JP4513796B2 - Abnormality monitoring device - Google Patents

Description

  The present invention relates to an abnormality monitoring device that determines the presence or absence of an abnormality in an operation to be inspected by classifying characteristics of an electric signal reflecting the operation to be inspected by a neural network.

  2. Description of the Related Art Conventionally, a technique for classifying a target signal obtained from a test object by using a classification function of a neural network (neurocomputer) is known. This type of technology is used in an abnormality monitoring apparatus that recognizes voice and determines whether an apparatus is operating normally or whether an apparatus has an abnormality. For example, in an abnormality monitoring apparatus that employs this type of technology, the operation sound of the device or the vibration of the device is converted into an electrical signal by a sensor unit (transducer), and the output of the sensor unit is used as the target signal. Various techniques for extracting feature vectors composed of a plurality of elements to be represented and classifying the feature vectors using a neural network have been proposed.

  Various configurations are known for neural networks. For example, it has been proposed to classify feature vector categories using a competitive learning type neural network (self-organizing map = SOM). The competitive learning type neural network is a neural network including two layers of an input layer and an output layer, and performs two operations of a learning mode and an inspection mode.

  In the learning mode, learning data is given without using a teacher signal. If a category is given to learning data, a category can be associated with a neuron in the output layer, and a cluster of neurons belonging to the same category can be formed. Accordingly, in the learning mode, a clustering map indicating a category can be associated with a cluster of neurons in the output layer.

In the inspection mode, the feature vector (input data) to be classified is given to the learned competitive learning type neural network, and the category of the cluster to which the fired neuron belongs in the clustering map is checked against the clustering map. Can be classified (see, for example, Patent Document 1).
JP 2004-354111 A

  By the way, although the thing of patent document 1 is considered that it can be used for various uses if the kind of sensor used for a signal input part is changed, or learning data is changed, learning data is Since it is generated using a signal taken from a signal input unit and an input unit is provided to set a category of learning data, there is a problem that the scale of the abnormality monitoring device increases. .

  On the other hand, this type of abnormality monitoring device is preferably used by being incorporated in various devices and various devices to be inspected, and is a small abnormality monitoring device that is intended for a specific application and can be incorporated in or attached to the inspection object. Is desired.

  The present invention has been made in view of the above reasons, and its purpose is to make it possible to receive setting data necessary for setting a weight vector of a neural network from an external device, which is independent of the external device in actual use. It is an object of the present invention to provide an anomaly monitoring apparatus that can be downsized by operating.

According to the first aspect of the present invention, at least one element of the range extracted as the target signal and the type of the feature vector extracted from the target signal among the electric signals reflecting the operation of the inspection target can be selected as the adjustment element, and the signal input unit A feature extraction unit that extracts a feature vector representing the feature of the target signal captured by the step, a category classification unit that classifies the category of the feature vector extracted by the feature extraction unit after setting a weight vector for each neuron using a neural network, and a service interface for exchanging with an external device configuration data required for setting the weight vector in the category classification unit for selecting the regulatory elements an abnormality monitoring device Ru provided integrally, the external device to be inspected An analysis signal input unit that captures electrical signals that reflect operation and an analysis signal input unit Select the signal storage unit that stores the electrical signal and the range to be extracted as the target signal from the electrical signal stored in the signal storage unit, and select and extract feature vectors representing the characteristics of the target signal from multiple types A feature extraction unit, a feature vector storage unit that stores feature vectors extracted by the analysis feature extraction unit, and a neural network that performs learning using the feature vectors stored in the feature vector storage unit. The analysis category classification unit for classifying the category of the feature vector extracted by the feature extraction unit, and the classification result in the analysis category classification unit for the electric signal captured by the analysis signal input unit during the operation of the inspection target are presented. Presenting means, and in the analysis feature extraction unit, the range from which the target signal is extracted from the electrical signal and the target signal are extracted Selects a combination of desired classification results of the combination of the types of symptoms vector selectable range as regulatory elements of the feature extraction unit, which gives the combination selected in the feature extraction unit, the setting data After receiving, the external device is removed from the service interface and operates independently of the external device.

According to the configuration of the first aspect of the present invention, the service interface for exchanging the setting data with the external device is provided, and after the setting data is received, the external device is detached from the service interface and operates independently of the external device. Therefore, after receiving the setting data from the external device, the external device can be removed from the service interface, and the function for collecting the setting data can be separated as the external device. As a result, it is not necessary to have a function to collect setting data for setting the weight vector in the category classification unit, and the configuration can be simplified and reduced in size, and the number of components can be reduced and created at low cost. It becomes possible to do. Also, since it is possible to share an external device with multiple abnormality monitoring devices, it is possible to share when using multiple abnormality monitoring devices by classifying the functions given to the external device from the abnormality monitoring device. The cost reduction effect due to the conversion becomes high. Furthermore, since it is only necessary to provide an input unit such as a keyboard for setting the category of learning data in the external device, it can be miniaturized and can be incorporated into the inspection object or attached to the inspection object. It becomes possible.

By the way, in the present invention, at least one element of the range in which the target signal is extracted from the electric signal and the type of the feature vector is an adjustment element. The range of the target signal includes a time range, a frequency range, an amplitude range, and the like. The type of feature vector means the type of feature vector whose elements are frequency components, amplitude components, and the like. Range taking out a target signal from an electric signal, because there at least one type of feature vectors as selectable as regulatory elements, it is possible to extract information that well reflects the operation of the test object, the determine Jogayo easily Thus, the determination accuracy is improved.

In addition, the external device captures the electrical signal from the analysis signal input unit and stores it in the signal storage unit, selects the range from which the target signal is extracted for the electrical signal stored in the signal storage unit, and selects the type of feature vector Therefore, feature vectors can be extracted under various conditions. By evaluating the results obtained by applying the feature vectors extracted under various conditions to the category classification unit, it is possible to easily find the conditions for obtaining the desired classification results. Moreover, since the combination of the obtained conditions is selected within a range that can be set by the feature extraction unit of the abnormality monitoring device, it is possible to set conditions suitable for obtaining a desired classification result in the abnormality monitoring device.

  In the embodiment described below, an example is shown in which whether the operation of the inspection target is normal or abnormal is determined based on the feature vector of the target signal generated by the operation of the inspection target. Moreover, although the installation apparatus provided with motive power sources like a motor is assumed as a test object, the kind of test object is not ask | required in particular.

  As shown in FIG. 1, the abnormality monitoring apparatus A described in the present embodiment uses a category classification unit 1 including an unsupervised competitive learning type neural network (hereinafter simply referred to as “neural network”). As shown in FIG. 2, the neural network as the category classification unit 1 includes two layers of an input layer 21 and an output layer 22, and each neuron N <b> 2 of the output layer 22 is connected to all the neurons N <b> 1 of the input layer 21. It has a combined configuration. The neural network as the category classification unit 1 is assumed to be realized by executing an appropriate application program on a sequential processing type computer, but a dedicated neurocomputer can also be used.

  The operation of the neural network as the category classification unit 1 generally includes a learning mode and an inspection mode. After learning using appropriate learning data in the learning mode, a plurality of items generated from actual target signals in the inspection mode are used. Classify feature vector (input data) categories consisting of elements.

  The coupling degree (weighting coefficient) between the neuron N1 of the input layer 21 and the neuron N2 of the output layer 22 is variable, and in the learning mode, the category classification unit 1 is learned by inputting learning data to the category classification unit 1. The weighting coefficient for each neuron N1 in the input layer 21 and each neuron N2 in the output layer 22 is determined. In other words, each neuron N2 in the output layer 22 is associated with a weight vector whose element is a weight coefficient between each neuron N1 in the input layer 21. Therefore, the weight vector has the same number of elements as the neuron N1 in the input layer 21, and the number of feature vector elements input to the input layer 21 matches the number of elements in the weight vector.

  On the other hand, in the inspection mode, when input data whose category is to be determined is given to the input layer 21 of the learned category classification unit 1, the Euclidean distance between the weight vector and the input data among the neurons N2 of the output layer 22 is the smallest. A neuron N2 fires. If a category is associated with the neuron N2 of the output layer 22 in the learning mode, the category of input data can be known from the category of the position of the fired neuron N2.

  A category is associated with each neuron N2 of the output layer 22 in the neural network that is the category classification unit 1 in the procedure described later. In this embodiment, the category classification unit 1 classifies two categories, normal and abnormal, and inputs only learning data of normal categories in the learning mode. That is, when the input data given in the inspection mode does not belong to the normal category set in the category classification unit 1, the input data is regarded as abnormal.

  The category of the learning data is reflected in the category of each neuron N2 of the output layer 22, and when a large number (for example, 150) of learning data is given, the learning data among the neurons N2 of the output layer 22 in the category classification unit 1 A weight vector having a small Euclidean distance from the learning data is set in the neuron N2 corresponding to the category. That is, the neuron N2 is fired by giving the learning data after learning.

  By the way, the target signal to be classified by the category classification unit 1 is an electrical signal obtained with the operation of the equipment device (hereinafter simply referred to as “device”) X. For example, vibration for detecting vibration generated during the operation of the device X The output of the sensor Y is used as an electrical signal that reflects the operation of the device X. As the vibration sensor Y, an acceleration pickup is used. This electrical signal is taken into the signal input unit 2. The vibration sensor Y basically includes a transducer that detects the operation of the device X and a front end (a preamplifier, a filter, and the like) that sends the output of the transducer to the subsequent stage. Various transducers such as a microphone, a TV camera, and an odor sensor that detect the operation sound of the device X can be used alone or in combination. Moreover, it is also possible to take in the load current etc. which generate | occur | produce from the apparatus X to the signal input part 2 as an electrical signal. The electrical signal captured by the signal input unit 2 is subjected to preprocessing for reducing noise by limiting the frequency band as necessary, and then given to the feature extraction unit 3 to extract a feature vector.

  The feature extraction unit 3 has a function of cutting out a range reflecting the operation of the device X from the electrical signal as a target signal. As a range reflecting the operation of the device, it is important that it is a period during which the device X is operating. Therefore, by using a timing signal (trigger signal) synchronized with the operation of the device X, or by using a waveform characteristic of the target signal (for example, a start point and an end point of a group of target signals), the signal input unit The timing for segmenting the target signal from the output of 2 is determined. For example, when the target signal output from the device X has periodicity, it is divided for each period by segmentation, and a feature vector for each period is extracted. The signal input unit 2 or the feature extraction unit 3 converts the target signal into a digital signal.

  An example in which the range of the target signal is defined by time is shown in FIG. FIG. 3A shows an example in which the time t0 when the trigger signal is generated from the device X is set as the start time, and the time t1 when the predetermined time T1 has elapsed since the trigger signal is generated is set as the end time of the target signal. Such a target signal is suitable when a trigger signal is generated from the device X and the operation speed of the device X is constant. FIG. 3B shows an example in which the start time t2 and the end time t3 for extracting the target signal are determined using the change in the amplitude of the electrical signal. Such a target signal can be extracted even when a trigger signal is not generated from the device X, and can be extracted even when the operation speed of the device X changes. FIG. 3C shows an example in which the start time is the time t0 when the trigger signal is generated from the device X, and the end time t3 of the target signal is determined using the amplitude change. This target signal can be used when the device X generates a trigger signal and the operation speed of the device X changes.

  The range to be extracted as the target signal in the feature extraction unit 3 may be not only the time range but also the frequency range or the amplitude range. By appropriately determining the range and cutting out the target signal, the operation of the device X is determined. Features are easier to extract. In the feature extraction unit 3, a method for determining a range to be extracted as an object signal from the electrical signal is instructed by setting data to be described later.

  An example in which the range of the target signal is defined by frequency is shown in FIG. FIG. 4A is an example in which the entire audio frequency is the target signal, and the elements of the feature vector are distributed over the entire audio frequency as shown in FIG. FIG. 4C shows an example in which the low frequency range of the audible frequency is used as the target signal. As shown in FIG. 4D, the feature vector element is only in the low frequency range of the audible frequency. Similarly, if the high frequency range of the audible frequency is set as the target signal as shown in FIG. 4E, the element of the feature vector is only in the high frequency range of the audible frequency as shown in FIG. 4F.

  The feature vector extracted from the target signal after performing segmentation is information reflecting the operation of the device X, and is appropriately extracted from the information included in the target signal. Which information is to be focused on can be selected, and can be selected by setting data in the same manner as the range for extracting the target signal. As the feature vector, power (frequency component) for each of a plurality of frequency bands can be extracted, and a vector having the frequency component as an element can be used as the feature vector. For the extraction of frequency components, an FFT (Fast Fourier Transform) technique or a filter bank made up of a large number of bandpass filters is used. Which frequency band power is used as an element of the feature vector is appropriately selected according to the target device X and the abnormality to be extracted. That is, the frequency band is selected by setting data.

  In the feature extraction unit 3, at least one of the range (or rule) to cut out the target signal and the type of feature vector (which frequency is an element) is an adjustable adjustment element, and there are a plurality of options for the adjustment element. Can be selected from. The adjustment element is selected by setting data described later.

  The feature vector obtained for each period from the feature extraction unit 3 is given to the category classification unit 1 every time the feature vector is extracted. In order to be able to use the category classification unit 1 composed of a neural network, it is necessary to set a weight vector for each neuron N2 of the output layer 22 in the category classification unit 1. As described above, when a weight vector is set to the neuron N2 of the category classification unit 1 and then the feature vector extracted by the feature extraction unit 3 is given to the category classification unit 1, the weight vector having the minimum Euclidean distance from the feature vector Since the neuron N2 to which is set fires, the category of the feature vector can be classified by setting the category for each neuron N2. If a clustering map associated with the output layer 22 is provided here and a category is associated with each neuron N2 in the clustering map, the category of the fired neuron can be known.

  By the way, the feature of this embodiment is that a service interface 4 for receiving a weight vector associated with each neuron N2 of the category classification unit 1 from the external device B is provided, and the weight received from the external device B through the service interface 4 The vector is set in each neuron N2 of the category classification unit 1 by the data setting unit 5. The category classification unit 1, the signal input unit 2, the feature extraction unit 3, the service interface 4, and the data setting unit 5 are housed in one package or container and are provided integrally. In addition, after the weight vector is set in the category setting unit 1, the external device B can be removed from the service interface 4, and can operate independently of the external device B.

  The abnormality monitoring apparatus A of the present embodiment does not have a function of learning the neural network 1 used in the category classification unit 1 and must directly set a weight vector for each neuron N2. Therefore, the generation of the weight vector is performed by the external apparatus B, and is transferred as setting data from the external apparatus B using the service interface 4.

  The external device B can be realized by executing an appropriate program on a general-purpose computer, but a dedicated device that realizes the functions described below can also be used. As shown in FIG. 5, the external device B includes a communication interface 14 and communicates between the external device B and the abnormality monitoring device A by connecting the communication interface 14 to the service interface 4 via the connection line C. Is possible.

  The external device B includes an analysis category classification unit 11 having the same configuration as the category classification unit 1 of the abnormality monitoring device A, and an analysis signal input unit 12 that takes in an electrical signal from the vibration sensor Y that detects the vibration of the device X. And an analysis feature extraction unit 13 that extracts the target signal from the electrical signal captured by the analysis signal input unit 12 and extracts a feature vector representing the feature of the target signal. The analysis feature extraction unit 13 can select a range in which the target signal is cut out from the electrical signal, and can select a type of feature vector representing the feature of the target signal.

  The range of the target signal and the type of feature vector in the analysis feature extraction unit 13 can be selected by the selection input unit 13a, and the content selected by the selection input unit 13a is given to the feature extraction unit 3 of the abnormality monitoring apparatus A. It becomes the contents of the setting data. However, the selection input unit 13a of the external device 3 can select a wide range so as to be compatible with various abnormality monitoring devices A. Normally, selection exceeding the range selectable by the feature extraction unit 3 of the abnormality monitoring device A is possible. It is possible. Therefore, the analysis feature extraction unit 13 of the external device B selects a range suitable for the abnormality monitoring device A by limiting the selection range according to the specifications of the feature extraction unit 3 of the abnormality monitoring device A. The limitation of the selection range is instructed by specifying the model number of the abnormality monitoring apparatus A in the selection input unit 13a.

  The feature vector extracted by the analysis feature extracting unit 13 is stored in the feature vector storage unit 15 with a known category. The category assigned to the feature vector stored in the feature vector storage unit 15 is given using the category input unit 15a. Here, the selection input unit 13a and the category input unit 15a are realized by sharing an input device such as a keyboard and a mouse.

  By the way, a feature vector corresponding to a specific abnormality of the device X (for example, abnormal noise of the motor due to contamination of foreign matter in the housing or lack of oil in the bearing of the motor) can be obtained relatively easily. In many cases, it is difficult to obtain a feature vector corresponding to an abnormality, and it is also difficult to set a category for an unexpected abnormality. Therefore, in this embodiment, a normal category is assigned only to a feature vector obtained when the operation of the device X is normal. In other words, a feature vector to which a normal category is not assigned belongs to an abnormal category. In this manner, the feature vector storage unit 15 stores feature vectors assigned with normal categories and feature vectors not assigned with categories.

  Similar to the category classification unit 1 of the abnormality monitoring device A, the analysis category classification unit 11 is configured using a competitive learning type neural network. In the abnormality monitoring device A, the weight vector is directly applied to the neuron N2 of the output layer 22. Although learning was not necessary because it was set, the external device B needs learning in order to set a weight vector for the neurons in the output layer of the analysis category classification unit 11. That is, the operation of the analysis category classification unit 11 has a learning mode and an inspection mode, and it is necessary to first set a weight vector in the learning mode.

  In the learning mode, only the feature vectors whose category is normal among the feature vectors stored in the feature vector storage unit 15 are input to the analysis category classification unit 11 as learning data. In a competitive learning type neural network having about 20 to 40 neurons in the output layer, by inputting an appropriate number (for example, 150) of feature vectors of the same category as learning data, a weight vector for classifying the category is obtained. Can be set for each neuron. Further, when a feature vector having a normal category is given, a normal category is set for a neuron whose Euclidean distance between the feature vector and the weight vector is equal to or less than a predetermined threshold.

  After the weight vector is set, the category classification unit 11 for analysis is set to the inspection mode, and the category of normal and abnormal can be classified by giving the feature vector. Here, in the present embodiment, feature vectors that are not normally classified by the analysis category classification unit 11 are classified abnormally. In actual classification, the Euclidean distance between the input feature vector and the weight vector set for the neuron in the output layer is obtained. If the category set for the neuron with the smallest Euclidean distance is normal, the feature vector is normal. Classify into: If the Euclidean distance is greater than or equal to the threshold specified for the weight vector of neurons belonging to the normal category, or the normal category is not set for the neuron with the minimum Euclidean distance, the input feature vector is Classify as abnormal.

  As the feature vectors input to the analysis category classification unit 11 in the inspection mode, those having a known category are used. That is, the feature vector stored in the feature vector storage unit 15 is used. The result classified in the analysis category classification unit 11 is presented to the user by a presentation means 16 such as a clustering map in which regions are associated with the neurons of the output layer of the analysis category classification unit 11. When a clustering map is used as the presentation means 16, the clustering map is displayed on the screen of the display device, the areas in the clustering map are associated with neurons, and when a normal category of neurons fires, the clustering map is normal. A display indicating the firing of neurons is performed in the area corresponding to.

  By the way, when the range of the target signal and the type of the feature vector are selected in the analysis feature extraction unit 13, if the input electrical signals are different, the difference in the classification result in the analysis category classification unit 11 is It cannot be distinguished whether it is caused by a difference in signal range or feature vector type or by an electric signal itself. Therefore, a signal storage unit 17 for storing an electrical signal is provided between the analysis signal input unit 12 and the analysis feature extraction unit 13.

  The analysis feature extraction unit 13 can extract the feature vector by changing the range of the target signal and the type of the feature vector for the same electrical signal by repeatedly using the electrical signal stored in the signal storage unit 17. become. That is, it becomes easy to compare the change of the classification result in the analysis category classification unit 11 due to the change of the range of the target signal and the type of the feature vector, and by confirming the result shown in the presentation means 16, A combination suitable for classification can be selected for a combination of a signal range and a feature vector type. However, this combination is selected within a selectable range of adjustment elements in the feature extraction unit 3 of the abnormality monitoring apparatus A.

  As described above, in the external device B, the analysis feature extraction unit 13 extracts the target signal from the electrical signal stored in the signal storage unit 17, extracts the feature vector of the target signal, and stores it in the feature vector storage unit 15. . Further, the analysis category classification unit 11 is set to the learning mode, the learning is performed on the analysis category classification unit 11 using the feature vector stored in the feature vector storage unit 15 as learning data, and then stored in the feature vector storage unit 15. The presenting means 16 is made to present the classification result in the analysis category classification unit 11 using the feature vector. When the range of the target signal and the type of the feature vector in the analysis feature extracting unit 11 are changed and the classification result presented to the presentation unit 16 is in a desired state, the range of the target signal and the feature vector at that time The combination with the type is used as setting data together with the weight vector.

  As described above, the external device B can communicate with the abnormality monitoring device A, and the feature extraction unit of the abnormality monitoring device A is sent to the abnormality monitoring device A by sending the setting data determined in the above procedure in the external device B. 3 defines the range of the target signal and the type of the feature vector, and sets the weight vector of the category classification unit 1. The data setting unit 5 performs processing for setting the setting data in the feature extraction unit 3 and the category classification unit 1. The external device B also has a function of reading the setting data set in the abnormality monitoring device A, and this function is used to adjust the setting data using the setting data already set in the abnormality monitoring device A. Or setting data can be set in another abnormality monitoring apparatus A.

(Embodiment 2)
In the first embodiment, the category classification unit 1 of the abnormality monitoring apparatus A does not have a learning function and receives a weight vector as setting data from the external apparatus B. However, in the present embodiment, the category of the abnormality monitoring apparatus A It is assumed that the classification unit 1 has a learning mode and an inspection mode in the same manner as the analysis category classification unit 11 of the external device B. Therefore, the external device B does not give the weight vector to the category classification unit 1 as setting data, but gives the feature vector used as learning data when the analysis category classification unit 11 is learned to the abnormality monitoring device A as setting data. It is like that.

  As shown in FIG. 6, the abnormality monitoring device A is provided with a learning data storage unit 6 so that the feature vector received from the external device B can be used as learning data, and the feature received from the external device B is used. The vector is stored in the learning data storage unit 6. The feature vector stored in the learning data storage unit 6 is the one used when a desired classification result is obtained in the external device B.

  In the configuration of the first embodiment, the abnormality monitoring device A can receive the setting data from the external device B, and the abnormality monitoring device A can be used if the data setting unit 5 sets the setting data in each unit. Then, the abnormality monitoring apparatus A cannot be used only by receiving the setting data from the external apparatus B. In the present embodiment, the feature vector stored in the learning data storage unit 6 among the setting data received by the abnormality monitoring apparatus A is used as learning data, and weighting is performed by learning with learning data using the category classification unit 1 as a learning mode. Set the vector.

  The present embodiment is different from the configuration of the first embodiment in that the category classification unit 1 has a learning mode and sets a weight vector in the learning mode. The category of the electrical signal can be classified by giving the feature vector of the electrical signal captured by the unit 2 to the category classification unit 1. Other configurations and operations are the same as those of the first embodiment.

1 is a block diagram illustrating a first embodiment. It is a schematic block diagram of the neural network used for the same as the above. It is a figure which shows the operation example of the feature extraction part used for the same as the above. It is a figure which shows the operation example of the feature extraction part used for the same as the above. It is a block diagram which shows the structural example of the external apparatus used for the same as the above. FIG. 6 is a block diagram illustrating a second embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Category classification | category part 2 Signal input part 3 Feature extraction part 4 Service interface 5 Data setting part 6 Learning data memory | storage part 11 Analysis category classification | category part 12 Analysis signal input part 13 Analysis feature extraction part 14 Communication interface 15 Feature vector storage part 16 Presentation means 17 Signal storage unit 21 Input layer 22 Output layer A Abnormality monitoring device B External device N1 neuron N2 neuron X device

Claims (1)

Of the electrical signal reflecting the operation of the inspection target, at least one element of the range extracted as the target signal and the type of the feature vector extracted from the target signal can be selected as the adjustment element, and the characteristics of the target signal captured by the signal input unit A feature extraction unit that extracts a feature vector representing a function, a category classification unit that classifies a category of the feature vector extracted by the feature extraction unit after setting a weight vector for each neuron using a neural network, and setting data for selecting an adjustment element a malfunction monitoring apparatus Ru integrally includes a service interface for exchanging with an external device, the external apparatus, an analysis signal an input unit for taking an electrical signal that reflects the behavior of the test object, analyzing signals A signal storage unit that stores electrical signals captured by the input unit, and an electrical signal stored in the signal storage unit A feature extraction unit for selecting a range to be extracted from the target signal and selecting and extracting a feature vector representing the feature of the target signal from a plurality of types, and a feature vector storage for storing the feature vector extracted by the analysis feature extraction unit A neural network in which learning is performed using feature vectors stored in the feature vector storage unit, and after learning, an analysis category classification unit that classifies the category of the feature vectors extracted by the analysis feature extraction unit, and an inspection A presentation means for presenting a classification result in the analysis category classification unit for the electrical signal captured by the analysis signal input unit in accordance with the operation of the target, and a range in which the target signal is extracted from the electrical signal in the analysis feature extraction unit; A combination that obtains a desired classification result among combinations of feature vectors extracted from the target signal The Align select a selectable range as regulatory elements of the feature extraction unit, which gives the combination selected in the feature extraction unit, and the external device is removed from the service interface after receiving the configuration data external device abnormality monitoring equipment, characterized in that operates independently.

Images