JP4760614B2 - Method for selecting learning data of signal identification device - Google Patents

Description

  The present invention relates to a signal identification device that classifies target signals including vibration components using a competitive learning type neural network, and selects a learning data suitable for learning of the competitive learning type neural network from the collected data. The present invention relates to a method for selecting learning data.

  Conventionally, a technique for classifying a target signal including a vibration component 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 a device or the vibration of a device is converted into an electrical signal by a sensor unit (transducer), and the output of the sensor unit is used as a target signal. Various techniques have been proposed for extracting feature quantities consisting of the above parameters and classifying the feature quantities using a neural network.

  Various configurations are known for neural networks. For example, it has been proposed to classify feature 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 quantity (input data) to be classified is given to a learned competitive learning 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, since the competitive learning type neural network does not use a teacher signal, the accuracy of category classification in the inspection mode is determined by the accuracy of the category of learning data input in the learning mode. That is, it is necessary to employ appropriate learning data to be used in the learning mode, but there is no appropriate standard for selecting learning data. Therefore, one neuron of the output layer in the competitive learning type neural network may fire with learning data of a plurality of categories. In other words, unless learning is performed using appropriate learning data, there arises a problem that neurons that cannot determine the category when fired are included in the output layer.

  The present invention has been made in view of the above-mentioned reasons, and its purpose is to improve the classification accuracy of the category of the signal identification device by optimizing the learning data used for learning of the competitive learning type neural network. It is to provide a method for selecting learning data.

  According to the first aspect of the present invention, a signal input unit that captures a target signal including a vibration component, a feature extraction unit that extracts a feature amount including a plurality of parameters for the target signal, and a feature amount extracted by the feature extraction unit is used as input data. It is expected to be classified into a single category, a competitive learning type neural network that classifies input data according to categories learned in advance using learning data, an output unit that outputs the classification result by the competitive learning type neural network, and A learning data selection method in a signal identification device including a learning data selection unit that selects data to be adopted as learning data from a data set composed of a plurality of data, wherein the learning set is a competitive learning type neural network using the data set To a competitive learning type neural network In the learning data selection unit, each of the data in the data set and the output layer of the competitive learning type neural network of the fired neuron are set in each neuron of the output layer. The magnitude of the difference vector with respect to the weight vector is obtained as the divergence degree for each data, and the divergence degree is specified when the average value and the variance value of the divergence degree are equal to or greater than the average threshold value and the variance threshold value, respectively. Data selection processing is performed to delete data that exceeds the divergence threshold from the data set. The average value of the divergence obtained from the data included in the data set is smaller than the average threshold, and the variance of the divergence is smaller than the dispersion threshold. The process of performing the data selection process until is repeated.

  According to this method, it is possible to evaluate the degree of attribution to the category expected for the data included in the data set, and it is possible to exclude data that may belong to other categories. In other words, data with a large deviation from the weight vector set for each neuron in the output layer of the competitive learning type neural network is deleted from the data set for use in learning. Data collected suddenly as noise when collecting data and data constituting the data set can be removed from the data set, and the accuracy of the category of data included in the data set used as learning data can be improved. it can. That is, it is possible to improve accuracy when classifying input data.

  According to the second aspect of the present invention, a signal input unit that captures a target signal including a vibration component, a feature extraction unit that extracts a feature amount including a plurality of parameters for the target signal, and a feature amount extracted by the feature extraction unit is used as input data. It is expected to be classified into a single category, a competitive learning type neural network that classifies input data according to categories learned in advance using learning data, an output unit that outputs the classification result by the competitive learning type neural network, and A learning data selection method in a signal identification device including a learning data selection unit that selects data to be adopted as learning data from a data set composed of a plurality of data, wherein the learning set is a competitive learning type neural network using the data set To a competitive learning type neural network After setting a weight vector for each neuron in the output layer as an element, a weight between each neuron in the output layer of the competitive learning neural network is set for each neuron in the learning data selection unit. Find either the minimum value or the sum of the vector Euclidean distances, and delete all data that fired the neuron from the data set when either the minimum value or the sum is greater than or equal to the specified threshold After performing category selection processing and category selection processing for all neurons in the output layer of the competitive learning type neural network, each data of the data set and the weight vector of the fired neuron in the output layer of the competitive learning type neural network For each data, the magnitude of the difference vector from Data selection processing for deleting from the data set the data whose divergence degree is equal to or greater than the prescribed divergence degree threshold when the average value and the variance value of the divergence degrees are equal to or greater than the prescribed average threshold value and equal to or greater than the dispersion threshold value. And the process of performing the data selection process is repeated until the average value of the divergence degree obtained from the data included in the data set is smaller than the average threshold value and the variance value of the divergence degree is smaller than the dispersion threshold value.

  According to this method, since the data selection process is performed in the same manner as in the first aspect of the invention, the same effect as that of the first aspect of the invention can be achieved, and the neurons fired by the data of different categories by performing the category selection process. By deleting the data that fired the neuron, it is possible to prevent data in a category different from the expected category from being used as learning data. In other words, it can be avoided that a neuron fired by data of different categories is mistaken as a neuron of an expected category.

  According to a third aspect of the present invention, a signal input unit that captures a target signal including a vibration component, a feature extraction unit that extracts a feature amount including a plurality of parameters for the target signal, and a feature amount extracted by the feature extraction unit is used as input data. It is expected to be classified into a single category, a competitive learning type neural network that classifies input data according to categories learned in advance using learning data, an output unit that outputs the classification result by the competitive learning type neural network, and A learning data selection method in a signal identification device including a learning data selection unit that selects data to be adopted as learning data from a data set composed of a plurality of data, the learning data selection unit including a data set Find the Euclidean distance between all data and set the data The average value and the variance value of the Euclidean distance obtained from all data are set as the first average value and the first variance value, and the remaining data obtained by excluding one piece of data of interest from the data set is obtained as a target. The average value and variance value of the Euclidean distance are defined as the second average value and the second variance value, and the difference between the first average value and the second average value is equal to or greater than a specified first threshold value, and the first The condition that the difference between the variance value and the second variance value is greater than or equal to the prescribed second threshold, the second average value is less than or equal to the first average value, and the second variance value is less than or equal to the first variance value. When satisfying, a data removal process of deleting one focused data from the data set is performed, and a process of performing the data removal process is repeated until there is no data to be deleted for all data included in the data set .

  According to the first and second aspects of the present invention, the competitive learning type neural network is trained with the data included in the data set. According to this method, the competitive learning type neural network is used using the data included in the data set. When selecting data to be used as learning data from among the data included in the data set, it is possible to evaluate the suitability as learning data using the data itself included in the data set. It is not necessary to operate a competitive neural network, and learning data can be selected in a short time with a small processing load. Similarly to the first and second aspects of the invention, it is possible to delete from the data set data that is different from the category that does not belong to the category expected from the learning data or data that is suddenly mixed as noise. The accuracy in classifying data can be increased.

  According to the method of the present invention, there is an advantage that data that can be attributed to other categories can be excluded by evaluating the degree of attribution to the expected category for the data included in the data set. As a result, the data set used as learning data can be removed from the data set, such as data that is different from the category that does not belong to the expected category or data that suddenly get mixed as noise when collecting the data constituting the data set. There is an advantage that the accuracy of the category of the data included in can be increased and the accuracy in classifying the input data can be increased.

  In the embodiment described below, an example in which the technique of the present invention is applied to an abnormality monitoring apparatus that determines whether an operation of a device is normal or abnormal based on a feature amount of a target signal obtained by the operation of the device will be described. That is, an abnormality monitoring device is exemplified as the signal identification device. Moreover, although the equipment apparatus provided with motive power sources, such as a motor, is assumed as an apparatus, the kind of apparatus in particular is not ask | required.

(Embodiment 1)
As shown in FIG. 1, the signal identification apparatus described in this embodiment uses an unsupervised competitive learning type neural network (hereinafter simply referred to as “neural network”) 1. As shown in FIG. 2, the neural network 1 includes two layers of an input layer 11 and an output layer 12, and each neuron N <b> 2 of the output layer 12 is coupled to all the neurons N <b> 1 of the input layer 11. Have. Although the neural network 1 is assumed to be realized by executing an appropriate application program on a sequential processing type computer, a dedicated neurocomputer can also be used.

  The operation of the neural network 1 has a learning mode and an inspection mode. After learning using appropriate learning data in the learning mode, a feature amount (input) generated from an actual target signal in the inspection mode. Data) category.

  The degree of connection (weighting coefficient) between the neuron N1 in the input layer 11 and the neuron N2 in the output layer 12 is variable. In the learning mode, the learning data is input to the neural network 1 to learn the neural network 1, and the input layer The weighting coefficients of the 11 neurons N1 and the neurons N2 of the output layer 12 are determined. In other words, each neuron N2 in the output layer 12 is associated with a weight vector whose element is a weight coefficient between each neuron N1 in the input layer 11. Therefore, the weight vector has the same number of elements as the neuron N1 of the input layer 11, and the number of parameters of the feature quantity input to the input layer 11 matches the number of elements of the weight vector.

  On the other hand, in the inspection mode, when the input data whose category is to be determined is given to the input layer 11 of the neural network 1, among the neurons N2 in the output layer 12, the neuron N2 having the smallest Euclidean distance between the weight vector and the input data is found. set a fire. If a category is associated with the neuron N2 of the output layer 12 in the learning mode, the category of the input data can be known from the category of the position of the fired neuron N2.

  The neuron N2 of the output layer 12 is associated with each region of the two-dimensional clustering map 5 having, for example, 6 × 6 regions on a one-to-one basis. Therefore, if the learning data category is associated with each area of the clustering map 5 in the learning mode, the category corresponding to the neuron N2 fired by the input data can be known from the clustering map 5. It functions as an output unit that outputs the classification result by the neural network 1.

  Categories are associated with each region of the clustering map 5 (substantially, each neuron N2 of the output layer 12) in the procedure described below. Further, in this embodiment, for simplicity of explanation, the categories to be classified by the neural network 1 are two types, normal and abnormal. In this embodiment, only the learning data of the normal category in the learning mode is used. input. The category of the learning data is reflected in the category of each neuron N2 of the output layer 12, and when a large number (for example, 150) of learning data is given, the category with high similarity is arranged at a close position on the clustering map 5. Is done. That is, the neuron N <b> 2 that is fired by learning data having similar feature quantities is arranged at a close position on the clustering map 5.

  Therefore, the neurons N2 fired corresponding to the learning data belonging to the same category among the neurons N2 in the output layer 12 gather at close positions on the clustering map 5 to form a cluster composed of the set of neurons N2. The learning data given to the neural network 1 in the learning mode is stored in the learning data storage unit 6 and is read from the learning data storage unit 6 and given to the neural network 1 as necessary.

  By the way, the target signal to be classified by the neural network 1 is an electric signal obtained from the device X, for example, a microphone 2a for detecting the operation sound of the device X, and a vibration sensor 2b for detecting vibration generated during the operation of the device X. The output of the sensor unit 2 consisting of at least one of the above is used. The configuration of the sensor unit 2 is appropriately selected according to the type of the device X, and various sensors such as a TV camera and an odor sensor can be used alone or in combination in addition to the microphone 2a and the vibration sensor 2b. Alternatively, a signal generated by the device X can be taken out and used as a target signal. Since the sensor unit 2 captures a target signal generated by the operation of the device X, it functions as a signal input unit.

  The target signal, which is an electrical signal obtained by the sensor unit 2, is given to the feature extraction unit 3, and the feature amount of the target signal is extracted. In the present embodiment, the target signal given from the sensor unit 2 to the feature extraction unit 3 is a signal including a vibration component, and has a plurality of parameters representing the vibration component of the target signal when input to the feature extraction unit 3. Feature quantities are extracted.

  The feature extraction unit 3 uses a timing signal (trigger signal) synchronized with the operation of the device X in order to extract a feature value from the target signal generated by the device X under the same conditions, or uses a waveform feature ( For example, the target signal is segmented by using a set of target signal start points and end points, and then divided into signals for each appropriate unit time, and feature quantities for each unit time To extract. Therefore, the feature extraction unit 3 includes a buffer that temporarily stores a target signal given from the sensor unit 2. Further, the feature extraction unit 3 performs preprocessing for reducing noise by limiting the frequency band as necessary. Furthermore, it also has a function of converting a target signal output from the sensor unit 2 into a digital signal.

  In order to simplify the explanation, it is assumed here that a plurality of frequency components (power for each frequency band) are extracted from the vibration components of the target signal after segmentation, and each frequency component is used as a parameter for the feature amount. explain. 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 component is used for the feature amount is appropriately selected according to the target device X and the abnormality to be extracted.

  The feature quantity obtained from the feature extraction unit 3 per unit time is given to the neural network 1 every time the feature quantity is extracted. The feature amount is also stored in the learning data storage unit 6 for use as learning data. The learning data storage unit 6 is a FIFO (first-in first-out), and holds, for example, 150 feature values as learning data at the maximum.

  Below, the learning data selection part 4 which is the characteristic of this invention is demonstrated. The learning data storage unit 6 stores a large number of learning data expected to be classified into normal categories. The learning data stored in the learning data storage unit 6 is obtained by operating the device X. Because it is collected, it is not always guaranteed whether the device X was operating normally at the time of learning data collection, and there is a possibility that environmental sound or sudden noise is mixed in the learning data There is also. Therefore, the category associated with each learning data is not always appropriate.

  Therefore, in the present embodiment, the learning data selection unit 4 is provided, the result when the data stored in the learning data storage unit 6 is given to the neural network 1 is evaluated, and the evaluation is stored in the learning data storage unit 6. By selecting data suitable for learning from the stored data, only appropriate learning data remains in the learning data storage unit 6.

  Here, a set of data stored in the learning data storage unit 6 is called a data set, and each data constituting the data set is expected to be classified into a specific category (that is, a normal category). It is assumed that That is, it is assumed that a category is associated with each data. Although it is assumed that the data set is the entire data stored in the learning data storage unit 6, a plurality of data sets may be stored in the learning data storage unit 6. Further, the number of data constituting the data set is arbitrary within a range that can be stored in the learning data storage unit 6.

  In the present embodiment, the learning data selection unit 4 deletes data that is inappropriate as learning data from the data set stored in the learning data storage unit 6 by performing the processing of the procedure shown in FIG. That is, first, the neural network 1 is set to the learning mode, and the neural network 1 is learned using all data included in the data set stored in the learning data storage unit 6 (S1). By this processing, a weight vector is set for each neuron N2 in the output layer 12 of the neural network 1.

Next, the learning data selection unit 4 obtains the degree of divergence between each data of the data set given to the neural network 1 and the firing neuron N2 among the neurons N2 of the output layer 12 of the neural network 1 for each data (S2). ). That is, the degree of divergence is a value that reflects the magnitude of a difference vector between input data (vector) given to the neural network 1 and a weight vector associated with the neuron N2 fired by the input data. Is [X], and the weight vector of the fired neuron N2 is [Wwin] ([a] means that a is a vector), the divergence degree Y is defined by the following equation.
Y = ([X] / X- [Wwin] / Wwin) ^T ([X] / X- [Wwin] / Wwin)
Here, T represents transposition, and X and Wwin without a square bracket indicate the norm of each vector. The divergence degree Y is normalized by dividing the elements of each vector by the norm.

  As described above, since the divergence degree is obtained for each data, the divergence degrees corresponding to the number of data are obtained. For the obtained degree of divergence, an average value and a variance value are obtained (S3), compared with an average threshold value defining the average value, and compared with a variance threshold value defining the variance value (S4). The average threshold value and the dispersion threshold value are appropriately set by the user. Here, if the average value of the divergence degree is smaller than the average threshold value and the variance value of the divergence degree is smaller than the dispersion threshold value, it is determined that the category is properly set, and the learning mode is ended (S5).

  On the other hand, when the average value of the divergence degree is equal to or greater than the average threshold value or the variance value is equal to or greater than the variance threshold value, it is determined that data whose category is not appropriate is included in the learning data, and each data included in the data set is set to 1 Individually selected (S6), and the divergence obtained for the selected data is compared with a prescribed divergence threshold (S7). When the divergence degree is equal to or less than the divergence degree threshold, the data is adopted as learning data, and when the divergence degree is equal to or greater than the divergence degree threshold, the data is deleted from the data set (S8). The processes in steps S6 to S9 are performed for all data included in the data set (S9).

  The above-described processing of steps S1 to S9 is called data selection processing. In the data selection processing, the average value of the divergence obtained from the data included in the data set is smaller than the average threshold, and the variance of the divergence is the dispersion threshold. Repeat until smaller. That is, it repeats until it branches to completion | finish (S5) of learning mode in step S4.

  By performing the data selection process described above, data deviating from the normal category data group can be excluded from the data set. Further, since the neural network 1 is learned by data that does not belong to the normal category, or data that has been removed from environmental sound or noise, the neural network 1 is learned with appropriate data at the end of the learning mode. Will be. As a result, it is possible to classify input data with high accuracy in the inspection mode.

  In the above-described example, the category is only normal, but other categories may be used, and when learning about a plurality of categories, the data of each data set is associated with one type of category. That's fine. Therefore, if a plurality of data sets associated with one type of category are used and the above learning is performed for each data set, the neural network 1 can be learned for a plurality of categories.

(Embodiment 2)
In the first embodiment, only the relationship between the input data at the time of learning and the neuron N2 fired in the neural network 1 is evaluated. In the present embodiment, the distance between neurons N2 having similar categories in the output layer 12 of the neural network 1 is evaluated. Note that the relationship between the neurons N2 in the output layer 12 is also evaluated.

  The neural network 1 is learned using the data set stored in the learning data storage unit 6 as in the first embodiment. As shown in FIG. 4, after learning of the neural network 1 (S1), the learning data selection unit 4 targets all the neurons N2 in the output layer 12 of the neural network 1 and performs Euclidean weight vectors with other neurons N2. The distance is obtained (S2). Here, in actual calculation, the square of the difference between the weight vectors of the two neurons N2 is substituted for the Euclidean distance (that is, the square of the Euclidean distance is used to ignore the sign).

Next, for each neuron N2, one of the minimum value and the sum of the obtained Euclidean distances (the square of the Euclidean distance) is obtained (S3). For each neuron N2, the neuron N2 having the smallest square of the Euclidean distance means the neuron N2 having the highest degree of category similarity, and the fact that the sum of the squares of the Euclidean distance is minimized is the category similarity. Therefore, the neurons N2 are concentrated and exist without being dispersed. Therefore, for each neuron N2, the minimum value or the sum obtained in step S3 is compared with a threshold value (S5), and the threshold value is exceeded. If it is, it is regarded as a neuron N2 whose category is not properly set, and all the data that fired the neuron N2 are deleted from the data set (S6). The process of steps S5 and S6 is called a category selection process, and the category selection process is performed for all the neurons N2 in the output layer 12 of the neural network 1 (S4, S7). When data is deleted from the data set in step S6, learning in step S1 is performed again using the data set composed of the remaining data (S8).

  If the data is not deleted from the data set in step S6, the same data selection process as that of the first embodiment is performed (S9 to S17), and the data whose category is deviated is removed from the data set.

  If the method of this embodiment is adopted, the data of the data set is evaluated based on the proximity of the category of the neuron N2 in the output layer 12 of the neural network 1, so that the neurons N2 belonging to the normal category are not dispersed and clustering is performed. In the map 13, it is possible to concentrate the existence positions of the neurons N2 belonging to the normal category. Furthermore, since the category selection process is a pre-process, and after the ambiguous data of the category is removed from the data set, the data selection process similar to that of the first embodiment is performed, the environmental noise or noise that does not belong to the normal category is generated. Data is also removed, and learning with more appropriate data than in the first embodiment becomes possible. Similar to the first embodiment, the category is not limited to normal, but may be other categories, and a plurality of categories can be learned. Other configurations and operations are the same as those of the first embodiment.

(Embodiment 3)
In the first embodiment and the second embodiment described above, the neural network 1 is learned and the weight vector is associated with the neuron N2 of the output layer 12, and then the category is evaluated. A technique for optimizing data used for learning by evaluating the data set itself stored in the storage unit 6 will be exemplified.

  In the present embodiment, as shown in FIG. 5, the learning data selection unit 4 first obtains the Euclidean distance between all the data included in the data set stored in the learning data storage unit 6 (S1). The Euclidean distance actually uses the square of the difference between the data in order to ignore the sign.

  Next, an average value and a variance value of the Euclidean distance (the square of the Euclidean distance) are obtained for all data in the data set, and set as a first average value and a first variance value, respectively (S2). Furthermore, one piece of data of interest is excluded from the data set (S3), and the average value and the variance value of the Euclidean distance (the square of the Euclidean distance) are obtained from the remaining data, and the second average value and The second variance value is set (S4). If the excluded data belongs to the same category, the difference between the first average value and the second average value is small, and the difference between the first variance value and the second variance value is also small. Can be estimated.

  Therefore, a difference between the first average value and the second average value and a difference between the first variance value and the second variance value are obtained (S5), and one of the differences is defined for each difference. If it is above the first and second threshold values, it can be estimated that the excluded data is not appropriate (S6). However, even if this condition is satisfied, there is still a possibility that the data is appropriate data. Therefore, the second average value is not more than the first average value (S7), and the second variance value is the first value. When the condition that the variance value is less than (S8) is satisfied, the excluded data is determined to be data that does not belong to the normal category, and is deleted from the data set (S9). If any one of steps S6 to S8 does not satisfy the condition, the excluded data is not deleted and is regarded as data suitable for learning. The processing of steps S3 to S9 is called data removal processing.

  When the data is deleted in step S9, the second average value and the second variance value are read as the first average value and the first variance value (S10), and the process returns to step S3 to store other data. The data removal process is repeated for all data (S11). Further, when data is deleted in step S9, the process returns to step S1 to repeat the above process, and the data removal process is repeated until there is no data to be deleted (S12).

  The data set stored in the learning data storage unit 6 is evaluated by the above processing, and if there is no data to be deleted, the data of the data set is regarded as appropriate and used for learning of the neural network 1 (S13).

  In this embodiment, the adequacy of the data included in the data set as learning data is evaluated before learning of the neural network 1, so that the neural network 1 is changed as in the method described in the first and second embodiments. It is not necessary to operate the learning mode many times, and the processing load can be reduced. Other configurations and operations of the present embodiment are the same as those of the first and second embodiments, and the methods of the first and second embodiments and the method of the present embodiment can be used in combination.

It is a block diagram common to each embodiment of this invention. It is a schematic block diagram of the neural network used for the same as the above. FIG. 3 is an operation explanatory diagram of the first embodiment. FIG. 9 is an operation explanatory diagram of the second embodiment. FIG. 9 is an operation explanatory diagram of Embodiment 3.

Explanation of symbols

1 Neural network (competitive learning type neural network)
2 Signal input unit 2a Microphone 2b Vibration sensor 3 Feature extraction unit 4 Learning data selection unit 5 Clustering map (output unit)
6 Learning Data Storage Unit 11 Input Layer 12 Output Layer N1 Neuron N2 Neuron X Device

Claims (3)

  A signal input unit that captures a target signal including a vibration component, a feature extraction unit that extracts a feature amount including a plurality of parameters for the target signal, and learning in advance using learning data using the feature amount extracted by the feature extraction unit as input data Data consisting of a competitive learning type neural network that classifies input data according to the selected category, an output unit that outputs the classification result of the competitive learning type neural network, and a plurality of data expected to be classified into a single category A learning data selection method in a signal identification device including a learning data selection unit that selects data to be adopted as learning data from a set, wherein competition is performed by learning a competitive learning type neural network using the data set Each level of the output layer in the learning type neural network After setting a weight vector whose element is a set of weight coefficients for each of the alons, in the learning data selection unit, the difference between each data of the data set and the weight vector of the firing neuron in the output layer of the competitive learning type neural network Obtain the magnitude of the vector as the divergence degree for each data, and the divergence degree is greater than or equal to the specified divergence degree threshold when the average value and the variance value of the divergence degree are equal to or greater than the specified average threshold value and the variance threshold value, respectively. The data selection process is performed until the average value of the divergence obtained from the data included in the data set is smaller than the average threshold and the variance of the divergence is smaller than the dispersion threshold. A learning data selection method for a signal identification device, characterized in that the processing to be performed is repeated.   A signal input unit that captures a target signal including a vibration component, a feature extraction unit that extracts a feature amount including a plurality of parameters for the target signal, and learning in advance using learning data using the feature amount extracted by the feature extraction unit as input data Data consisting of a competitive learning type neural network that classifies input data according to the selected category, an output unit that outputs the classification result of the competitive learning type neural network, and a plurality of data expected to be classified into a single category A learning data selection method in a signal identification device including a learning data selection unit that selects data to be adopted as learning data from a set, wherein competition is performed by learning a competitive learning type neural network using the data set Each level of the output layer in the learning type neural network After setting a weight vector whose elements are each a pair of weighting factors, the Euclidean distance of the weight vector between each neuron in the output layer of the competitive learning type neural network and other neurons in the learning data selection unit Performs a category selection process that calculates either the minimum value or the sum, and deletes all data that fired that neuron from the data set when either the minimum value or the sum is greater than the specified threshold. After performing category selection processing for all neurons in the output layer of the competitive learning type neural network, the difference vector between each data of the data set and the weight vector of the firing neuron in the output layer of the competitive learning type neural network Obtain the magnitude as the degree of divergence for each piece of data, Data selection processing is performed to delete from the data set data whose divergence level is equal to or higher than the specified divergence degree threshold when the value and the variance value are equal to or greater than the specified average threshold value and the variance threshold value. Selection of learning data of a signal identification device characterized by repeating a data selection process until an average value of divergence obtained from contained data is smaller than an average threshold and a variance of divergence is smaller than a dispersion threshold Method.   A signal input unit that captures a target signal including a vibration component, a feature extraction unit that extracts a feature amount including a plurality of parameters for the target signal, and learning in advance using learning data using the feature amount extracted by the feature extraction unit as input data Data consisting of a competitive learning type neural network that classifies input data according to the selected category, an output unit that outputs the classification result of the competitive learning type neural network, and a plurality of data expected to be classified into a single category A learning data selection method in a signal identification device including a learning data selection unit that selects data to be adopted as learning data from a set, and in the learning data selection unit, all data included in the data set Find the Euclidean distance and target all data in the dataset The average value and variance value of the Euclidean distance obtained in this way are used as the first average value and the first variance value, and the average of the Euclidean distance obtained from the remaining data excluding one data item of interest from the data set. The value and the variance value are defined as the second average value and the second variance value, and the difference between the first average value and the second average value is equal to or greater than a prescribed first threshold value, and the first variance value and the second variance value When the condition that the difference from the variance value is equal to or greater than the prescribed second threshold, the second average value is equal to or less than the first average value, and the second variance value is equal to or less than the first variance value is noted Learning of a signal identification device characterized by performing data removal processing for deleting one piece of data from a data set, and repeating data removal processing until there is no data to be deleted for all data included in the data set Data択方 method.

Images