JP6641056B1 - Device abnormality diagnosis method and device abnormality diagnosis system - Google Patents

Abstract

To provide a device abnormality diagnosis technique capable of performing diagnosis without depending on the skill of an engineer. An apparatus abnormality diagnosis method includes the steps of: adjusting target data so that the number of dimensions of target data to be diagnosed included in data indicating the state of the device is a specific dimension; The method includes a step of analyzing the adjusted target data using an algorithm that handles data of a specific number of dimensions, and a step of displaying the target data in a form in which the presence or absence of an abnormality can be identified based on the analysis of the target data. [Selection diagram] FIG.

Description

  An embodiment of the present invention relates to a device abnormality diagnosis method and a device abnormality diagnosis system.

  If the equipment provided in the plant stops or fails unplannedly, the operation of the plant is affected and the damage is increased. Therefore, it is required to monitor the state of the device and appropriately evaluate its soundness. Also, if there is a sign of abnormality as a result of the evaluation, it is necessary to clarify the cause and take an early action.

  Plant equipment includes rotating equipment such as a motor or a pump, piping, valves, cables, and the like. These soundnesses are evaluated by measuring vibration, temperature, current, pressure, flow rate, sound, etc. according to the type of equipment and monitoring the changes. However, it is difficult for a human to precisely diagnose each device because the number of devices to be monitored in the plant or the number of monitoring items is large. Further, depending on the device, noise or individual difference is large, and an appropriate diagnosis requires experience and skill of an engineer.

  For this reason, a method for performing an efficient diagnosis regardless of the skill level of a technician has been considered. For example, a technique is known in which a set of a time change of a vibration or an acoustic spectrum at the time of abnormality of a rotating device and a peak frequency is registered in advance as reference data, and the presence or absence of an abnormality or a cause is identified by collating with a spectrum of measurement data. I have. Further, there is known a technique of sensing and collecting a drive signal of a motor-operated valve and performing abnormality diagnosis using a preset allowable value of a diagnostic item and a diagnostic record. In recent years, development of an efficient and effective abnormality diagnosis technique using a diagnosis method using AI has been promoted. For example, a technique is known in which normal or abnormal data is accumulated and the presence or absence of an abnormality is determined by machine learning.

JP-A-10-274558 JP 2000-65246 A JP 2019-67069 A

  In the above-described technique, a threshold value serving as a reference is used for determining normality or abnormality. As for the standard, an engineer needs to set a standard value for each device. The setting of this reference value is performed by a skilled engineer while considering past measurement data or the specifications of the target device, and therefore requires time and effort. For this reason, the conventional technique still has a dependency on the experience and skill of the engineer.

  An embodiment of the present invention has been made in consideration of such circumstances, and has as its object to provide a device abnormality diagnosis technique capable of performing a diagnosis without depending on the skill of an engineer.

The device abnormality diagnosis method according to the embodiment of the present invention is a method for extracting, from data indicating at least one of a time waveform and a frequency spectrum of a signal which is data indicating a state of the device, a diagnosis among data indicating the state of the device. individual comprising the steps of adjustment to a specific number of dimensions the number of dimensions is the number of data, the target data adjusted to the specific number dimension, the specific number of dimensions of data included in the first target data to be Performing a topological data analysis using a mapper , which is an algorithm that handles the data processing, and displaying the analysis result using the mapper in a cluster display or a time-series display in a mode in which the presence or absence of an abnormality can be identified. .

  According to the embodiments of the present invention, a device abnormality diagnosis technique capable of performing diagnosis without depending on the skill of an engineer is provided.

FIG. 1 is a configuration diagram illustrating a device abnormality diagnosis system according to an embodiment. FIG. 2 is a block diagram showing an analysis processing unit. FIG. 3 is a block diagram showing a data preprocessing unit. FIG. 3 is a block diagram showing a process setting unit. FIG. 3 is a block diagram showing a parameter tuning unit. FIG. 4 is a block diagram showing a result display control unit. FIG. 3 is a block diagram illustrating a display adjustment unit. FIG. 3 is a block diagram illustrating an abnormality cause analysis unit. Explanatory drawing which shows an example of the data of the acceleration amplitude after preprocessing. FIG. 4 is an explanatory diagram showing an example of frequency data after preprocessing. 9 is a table illustrating an example of a correlation between a parameter and an analysis result. FIG. 4 is a screen diagram showing an example of a cluster display. FIG. 4 is a screen diagram showing an example of a time-series display. 5 is a flowchart illustrating a method for diagnosing device abnormality. 9 is a flowchart illustrating abnormality diagnosis processing. 9 is a flowchart showing data preprocessing. 9 is a flowchart illustrating a display adjustment process. FIG. 9 is an explanatory diagram showing an analysis result using hierarchical clustering of a modified example. FIG. 9 is an explanatory diagram showing an analysis result using hierarchical clustering of a modified example. FIG. 9 is an explanatory diagram showing an analysis result using hierarchical clustering of a modified example.

  Hereinafter, embodiments of a device abnormality diagnosis method and a device abnormality diagnosis system will be described in detail with reference to the drawings.

  Reference numeral 1 in FIG. 1 denotes an apparatus abnormality diagnosis system according to the present embodiment. This abnormality diagnosis system 1 is a system that diagnoses an abnormality of a rotating machine 2 as a device provided in a facility such as a power plant. The rotating machine 2 is, for example, a motor mounted on a pump or a blower.

  For example, when there is an abnormality such as a failure in the rotating machine 2 or when there is a sign of the abnormality, the acceleration amplitude or the specific frequency component of the vibration is different from that in the normal state. If the difference between the normal state and the abnormal state is large, it is easy to find the abnormality. If the difference is small, it is difficult to find the abnormality. In addition, the amount of data relating to the phenomena occurring during the operation of the rotating machine 2 as an apparatus is enormous, and it is difficult to manually analyze such a large amount of data.

  In recent years, with the development of technologies related to machine learning and artificial intelligence, it has become possible to extract useful information from a large amount of data. As an algorithm for such data analysis, for example, there is a topological data analysis (TDA) or a cluster analysis (Clustering). The abnormality diagnosis system 1 of the present embodiment diagnoses an abnormality of the rotating machine 2 (device) using an algorithm that can handle a large amount of data.

  In the conventional analysis technology, processing of dimension reduction is necessary when handling large-scale, multidimensional data. For example, in dimension reduction by a statistical method, there is a problem that information in data is lost. In addition, the abnormality diagnosis using supervised learning also has a problem that it is difficult to construct a learning model if there are few abnormal data labels.

  Therefore, in the present embodiment, an abnormality is diagnosed from a data group by using a data analysis technique that can analyze data without a teacher and focus on topological features of the data group.

  For example, in conventional topological data analysis, manual judgment is required for setting parameters for analysis or interpreting analysis results. On the other hand, in the present embodiment, characteristic data (abnormality) is obtained by sweeping analysis parameters used in topological data analysis and analyzing output results (for example, the number of data, the number of edges, and the number of nodes). Data) can be extracted. Further, it is possible to estimate which parameter contributes to the abnormal event.

  As shown in FIG. 1, a user of the abnormality diagnosis system 1 obtains vibration data of the rotating machine 2 by attaching an inspection sensor 3 (for example, a vibration meter) to the rotating machine 2 provided in the plant. Note that the monitoring data, which is data indicating the operating state of the rotating machine 2, includes data such as sensors for collecting operation sounds and thermography, in addition to vibration data. The inspection sensor 3 may be a permanent one, or may be installed every time a measurement is performed.

  The monitoring data is at least one of a time waveform and a frequency spectrum of a signal generated from the inspection sensor 3 attached to the rotating machine 2 as an apparatus. Further, the monitoring data may be an absolute value of a time waveform or a frequency spectrum of the signal, or an amount representing a characteristic of the signal or a combination thereof.

  It should be noted that the state quantities targeted by the vibration data include, for example, displacement, speed, acceleration, and the like. Further, the user can select an arbitrary measurement condition such as the number of sampling points or the frequency range. In this way, the abnormality of the rotating machine 2 can be diagnosed based on the vibration data that can be easily acquired by the inspection sensor 3.

  As the monitoring data, for example, vibration data measured by the inspection sensor 3 is exemplified. Note that the monitoring data includes not only continuous data that constantly monitors the state of the rotating machine 2 but also intermittent data that is periodically measured. In the following description, vibration data will be described as an example, but other data may be used.

  It should be noted that the monitoring data includes, in addition to the primary data indicating the main purpose event achieved by the operation of the rotating machine 2, secondary secondary data generated by the operation of the rotating machine 2. . For example, in the case of the rotating machine 2, the number of rotations is primary data. Then, vibration data other than the main purpose event generated by the rotation operation becomes secondary data. In the present embodiment, an example in which abnormality of the rotating machine 2 is diagnosed using secondary data will be described. Note that the abnormality of the rotating machine 2 may be diagnosed using the primary data.

  Then, the monitoring data is input to the abnormality diagnosis system 1 for analysis. For example, the monitoring data is input to the abnormality diagnosis system 1 via a user terminal 5 (for example, a PC) connected to the Internet 4. The monitoring data may be directly input to the abnormality diagnosis system 1.

  This abnormality diagnosis system 1 may be mounted on a cloud server outside the plant, or may be mounted on a personal computer or server inside the plant. Moreover, after collecting the monitoring data in an electric device such as a server or a predetermined memory, the collected data may be transferred to the abnormality diagnosis system 1.

  The inspection sensor 3 detects, for example, accelerations in three X-axis, Y-axis, and Z-axis directions. Thereby, the vibration (data) waveform of the acceleration of the rotating machine 2 can be obtained.

  Next, the system configuration of the abnormality diagnosis system 1 will be described with reference to the block diagrams shown in FIGS.

  As shown in FIG. 1, the abnormality diagnosis system 1 includes a main control unit 6, an information input unit 7, a communication unit 8, a display output unit 9, and an analysis processing unit 10. These are realized by executing a program stored in the memory or the HDD by the CPU.

  Further, the abnormality diagnosis system 1 includes a monitoring database 11, a process database 12, a specification database 13, and an abnormality cause database 14. These are collections of information stored in a memory or HDD and organized so that they can be searched or stored.

  The abnormality diagnosis system 1 according to the present embodiment has hardware resources such as a CPU, a ROM, a RAM, and an HDD. When the CPU executes various programs, information processing by software is realized using the hardware resources. It is composed of a computer. Further, the abnormality diagnosis method according to the present embodiment is realized by causing a computer to execute a program.

  The main control unit 6 controls the abnormality diagnosis system 1 overall.

  The information input unit 7 receives predetermined information according to an operation of a user who uses the abnormality diagnosis system 1. The information input unit 7 includes an input device such as a mouse or a keyboard. That is, predetermined information is input to the information input unit 7 in accordance with the operation of these input devices. Predetermined information such as monitoring data is input to the information input unit 7.

  The communication unit 8 communicates with an electronic device such as the user terminal 5 via a communication line such as the Internet 4. This communication unit 8 may be mounted on a wireless communication device. The communication unit 8 may be mounted on a predetermined network device, for example, a wireless LAN access point or an antenna. Note that the communication unit 8 may communicate with the user terminal 5 via a WAN (Wide Area Network) or a mobile communication network.

  The display output unit 9 displays various information indicating the analysis result. For example, screens such as a cluster display (see FIG. 12) and a time-series display (see FIG. 13) are displayed. That is, the display output unit 9 is a graphic user interface (GUI) for displaying the analysis result.

  The abnormality diagnosis system 1 according to the present embodiment includes a display device such as a display that outputs an analysis result. That is, the display output unit 9 controls a screen displayed on the display. The display may be separate from the computer main body, or may be integral with the computer. Further, the display output unit 9 may control a screen displayed on a display of another computer connected via a network.

  In the present embodiment, a display is exemplified as a display device, but other embodiments may be used. For example, information may be displayed using a head-mounted display or a projector. Further, a printer that prints information on a paper medium may be used instead of a display. That is, a target mounted on the display output unit 9 may include a head mounted display, a projector, or a printer.

  The analysis processing unit 10 analyzes target data or past data included in the monitoring data using an algorithm that handles data of a specific number of dimensions (see FIG. 2). In the present embodiment, the monitoring data is exemplified as the data indicating the operating state of the rotating machine 2. However, a data set created by combining the monitoring data and the process data is used during the operation of the rotating machine 2. May be data indicating the state of. In the following description, the term monitoring data may include process data.

  Here, the number of specific dimensions in the present embodiment is four or more. Note that, for example, the analysis processing unit 10 performs analysis using at least one of topological data analysis and cluster analysis. In this way, it is possible to perform analysis using data of a specific number of dimensions.

  The target data is data to be diagnosed included in the monitoring data. Note that the process data acquired at the same time as or around the monitoring data may be included in the target data.

  The past data is data in which the presence or absence of an abnormality included in the monitoring data is known. For example, monitoring data acquired from the rotating machine 2 that is currently operating does not know whether there is any abnormality.

  This monitoring data is to be diagnosed as abnormal data as target data. In addition, the monitoring data acquired from the rotating machine 2 several months or several years ago can be grasped after the data is acquired, whether the rotating machine 2 actually operates normally or is in an abnormal state. I have. Such monitoring data is treated as past data. It should be noted that the process data acquired at the same time as or around the same time as the past monitoring data may be included in the past data.

  The monitoring database 11 stores monitoring data obtained from the rotating machine 2 by the user using the inspection sensor 3.

  The process database 12 stores process data indicating a process related to the rotating machine 2. The process data includes, for example, operating conditions such as rotation speed, flow rate, pressure, temperature, and current, and operating conditions such as an operating period and a use period. The process data may be stored by the user or may be stored in advance.

  The specification database 13 stores specification data indicating the specifications of the rotating machine 2. The specification data includes, for example, in the case of the rotating machine 2, information such as a model, a maker, a size, a head, a flow type, a bearing type, and the number of blades. The specification data may be stored by the user or may be stored in advance.

  The abnormality cause database 14 stores cause information (abnormal cause data) indicating the correspondence between the feature amount of the target data extracted based on the analysis of the analysis processing unit 10 and the abnormal cause of the rotating machine 2. The cause information indicates a correlation between the feature amount of the abnormality and the cause of the abnormality.

  The cause information may be input and set by the user, or may be stored in advance. In this manner, the cause of the abnormality of the rotating machine 2 can be easily estimated based on the correspondence between the feature value and the cause of the abnormality.

  As shown in FIG. 2, the analysis processing unit 10 includes a data acquisition unit 15, a data preprocessing unit 16, a processing setting unit 17, a data analysis unit 18, a parameter tuning unit 19, a result organizing unit 20, a result display control unit 21, The apparatus includes a display adjustment unit 22, an abnormality cause analysis unit 23, a cause information acquisition unit 24, and a cause display control unit 25.

  The data acquisition unit 15 acquires various data from the monitoring database 11, the process database 12, and the specification database 13. In the following description, the term monitoring data may include process data or specification data acquired by the data acquisition unit 15.

  The data preprocessing unit 16 extracts data used for analysis of target data or past data from the monitoring data, and performs preprocessing of these data before being analyzed by the data analysis unit 18 (see FIG. 3). Here, the number of dimensions of the target data or the past data is adjusted to be a specific number of dimensions. Note that the target data and the past data are adjusted so as to have the same specific number of dimensions.

  The process setting unit 17 performs various settings or an algorithm for analysis used when the data analysis unit 18 performs analysis (see FIG. 4).

  The data analysis unit 18 analyzes the target data or the past data adjusted to the specific dimension number using at least one of a topological data analysis and a cluster analysis, which are algorithms for handling data of the specific dimension number. The data analyzer 18 can detect target data indicating an abnormal state from target data included in the monitoring data.

  The parameter tuning unit 19 tunes various settings or analysis algorithm settings used when the data analysis unit 18 performs analysis (see FIG. 5).

  The result organizing unit 20 organizes whether or not the analysis result of the data obtained by the data analyzing unit 18 is optimal. That is, for the results analyzed by the data analysis unit 18, the correspondence between the analysis parameters and the analysis results is arranged.

  If the data analysis result is not optimal, the parameter tuning unit 19 tunes the setting. In other words, the tuning and the data analysis by the parameter tuning unit 19 are performed until an analysis result capable of displaying the target data in a form in which the presence or absence of abnormality of the rotating machine 2 can be identified based on the data analysis by the data analysis unit 18 is obtained. The analysis of the data by 18 is repeated.

  The result display control unit 21 controls the display output unit 9 to display a screen showing a result of analysis of the data obtained by the data analysis unit 18 in a form in which the presence or absence of an abnormality of the rotating machine 2 can be identified. (See FIG. 6). For example, control is performed to cause the display output unit 9 to display a screen such as a cluster display (see FIG. 12) and a time-series display (see FIG. 13).

  Here, an aspect of the present embodiment in which the presence or absence of an abnormality can be identified is, for example, a three-dimensional or lower graph. For example, the cluster display (see FIG. 12) is illustrated as a two-dimensional graph. This cluster display may be a three-dimensional graph. With this configuration, target data of four or more dimensions that cannot be visualized by a normal display technique can be visualized as a graph of three or less dimensions, and can be used to determine whether there is an abnormality in the rotating machine 2. The time-series display (see FIG. 13) is included in a mode in which the presence or absence of an abnormality can be identified.

  When the topological data analysis is used, the result display control unit 21 sorts the abnormality detection results including the number of nodes, the number of edges, the total number of input data, the number of unique data included in the nodes, the number of node groups, and the like. I do. In the following description, a node group may be referred to as a cluster. The result display control unit 21 also arranges analysis parameters for topological data analysis including a metric of distance, a filter function, an interval, an overlap, a clustering method, and the like.

  The display adjustment unit 22 controls a screen to be displayed on the display output unit 9 according to a user operation (see FIG. 7).

  The abnormality cause analysis unit 23 analyzes the cause of the abnormality when the target data has an abnormality based on the analysis result of the data obtained by the data analysis unit 18 (see FIG. 8).

  The cause information acquisition unit 24 acquires cause information from the abnormality cause database 14. An analysis is performed by the abnormality cause analysis unit 23 based on the cause information.

  The cause display control unit 25 displays on the display output unit 9 a screen showing an analysis result of the data obtained by the abnormality cause analysis unit 23, in which the user can diagnose the cause of the abnormality of the rotating machine 2. Is performed.

  As shown in FIG. 3, the data pre-processing unit 16 includes a target data adjustment unit 26, a past data adjustment unit 27, a loss information interpolation unit 28, a data label assignment unit 29, a data scaling unit 30, a spectrum analysis unit 31, and an amplitude value. A calculation unit 32 and a data format conversion unit 33 are provided.

  The target data adjustment unit 26 adjusts the target data so that the number of dimensions of the target data to be diagnosed included in the monitoring data or the process data becomes a specific number of dimensions.

  The past data adjustment unit 27 adjusts the past data so that the number of dimensions of the past data for which the presence or absence of an abnormality included in the monitoring data or the process data is known is a specific number of dimensions.

  Note that “adjusting the data to have a specific number of dimensions” in the present embodiment may include the user thinning out the data or compressing the dimension to have an arbitrary dimension. Furthermore, extracting only data that easily contributes to an abnormality from the data to be analyzed based on physical knowledge of the vibration of the rotating machine 2 is included. For example, when there is 2000-dimensional frequency data, a pre-process of extracting (adjusting the dimensions) only 48-dimensional data that is likely to contribute to an abnormality may be performed.

  The missing information interpolating unit 28 interpolates the missing information when the target data or the past data has a specific number of dimensions and, when the information of some of the information of the respective dimensions is missing. By doing so, it is possible to prepare target data or past data that can be analyzed by an algorithm that handles data of a specific number of dimensions, by aligning the information of the respective dimensions of the target data or past data.

  The data label assigning unit 29 assigns a predetermined label to target data or past data whose specific dimension number has been adjusted. For example, with respect to data analyzed in the past or data in which a significant feature was observed during data collection, there are known data that are normal data or abnormal data. A label L is attached to such past data as information that can identify the presence or absence of an abnormality (see FIGS. 9 and 10). That is, the label L that can identify the presence or absence of the abnormality is associated with the past data. It should be noted that the label L is blank for the target data.

  The data scaling unit 30 performs a process of normalizing or standardizing the monitoring data. By this processing, the scale of data having different scales for each dimension can be made uniform. For example, when the normalization processing is performed, the range of the value of the feature amount falls within a certain range. It is often within the range [0, 1] or [-1, 1]. In the standardization, conversion is performed so that the average of the feature amounts is 0 and the variance is 1.

  The spectrum analysis unit 31 extracts an arbitrary frequency component from the vibration data. Based on this analysis, vibration spectrum data that becomes target data or past data can be generated. The spectrum analysis unit 31 extracts a specific frequency component of the monitoring data, performs an envelope process, and extracts a specific periodic frequency component using, for example, a fast Fourier transform (FFT). Or

  The amplitude value calculator 32 analyzes the amplitude value included in the monitoring data. It should be noted that amplitude value data that becomes target data or past data can be created based on this analysis.

  That is, in the present embodiment, amplitude value data or vibration spectrum data, which is information relating to vibration of the rotating machine 2, can be acquired based on the monitoring data.

  The data format conversion unit 33 converts the data format of the target data or the past data so that the data format conforms to the algorithm used in the data analysis unit 18. The conversion of the data format is to adjust the shape of the data in order to perform the abnormality analysis.

  As shown in FIG. 4, the processing setting unit 17 includes a preprocessing condition setting unit 34 and an algorithm setting unit 35.

  The pre-processing condition setting unit 34 sets pre-processing conditions. For example, settings for the above-described interpolation or normalization processing are changed.

  The algorithm setting unit 35 sets an algorithm handled by the data analysis unit 18. For example, it is set whether to use topological data analysis, use cluster analysis, or use both analyses. It also sets hyperparameters for those algorithms. For example, when the topological analysis is used, a filter function, an overlap ratio, and the like are changed by setting the hyperparameter.

  As shown in FIG. 5, the parameter tuning unit 19 includes a specific dimension number setting unit 36 and a parameter optimizing unit 38.

  The specific dimension number setting unit 36 sets a specific dimension number used when analyzing target data or past data. By doing so, it is possible to set the specific number of dimensions so that the presence or absence of the abnormality can be identified. The setting of the specific number of dimensions may be performed by the preprocessing condition setting unit 34 or the algorithm setting unit 35.

  In addition, the setting of the specific dimension number is repeated so that the display of the target data or the past data obtained after the analysis has a form in which the presence or absence of the abnormality can be identified. In this way, the setting of the specific dimension number can be repeatedly performed until the presence / absence of the abnormality can be identified. This setting may be performed manually or automatically.

  The parameter optimizing unit 38 tunes the setting of parameters used for analysis in the data analyzing unit 18 based on the correspondence between the analysis parameters arranged in the result organizing unit 20 and the analysis results. In this tuning, parameter change is repeated and parameter optimization is performed so that the display of target data or past data obtained after analysis is in a form in which the presence or absence of an abnormality can be identified.

  As shown in FIG. 6, the result display control unit 21 includes a cluster display control unit 39 and a time-series display control unit 40.

  The cluster display control unit 39 controls the display output unit 9 to display a screen for cluster display (see FIG. 12) based on the analysis result of the data obtained by the data analysis unit 18. The cluster display is a screen displaying the target data as the node N. By doing so, the target data can be visualized as a cluster display, and it is easy to identify whether or not there is an abnormality in the rotating machine 2.

  The time-series display control unit 40 controls the display output unit 9 to display a screen of a time-series display (see FIG. 13) showing the analysis result of the data obtained by the data analysis unit 18. The time-series display is a screen in which the target data is displayed in the order of the monitored time series, and the nodes N including the target data are displayed in an identifiable manner. In this way, the target data can be displayed in chronological order, so that it is easy to identify whether or not there is an abnormality in the rotating machine 2.

  As illustrated in FIG. 7, the display adjustment unit 22 includes a node selection reception unit 41 and a target information display unit 42.

  The node selection accepting unit 41 performs a process of accepting selection of a node N or a cluster C on a cluster-displayed screen.

  The target information display unit 42 displays the breakdown information B received by the node selection reception unit 41 and included in the selected node N (see FIG. 12). The breakdown information B includes the target information of the target data included in the selected node N. This makes it possible to easily obtain information on the target data corresponding to the node N.

  As illustrated in FIG. 8, the abnormality cause analysis unit 23 includes a feature amount extraction unit 43 and an abnormality cause estimation unit 44.

  The feature amount extraction unit 43 extracts a feature amount of the target data analyzed by the data analysis unit 18.

  The abnormality cause estimating unit 44 estimates an abnormality cause based on the feature amount extracted by the feature amount extracting unit 43. That is, based on the feature amount extracted by the feature amount extraction unit 43, the data is arranged so that the target data can be displayed in a mode in which the cause of the abnormality can be diagnosed. In this way, when an abnormality is found in the rotating machine 2, the cause can be estimated.

  Next, target data or past data analyzed by the data analysis unit 18 will be described with reference to FIGS.

  FIG. 9 is a table including an example of acceleration amplitude data that has been preprocessed by the data preprocessing unit 16. In this table, the acceleration amplitude of a specific number of dimensions is registered in association with the monitored date (measurement date, data collection date and time). It should be noted that any number of specific dimensions of the acceleration amplitude can be selected. Here, the data of the latest date is the target data T to be diagnosed as to whether there is an abnormality.

  In addition, a plurality of columns S of the acceleration amplitude of a specific dimension number are provided. The number of these columns S is the number of dimensions of the data. In the table of FIG. 9, five-dimensional or more, for example, 50-dimensional acceleration amplitude is registered in association with a predetermined date (measurement date, data collection date and time). Further, a label L capable of identifying the presence or absence of an abnormality is registered in association with the past data P. For example, the label L of “A” is registered in the normal past data P, and the label L of “B” is registered in the abnormal past data P.

  FIG. 10 is a table including an example of vibration spectrum data that has been preprocessed by the data preprocessing unit 16. In this table, vibration values for each specific number of dimensions are registered in association with the monitored date. Note that an arbitrary number can be selected as the specific dimension number of the frequency. Here, the data of the latest date is the target data T to be diagnosed as to whether there is an abnormality.

  Note that a plurality of frequency columns F having a specific number of dimensions are provided. The number of these columns F is the number of dimensions of the data. In the table of FIG. 10, five-dimensional or more, for example, 50-dimensional acceleration spectra are registered in association with a predetermined date. Further, a label L capable of identifying the presence or absence of an abnormality is registered in association with the past data P. For example, the label L of “A” is registered in the normal past data P, and the label L of “B” is registered in the abnormal past data P.

  9 and 10, the number of specific dimensions of the target data and the past data is unified. Here, there are cases where data corresponding to each of the dimensions cannot be extracted in association with each date. For example, data of several dimensions out of 50-dimensional data may be lost. The missing data is interpolated by the missing information interpolation unit 28. The value of the data to be interpolated may be zero, the average value of the data included in the same row as the missing item, or the interpolation based on the data before and after this average value. Or, it may be extrapolated, the average value of the data included in the same column as the missing item, or the one interpolated or extrapolated based on the data before and after this average value, Other arbitrarily determined values may be used.

  As the past data of the present embodiment, a mode using data acquired from the rotating machine 2 to be diagnosed as abnormal is exemplified, but other modes may be used. For example, when there is a rotating machine 2 having the same configuration, manufacturer, specifications, and model, data acquired from the other rotating machine 2 may be used as past data of the rotating machine 2 to be diagnosed as abnormal. .

  The data pre-processing unit 16 performs data complement and format conversion so that the specific dimension number has the same dimension number for all dates. The processing of the data pre-processing unit 16 may be performed automatically or may be set arbitrarily by hand.

  A conventional statistical method using machine learning is based on a statistical analysis, and it is assumed that data to be handled follows some distribution such as a normal distribution. Therefore, it is necessary to construct an analysis method according to the characteristics of individual data. Further, when handling large-scale, multi-dimensional data, processing such as reduction of the dimension of the data is required. In the dimension reduction by such a statistical method, there is a problem that information in data is lost. The present embodiment can solve such a problem.

  Next, a topological data analysis, a so-called topological data analysis (hereinafter, referred to as “TDA”) used in the present embodiment will be described. TDA is a technique for analyzing data features of large-scale multi-dimensional data by focusing on topological information (data form) of the data instead of statistical features. In this TDA, there are two types of methods, a persistent homology and a mapper.

  Persistent homology is a technique that can capture geometric invariants, and can quantitatively evaluate the details of data shapes. The mapper is a technique for expressing high-dimensional data with a simple graph that can grasp geometric information.

  In the present embodiment, a mode using TDA Mapper will be exemplified. Note that other analysis techniques may be used.

  With TDA Mapper, data with a high number of dimensions (specific number of dimensions) can be visualized as a graph of three dimensions or less. In addition, important parts for grasping data can be collectively expressed as one node N. For example, nodes N having continuous data are connected by an edge E, and a set of data is converted into a graph (see FIG. 12).

  Note that “visualization” in the present embodiment indicates a mode in which a human can recognize a shape. For example, a three-dimensional graph displayed on a display is a two-dimensional image, but a human can recognize the three-dimensional shape.

  The TDA Mapper has a feature that the analysis result hardly depends on the coordinate system of the target data, and has a feature that data in different coordinate systems can be compared without depending on the selected coordinate system. Further, the TDA Mapper has a feature that it is invariable with respect to a small displacement. For example, it has a feature that it is hardly affected by noise.

  Further, the TDA Mapper has a feature that multidimensional data can be compressed into a low dimension and the shape of the data can be expressed. For example, when a set of thousands of data is input, it can be represented as a network having 13 vertices and 12 edges.

  In TDA Mapper, multidimensional data is input, and a distance metric of the input data is set. Next, a filter function for mapping the data for which the distance has been set to a low dimension is set. One or more filter functions can be set. For example, a function that projects data using the U-grid distance as a metric on the Y axis can be used as a filter. Next, the projected data is divided into bins at determined intervals. The interval (interval) and the overlap rate (overlap) between the intervals are important in the relationship between the node and the edge described later. Next, for each interval, the data contained therein is clustered. An example of the clustering algorithm is the k-means method. In this case, the number k of clusters is also a parameter. The cluster at each interval becomes a node at the time of visualization. If data in the cluster is duplicated, draw a line between the nodes. This line is called an edge, and means that there is a connection between the nodes.

In the course of the above processing, the following hyper-parameters of Mapper that can be set (need to be set) by the user are as follows.
Metric: Definition of distance.
Filter functions: density, eccentricity, graph Laplacian, projection on each axis, SVD (singular value decomposition), Isolation forest.
Interval: If the number of intervals is increased, the number of clusters when performing the clustering process is increased. However, many empty clusters may be created (the number of data per cluster is small).
Overlap: When the number of overlaps is increased, connections (edges) between nodes increase.
Clustering method: Hierarchical clustering, k-means, and other methods.

  The parameter tuning unit 19 and the result sorting unit 20 sort out the correlation between these control parameters and the analysis results (see FIG. 11). Then, the parameter tuning unit 19 sweeps the analysis parameters until the data point of interest is plotted on a node of the graph.

  This parameter tuning operation ensures that the data point of interest is represented on the node. Next, it is checked whether the node including the data point of interest includes a data point of another measurement date, and has a relationship with another node, that is, whether the node is connected by an edge. At this time, if the data on another measurement day has a normal label L, it is determined that the data of interest is also likely to be normal. On the other hand, when there is no connection with the normal data, it is determined that the focused data is not normal (the possibility of abnormality is high).

  According to the above procedure, it is possible to automatically determine the presence or absence of the abnormality without setting the analysis parameters by the user.

  As shown in FIG. 12, a cluster display (graph) generated by the data analysis unit 18 and resulting from the TDA Mapper is displayed on the display. The cluster display is controlled by the cluster display control unit 39 of the result display control unit 21.

  The cluster display includes a graph display area G in which the node N and the cluster C are displayed, a breakdown information display area D in which the breakdown information B included in the node N is displayed, and a color label contour indicating the color classification of the node N. R is provided.

  In the example of FIG. 12, a node N is displayed as a circle in the graph display area G. For example, the larger the number of target data or past data included in one node N, the larger the displayed circle. Further, a plurality of nodes N are connected by an edge E. The node group connected by the edge E is a cluster C. Then, the cluster C1 of the node N containing normal data and the clusters C2, C3, C4 of the node N containing abnormal data are displayed separately.

  For example, one cluster C1 including normal data is displayed, and three clusters C2 including abnormal data are displayed. Note that the cluster C of the present embodiment indicates, in principle, a plurality of nodes N connected by an edge E. However, even if a single node N is separated from another cluster C, It is assumed that it is handled as one cluster C.

  In the cluster display, a graph generated based on both the target data and the past data is displayed. Note that there is a case where both the target data and the past data are included in one predetermined node N.

  The user can cause the breakdown information B included in the selected node N to be displayed in the breakdown information display area D by selecting a predetermined node N displayed in the graph display area G with the mouse cursor M. .

  For example, when a plurality of target data are included in the selected node N, a date (measurement date, data collection date and time) corresponding to each target data is displayed. Note that attribute information such as the number of target data items and data labels included in the selected node N may be displayed in the breakdown information display area D. Also, the statistics included in the node N may be displayed in the breakdown information display area D.

  In addition, what is selected by the mouse cursor M may be the cluster C. In that case, the breakdown information B included in the selected cluster C can be displayed in the breakdown information display area D.

  In the color label contour R, the setting content of the display color of the node N is displayed. For example, a node N including past data in which a label L of “A” indicating normal is registered is displayed in yellow, and a node N including past data in which a label L of “B” indicating abnormal is registered is displayed in blue. I do. In this way, based on the analysis of the past data, the past data can be displayed in such a manner that the associated label L can be identified.

  Note that the user may perform color coding for each node N for which confirmation is desired, or may perform color coding at a frequency corresponding to each node N. Further, for example, a label indicating that the rotating machine 2 has been disassembled and inspected and a label that has not been disassembled and inspected may be attached to the data, and color-coded. Further, the color of the node N may be classified for each parameter affected at the time of analysis.

  Further, in the cluster display, when it is determined that the data of interest has a high possibility of abnormality, the node N including the target data having the abnormality is highlighted on the GUI by the unique node highlighting function (the possibility of abnormality is displayed). Display).

  As shown in FIG. 13, a time-series display generated by the data analysis unit 18 and being a result of the TDA Mapper is displayed on the display. The control of the time-series display is performed by the time-series display control unit 40 of the result display control unit 21.

  In this time-series display, data that has been graphed in the cluster display is displayed in chronological order. For example, the time series display includes a column K1 of monitored dates (measurement date, data collection date and time), a column K2 of a node N corresponding to each date, a column K3 of a cluster C corresponding to each node N, A column K4 indicating the possibility of abnormality of each node N or cluster C is provided.

  By referring to the time-series display, the user can display whether or not each node N or cluster C has a possibility of abnormality. For example, “normal” is displayed corresponding to the node N including the past data in which the label “A” indicating normal is registered, and the past data in which the label “B” indicating abnormal is registered is displayed. "Abnormal" is displayed corresponding to the included node N.

  That is, based on the analysis of the target data, the result display control unit 21 displays the target data in a mode (for example, time-series display) in which the presence or absence of an abnormality can be identified. Further, based on the analysis of the past data, the past data is displayed in such a manner that the associated label L can be identified. By doing so, it is possible to identify the presence or absence of an abnormality in the target data by referring to the display mode of the past data.

  Note that the cluster display (see FIG. 12) and the time-series display (see FIG. 13) can be arbitrarily switched by a user operation.

  The abnormality cause analysis unit 23 extracts a characteristic amount of data determined to have a possibility of abnormality by using a characteristic amount extracting function of abnormal data.

  Here, a case will be described in which data to be analyzed is frequency component data. For example, a difference between each frequency component is calculated for the data determined to be abnormal and the data with a normal label, and a frequency having a large difference between the frequency components is extracted. The difference between the extracted frequency and the frequency component is used as the feature amount of the abnormal data. For the definition of the frequency component having a large difference, an extraction method that exceeds a preset threshold value or an arbitrary extraction method such as extraction of a difference of several% at the top can be selected. Then, the extracted feature amounts (the difference between the frequency and the frequency component) are input to the abnormality cause analysis unit 23.

  In the present embodiment, by inputting the characteristic amount to the abnormality cause analysis unit 23, the degree of influence of the characteristic amount on the abnormality cause can be known, and the abnormality cause can be estimated. Further, by automating parameter tuning necessary for abnormality detection, it is possible to estimate the cause of the abnormality without manually performing the abnormality detection. Further, the cause of the abnormality can be estimated from the result of the abnormality detection.

  Next, a method of diagnosing abnormality of the rotating machine 2 will be described with reference to the flowchart of FIG. Note that FIG. 1 to FIG. 13 are appropriately referred to.

  As shown in FIG. 14, first, in step S11, the user of the abnormality diagnosis system 1 attaches the inspection sensor 3 to the rotating machine 2.

  In the next step S12, the user monitors the operating state of the rotating machine 2 using the inspection sensor 3 attached to the rotating machine 2 to obtain monitoring data.

  In the next step S13, the user inputs the monitoring data acquired using the inspection sensor 3 to the monitoring database 11 of the abnormality diagnosis system 1 using the user terminal 5.

  In the next step S <b> 14, the user inputs the process data of the rotating machine 2 to be diagnosed to the process database 12 of the abnormality diagnosis system 1 using the user terminal 5.

  In the next step S15, the user inputs the specification data or process data of the rotating machine 2 to be diagnosed to the specification database 13 of the abnormality diagnosis system 1 using the user terminal 5.

In the next step S16, the user inputs the abnormality cause data of the rotating machine 2 to be diagnosed to the abnormality cause database 14 of the abnormality diagnosis system 1 using the user terminal 5.

  In the next step S17, the user sets the abnormality cause data, that is, sets the relationship between the characteristic amount of the target data that can be extracted based on the analysis of the analysis processing unit 10 and the abnormality cause of the rotating machine 2. I do. That is, in the abnormality cause database 14, the correspondence between the feature amount of the target data that can be extracted and the abnormality cause is stored in advance.

  In the next step S <b> 18, the user uses the abnormality diagnosis system 1 to analyze target data or past data included in the monitoring data.

  In the next step S19, the user performs a predetermined operation of displaying the analysis result on the display of the abnormality diagnosis system 1. The user performs, for example, an operation of displaying a cluster display screen, an operation of displaying a time-series display screen, an operation of selecting a node N or a cluster C, and an operation of displaying a screen showing a result of analyzing the cause of abnormality.

  In the next step S20, the user identifies the presence or absence of an abnormality in the rotating machine 2 based on the information displayed on the display of the abnormality diagnosis system 1.

  In the next step S21, the user specifies the cause of the abnormality of the rotating machine 2 based on the information displayed on the display of the abnormality diagnosis system 1. Then, the abnormality diagnosis method ends.

  Next, the abnormality diagnosis processing executed by the analysis processing unit 10 will be described with reference to the flowchart of FIG. Note that FIG. 1 to FIG. 13 are appropriately referred to.

  As shown in FIG. 15, first, in step S31, the data acquisition unit 15 of the analysis processing unit 10 acquires monitoring data stored in the monitoring database 11. Here, the process data stored in the process database 12 and the specification data stored in the specification database 13 may be obtained.

  In the next step S32, the preprocessing condition setting unit 34 of the processing setting unit 17 of the analysis processing unit 10 sets preprocessing conditions to be executed by the data preprocessing unit 16. Here, the specific dimension number setting unit 36 of the parameter tuning unit 19 sets the specific dimension number of the target data or the past data based on the setting of the preprocessing condition setting unit 34. The algorithm setting unit 35 of the processing setting unit 17 sets an algorithm used in the data analysis unit 18. For example, TDA Mapper is used as the algorithm.

  In the next step S33, the data pre-processing unit 16 of the analysis processing unit 10 executes data pre-processing.

  In the next step S34, the data analysis unit 18 of the analysis processing unit 10 executes a data analysis process. Here, the target data adjusted to the specific number of dimensions is analyzed using an algorithm that handles data of the specific number of dimensions. Further, the past data adjusted to the specific number of dimensions is analyzed using an algorithm.

  When TDA Mapper is used as an algorithm, data input processing, mapping processing, cover processing, clustering / node conversion processing, and graph processing are executed in the data analysis processing of the data analysis unit 18.

  In the next step S35, the result organizing unit 20 of the analysis processing unit 10 executes an analysis result organizing process. Here, whether or not the analysis result of the data obtained by the data analysis unit 18 is optimal is arranged.

  In the next step S36, the result organizing unit 20 determines whether or not the analysis result is optimal. The determination as to whether or not the data is optimal is set based on, for example, whether or not the display of the target data is in a form in which the presence or absence of an abnormality can be identified. Here, when the analysis result is optimal (step S36 is YES), the process proceeds to step S38 described later. On the other hand, if the analysis result is not optimal (NO in step S36), the process proceeds to step S37.

  In step S37, the parameter tuning unit 19 of the analysis processing unit 10 performs a parameter tuning process. Here, tuning of various settings used for analysis by the data analysis unit 18 or setting of an algorithm for analysis is performed. Then, the process returns to step S33. That is, in the present embodiment, the setting of the specific dimension number is repeated so that the display of the target data is in a form in which the presence or absence of the abnormality can be identified.

  In step S38, the abnormality cause analysis unit 23 of the analysis processing unit 10 executes an abnormality cause analysis process. Here, the feature value extraction unit 43 extracts the feature value of the target data analyzed by the data analysis unit 18. Further, the abnormality cause estimating unit 44 estimates the abnormality cause based on the characteristic amount extracted by the characteristic amount extracting unit 43, so that the target data can be displayed in such a manner that the abnormal cause can be diagnosed. Organize your data.

  In the next step S39, the display adjustment unit 22 of the analysis processing unit 10 executes a display adjustment process. Here, the display adjustment unit 22 controls a screen displayed on the display based on a user operation. Then, the abnormality diagnosis processing ends.

  Next, data pre-processing performed by the data pre-processing unit 16 will be described with reference to the flowchart in FIG. Note that FIG. 1 to FIG. 13 are appropriately referred to.

  As shown in FIG. 16, first, in step S41, the target data adjustment unit 26 of the data pre-processing unit 16 sets the target data so that the number of dimensions of the target data to be diagnosed included in the monitoring data becomes the specific number of dimensions. Perform data adjustment processing.

  In the next step S42, the past data adjusting unit 27 of the data pre-processing unit 16 performs a process of adjusting the past data so that the number of dimensions of the past data for which the presence or absence of an abnormality included in the monitoring data is known is a specific number of dimensions. Do.

  In the next step S43, the missing information interpolating unit 28 of the data pre-processing unit 16 determines whether or not some of the information of the respective dimensions is missing when the target data has the specified number of dimensions. , A process of interpolating the missing information.

  In the next step S44, the data label assigning unit 29 of the data pre-processing unit 16 performs a process of associating the past data adjusted to the specific number of dimensions with a label L capable of identifying the presence or absence of an abnormality.

  In the next step S45, the data preprocessing unit 16 performs other processing. For example, the data scaling unit 30 normalizes the data. The spectrum analyzer 31 calculates a spectrum component. The amplitude value calculator 32 calculates an amplitude value. The data format conversion unit 33 converts the data format. Then, the data pre-processing ends.

  Next, the display adjustment processing executed by the display adjustment unit 22 will be described with reference to the flowchart in FIG. Note that FIG. 1 to FIG. 13 are appropriately referred to.

  As shown in FIG. 17, first, in step S51, the display adjustment unit 22 determines whether or not a cluster display (graph) screen is being displayed on the display. If the graph is not being displayed (NO in step S51), the process proceeds to step S54 described below. On the other hand, when the graph is being displayed (YES in step S51), the process proceeds to step S52.

  In step S52, the node selection receiving unit 41 of the display adjusting unit 22 determines whether the selection of the node N or the cluster C on the cluster-displayed screen has been received. Here, when the selection of the node N or the cluster C is not received (step S52 is NO), the process proceeds to step S54. On the other hand, when the selection of the node N or the cluster C is received (YES in step S52), the process proceeds to step S53.

  In step S53, the target information display unit 42 of the display adjustment unit 22 displays the target data included in the selected node N or cluster C or the breakdown information B (target information) of past data in the breakdown information display area D. I do. Then, the process proceeds to step S54.

  In step S54, the display adjustment unit 22 determines whether or not an operation for displaying a time-series display screen on the display has been received. Here, when the display operation of the time series display is not received (NO in step S54), the process proceeds to step S56 described later. On the other hand, when the display operation of the time-series display is received (YES in step S54), the process proceeds to step S55.

  In step S55, the time-series display control unit 40 of the result display control unit 21 displays a time-series display screen on the display based on the processing of the display adjustment unit 22. The time-series display control unit 40 displays the target data in the order of the monitored time series, and displays the nodes including the target data in an identifiable manner. In addition, the time-series display control unit 40 displays the past data in a form in which the associated label L can be identified based on the analysis of the past data. For example, a display indicating the possibility of abnormality is performed (see FIG. 13). Then, the process proceeds to step S56.

  In step S56, the display adjustment unit 22 determines whether or not an operation to display a cluster display screen on the display has been received. Here, when the display operation of the cluster display is not received (NO in step S56), the process proceeds to step S58 described later. On the other hand, if a cluster display operation has been received (YES in step S56), the process advances to step S57.

  In step S57, based on the processing of the display adjustment unit 22, the cluster display control unit 39 of the result display control unit 21 displays a cluster display screen on the display. The cluster display control unit 39 displays the target data in a manner that allows the presence or absence of an abnormality to be identified based on the analysis of the target data. The cluster display control unit 39 displays the past data in such a manner that the associated label L can be identified based on the analysis of the past data. For example, the nodes N are displayed in different colors depending on the possibility of abnormality (see FIG. 12). Then, the process proceeds to step S58.

  In step S58, the display adjustment unit 22 determines whether or not an operation to display a screen indicating the result of the analysis of the cause of the abnormality on the display has been received. Here, when the display operation of the abnormality cause analysis result is not received (NO in step S58), the process proceeds to step S60 described later. On the other hand, when the display operation of the abnormality cause analysis result is received (YES in step S58), the process proceeds to step S59.

  In step S59, based on the processing of the display adjustment unit 22, the cause display control unit 25 causes the display to display a screen showing the result of the analysis of the cause of the abnormality. Then, the process proceeds to step S60.

  In step S60, the display adjustment unit 22 determines whether or not an operation for terminating the display of the screen showing the analysis result being displayed on the display has been received. Here, when the operation to end the display is not received (NO in step S60), the process proceeds to step S51 described above. On the other hand, when an operation to end the display is received (YES in step S60), the process proceeds to step S61.

  In step S61, based on the processing of the display adjustment unit 22, the result display control unit 21 or the cause display control unit 25 ends the display of the screen showing the analysis result being displayed on the display. Then, the display adjustment processing ends.

  Note that, in the flowchart of the present embodiment, an example in which each step is executed in series is illustrated, but the order of each step is not necessarily fixed, and even if the order of some steps is switched. good. Also, some steps may be executed in parallel with other steps.

  The system of the present embodiment includes a control device in which processors such as a dedicated chip, an FPGA (Field Programmable Gate Array), a GPU (Graphics Processing Unit), or a CPU (Central Processing Unit) are highly integrated, and a ROM (Read Only). Memory) or a storage device such as a random access memory (RAM), an external storage device such as a hard disk drive (HDD) or a solid state drive (SSD), a display device such as a display, and an input device such as a mouse or a keyboard. , A communication interface. This system can be realized by a hardware configuration using an ordinary computer.

  The program executed by the system according to the present embodiment is provided by being incorporated in a ROM or the like in advance. Alternatively, this program is a non-transitory computer-readable storage medium such as a CD-ROM, a CD-R, a memory card, a DVD, and a flexible disk (FD) in a file in an installable or executable format. And may be provided.

  Further, the program executed by this system may be stored on a computer connected to a network such as the Internet, and downloaded and provided via the network. In addition, the system can be configured by combining separate modules that independently perform the functions of the components by mutually connecting them via a network or a dedicated line.

  Note that, in the present embodiment, cluster display using topological data analysis is exemplified as a mode in which the presence or absence of an abnormality can be identified, but other modes may be used. For example, a mode in which the presence or absence of an abnormality can be identified may be a hierarchical cluster display based on a hierarchical cluster analysis, or a non-hierarchical cluster display based on a non-hierarchical cluster analysis. good. In the non-hierarchical cluster analysis, for example, a technique such as the K-means method may be used.

  Hereinafter, a modified example using hierarchical clustering will be described with reference to FIGS. 18, 19, and 20.

  For example, N input data items that are not subjected to hierarchical clustering are merged in descending order of similarity to form a gradually larger cluster. Finally, the process of integrating data or clusters is represented by a tree called a dendrogram by a method of integrating N pieces of data into one cluster.

  Here, as parameters of the hierarchical clustering, there are a scale and a method for measuring the similarity of data, and a scale for measuring the similarity includes a U-grid distance and a Minkowski distance. In addition, as a method of measuring the distance between clusters, there are a Ward method, a group average method, and the like. These parameters are optimized by the parameter tuning unit 19.

  An index used for parameter tuning and optimization includes, for example, a correlation coefficient between a cluster distance and a Corfuen matrix (Corfuen correlation coefficient).

  FIG. 18 shows a display result by a hierarchical cluster. For example, the analysis target data processed by the data preprocessing unit 16 is input to the data analysis unit 18, analyzed using a hierarchical clustering method, and the analysis result is displayed on the display output unit 9. Among the results, it is the result of cluster display. Here, 57 pieces of amplitude value data of k dimensions are input. Among the 57 amplitude value data, there are data in the normal range and data showing an abnormality.

  For example, in FIG. 18, when each individual data is a node N, a plurality of nodes N or a node group including at least one node N can be defined as a cluster C. Also in the case of using this hierarchical clustering, similarly to the above-described TDA, it is possible to separately display the clusters C5 and C6 of the node N including normal data and the cluster C7 of the node N including abnormal data. It is.

  As shown in FIG. 19, in the hierarchical clustering, in the process of fusing various data and clusters, normal data has an average branching mode. On the other hand, as shown in FIG. 20, the abnormal data is branched and displayed in the upper layer. In the above description, the case where the number of normal data is larger than the number of abnormal data is exemplified. However, the number of normal data may be smaller than the number of abnormal data.

  According to the embodiment described above, diagnosis is performed without depending on the skill of an engineer by including a step of analyzing target data adjusted to a specific number of dimensions using an algorithm that handles data of a specific number of dimensions. be able to.

  Although several embodiments of the present invention have been described, these embodiments are provided by way of example and are not intended to limit the scope of the invention. These embodiments can be implemented in other various forms, and various omissions, replacements, changes, and combinations can be made without departing from the gist of the invention. These embodiments and their modifications are included in the scope and gist of the invention, and are also included in the invention described in the claims and equivalents thereof.

DESCRIPTION OF SYMBOLS 1 ... Abnormality diagnosis system, 2 ... Rotating machine, 3 ... Inspection sensor, 4 ... Internet, 5 ... User terminal, 6 ... Main control part, 7 ... Information input part, 8 ... Communication part, 9 ... Display output part, 10 ... Analysis processing unit, 11 ... Monitoring database, 12 ... Process database, 13 ... Spec database, 14 ... Abnormal cause database, 15 ... Data acquisition unit, 16 ... Data preprocessing unit, 17 ... Process setting unit, 18 ... Data analysis unit , 19: Parameter tuning unit, 20: Result arrangement unit, 21: Result display control unit, 22: Display adjustment unit, 23: Abnormal cause analysis unit, 24: Cause information acquisition unit, 25: Cause display control unit, 26: Target Data adjustment unit, 27: Past data adjustment unit, 28: Loss information interpolation unit, 29: Data label assignment unit, 30: Data scaling unit, 31: Spectrum Analysis unit, 32: amplitude value calculation unit, 33: data format conversion unit, 34: preprocessing condition setting unit, 35: algorithm setting unit, 36: specific dimension number setting unit, 38: parameter optimization unit, 39: cluster display Control unit, 40: time-series display control unit, 41: node selection reception unit, 42: target information display unit, 43: feature amount extraction unit, 44: abnormality cause estimation unit, B: breakdown information, C: cluster, D ... Breakdown information display area, E: edge, F: frequency column, G: graph display area, K1 to K4: time series display column, L: label, M: mouse cursor, N: node, P: past data, R ... color label contour, S: column of acceleration amplitude, T: target data.

Claims (14)

Individual data extracted from at least one of the time waveform and the frequency spectrum of a signal that is data indicating the state of the device and included in one target data to be diagnosed among the data indicating the state of the device a step of adjustment to a specific number of dimensions the number of dimensions is the number of,
The target data adjusted to the specific number of dimensions , performing a topological data analysis using a mapper that is an algorithm that handles the data of the specific number of dimensions,
A step of displaying the analysis result using the mapper in a cluster display or a time-series display in a mode in which the presence or absence of an abnormality can be identified,
including,
Device abnormality diagnosis method. The specific dimension number is four or more dimensions,
An aspect in which the presence or absence of the abnormality can be identified is a three-dimensional or less graph as the cluster display .
The method for diagnosing abnormality of a device according to claim 1. Adjusting the past data so that the number of dimensions of the known past data, the presence or absence of the abnormality included in the data indicating the state of the device, is the specific dimension number,
A step of associating the past data adjusted to the specific number of dimensions with a label capable of identifying the presence or absence of the abnormality,
Analyzing the past data adjusted to the specific number of dimensions using the mapper ,
And displaying the analysis results the labels applied corresponds is using the mapper in an identifiable manner,
The method for diagnosing a device abnormality according to claim 1 or 2. Including setting the number of specific dimensions,
The method for diagnosing an abnormality of an apparatus according to claim 1. Including a step of repeating the setting of the specific number of dimensions so that the display of the target data is in a form in which the presence or absence of the abnormality can be identified.
An apparatus abnormality diagnosis method according to claim 4. When the target data is the specific number of dimensions, when information of some of the number of dimensions of the information of each dimension is missing, including the step of interpolating the missing information,
The method for diagnosing an abnormality of a device according to any one of claims 1 to 5. Extracting a feature amount of the target data;
Displaying the target data in a mode in which the cause of the abnormality can be diagnosed based on the extracted feature amount;
including,
The method for diagnosing abnormality of a device according to any one of claims 1 to 6. Including a step of storing in advance a correspondence relationship between the feature amount and the cause of the abnormality,
A method for diagnosing abnormality of a device according to claim 7. An aspect in which the presence or absence of the abnormality can be identified is a cluster display in which the target data is a node.
The method for diagnosing an abnormality of a device according to any one of claims 1 to 8 . Receiving a selection of the node displayed in the cluster,
Displaying information of the target data included in the selected node;
including,
The method for diagnosing abnormality of a device according to claim 9 . Displaying the target data in a monitored chronological order, and including displaying the node including the target data in an identifiable manner,
The method for diagnosing a device abnormality according to claim 9 or claim 10 . Data indicating the state of the device is data indicating vibration during operation of the device,
An abnormality diagnosis method for a device according to any one of claims 1 to 11 . Individual data extracted from at least one of the time waveform and the frequency spectrum of a signal that is data indicating the state of the device and included in one target data to be diagnosed among the data indicating the state of the device a data preprocessing unit for adjustment to a particular number of dimensions the number of dimensions is the number of,
The target data adjusted to the specific dimension number, a data analysis unit that performs topological data analysis using a mapper that is an algorithm that handles the data of the specific dimension number,
A result display control unit that displays the analysis result using the mapper in a cluster display or a time-series display that is a mode in which the presence or absence of an abnormality can be identified,
Comprising,
Equipment diagnosis system. A device abnormality diagnosis method using the device abnormality diagnosis system according to claim 13 ,
Using an inspection sensor attached to the device, monitoring the state of the device to obtain monitoring data,
Inputting the monitoring data to the abnormality diagnosis system;
including,
Device abnormality diagnosis method.

Images