... Diagnostic data storage unit, 122, 422... Operation data storage unit, 123, 423... Weight storage unit, 124, 424... Model storage unit, 125... Similar case storage unit, 126... Estimation result result storage unit, 127...output information storage unit, 131...diagnosis data acquisition unit, 132...driving data acquisition unit, 133...similar case extraction unit, 134...faulty part estimation unit, 135...output information generation unit, 322...input unit, 323...display unit, 324...terminal storage unit, 325...terminal control unit, 421...maintenance record information storage unit, 431...maintenance record information collection unit, 432...driving data collection unit, 433...weight learning unit, 434...model learning unit, NW1, NW2...network

Claims

1. a maintenance record information storage unit that stores maintenance record information obtained by diagnosing a target device in the past, the maintenance record information including data of a plurality of data items and information on faulty parts; an importance learning unit that divides the maintenance record information stored in the maintenance record information storage unit according to predetermined division conditions based on the data items, learns the relationship between the data items in the maintenance record information and faulty parts for each division, and generates the importance of the data items for each division condition; a similar case extraction unit that classifies diagnostic data for diagnosing the diagnosis target device according to the division conditions and extracts similar cases that are similar to the diagnostic data from the maintenance record information storage unit based on the importance of the data items corresponding to the classification; a faulty part inferring unit that infers a faulty part of the device to be diagnosed based on the similar cases extracted by the similar case extracting unit; A maintenance support system equipped with:

2. a model learning unit that learns the maintenance record information as learning data and generates a faulty part detection model that infers the faulty part from the similar cases; The faulty part inferring unit infers the faulty part from the similar cases using the faulty part detection model. The maintenance support system according to claim 1 .

3. an output information generating unit that generates output information based on the estimation result obtained by the failed part estimation unit; the similar case extraction unit extracts a plurality of similar cases; the failed part estimation unit estimates the failed part and the failure probability for each of the plurality of similar cases using the failed part detection model; The output information generation unit calculates an average value of the failure probability for the failed part in a plurality of the similar cases, selects a specific number of candidates for the failed part in descending order of the average value of the failure probability, and generates the output information including the selected candidates for the failed part. The maintenance support system according to claim 2 .

4. the similar case extraction unit extracts the similar case that is similar to the diagnostic data received from the maintenance terminal; The output information generation unit generates the output information including the selected candidate faulty part and the average value of the failure probability, and transmits the generated output information to the maintenance terminal. The maintenance support system according to claim 3 .

5. the importance learning unit generates importance of the data item for each division condition using the learning data including operation data of the diagnosis target device previously collected from the diagnosis target device and the maintenance record information; The model learning unit generates the faulty part detection model using the learning data including the operation data of the device to be diagnosed and the maintenance record information. The maintenance support system according to any one of claims 2 to 4.

6. the importance learning unit divides the maintenance record information according to the region in which the diagnosis target device is installed, and generates importance of the data item for each region; The similar case extraction unit classifies the diagnostic data by region and extracts the similar cases based on the importance of the data items corresponding to the classified regions. The maintenance support system according to claim 1 .

7. A maintenance support method for a maintenance support system including a maintenance record information storage unit that stores maintenance record information obtained by diagnosing a diagnosis target device in the past, the maintenance record information including data of a plurality of data items and faulty part information, an importance learning unit divides the maintenance record information stored in the maintenance record information storage unit according to predetermined division conditions based on the data items, learns a relationship between the data items in the maintenance record information and faulty parts for each division, and generates an importance of the data items for each division condition; a similar case extraction unit classifies diagnostic data for diagnosing the diagnosis target device according to the division conditions, and extracts similar cases that are similar to the diagnostic data from the maintenance record information storage unit based on the importance of the data items corresponding to the classification; A faulty part estimation unit estimates a faulty part of the device to be diagnosed based on the similar cases extracted by the similar case extraction unit. Maintenance support method.