Diagnostic device, diagnostic system, and diagnostic method

Abstract

A control unit (3) of a diagnostic device (10) selects one or more first features (T1) from time series data obtained from operating data about components (42, 43, 44) constituting a rotating machinery system (40) and one or more second features (T2) from frequency data obtained through frequency analysis of the operating data, and diagnoses the state of the rotating machinery system (40) on the basis of the selected first feature(s) (T1) and the selected second feature(s) (F1).

Description

Diagnostic device, diagnostic system, and diagnostic method

[0001] The present disclosure relates to a diagnostic device, a diagnostic system, and a diagnostic method.

[0002] Conventionally, there have been abnormality diagnosis devices and methods that diagnose abnormalities in rotating mechanical systems consisting of rotating machines, power transmission mechanisms, load equipment, etc., using acquired sensor signal information. Such abnormality diagnosis devices and methods require processing of huge amounts of data, such as vibration, temperature, pressure, and sound, obtained from the sensors, which can reduce the efficiency and accuracy of diagnosis. Therefore, in order to achieve efficient and accurate diagnosis of rotating mechanical systems, a diagnosis device using AI (Artificial Intelligence) as described below has been disclosed.

[0003] That is, in conventional diagnostic devices, the current of the rotating machine to be diagnosed is measured, noise is removed using a sequential cyclic statistical information filter, and the noise-removed current signal in the frequency domain or the current signal in the time domain is used to extract features that reflect the state of the rotating machine system to be diagnosed from the current signal using deep learning, thereby identifying abnormalities in the rotating machine (see, for example, Patent Document 1).

[0004] Japanese Patent Application Laid-Open No. 2021-76564

[0005] In the abnormality diagnosis method for diagnosing abnormalities in a rotating machine disclosed in the above-mentioned conventional diagnostic device, noise is removed from a measured current signal, and feature quantities reflecting the state of the rotating machine are extracted from data obtained from a sensor, and abnormalities are diagnosed by deep learning. However, among the multiple feature quantities obtained from data acquired by the sensor, the feature quantities reflecting abnormalities differ depending on the type, size, etc. of the abnormality, which may result in reduced diagnostic accuracy.

[0006] The present disclosure discloses a technique for solving the above-mentioned problems, and aims to provide a diagnostic device, a diagnostic system, and a diagnostic method that are highly accurate and efficient.

[0007] A diagnostic device according to the present disclosure includes an acquisition unit that acquires operation data of components that constitute a rotating mechanical system, and a control unit that diagnoses the state of the components that constitute the rotating mechanical system based on feature quantities obtained from the acquired operation data, wherein the control unit selects at least one first feature quantity as the feature quantity from time-series data obtained from the operation data and at least one second feature quantity as the feature quantity from frequency data obtained by frequency analysis of the operation data, and diagnoses the state of the rotating mechanical system based on the selected first feature quantity and second feature quantity.A diagnostic system according to the present disclosure includes the diagnostic device configured as described above, and a notification unit that is connected to the diagnostic device via a network and that notifies an administrator of a diagnosis result by the diagnostic device. The diagnostic method disclosed herein is a diagnostic method using a diagnostic device configured as described above, which selects one or more first feature quantities as the feature quantities from the time-series data of the operating data and one or more second feature quantities as the feature quantities from frequency data obtained by frequency analysis of the operating data, and diagnoses the state of the rotating mechanical system based on the selected first feature quantities and second feature quantities.

[0008] According to the diagnostic device, diagnostic system, and diagnostic method of the present disclosure, it is possible to obtain a diagnostic device, diagnostic system, and diagnostic method that are highly accurate and efficient.

[0009] 11A is a block diagram showing a schematic configuration of a diagnostic system including a diagnostic device according to embodiment 1. FIG. 11B is a diagram showing a schematic configuration of a rotating mechanical system that is a diagnosis target of the diagnostic device according to embodiment 1. FIG. 11C is a flowchart showing a flow of state diagnosis processing performed by the diagnostic device according to embodiment 1. FIG. 11D is a conceptual diagram showing operation data stored by a feature calculation unit according to embodiment 1. FIG. 11E is a conceptual diagram showing feature quantities extracted by the feature calculation unit according to embodiment 1. FIG. 11F is a conceptual diagram showing feature quantity conversion processing according to embodiment 1. FIG. 11G is a diagram showing a detailed configuration of map data according to embodiment 1. FIG. 11H is a current frequency spectrum obtained by FFT analysis of a current of the rotating mechanical system according to embodiment 1. FIG. 11H is a conceptual diagram for explaining an AI function in the abnormality diagnosis unit according to embodiment 1. FIG. 11B is a diagram showing an example of the accuracy of abnormality diagnosis when a threshold value is not set. FIG. 11B is a diagram showing an example of the accuracy of abnormality diagnosis when a threshold value is set. FIG. 11F is a flowchart showing another example of the flow of state diagnosis processing performed by the diagnostic device according to embodiment 1. FIG. 11G is a diagram showing a hardware configuration of a control device of the diagnostic device according to embodiment 1.

[0010] Embodiment 1. Fig. 1 is a block diagram showing the schematic configuration of a diagnostic system 30 including a diagnostic device 10 according to embodiment 1. Fig. 2 is a diagram showing the schematic configuration of a rotating mechanical system 40 that is the target of diagnosis by the diagnostic device 10 according to embodiment 1. The diagnostic system 30 of this embodiment diagnoses the state of the rotating mechanical system 40, such as an abnormality, and notifies the diagnosis result to a manager who performs maintenance.

[0011] First, a description will be given of the rotating mechanical system 40 to be diagnosed. As shown in Fig. 2, the rotating mechanical system 40 includes a power supply 41, a rotating machine 42, a power transmission mechanism 43, and a load facility 44 as components.

[0012] The power supply 41 supplies power from a power grid. Note that the power supply 41 may convert the power supplied from the power grid to any frequency, any current, or any voltage using an inverter and supply the converted power. The rotating machine 42 refers to a machine that performs a rotating operation, including an electric motor, a single-phase / three-phase induction machine, a synchronous machine, etc., and has a function of converting electrical energy into mechanical energy.

[0013] The power transmission mechanism 43 is a belt, gear, coupling, or the like, and is a mechanism that connects the rotating machine 42 and the load equipment 44. The load equipment 44 is a pump, fan, compressor, or the like, and is equipment that is driven using the mechanical energy converted by the rotating machine 42 as a power source. It is also possible to directly connect the rotating machine 42 and the load equipment 44 without providing the power transmission mechanism 43. The components that make up the rotating mechanical system 40 are not limited to those listed above, and may be any equipment related to the operation of the rotating machine 42.

[0014] A plurality of sensors 1 are provided near each component of the rotating mechanical system 40 to detect current, voltage, vibration (vibration velocity, vibration acceleration), sound, strain, temperature, pressure, AE (Acoustic Emission), power, and the like, as operating data related to the operating state of each component. These sensors 1 can detect appropriate operating data for any type of rotating mechanical system 40. The detected operating data is input to a diagnostic device 10.

[0015] The operating data acquired by the sensor 1 is not limited to the operating data described above, as long as it indicates the state of each component of the rotating mechanical system 40. The operating data acquired by the sensor 1 is affected by the environmental conditions of the rotating mechanical system 40. For example, if the load equipment 44 is a pump, the pump may be submerged, in which case diagnosis based on vibration, sound, etc. is difficult. Therefore, in such cases, acquisition of operating data such as current and voltage takes priority over vibration, sound, etc.

[0016] Next, a description will be given of the configuration of the diagnostic system 30. As shown in Fig. 1, the diagnostic system 30 includes a diagnostic device 10 that diagnoses abnormalities in a rotating mechanical system 40, and a display unit 20 that serves as a notification unit that notifies a manager of the diagnosis results.

[0017] The diagnostic device 10 includes a data acquisition unit 2 as an acquisition unit, and a control unit 3. The data acquisition unit 2 acquires driving data detected by the sensor 1. The data acquisition unit 2 also includes a filter function for removing noise from the detected driving data, an AD conversion function, and a memory function for storing data.

[0018] The operating data stored by the memory function may be stored in another external computer by the communication function. The memory function may also be configured to include a function for deleting stored operating data and a communication function, and may acquire, for example, past operating data, simulation data, etc., using the communication function and store them in the memory function.

[0019] The feature calculation unit 4 calculates feature quantities for abnormality determination from each piece of operating data acquired by the data acquisition unit 2, and includes an extraction unit 4A, a time series data storage unit 4B, a frequency data storage unit 4C, a conversion unit 4D, a first database DB1, and a second database DB2.

[0020] The abnormality diagnosis unit 5 performs abnormality diagnosis using AI based on the feature amounts calculated by the feature calculation unit 4, and includes a model generation unit 5A, an AI diagnosis unit 5B, and a prediction unit 5C. Details of the components constituting the feature calculation unit 4 and the abnormality diagnosis unit 5 will be described later.

[0021] The display unit 20 is a device including a display device such as a display that can display the diagnostic results from the diagnostic device 10 and set information, and controls the images displayed on the display. The display unit 20 may be integrated with the diagnostic device 10, or may be provided with a communication function and be separate from the diagnostic device 10. In this case, the display may be controlled by a separate computer connected via the communication function, or by the display unit 20.

[0022] The display unit 20 may have a function to store information on the diagnostic results obtained by the diagnostic device 10 and the set information, and may also have a function to organize and search this information in chronological order. For example, the abnormality diagnostic results may be displayed as a graph showing a time-series trend.

[0023] Although the display unit 20 having a display has been exemplified as the notification unit in the above, the present invention is not limited to this. The notification unit may use a printer or the like that prints on paper media to notify the administrator of the diagnosis results and the set information on paper media, or may use a light or the like to notify the administrator of only normal and abnormal cases. Alternatively, the notification unit may be a speaker or the like with an audio output function, as long as it has the function of notifying the administrator of the abnormality diagnosis results of the diagnostic device 10.

[0024] Here, abnormalities in the rotating mechanical system 40 include abnormalities in the rotating machine 42, abnormalities in the power transmission mechanism 43, and abnormalities in the load equipment 44. Abnormalities in the rotating machine 42 include, for example, bearing abnormalities, rotor bar abnormalities, eccentricity abnormalities, imbalance abnormalities, layer short abnormalities, torque abnormalities, etc. Abnormalities in the power transmission mechanism 43 include, for example, loose or broken belts, cracks or wear in belts, wear of gear teeth, chipped or broken teeth, etc. Abnormalities in the load equipment 44, in the case of a pump, for example, include cavitation abnormalities, air entrapment abnormalities, seal wear or damage, foreign matter contamination, bearing abnormalities, etc. Note that abnormalities in the rotating mechanical system 40 include, in addition to the abnormalities listed above, abnormalities in fans, compressors, etc.

[0025] The following describes the state diagnosis process for abnormalities and the like of the rotating mechanical system 40 performed by the diagnosis device 10. Fig. 3 is a flow chart showing the flow of the state diagnosis process performed by the diagnosis device 10 of this embodiment. Fig. 4 is a conceptual diagram showing operation data stored in the time series data storage unit 4B and the frequency data storage unit 4C provided in the feature calculation unit 4 of this embodiment. Fig. 5 is a conceptual diagram showing feature amounts extracted by the extraction unit 4A provided in the feature calculation unit 4 of this embodiment. Fig. 6 is a conceptual diagram showing feature amount conversion processing in the conversion unit 4D provided in the feature calculation unit 4 of this embodiment.

[0026] First, when the condition diagnosis process is started, the diagnosis device 10 acquires the operation data of each component of the rotating mechanical system 40 detected by each sensor 1 using the data acquisition unit 2 (step S001).

[0027] Next, the feature calculation unit 4 extracts one or more time-series data from the acquired operating data using the extraction unit 4A (step S002), and stores the extracted time-series data in the time-series data storage unit 4B as shown in Fig. 4. Fig. 4 shows an example in which the current flowing through the rotating mechanical system 40, vibration acceleration, AE (Acoustic Emission) which is a sound wave generated when a material is deformed or broken, and sound are stored as time-series data.

[0028] The extraction unit 4A also performs frequency analysis such as FFT (Fast Fourier Transform) on the operation data stored in the time-series data storage unit 4B (step S003). The extraction unit 4A extracts one or more frequency data from the frequency data converted into the frequency scale in this way (step S004), and stores the extracted frequency data in the frequency data storage unit 4C as shown in Fig. 4. Fig. 4 shows an example in which frequency spectrum data obtained by performing FFT on the current flowing through the rotating machine 42, vibration acceleration, AE, and sound are stored as the frequency data.

[0029] 4 shows current, vibration acceleration, AE, and sound as examples of time-series data and frequency data extracted from the operation data, but voltage, strain, pressure, power, etc. may also be used. Furthermore, although an example using FFT (Fast Fourier Transform) for frequency analysis has been shown, STFT (Short Time Fourier Transform), Wavelet Transform, etc. may also be used.

[0030] Next, the extraction unit 4A extracts one or more pieces of time series data from the time series data stored in the time series data storage unit 4B as time series feature amounts T1, which are first feature amounts, as shown in Fig. 5, as feature amounts to be used in the abnormality determination process by the abnormality diagnosis unit 5, which will be described later (step S005). Furthermore, the extraction unit 4A extracts one or more pieces of frequency data from the frequency data storage unit 4C as frequency feature amounts F1, which are second feature amounts, as shown in Fig. 5, as feature amounts to be used in the abnormality determination process by the abnormality diagnosis unit 5, which will be described later (step S006).

[0031] Here, the extraction unit 4A extracts the time series feature value T1 on the time series scale and the frequency feature value F1 on the frequency scale based on a first database DB1 described below. The feature values ​​obtained from the components constituting the rotating mechanical system 40 reflect the respective anomalies to different degrees depending on the type, degree, etc. of the anomaly occurring in the rotating mechanical system 40. The first database DB1 stores the frequency feature value F1 correlated with each time series feature value T1 in association with the respective feature value. For example, a first feature value that is affected by a certain anomaly is stored in association with a second feature value that is affected in a similar manner to the first feature value.

[0032] The correspondence between the time-series feature values ​​T1 and the frequency feature values ​​F1 in the first database DB1 is not limited to one-to-one correspondence. For example, three frequency feature values ​​F1, i.e., a current frequency spectrum, a vibration frequency spectrum, and a sound wave frequency spectrum, may be recorded in association with one current time-series data as the time-series feature value T1. Such association of the frequency feature values ​​F1 with the time-series feature values ​​T1 may be based on verification through experiments, simulations, or the like, or on estimation using AI. Furthermore, the first database DB1 may record time-series feature values ​​T1 and frequency feature values ​​F1 that are uncorrelated with each other. For example, the first database DB1 may record time-series feature values ​​T1 and frequency feature values ​​F1 that are related to but different components of the rotating mechanical system 40. In this way, the first database DB1 records correlations, including whether or not there is a correlation between the time-series feature values ​​T1 and the frequency feature values ​​F1.

[0033] The extraction unit 4A may extract all the time series data and frequency data stored in the time series data storage unit 4B and the frequency data storage unit 4C as feature quantities.

[0034] Next, the feature calculation unit 4 uses the conversion unit 4D to generate map data M as two-dimensional array data in which the time-series feature value T1 and the frequency feature value F1 are arranged in a two-dimensional space as shown in FIG. 6 (step S007). In this way, the conversion unit 4D converts the time-series feature value T1 and the frequency feature value F1 into two-dimensional map data M by plotting them as a heat map. Note that although a heat map is used as an example of the method of conversion into two-dimensional array data, this is not intended to be limited to a heat map. For example, the method may be converted into two-dimensional array data such as a bubble chart or an area graph. When the map data M is used, if an abnormality is detected in machine learning using AI, which will be described later, the determination result of the abnormality is visualized using a heat map.

[0035] The following describes the detailed configuration of the map data M. Fig. 7 is a diagram showing the detailed configuration of the map data M according to this embodiment. Fig. 8 shows a current frequency spectrum obtained by FFT analysis of the current flowing through the rotating machine 42 as time-series data, and shows each frequency data used in the map data M.

[0036] The map data M is divided into a plurality of cells, each having a set number of rows and columns; in this example, it is configured with 6 rows and 6 columns. A time-series feature value T1 is input to each cell of the first and second rows and columns of the map data M. Here, the time-series feature value T1 is calculated based on statistics of each time-series data. For example, the statistics include at least one of the root mean square, maximum value, mean value, variance, standard deviation, crest factor, sum, clearance coefficient, skewness, kurtosis, total harmonic distortion (THD), variance of the effective value in any period, and variance of the maximum value in any period.

[0037] A frequency feature value F1 is input to each cell in each column from row 3 to row 6 of the map data M. The input frequency feature value F1 is a value related to the spectrum that reflects an abnormality in the rotating mechanical system 40, such as the sideband of the maximum spectrum peak, the sideband frequency, the floor level, the sum of the sidebands, etc., as will be described below.

[0038] For example, if the acquired operating data is current or voltage, the above-mentioned maximum spectrum will be the power supply frequency fs of the rotating machine 42, as shown in Figure 8. In other words, if the acquired data is current or voltage, sidebands appear on the low-frequency side and high-frequency side of the spectrum peak of the power supply frequency fs, and these sidebands are spectra caused by abnormalities. As shown in Figure 8, sidebands f1 (f1(C) and f1(D)), f2 (f2(B) and f2(E)), and f3 (f3(A) and f3(F)) appear on both sides of the power supply frequency fs of the rotating machine 42.

[0039] For example, when the bearing of the rotating machine 42 is abnormal, the sideband wave fbearing caused by the bearing abnormality is expressed by the following equation, where fs is the power supply frequency, fr is the actual rotational frequency of the rotating machine 42, and n is the natural number of the order of the harmonic.

[0040]

[0041] Furthermore, for example, when a rotor bar of the rotating machine 42 is abnormal, and the slip is s, the sideband f(rotor_bar) caused by the rotor bar abnormality is expressed by the following equation.

[0042]

[0043] Furthermore, for example, when the belt of the power transmission mechanism 43 loosens or breaks, the sideband fbelt caused by the loosening or breakage of the belt is expressed by the following equation, where fb is the rotation frequency of the belt.

[0044]

[0045] In this way, the extraction unit 4A derives the rotation frequency of the rotating machine 42, the rotation frequency of the belt, etc. Then, the extraction unit 4A uses, as the second feature amount, the frequencies or spectrum peak values ​​of sidebands f1, f2, and f3 generated around the power supply frequency, which are determined by the difference value or sum value between the power supply frequency fs that drives the rotating machine 42 and the rotation frequencies fr and fb. As a result, the feature amounts that reflect each of the above-mentioned abnormalities, such as bearing abnormality, rotor bar abnormality, and belt abnormality, are arranged in the map data M.

[0046] Here, the frequency feature value F1 arranged in each column of the fifth row of the map data M is the average value of the floor level in the current frequency spectrum shown in Fig. 8. Specifically, the extraction unit 4A derives the average of the floor level spectrum values ​​for each set frequency band. In this example, the extraction unit 4A derives the average value of the floor level for each of the set frequency bands: 0 Hz to 30 Hz, 10 Hz to 40 Hz, 20 Hz to 50 Hz, 30 Hz to 60 Hz, 40 Hz to 70 Hz, and 50 Hz to 80 Hz.

[0047] In deriving the average value, the extraction unit 4A sorts the spectral values ​​of the frequency components in each set frequency band in ascending order according to their magnitude. Then, to exclude spectral values ​​that deviate significantly, the extraction unit 4A sets a threshold range for the spectral values ​​for each set frequency band. Then, the extraction unit 4A adjusts the spectral values ​​of each frequency component in each frequency band to a value within the threshold range, and then derives the average value of the spectral values.

[0048] Although the above example illustrates the use of the frequencies or spectral peak values ​​of sidebands occurring on both sides of the power supply frequency fs of the rotating machine 42, the present invention is not limited to this. Alternatively, the frequency or spectral peak value of an nth-order harmonic sideband that is n times the power supply frequency fs (n is a natural number), such as a sideband (nfs-fb), may be used. The feature calculation unit 4 may also have a memory function for storing the extracted first and second feature amounts and a function for chronologically organizing and searching these feature amounts. The feature calculation unit 4 may also have a memory function for storing the converted two-dimensional array data and a function for chronologically organizing and searching the data.

[0049] The data stored by the memory function may be stored in another computer by the communication function. The memory function may also include a function for deleting stored data and a function for acquiring past data by the communication function and storing it in the memory function. The frequency feature F1 may be not only the spectral value caused by the anomaly itself, but also a value calculated based on statistics. For example, the statistics may include at least one of variance, standard deviation, kurtosis, and skewness.

[0050] 3 , next, if the state determination process is in the learning phase, the abnormality diagnosis unit 5 constructs a trained model using machine learning by the model generation unit 5A (step S008). That is, the abnormality diagnosis unit 5 has an AI function and performs machine learning and deep learning. Note that if the state determination process is not in the learning phase in step S007, that is, if a trained model has already been constructed, the learning process in step S008 is not performed.

[0051] Next, the abnormality diagnosis unit 5 diagnoses abnormalities in the rotating mechanical system 40 using the AI ​​diagnosis unit 5B and the trained model configured as described above (step S009). Next, the result of the diagnosis by the diagnosis device 10 is displayed on the display unit 20 (step S010). If the diagnosis result indicates an abnormality (step S011, YES), the abnormality diagnosis unit 5 stops operation of the rotating mechanical system 40, and if the diagnosis result indicates no abnormality (step S011, YES), the abnormality diagnosis unit 5 continues.

[0052] In the above step S006, the extraction unit 4A extracts the time-series feature value T1 on the time-series scale and the frequency feature value F1 on the frequency scale based on the first database DB1, but the extraction may be performed without using the first database DB1. As long as one or more time-series feature value T1 on the time-series scale and one or more frequency feature value F1 on the frequency scale are extracted, the effects described below can be obtained. Furthermore, in the present embodiment, an example has been given in which, if the diagnosis result indicates an abnormality, operation is immediately stopped. However, if the diagnosis result indicates an abnormality, an operation may be performed in which an abnormality is reported to a manager, such as an operator or maintenance person, and the operation is immediately stopped.

[0053] In this embodiment, an abnormality location prediction unit 5C is provided that has a simple function of predicting abnormality locations using the diagnosis results obtained by the abnormality diagnosis unit 5. For example, when diagnosis is performed using an autoencoder that inputs feature quantities, the input / output errors are sorted in ascending order, and the abnormality location is predicted based on that order. In addition, when predicting the abnormality location, the failure probability may be calculated based on the fault location. The calculated abnormality prediction data is displayed on the display unit 20.

[0054] An example of the AI ​​function provided in the abnormality diagnosis unit 5 of this embodiment will be described below. Fig. 9 is a conceptual diagram for explaining the AI ​​function in the abnormality diagnosis unit 5. Fig. 10 is a conceptual diagram for explaining the AI ​​function in the abnormality diagnosis unit 5.

[0055] The abnormality diagnosis unit 5 performs diagnosis using a decision tree, a support vector machine (SVM), a local outlier factor (LOF), a neural network, a convolutional neural network, or an autoencoder, as shown in Fig. 9. Clustering, regression analysis, etc. may also be used.

[0056] A decision tree identifies data based on the relationship between the data's features and a threshold value, and branches out using conditional branching at each node to identify the data. A random forest, which uses multiple decision trees for diagnosis, may also be used. An SVM finds boundaries and interfaces that distinguish different classes and categories, and constructs and identifies an optimal discriminant function for classifying two or more classes. The local outlier factor method estimates local density from the distance to nearby points, compares the local density of each data point, and identifies anomalies.

[0057] Neural networks are a type of artificial intelligence that teaches computers to process data in a way that mimics the way the human brain works. They weight multiple features (inputs) and learn weights to minimize the error between the resulting predictions and the label data, allowing for classification.

[0058] Alternatively, a convolutional neural network, which adds a convolutional layer and a pooling layer to a neural network, may be used. Convolutional neural networks perform specific mathematical functions, such as filtering, to extract features from images that are relevant to image recognition and classification, and have a proven track record in image classification.

[0059] The convolutional layer obtains a feature map by filtering the nodes, and the pooling layer further reduces the feature map output from the convolutional layer to create a new feature map. An autoencoder is composed of an encoder and a decoder, and learns to reduce the dimensionality of the input data, extract features, and reconstruct them to return to the original data. Reducing the dimensionality of the data causes some information to be lost, but it is possible to retain important information so that it can be reproduced even without the missing parts.

[0060] It is also possible to add a convolutional layer and a pooling layer to the autoencoder to perform dimensionality reduction and reconstruction. This improves adaptability when used for image classification. Furthermore, anomalies can be diagnosed using an autoencoder based on the error between input and output. When diagnosing based on the reconstruction error, which is the error between input and output, diagnosis using a threshold or local outlier factor method, for example, is used. Diagnosis using a threshold diagnoses an anomaly when the reconstruction error exceeds a threshold. Diagnosis using the local outlier factor method is a clustering technique that diagnoses anomalies by performing clustering using the reconstruction error output by the autoencoder as input.

[0061] Here, when the above-mentioned convolutional neural network and autoencoder are used, the two-dimensional array data of the features converted by the conversion unit 4D is used, so that the features can be recognized as image data, thereby further improving the accuracy of diagnosis.

[0062] Note that the abnormality diagnosis unit 5 may use multiple algorithms for diagnosis. For example, diagnosis may be performed using multiple algorithms and a majority vote may be performed on the diagnosis results, or arbitrary weights may be set according to a model and diagnosis may be performed using a threshold value, etc. Here, AI diagnosis requires prior learning, and a learning phase must be set as shown in step S008 above, or a learned model must be implemented in the AI ​​diagnosis unit. In other words, a function for setting a learning phase and a function for incorporating a learned model trained by an external computer are required.

[0063] Furthermore, AI learning can be divided into supervised learning and unsupervised learning. Supervised learning requires prior learning including correct answer labels, whereas unsupervised learning can discover patterns and trends without the need for correct answer labels. If labels are required, a function may be provided that allows labels to be set manually as needed. For example, an administrator may assign labels during learning. Re-labeling may also be performed after learning.

[0064] As described above, the abnormality diagnosis unit 5 performs abnormality diagnosis using AI. However, it is not intended that the state diagnosis of the rotating mechanical system 40 by the abnormality diagnosis unit 5 be limited to diagnosis using AI. For example, the abnormality diagnosis unit 5 may be configured without an AI function and may diagnose the above feature quantities using a threshold value. In this way, the abnormality diagnosis by the abnormality diagnosis unit 5 is performed using one or more of machine learning, deep learning, and judgment using a threshold value.

[0065] 3 may be provided from an external computer. For example, operation data of the rotating mechanical system 40 detected by a sensor equipped with a communication function may be input to the external computer via a dedicated line or a network, and the rotating mechanical system 40 may be diagnosed for abnormality according to the above-described flow chart.

[0066] The following describes control that further improves the diagnostic accuracy of the rotating mechanical system 40. In order to unify the scales of the extracted feature quantities in the processes of steps S005 and S006, the extraction unit 4A may perform scale unification, for example, by calculating normalization and standardization.

[0067]

[0068] That is, the extraction unit 4A adjusts the value of the feature amount that exceeds the threshold range so that it is within the upper or lower limit of the threshold range, thereby emphasizing slight differences in the feature amount and enabling accurate abnormality diagnosis.

[0069] FIG. 11A is a diagram showing an example of the accuracy of abnormality diagnosis when the threshold value Kc is not set. FIG. 11B is a diagram showing an example of the accuracy of abnormality diagnosis when the threshold value Kc is set. The vertical axis represents the mean absolute error (MAE), and the horizontal axis represents the belt tension. The belt tension is considered to be appropriate (normal) in the range of 75% to 125%, and any tension outside this range is considered to be abnormal. In this example, the belt tension is considered to be appropriate (normal) in the range of 75% to 125%, but the appropriate tension range may be changed as desired. For example, the appropriate range may be 80% to 120%, or may be changed to 70% or above.

[0070] As shown in Figure 11A, when the threshold value Kc is not set, the MAE value is large, indicating a large error, even when the belt tension is within the appropriate range of 75% to 125%. In contrast, as shown in Figure 11B, when the threshold value Kc is set, the MAE value is small, indicating a small error, when the belt tension is within the appropriate range of 75% to 125%. As described above, by setting the threshold value Kc that defines the threshold range, it is possible to accurately diagnose the rotating mechanical system 40. Note that the present invention is not limited to control in which spectral peak values ​​that exceed the threshold value Kc are cut off to the same value as the threshold value Kc; as long as they do not exceed the threshold value Kc, they may be set to a value less than the threshold value Kc.

[0071] Note that the following scaling of feature quantities may be performed differently from the above. The data acquisition unit 2 acquires status signals indicating the operating status of each component of the rotating mechanical system 40. The status signals may be load information or the like. Alternatively, the operator may acquire belt tension or the like and input this as a status signal to the data acquisition unit 2.

[0072] The feature calculation unit 4 includes a second database DB2 shown in FIG. 1 , which stores correction values ​​for correcting the feature values ​​acquired at the 30% reduced belt tension to the feature values ​​acquired at the 100% belt tension, which is the first operating state, when the belt tension indicated by the status signal as the operating state is 30% less than the 100% belt tension, which is the first operating state. The second database DB2 may be generated, for example, by calculating average values ​​of feature values ​​acquired multiple times in each operating state based on the status signal. This enables accurate diagnosis regardless of the state of each component, such as the belt, that constitutes the rotating mechanical system 40.

[0073] Furthermore, for example, the ratio between the feature quantity at the initial stage when the setup of the rotating mechanical system 40 is completed and the feature quantity during operation may be derived as shown below, and the derived ratio may be used as a correction amount for correcting the feature quantity. For example, the control unit 3 may define the initial stage as several hours after the start of operation of the rotating mechanical system 40, and obtain average values ​​F_mean1, F_mean2, ... of each feature quantity at this initial stage. Next, the control unit 3 may obtain each feature quantity F1, F2, ... during operation to be diagnosed, derive F1 / F_mean1, F2 / F_mean2, ..., and use this ratio as a correction amount for scaling the feature quantity. That is, the extraction unit 4A may cut off the extracted feature quantity at a threshold value Kc so that it does not exceed the threshold range. When the cutoff threshold Kc and the feature quantity are denoted by x, the following equation is given:

[0074] A diagnostic device and diagnostic method for a rotating machinery system that are different from those described above will now be described. Fig. 12 is a flow chart showing another example of the flow of the condition diagnostic process performed by the diagnostic device 10 of this embodiment. Steps corresponding to those shown in Fig. 3 are assigned the same step numbers as in Fig. 3. For simplification, some steps shown in Fig. 3 are omitted from the illustration.

[0075] In the present diagnosis device 10, the sensor 1 and the data acquisition unit 2 acquire operation data when the rotating mechanical system 40 is in operation (step S001). One or more time-series feature quantities are extracted from the acquired operation data, and one or more frequency feature quantities are extracted from the data converted to a frequency scale (steps S005 and S006). At this time, the feature calculation unit 4 extracts each feature quantity as a feature quantity of the rotating machine 42, a feature quantity of the power transmission mechanism 43, and a feature quantity of the load equipment 44, respectively.

[0076] The abnormality diagnosis unit 5 performs an abnormality diagnosis using the extracted feature amounts of the rotating machine 42, the feature amounts of the power transmission mechanism 43, and the feature amounts of the load equipment 44 (step S009). The abnormality diagnosis result is displayed on the display unit 20 (step S010). If the diagnosis result indicates an abnormality, the abnormality diagnosis unit 5 stops operation (step S011, YES), and if the diagnosis result indicates no abnormality (step S011, NO), the abnormality diagnosis is continued.

[0077] In addition, the characteristic quantities of the rotating machine 42, the characteristic quantities of the power transmission mechanism 43, and the characteristic quantities of the load equipment 44 may be calculated from operating data acquired by the same sensor 1, or multiple sensors 1 may be used, with different sensors 1 being used for each component.

[0078] Note that different frequency analyses may be performed when calculating the characteristic amounts of the rotating machine 42, the power transmission mechanism 43, and the load equipment 44. In other words, the frequency analyses performed when calculating the characteristic amounts of the rotating machine 42, the power transmission mechanism 43, and the load equipment 44 may be the same or different.

[0079] The abnormality diagnosis of the rotating machine 42, the abnormality diagnosis of the power transmission mechanism 43, and the abnormality diagnosis of the load equipment 44 may be performed using the same algorithm or different algorithms. Note that the time-series feature value T1 and the frequency feature value F1 may be extracted differently for the rotating machine 42, the power transmission mechanism 43, and the load equipment 44. Also, different numbers of feature values ​​may be extracted for each. In other words, the feature values ​​of the rotating machine 42, the power transmission mechanism 43, and the load equipment 44 may be the same feature or different features. Even with this diagnostic process, as with the diagnostic processing described above, it is possible to diagnose abnormalities in the rotating mechanical system 40, which is made up of the rotating machine 42, the power transmission mechanism 43, and the load equipment 44, with high accuracy.

[0080] Although the method of diagnosing abnormalities in the rotating machine 42, the power transmission mechanism 43, and the load equipment 44 individually has been described, the present invention is not limited to this. The devices that constitute the rotating mechanical system 40 may be divided into a plurality of groups and each may be diagnosed for abnormalities. For example, the power transmission mechanism 43 may be further divided into a belt-related mechanism and a gear-related mechanism and each may be diagnosed for abnormalities. Furthermore, the rotating mechanical system 40 may be divided into the rotating machine 42 and machines other than the rotating machine 42, and the power transmission mechanism 43 and the load equipment 44 may be diagnosed for abnormalities together. Such an abnormality diagnosis device for a rotating mechanical system can individually diagnose abnormalities in the rotating machine, the power transmission mechanism, and the load equipment, and can more appropriately diagnose abnormalities in the rotating mechanical system.

[0081] The hardware configuration of the control unit 3 will be described below. The control device as the control unit 3 is configured with a processor 3A and a storage device 3B serving as the memory function described above, as shown in FIG. 13 , which is an example of hardware. The storage device 3B includes a volatile storage device such as a random access memory and a non-volatile auxiliary storage device such as a flash memory, both not shown. A hard disk auxiliary storage device may also be provided instead of the flash memory. The processor 3A executes a program input from the storage device 3B. In this case, the program is input to the processor 3A from the auxiliary storage device via the volatile storage device. The processor 3A may output data such as calculation results to the volatile storage device of the storage device 3B, or may store the data in the auxiliary storage device via the volatile storage device.

[0082] The diagnostic device of this embodiment configured as described above is a diagnostic device comprising an acquisition unit that acquires operating data of components that constitute a rotating mechanical system, and a control unit that diagnoses the state of the components that constitute the rotating mechanical system based on feature quantities obtained from the acquired operating data, wherein the control unit selects one or more first feature quantities as the feature quantities from time series data obtained from the operating data, and one or more second feature quantities as the feature quantities from frequency data obtained by frequency analysis of the operating data, and diagnoses the state of the rotating mechanical system based on the selected first feature quantities and second feature quantities.

[0083] Physical phenomena vary depending on the type of abnormality occurring in a rotating mechanical system. Therefore, a diagnosis using only one type of signal, such as time-series features or frequency features, may not fully reflect the abnormal state of the rotating mechanical system, resulting in reduced diagnostic accuracy. For example, when belt tension decreases and an abnormal state occurs, the rotational motion of the rotating machine is not directly transmitted to the belt, causing the belt to vibrate. In this case, the force on the belt increases when torque is high and decreases when torque is low, resulting in the belt's vibrating string vibration. When the intensity of the spectral value indicating an abnormality changes with torque, a diagnosis using only a threshold value for frequency features may result in an erroneous diagnosis, where a normal value is detected as an abnormality. Therefore, a diagnosis using both time-series features and frequency features is required. In this way, diagnosis using both time-series features and frequency features improves the diagnostic accuracy of abnormalities in, for example, the power transmission mechanism and load that transmit the rotational energy of the rotating machine, which are susceptible to torque changes.

[0084] In this way, the diagnostic device of this embodiment selects one or more first feature quantities which are time-series feature quantities and one or more second feature quantities which are frequency feature quantities, thereby enabling efficient condition diagnosis of a rotating machinery system with high diagnostic accuracy.

[0085] Furthermore, in the diagnostic device of this embodiment configured as described above, the control unit includes a first database indicating the correlation between the time series data and the frequency data, selects one or more first feature amounts and one or more second feature amounts based on the first database, and diagnoses the state of the rotating mechanical system based on the selected first feature amounts and second feature amounts.

[0086] In this way, a first database indicating the correlation between time-series data and frequency data may be provided, and control may be performed to extract the first feature value and the second feature value based on this first database. Each feature value obtained from each component of the rotating mechanical system reflects the abnormality to a different extent depending on the type, severity, etc., of the abnormality occurring in the rotating mechanical system. Therefore, for example, if the time-series data is a torque waveform indicating temporal fluctuations in torque, the torque fluctuations are related to the motor current, and therefore the first database may associate current spectrum data as frequency data with the torque waveform as time-series data. This enables more accurate and efficient condition diagnosis of the rotating mechanical system than when feature values ​​are randomly extracted from time-series data and frequency data.

[0087] In addition, in the diagnostic device of this embodiment configured as described above, the control unit constructs a trained model for diagnosing the condition of the rotating mechanical system by machine learning using two-dimensional array data in which the first feature amount and the second feature amount are arranged, and diagnoses the condition of the rotating mechanical system based on the trained model and the array data generated using the acquired first feature amount and the second feature amount. This enables highly accurate and efficient condition diagnosis of the rotating mechanical system even in cases where extensive knowledge about rotating machines, power transmission mechanisms, load equipment, etc. is required, there are many condition changes to be considered in condition diagnosis, and a large amount of diverse operating data is acquired. Furthermore, even in cases where condition changes in the rotating mechanical system occur that are likely to affect the feature amount, such as fluctuations in load state, by configuring the array data to arrange at least feature amounts related to load data, accurate condition diagnosis of the rotating mechanical system is possible regardless of such load fluctuations.

[0088] In the diagnostic device of the present embodiment configured as described above, the array data is configured by being divided into a plurality of cells having a set number of rows and columns, and includes one or more cells to which the first feature amount is input and one or more cells to which the second feature amount is input, and the control unit adjusts the first feature amount and the second feature amount input into the cells to values ​​within a threshold range derived according to the variance of each value. In this way, by providing array data configured by being divided into a plurality of cells having a set number of rows and columns, adjusting the feature amount used in each cell to a value within a threshold range, and excluding outliers of the feature amount, it is possible to efficiently diagnose the condition of a rotating machinery system with high diagnostic accuracy, even when even a larger amount and a wider variety of operating data are used.

[0089] In the diagnostic device of the present embodiment configured as described above, the control unit sets the threshold range for the spectral values ​​of frequency components in each frequency band in the frequency data, and uses an average of the spectral values ​​of the frequency components adjusted to values ​​within the threshold range in each frequency band as the second feature. This enables condition diagnosis of the rotating machinery system taking the floor level into consideration, thereby enabling more accurate and efficient condition diagnosis of the rotating machinery system.

[0090] Furthermore, in the diagnostic device of this embodiment configured as described above, the control unit derives the rotational frequency of the component, and uses as the second feature value the frequency or spectrum peak value of a sideband component generated around a frequency that is n times (n is a natural number) the power supply frequency, the frequency or spectrum peak value being determined by the difference or sum between the power supply frequency that drives the rotating machine as the component and the rotational frequency. In this way, by using as the feature value the frequency or spectrum peak value of a sideband component caused by an abnormality, it is possible to diagnose the condition of the rotating machine system with even higher diagnostic accuracy and efficiency.

[0091] Furthermore, in the diagnostic device of this embodiment configured as described above, the control unit derives an average value of the feature quantity for multiple times in each of the operating states based on a state signal indicating the operating state of each of the components constituting the rotating mechanical system, includes a second database containing a correction value for correcting the acquired value of the feature quantity based on the derived average value of the feature quantity from the operating state when the feature quantity was acquired to a value of the feature quantity in a first operating state as the preset operating state, and corrects the acquired value of the feature quantity to the value of the feature quantity in the first operating state based on the second database. This enables efficient condition diagnosis of the rotating mechanical system with high diagnostic accuracy, regardless of changes in the condition of the rotating mechanical system, such as changes in belt tension and load.

[0092] Although exemplary embodiments are described in the present disclosure, the various features, aspects, and functions described in the embodiments are not limited to the application of a particular embodiment, but may be applied to the embodiments alone or in various combinations. Therefore, countless variations not illustrated are anticipated within the scope of the technology disclosed in this specification. For example, variations in, addition to, or omission of at least one component are included.

[0093] 2 Data acquisition unit (acquisition unit), 3 Control unit, 10 Diagnostic device, 20 Display unit (alert unit), 30 Diagnostic system, 40 Rotating mechanical system, 42 Rotating machine (component), 43 Power transmission mechanism (component), 44 Load equipment (component), T1 Time series feature (first feature), F1 Frequency feature (second feature), DB1 First database, DB2 Second database, M Map data (array data).

Claims

1. A diagnostic device comprising: an acquisition unit that acquires operating data of components that make up a rotating mechanical system; and a control unit that diagnoses the state of the components that make up the rotating mechanical system based on feature quantities obtained from the acquired operating data, wherein the control unit selects one or more first feature quantities as the feature quantities from time series data obtained from the operating data, and one or more second feature quantities as the feature quantities from frequency data obtained by frequency analysis of the operating data, and diagnoses the state of the rotating mechanical system based on the selected first feature quantities and second feature quantities.

2. The diagnostic device described in claim 1, wherein the control unit constructs a trained model for diagnosing the state of the rotating mechanical system by machine learning using two-dimensional array data in which the first feature amount and the second feature amount are arranged, and diagnoses the state of the rotating mechanical system based on the trained model and the array data generated using the acquired first feature amount and second feature amount.

3. The diagnostic device of claim 2, wherein the array data has a set number of rows and columns and is divided into a plurality of cells, and includes one or more cells to which the first feature amount is input and one or more cells to which the second feature amount is input, and the control unit adjusts the first feature amount and the second feature amount input into the cells to values ​​within a threshold range derived according to the variance of each value.

4. The diagnostic device according to claim 3, wherein the control unit sets the threshold range for the spectral values ​​of the frequency components in each frequency band in the frequency data, and uses the average of the spectral values ​​of the frequency components adjusted to values ​​within the threshold range in each frequency band as the second feature.

5. A diagnostic device according to any one of claims 2 to 4, wherein the control unit derives the rotational frequency of the component, and uses as the second feature the frequency or spectrum peak value of a sideband component generated around a frequency that is n times (n is a natural number) the power supply frequency, determined by the difference or sum between the power supply frequency that drives the rotating machine serving as the component and the rotational frequency.

6. A diagnostic device according to any one of claims 2 to 5, wherein the control unit comprises a first database indicating the correlation between the time series data and the frequency data, selects one or more of the first feature amounts and one or more of the second feature amounts based on the first database, and diagnoses the state of the rotating mechanical system based on the selected first feature amounts and second feature amounts.

7. A diagnostic device as claimed in any one of claims 2 to 6, wherein the control unit derives an average value of the feature quantity for multiple times in each of the operating states based on a state signal indicating the operating state of each of the components that make up the rotating mechanical system, and is provided with a second database containing a correction value that corrects the value of the feature quantity obtained based on the derived average value of the feature quantity from the operating state when the feature quantity was obtained to the value of the feature quantity in a first operating state that is a pre-set operating state, and corrects the value of the feature quantity obtained to the value of the feature quantity in the first operating state based on the second database.

8. A diagnostic device according to any one of claims 2 to 7, wherein the control unit uses an autoencoder, which is unsupervised machine learning, as the machine learning, derives a restoration error between an input value of the array data and an output value of the array data output from the autoencoder, and diagnoses the state of the rotating mechanical system based on the restoration error.

9. A diagnostic device according to any one of claims 2 to 8, wherein the control unit extracts the first feature amount and the second feature amount that indicate at least a load state of the component part.

10. A diagnostic device as described in any one of claims 2 to 9, wherein the control unit diagnoses the state of the rotating mechanical system for each component based on the first feature and the second feature selected for each component.

11. A diagnostic device as claimed in any one of claims 1 to 10, wherein the control unit uses at least one of the variance, standard deviation, kurtosis, skewness, current, voltage, vibration velocity, vibration acceleration, sound, distortion, temperature, pressure, power and AE of the maximum value of each of the operating data in a set period as the operating data to be frequency analyzed as the second feature.

12. The diagnostic device according to any one of claims 1 to 11, wherein the control unit selects, from the time series data, the first feature amount to include at least one of the variance of the maximum value of each of the operating data in a set period, standard deviation, kurtosis, skewness, maximum value, average value, sum, root mean square value, crest factor, clearance coefficient, total harmonic distortion (THD), variance of the effective value in a set period, and variance of the maximum value in a set period.

13. A diagnostic system comprising: a diagnostic device according to any one of claims 1 to 12; and a notification unit connected to said diagnostic device via a network and notifying an administrator of the diagnostic results obtained by said diagnostic device.

14. A diagnostic method using the diagnostic device according to any one of claims 1 to 12, comprising: selecting one or more first feature quantities as the feature quantities from the time-series data of the operating data; and selecting one or more second feature quantities as the feature quantities from frequency data obtained by frequency analysis of the operating data; and diagnosing the state of the rotating mechanical system based on the selected first feature quantities and second feature quantities.

Images