Simulation device, simulation method, and simulation program

Abstract

This simulation device (110) comprises a distribution model estimation unit (112) and a reference value setting unit (113). On the basis of an existing distribution model corresponding to the difference between the output of a physical model corresponding to an existing specification and data indicating the operating status of each existing specification product, and a test distribution model corresponding to the difference between the output of a physical model corresponding to a target specification and data indicating the operating status of each test machine, the distribution model estimation unit (112) estimates a distribution model corresponding to the difference between the output of the target specification model and the data indicating the operating status of each target specification product. On the basis of the outliers in the estimated distribution model, a reference value setting unit (113) sets an abnormality determination reference for determining whether the data indicating the operation status of each target specification product is abnormal.

Description

Simulation apparatus, simulation method, and simulation program

[0001] The present disclosure relates to a simulation apparatus, a simulation method, and a simulation program.

[0002] There is a technology for remotely diagnosing equipment based on physical models and field data. Patent Literature 1 discloses a calculation method characterized by collecting machine learning data used for elevator state diagnosis using a physical model (Synthetic data), field data (History data), and data (Calibrated synthetic data) generated by incorporating at least one type of abnormality into the physical model. Note that Patent Literature 1 discloses a data collection method but does not disclose a specific diagnosis method. In the technology disclosed in Patent Literature 1, diagnosis target data is determined to be normal when the diagnosis target data is closer to normal data than to abnormality reproduction data, and is determined to be abnormal when the diagnosis target data is closer to abnormality reproduction data than to normal data.

[0003] International Publication No. 2023 / 066484 Pamphlet

[0004] Although it is desired to realize high-precision remote diagnosis utilizing AI (Artificial Intelligence), the remote diagnosis cannot be realized when there is insufficient learning data corresponding to the target specifications. Here, it is possible to generate learning data corresponding to the target specifications using a physical model. However, specific diagnostic criteria cannot be determined only with learning data generated using a physical model.

[0005] Here, the main methods for diagnosing anomalies are the following [Method 1] and [Method 2]. [Method 1] is a method in which normal data and abnormal data are prepared, and the data to be diagnosed is judged as abnormal when the distance between the data to be diagnosed and the abnormal data is relatively close. An advantage of [Method 1] is that the abnormal data is clearly abnormal, so the reliability of the result of judging it as abnormal is high. [Method 2] is a method in which data corresponding to outliers that exist in a certain proportion of the total data is judged as abnormal. An advantage of [Method 2] is that it is possible to extract anomalies that have not been collected as abnormal data.

[0006] The technology disclosed in Patent Document 1 utilizes [Method 1]. Therefore, this technology has the problem that it cannot properly detect anomalies that cannot be reproduced in the physical model. Thus, we would like to use [Method 2], which can diagnose anomalies including those that cannot be reproduced in the physical model, but using [Method 2] requires a method for estimating the distribution of data related to the target specifications.

[0007] This disclosure aims to estimate the distribution of data related to the target specifications in anomaly diagnosis that utilizes a method of identifying data corresponding to outliers present in a certain proportion of the total data as anomalies.

[0008] The simulation device relating to this disclosure includes: an existing distribution model corresponding to the distribution of the difference between the output of an existing specification model, which is a physical model corresponding to an existing specification corresponding to a product with a track record of operation exceeding a reliable number of units, and data indicating the operating status of each existing specification product in a group of existing specification products consisting of each existing specification product on which the existing specification is implemented; a test distribution model corresponding to the difference between the output of a target specification model, which is a physical model corresponding to a target specification, and data indicating the operating status of each test unit in a group of test units consisting of each test unit on which the target specification is implemented; a distribution model estimation unit that estimates a distribution model corresponding to the distribution of the difference between the output of the target specification model and data indicating the operating status of each target specification product in a group of target specification products consisting of each target specification product on which the target specification is implemented, as an estimated target distribution model; and a reference value setting unit that sets an abnormality determination criterion for determining whether the data indicating the operating status of each target specification product in the group of target specification products is abnormal based on outliers in the estimated target distribution model.

[0009] According to this disclosure, the distribution model estimation unit estimates the target distribution model based on an existing distribution model and a test distribution model. The reference value setting unit sets an anomaly judgment criterion based on outliers in the estimated target distribution model. Here, the estimated target distribution model is a distribution model corresponding to the distribution of the difference between the output of the target specification model and the data showing the operating status of each target specification product in the target specification product group, which consists of each target specification product on which the target specification is implemented. Therefore, according to this disclosure, in an anomaly diagnosis that uses a method to judge data corresponding to outliers that exist in a certain proportion of the total data as an anomaly, the distribution of data related to the target specification can be estimated.

[0010] A diagram showing an example configuration of the simulation system 90 according to Embodiment 1. A diagram showing an example configuration of the simulation system 90 according to Embodiment 1. A diagram showing an example hardware configuration of the simulation device 110 according to Embodiment 1. A flowchart showing the operation of the simulation device 110 according to Embodiment 1. A diagram explaining the processing of the distribution model estimation unit 112 according to a modified example of Embodiment 1. A diagram showing an example hardware configuration of the simulation device 110 according to a modified example of Embodiment 1.

[0011] In the description and drawings of the embodiments, the same elements and corresponding elements are denoted by the same reference numeral. The descriptions of elements denoted by the same reference numeral are omitted or simplified as appropriate. The arrows in the figures mainly indicate the flow of data or processing. Also, "part" may be read as "circuit," "device," "equipment," "process," "step," "procedure," "processing," or "circuitry" as appropriate. The functions of each part of each device may be realized by firmware, software, hardware, or a combination thereof.

[0012] The following explains the concepts necessary to understand this disclosure. A physical model refers to a mechanism that simulates the physical behavior of a device using a physical simulator such as Simulink (registered trademark). Basically, any designer can construct a physical model. However, in simulations using a physical model, it is necessary to separately identify some physical parameters such as friction. Bayesian estimation is a method of estimating the posterior probability (distribution) by combining two distributions: prior probability and likelihood. Field data is various data collected from equipment actually in operation at customer sites, etc., through remote monitoring, etc. Field data is data that shows the operating status of each piece of equipment. Field data consists of, as a concrete example, data from each sensor and data showing each internal signal. Outliers refer to extremely large and extremely small values ​​in all the data. Here, it is a common practice to determine an anomaly when an outlier is measured. Although the specific criteria for outliers differ depending on the operational policy, it is common to consider 0.3% of the data as outliers (abnormal values).

[0013] The following explains the definitions used in this specification. Existing specification products are products that correspond to specifications that enable the collection of a large amount of various data (field data) related to the operation of equipment through remote monitoring. A certain number or more of products corresponding to existing specifications have already been introduced to the market. A specific example of this number is 1000. Existing specifications are specifications that correspond to existing specification products. Existing specifications only need to be specifications that enable the construction of a corresponding physical model. Target specification products are products that correspond to specifications for which sufficient field data has not been collected. Target specifications are specifications that correspond to products that are expected to increase in the market in the future, or specifications that differ from the normal specifications due to customization, etc. Target specifications are specifications that correspond to target specification products. Target specifications only need to be specifications that enable the construction of a corresponding physical model. Some parts of the target specifications differ from existing specifications. Target specifications may also be specifications that modify some parts of existing specifications. Distribution is the variability of the difference between the output of the physical model and the field data. Distributions may be obtained for each type of data, or distributions may be obtained based on features related to multiple types of data.

[0014] Embodiment 1. This embodiment will be described in detail below with reference to the drawings.

[0015] ***Configuration Description*** Figure 1 shows an example configuration of the simulation system 90 according to this embodiment when using edge devices. As shown in Figure 1, the simulation system 90 comprises a cloud system 100 and a target specification product 200. The cloud system 100 comprises a simulation device 110. The cloud system 100 also stores existing specification field data 191, test machine data 192, target specification field data 193, and judgment criterion values ​​194. The cloud system 100 may be a computer used in a remote monitoring center. Multiple elements of the cloud system 100 may be configured integrally as appropriate. The existing specification product is a product that implements the existing specifications and has a track record of operation in more than a reliable number of units. The reliable number of units is determined according to whether or not a sufficient amount of training data can be obtained. The reliable number of units may be any number. The target specification product 200 comprises a remote monitoring device 210. The target specification product 200 is a product that implements the target specifications. The number of target specification products 200 with a proven track record of operation is generally significantly less than the reliable number. The remote monitoring device 210 comprises a data acquisition device 211, an anomaly detection device 212, and a notification device 213. The remote monitoring device 210 may be incorporated into the target specification product 200, or it may be a separate device from the target specification product 200.

[0016] The simulation device 110 includes a distribution model acquisition unit 111, a distribution model estimation unit 112, and a reference value setting unit 113.

[0017] The distribution model acquisition unit 111 acquires distribution models corresponding to existing specifications. Specifically, first, the distribution model acquisition unit 111 performs a simulation using the existing specification model 181 and acquires the output of the existing specification model 181. Next, the distribution model acquisition unit 111 acquires each distribution model corresponding to the existing specifications based on the difference between each output of the existing specification model 181 and each data contained in the existing specification field data 191. In this case, the distribution model acquisition unit 111 may acquire multiple distribution models. A distribution model is a model of a distribution. A distribution model may be a probability density function, a distribution function, a histogram, or a set of points corresponding to the difference. The distribution model acquisition unit 111 may also acquire distribution models relating to the features or feature vectors of multiple types of data related to a series of operations such as opening and closing elevator doors. The distribution model acquisition unit 111 may acquire both distribution models relating to each data of multiple types of data and distribution models relating to the features of multiple types of data. Multiple distribution models include, as specific examples, a distribution model for the opening and closing speed of elevator doors and a distribution model for the torque of the motor during elevator door opening and closing. For example, in situations where it is difficult to open and close the elevator doors due to foreign objects, torque data and speed waveform data that deviate from the average values ​​for door opening and closing are obtained. Therefore, it is appropriate to learn the range in which most of the data regarding door opening and closing is observed, and to judge data that deviates by a certain value or more from the learned range as abnormal. Here, it is assumed that the majority of all data is data related to normal door opening and closing.

[0018] The distribution model estimation unit 112 estimates a distribution model corresponding to the distribution of the difference between the output of the target specification model and data indicating the operating status of each target specification product in the target specification product group, based on the existing distribution model and the test distribution model. Here, the estimated target distribution model, the output of the physical model used when generating the existing distribution model and the test distribution model corresponding to the estimated target distribution model, and the data indicating the operating status are all of the same type. The distribution model estimation unit 112 may generate multiple estimated target distribution models. The existing distribution model is a model corresponding to the distribution of the difference between the output of the existing specification model 181 and data indicating the operating status of each existing specification product in the existing specification product group. The test distribution model is a model corresponding to the difference between the output of the target specification model 182 and data indicating the operating status of each test machine in the test machine group. The existing specification product group consists of each existing specification product. The test machine group consists of each test machine. The target specification product group consists of each target specification product. At least a part of the test machine group may overlap with the target specification product group. Specifically, the distribution model estimation unit 112 estimates the distribution models corresponding to the target specifications by correcting the existing distribution models based on the existing distribution models and the distribution models corresponding to the test aircraft, when there is a discrepancy between the distribution models corresponding to the target specifications and the existing distribution models. The distribution model estimation unit 112 incorporates the characteristics of the distribution models corresponding to the target specifications into the existing distribution models. Therefore, basically, the distribution models corresponding to the existing specifications and the distribution models corresponding to the target specifications do not differ significantly. More specifically, first, the distribution model estimation unit 112 performs a simulation using the target specification model 182 and obtains the output of the target specification model 182. Next, the distribution model estimation unit 112 obtains the distribution models corresponding to the test aircraft based on the output of the target specification model 182 and the test aircraft data 192, similar to the distribution model acquisition unit 111. Next, the distribution model estimation unit 112 estimates the distribution models corresponding to the target specifications by combining the existing distribution models with the distribution models corresponding to the test aircraft.In this case, the distribution model estimation unit 112 may add the distribution models corresponding to the existing specifications and the distribution models corresponding to the test machine. Furthermore, if the difference between the existing specifications and the target specifications is relatively small, the distribution model estimation unit 112 may use the distribution model corresponding to the existing specifications as is, and use that as the distribution model corresponding to the target specifications. Also, for mechanisms with the same structure between the existing specifications and the target specifications, the distribution model estimation unit 112 may reuse the distribution model corresponding to the existing specifications as the distribution model corresponding to the target specifications. For each mechanism in the target specifications, the distribution model estimation unit 112 may select an existing specification that has the same mechanism as the mechanism in the target specifications, and reuse the distribution model corresponding to that same mechanism in the selected existing specification. For each mechanism in the target specifications that does not have an identical or similar mechanism in the existing specifications, the distribution model estimation unit 112 may use a distribution model obtained based on the physical model for each mechanism and the results of individual tests for each mechanism as the distribution model corresponding to the existing specifications for that mechanism.

[0019] The reference value setting unit 113 sets abnormality determination criteria to determine whether the data indicating the operating status of each target specification product in the target specification product group is abnormal, based on outliers in the estimated target distribution model. Specifically, the reference value setting unit 113 identifies outliers for each distribution model corresponding to the target specification and sets abnormality determination criteria based on the range in which the outliers exist. The definition of an outlier is assumed to be predetermined. Specific examples of outlier definitions include the total number of outliers being a certain percentage of the total number of data, or the distance from the mean or centroid of all data being greater than or equal to a certain value. The reference value setting unit 113 may also use the values ​​of the operating status data corresponding to each boundary, which are set such that the total number of data outside each boundary in the distribution model is a certain percentage of the total data in the distribution model, as abnormality determination criteria. There may be only one boundary or there may be multiple boundaries. The reference value setting unit 113 calculates the centroid of a feature vector that aggregates multiple types of data related to a series of operations, and may use a value of the operating status data corresponding to a threshold set such that the total number of data whose distance from the centroid is greater than or equal to a threshold is a certain percentage of the total data of the distribution model as the abnormality judgment criterion. The operating status data is data that indicates the operating status. When a field data is determined to be abnormal based on the abnormality judgment criterion, the difference corresponding to that field data is an outlier.

[0020] Existing specification model 181 is a physical model corresponding to the existing specification. Target specification model 182 is a physical model corresponding to the target specification.

[0021] Physical models include, for example, a model of the entire elevator, a model of the elevator belt, a model of the elevator door, or a model of the elevator motor. The types of data output by the physical model are the same as the types of data included in the field data. The physical model uses elements within the physical model to create variability in the output of the physical model. If it is not possible to create variability using elements within the physical model, variability may be added to the output of the physical model. Specific examples of variability include individual differences in each mechanism, variability in behavior based on the elasticity of parts, variability in behavior based on friction between parts, or a combination of these. Various properties of variability may be set by setting the variance of the distribution or the shape of the distribution function. Specific examples of the output of the physical model include the door speed and motor torque during elevator opening and closing. In this case, a specific example of a variability factor is friction (travel loss). Specific examples of door speed vary by approximately ±5 km / h due to friction.

[0022] The existing specification field data 191 consists of field data for each existing specification product. In other words, the existing specification field data 191 consists of data indicating the operating status of each existing specification product in the existing specification product group.

[0023] The test aircraft data 192 consists of data acquired from each test aircraft. In other words, the test aircraft data 192 consists of data showing the operational status of each test aircraft in the test aircraft group. The types of data included in the test aircraft data 192 are the same as the types of data included in the existing specification field data 191. A test aircraft is an aircraft that implements the target specification and is prepared for testing related to the target specification. A test aircraft may be located at the development site or at the sales site of the test aircraft. There may be multiple test aircraft. However, basically, it is not possible to obtain sufficient training data from test aircraft alone.

[0024] The target specification field data 193 consists of field data for each target specification product 200. In other words, the target specification field data 193 consists of data indicating the operating status of each target specification product in the target specification product group. The target specification field data 193 corresponds to the target specification database. The types of data included in the target specification field data 193 are the same as the types of data included in the existing specification field data 191. The target specification field data 193 is data used when updating the distribution model corresponding to the target specification.

[0025] The judgment criterion value 194 consists of various anomaly judgment criteria. Each anomaly judgment criterion corresponds to each target distribution model and is a criterion for determining whether the data indicating the operating status of the target specification product is abnormal or not.

[0026] The simulation device 110 can be applied to any machine having a drive unit. However, it is a condition that the physical models of each mechanism of the machine can be constructed; however, at the time of filing this application, it can be assumed that there are virtually no machines that cannot be modeled. In particular, applying the simulation device 110 to products that have both standard and rare specifications is effective. Standard specifications correspond to models for which a sufficient number of machines are in operation. Standard specifications correspond to existing specifications. Rare specifications correspond to models for which a sufficient number of machines are not in operation. Rare specifications correspond to target specifications. Note that rare specifications may change to standard specifications if a large number of machines corresponding to the rare specifications are sold.

[0027] The following are examples of products to which the simulation device 110 is expected to be applied, and the rare specifications corresponding to each product. Existing specifications and target specifications may also correspond to any of the following products. If the product is an elevator, specifications corresponding to a new model that has improved the mechanism and control of an existing model will be considered rare specifications immediately after the release of the new model. Also, in models with customizable elements, specifications corresponding to an elevator with special customizations selected by a small number of users will be considered rare specifications. As specific examples, regarding door specifications, specifications with a wider width than usual, specifications with a greater mass than usual due to decoration, specifications for doors that can accommodate higher ceilings than usual, and specifications for doors with higher airtightness than usual will be considered rare specifications. If the product is an escalator, specifications with a greater step mass than usual due to the incorporation of lighting equipment, and specifications for operating at a faster or slower speed than usual will be considered rare specifications. If the product is an air conditioner, specifications with a higher output than usual to accommodate a relatively large room, and specifications with longer piping than usual will be considered rare specifications. If the product is a refrigerator, specifications that allow it to operate at a lower temperature than usual, or specifications for use in keeping a larger space than a typical warehouse cool, are considered rare specifications. If the product is a railway motor, specifications that accommodate railway motors used with a different control method than usual in order to reuse existing control devices are considered rare specifications. If the product is an automobile motor, specifications that accommodate automobile motors for vehicles that have been modified to increase their weight are considered rare specifications. If the product is a production device or robot arm, specifications that accommodate products with a larger stroke than usual for the production of large parts are considered rare specifications. If the product is a generator, specifications that accommodate products with components made of different materials than many other products of the same type due to differences in the production period of the generator are considered rare specifications. If the product is a self-propelled robot, specifications that allow it to operate in special environments, or specifications that make the weight of the movable parts heavier than usual due to decoration or load, are considered rare specifications. A special environment is, specifically, an environment in which at least one of the following is outside the normal range: the amount of dust, humidity, or temperature.

[0028] The data acquisition device 211 periodically uploads the collected field data to the cloud system 100.

[0029] The anomaly detection device 212 determines whether the data to be diagnosed is abnormal by referring to the judgment criterion value 194. Note that when both a distribution model for each of the multiple types of data and a distribution model for the features of the multiple types of data are obtained, even if each of the multiple types of data included in the data to be diagnosed is judged to be normal when evaluated individually, it may be judged to be abnormal when the multiple types of data are evaluated together. In this case, the anomaly detection device 212 determines that the data to be diagnosed is abnormal.

[0030] The notification device 213 notifies the call center if the abnormality detection device 212 determines that the data to be diagnosed is abnormal. The notification device 213 may also notify a communication device owned by the user. After the notification, maintenance personnel may be dispatched to the site.

[0031] Figure 2 shows an example configuration of the simulation system 90 when edge devices are not being used. The differences between Figure 1 and Figure 2 will be explained below. The simulation device 110 further includes an anomaly detection device 120. The anomaly detection device 120 corresponds to a device that combines the anomaly detection device 212 and the notification device 213. The target specification product 200 includes a data collection device 221. The data collection device 221 is the same as the data collection device 211. However, the data collection device 221 uploads the collected field data to the cloud system 100 at a relatively high frequency.

[0032] Figure 3 shows an example of the hardware configuration of the simulation device 110 according to this embodiment. The simulation device 110 consists of a computer. The simulation device 110 may consist of multiple computers.

[0033] As shown in this figure, the simulation device 110 is a computer equipped with hardware such as a processor 11, memory 12, auxiliary storage device 13, input / output interface (IF) 14, and communication device 15. These hardware components are appropriately connected via signal lines 19.

[0034] The processor 11 is an IC (Integrated Circuit) that performs arithmetic processing and controls the hardware of the computer. Specific examples of the processor 11 include a CPU (Central Processing Unit), a DSP (Digital Signal Processor), or a GPU (Graphics Processing Unit). The simulation device 110 may have multiple processors that replace the processor 11. The multiple processors share the role of the processor 11.

[0035] Memory 12 is typically a volatile storage device, specifically RAM (Random Access Memory). Memory 12 is also called main memory. Data stored in memory 12 is saved to auxiliary storage device 13 as needed.

[0036] The auxiliary storage device 13 is typically a non-volatile storage device, specifically a ROM (Read Only Memory), HDD (Hard Disk Drive), or flash memory. Data stored in the auxiliary storage device 13 is loaded into the memory 12 as needed. The memory 12 and the auxiliary storage device 13 may be configured as an integrated unit.

[0037] Input / Output IF14 is a port to which input and output devices are connected. A specific example of an input / output IF14 is a USB (Universal Serial Bus) terminal. Specific examples of input devices include a keyboard and mouse. Specific examples of output devices include a display.

[0038] The communication device 15 consists of a receiver and a transmitter. A specific example of the communication device 15 is a communication chip or a NIC (Network Interface Card).

[0039] Each part of the simulation device 110 may use the input / output IF 14 and the communication device 15 as appropriate when communicating with other devices.

[0040] The auxiliary storage device 13 stores the simulation program. The simulation program is a program that allows the computer to implement the functions of each part of the simulation device 110. The simulation program is loaded into memory 12 and executed by the processor 11.

[0041] Data used when running the simulation program, and data obtained by running the simulation program, are appropriately stored in the memory device. Each part of the simulation device 110 makes appropriate use of the memory device. The memory device consists of, as a specific example, memory 12, auxiliary storage device 13, registers in the processor 11, and at least one of the cache memory in the processor 11. Note that the terms data and information may have the same meaning. The memory device may be independent of the computer. The functions of memory 12 and auxiliary storage device 13 may be implemented by other memory devices.

[0042] The simulation program may be recorded on a computer-readable non-volatile recording medium. Specific examples of non-volatile recording media include optical discs or flash memory. The simulation program may also be provided as a program product.

[0043] ***Explanation of Operation*** The operation procedure of the simulation device 110 corresponds to the simulation method. The program that implements the operation of the simulation device 110 corresponds to the simulation program.

[0044] FIG. 4 is a flowchart showing an example of the operation of the simulation apparatus 110. The operation will be described with reference to FIG. 4. As a specific example, the processing shown in this flowchart is executed after the start of sales of the target specification product which is a new model. As a specific example, by executing the processing shown in this flowchart, the operation of a diagnostic service for the target specification product can be started.

[0045] (Step S101) A distribution model acquisition unit 111 acquires each distribution model corresponding to an existing specification based on an existing specification model 181 and existing specification field data 191.

[0046] (Step S102) A user constructs a target specification model 182.

[0047] (Step S103) First, a distribution model estimation unit 112 acquires each distribution model corresponding to a test airframe based on a target specification model 182 and test airframe data 192. Next, the distribution model estimation unit 112 estimates each distribution model corresponding to a target specification based on each distribution model corresponding to an existing specification and each distribution model corresponding to the test airframe.

[0048] (Step S104) A reference value setting unit 113 sets an abnormality determination criterion corresponding to each distribution model based on each distribution model corresponding to a target specification and a predefined definition of an outlier, and registers each abnormality determination criterion in a determination reference value 194.

[0049] ***Description of Effects of Embodiment 1*** In the present embodiment, each distribution model corresponding to an existing specification is utilized to estimate each distribution model corresponding to a target specification. As a specific example, for elevator doors, when hundreds of types of doors can be selected, sufficient field data corresponding to each target specification cannot be collected. Meanwhile, according to the present embodiment, each distribution model corresponding to each target specification can be estimated in such a case. By utilizing the present embodiment, an AI (Artificial Intelligence)-based failure diagnosis service can be introduced at an early stage even for an airframe corresponding to a specification for which the amount of collected field data is relatively small.

[0050] In this embodiment, anomalies are detected based on the distribution of differences. Therefore, according to this embodiment, anomalies can be detected even when the mechanism of failure is unknown. For example, if the dimensions and weight of a door are changed, a physical model corresponding to the door can be created. However, the mechanism of failure is unknown regarding the dimensions and weight of the door after the change, and outliers cannot be identified from the physical model. On the other hand, according to this embodiment, anomalies can be detected even in such cases.

[0051] ***Other Configurations*** <Modification 1> The dimensions and speed differ between the existing specifications and the target specifications. Therefore, the distribution model corresponding to the existing specifications cannot be used as is as the distribution model corresponding to the target specifications. For example, if the distribution model corresponding to the existing specifications is used as is as the distribution model corresponding to the target specifications, a 1m deviation when an object moves 100m and a 1m deviation when an object moves 90m may be treated as equivalent. Here, these two types of deviations may have different meanings. It is also conceivable to convert the distribution model by multiplying the distribution corresponding to the existing specifications by a certain ratio, but there is no way to determine the conversion ratio. Here, it is considered that an effective means is to combine a distribution model based on data related to the target specifications, which exists but is not in sufficient quantity, with the distribution model corresponding to the existing specifications, using Bayesian estimation or the like. Therefore, the distribution model estimation unit 112 in this modification estimates the distribution model corresponding to the target specifications by combining at least one distribution model corresponding to the existing specifications and a distribution model corresponding to the test machine using Bayesian estimation.

[0052] The distribution model estimation unit 112 estimates the target distribution model using Bayesian estimation. Figure 5 is a diagram illustrating the processing of the distribution model estimation unit 112. The distribution model estimation unit 112 uses the distribution corresponding to the existing specifications as the prior probability and the distribution corresponding to the test machine as the likelihood to calculate the posterior probability. The prior distribution is a distribution generated based on a large amount of data. Therefore, the reliability of the prior distribution is relatively high. However, due to differences between the existing specifications and the target specifications, the prior distribution may not fit the target specifications. On the other hand, the likelihood is a distribution generated based on a small amount of data. However, since the specifications corresponding to the likelihood are the target specifications, the reliability of the likelihood as a distribution corresponding to the target specifications is relatively high. Therefore, by combining the prior distribution and the likelihood, the reliability of the distribution model corresponding to the target specifications can be increased. Note that if there are differences in dimensions such as aperture width between specifications, the distribution may change depending on the proportion of the difference. Therefore, it is appropriate to combine a distribution generated based on a large amount of data corresponding to specifications different from the target specifications with a distribution generated based on a small amount of data corresponding to the target specifications.

[0053] In this modified example, Bayesian estimation, which has many common applications, is used as a method to specifically estimate the data distribution of the target specifications. Therefore, according to this embodiment, the reliability of the distribution model corresponding to the target specifications can be improved.

[0054] <Modification 2> The target specification product 200 is not sold in large quantities like the existing specification product. However, as the target specification product 200 is in operation, data accumulates in the target specification field data 193. Therefore, in this modification, when sufficient data has accumulated in the target specification field data 193 beyond a predetermined standard, the distribution model estimation unit 112 reconstructs each distribution model corresponding to the target specification based on each distribution model corresponding to the existing specification, the target specification model 182, and the target specification field data 193. Furthermore, the standard value setting unit 113 in this modification updates the judgment standard value 194 based on each reconstructed distribution model. That is, in this modification, when the data indicating the operating status of each target specification product in the target specification product group has accumulated in the target specification database beyond the update standard amount, the distribution model estimation unit 112 updates the estimated target distribution model to an updated distribution model based on the existing distribution model, the output of the target specification model 182, and the data indicating the operating status of each target specification product in the target specification product group accumulated in the target specification database. The reference value setting unit 113 updates the abnormality judgment criteria based on outliers in the updated distribution model. The updated reference amount can be defined in any way.

[0055] As a specific example, the distribution model estimation unit 112 acquires each distribution model corresponding to the target specification field data 193 based on the target specification model 182 and the target specification field data 193, every time the number of field data corresponding to the target specification in the target specification field data 193 increases by 1000. Subsequently, the distribution model estimation unit 112 reconstructs each distribution model corresponding to the target specification by combining each distribution model corresponding to the target specification field data 193 and each distribution model corresponding to the existing specification using Bayesian estimation. After that, the reference value setting unit 113 updates the judgment reference value 194 based on each reconstructed distribution model. Note that if sufficient data has been accumulated in the target specification field data 193, the distribution model estimation unit 112 may acquire each distribution model corresponding to the target specification based on the target specification model 182 and the target specification field data 193.

[0056] According to this modified version, the judgment criterion value 194 can be set by giving more weight to the data collected for each aircraft corresponding to the target specification. Furthermore, the judgment criterion value 194 is a standard that takes into account the smaller amount of field data corresponding to the target specification compared to existing specifications.

[0057] <Modification 3> In this modification, the existing specification field data 191, the test unit data 192, and the target specification field data 193 each contain attribute information about the unit from which data is collected. Specific examples of attribute information include information about the unit's operating status and usage environment. Specific examples of information about the unit's operating status include information indicating the unit's operating frequency or duration, and the load on the unit. Specific examples of information about the unit's usage environment include information indicating the temperature and humidity of the unit's installation location, the region where the unit is installed, and whether the installation location is indoors. If the unit is an elevator, the attribute information includes information indicating the unit's travel frequency, whether it is used to transport cargo exceeding a certain mass, and whether the landing area is exposed to the outdoors. Note that the parameters of each physical model may be changed according to the attributes.

[0058] The distribution model acquisition unit 111 in this modified example appropriately classifies the aircraft according to attribute information and acquires a distribution model corresponding to each classification based on the field data collected for each aircraft corresponding to each classification.

[0059] The distribution model estimation unit 112 in this modified example selects each distribution model corresponding to the existing specifications according to the classification of attributes corresponding to the test machine, and estimates each distribution model corresponding to the target specifications based on each selected distribution model. The distribution model estimation unit 112 may also select each distribution model corresponding to the existing specifications according to the similarity of attributes.

[0060] In other words, in this modified example, each existing specification product in the existing specification product group has a target attribute. Each test unit in the test unit group has a target attribute. Each target specification product in the target specification product group has a target attribute. A target attribute is an attribute that corresponds to at least one of the operating conditions and the usage environment. A target attribute may be a collective term for multiple attributes that are similar to each other.

[0061] As a concrete example, the processing of each part in a case where a specific event is likely to occur in cold regions and less likely to occur outside of cold regions will be explained. The distribution model acquisition unit 111 acquires each distribution model corresponding to the existing specifications, which is based on field data collected from each unit installed in a cold region and each distribution model based on field data collected from each unit installed outside of cold regions. The distribution model estimation unit 112 estimates each distribution model corresponding to cold regions and the target specifications based on each distribution model corresponding to cold regions and the existing specifications. The distribution model estimation unit 112 estimates each distribution model corresponding to outside of cold regions and the target specifications based on each distribution model corresponding to outside of cold regions and the existing specifications. The reference value setting unit 113 sets each abnormality judgment criterion corresponding to cold regions based on each distribution model corresponding to cold regions and the target specifications, and sets each abnormality judgment criterion corresponding to outside of cold regions based on each distribution model corresponding to outside of cold regions and the target specifications.

[0062] In this modified version, when each distribution model differs significantly depending on the attribute, the distribution model corresponding to the target specification is estimated by considering the attribute. Therefore, according to this modified version, a more accurate judgment criterion value 194 can be set.

[0063] <Modification 4> In this modification, the parameters of the physical model are classified into parameters affected by the specifications and parameters not affected by the specifications. Parameters not affected by the specifications are identified in such a way that the difference between the output of the existing specification model 181 and the mean of the field data is small.

[0064] It is possible to determine in advance which parts are common to the existing specifications and which are not common to the target specifications. Here, it is presumed that parts with the same structure or components between the existing specifications and the target specifications will have the same characteristics. Therefore, using the parameter values ​​(such as friction or elasticity) identified for the existing specifications in the target specification model 182 contributes to improving the accuracy of simulations related to the target specifications. On the other hand, if parameters identified based on the existing specifications are used in the target specification model 182 for parts with different structures or components between the existing specifications and the target specifications, the accuracy of the target specification model 182 may decrease. That is, in this modified example, the values ​​of each parameter in the target specification model 182 that correspond to the common parts are the values ​​of each parameter in the existing specification model 181 that correspond to the common parts. The common parts are the parts that are common between the existing specifications and the target specifications. For each parameter in the target specification model 182 that relates to parts that are not common between the existing specifications and the target specifications, they may be individually identified by other means such as unit testing, or standard values ​​may be substituted.

[0065] This modified version allows for simulations that more closely resemble reality, thereby improving the accuracy of AI-based judgments.

[0066] <Modification 5> Figure 6 shows an example of the hardware configuration of the simulation device 110 according to this modification. The simulation device 110 includes a processing circuit 18 instead of a processor 11, a processor 11 and memory 12, a processor 11 and auxiliary storage device 13, or a processor 11, memory 12 and auxiliary storage device 13. The processing circuit 18 is hardware that realizes at least a part of each part of the simulation device 110. The processing circuit 18 may be dedicated hardware, or it may be a processor that executes a program stored in memory 12.

[0067] If the processing circuit 18 is dedicated hardware, specific examples of the processing circuit 18 include a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a combination thereof. The simulation device 110 may also include multiple processing circuits that substitute for the processing circuit 18. The multiple processing circuits share the role of the processing circuit 18.

[0068] In the simulation device 110, some functions may be implemented by dedicated hardware, while the remaining functions may be implemented by software or firmware.

[0069] The processing circuit 18 can be implemented, in specific examples, by hardware, software, firmware, or a combination thereof. The processor 11, memory 12, auxiliary storage device 13, and processing circuit 18 are collectively referred to as the "processing circuitry." In other words, the functions of each functional component of the simulation device 110 are implemented by the processing circuitry.

[0070] ***Other Embodiments*** Although Embodiment 1 has been described, multiple parts of this embodiment may be combined and implemented. Alternatively, this embodiment may be implemented partially. Furthermore, this embodiment may be modified in various ways as needed, and may be implemented as a whole or in any combination. Note that the embodiments described above are essentially preferred examples and are not intended to limit this disclosure, its applications, or the scope of use. Procedures described using flowcharts, etc., may be modified as appropriate.

[0071] 11 Processor, 12 Memory, 13 Auxiliary storage device, 14 Input / Output IF, 15 Communication device, 18 Processing circuit, 19 Signal line, 90 Simulation system, 100 Cloud system, 110 Simulation device, 111 Distribution model acquisition unit, 112 Distribution model estimation unit, 113 Reference value setting unit, 120 Anomaly detection device, 181 Existing specification model, 182 Target specification model, 191 Existing specification field data, 192 Test unit data, 193 Target specification field data, 194 Judgment reference value, 200 Target specification product, 210 Remote monitoring device, 211 Data acquisition device, 212 Anomaly detection device, 213 Notification device, 221 Data acquisition device.

Claims

1. A simulation device comprising: an existing distribution model that corresponds to the distribution of the difference between the output of an existing specification model, which is a physical model corresponding to an existing specification corresponding to a product with a track record of operation exceeding a reliable number of units, and data indicating the operating status of each existing specification product in a group of existing specification products consisting of each existing specification product on which the existing specification is implemented; a test distribution model that corresponds to the difference between the output of a target specification model, which is a physical model corresponding to a target specification, and data indicating the operating status of each test unit in a group of test units consisting of each test unit on which the target specification is implemented; a distribution model estimation unit that estimates a distribution model that corresponds to the distribution of the difference between the output of the target specification model and data indicating the operating status of each target specification product in a group of target specification products consisting of each target specification product on which the target specification is implemented, as an estimated target distribution model; and a reference value setting unit that sets an anomaly determination criterion for determining whether the data indicating the operating status of each target specification product in the group of target specification products is abnormal, based on outliers in the estimated target distribution model.

2. The simulation apparatus according to claim 1, wherein the distribution model estimation unit estimates the target distribution model by Bayesian estimation.

3. The simulation apparatus according to claim 1 or 2, wherein the distribution model estimation unit updates the estimated target distribution model to an updated distribution model based on the existing distribution model, the output of the target specification model, and the data indicating the operating status of each target specification product in the target specification product group stored in the target specification database, when the data indicating the operating status of each target specification product in the target specification product group exceeds an update threshold amount in the target specification database, and the threshold value setting unit updates the abnormality judgment criterion based on outliers in the updated distribution model.

4. The simulation apparatus according to any one of claims 1 to 3, wherein each existing specification product in the existing specification product group has a target attribute corresponding to at least one of the operating status and the usage environment, each test unit in the test unit group has the target attribute, and each target specification product in the target specification product group has the target attribute.

5. The simulation apparatus according to any one of claims 1 to 4, wherein the values ​​of each parameter in the target specification model that correspond to the common portion between the existing specification and the target specification are the values ​​of each parameter in the existing specification model that correspond to the common portion.

6. A simulation method in which a computer estimates a distribution model corresponding to the distribution of the difference between the output of an existing specification model, which is a physical model corresponding to an existing specification for a product with a proven track record of operation in more than a reliable number of units, and data indicating the operating status of each existing specification product in a group of existing specification products, which consists of each existing specification product on which the existing specification is implemented, as an estimated target distribution model, based on the output of a target specification model, which is a physical model corresponding to a target specification, and data indicating the operating status of each test unit in a group of test units, which consists of each test unit on which the target specification is implemented, and sets an anomaly determination criterion for determining whether the data indicating the operating status of each target specification product in the group of target specification products is abnormal or not, based on outliers in the estimated target distribution model.

7. A simulation program that causes a computer simulation device to execute the following: an existing distribution model that corresponds to the distribution of the difference between the output of an existing specification model, which is a physical model corresponding to an existing specification corresponding to a product with a track record of operation exceeding a reliable number of units, and data indicating the operating status of each existing specification product in a group of existing specification products consisting of each existing specification product on which the existing specification is implemented; a test distribution model that corresponds to the difference between the output of a target specification model, which is a physical model corresponding to a target specification, and data indicating the operating status of each test unit in a group of test units consisting of each test unit on which the target specification is implemented; a distribution model estimation process that estimates a distribution model corresponding to the distribution of the difference between the output of the target specification model and data indicating the operating status of each target specification product in a group of target specification products consisting of each target specification product on which the target specification is implemented, as an estimated target distribution model; and a reference value setting process that sets an anomaly determination criterion for determining whether the data indicating the operating status of each target specification product in the group of target specification products is abnormal, based on outliers in the estimated target distribution model.

Images