JP2019113914A - Data identification device and data identification method - Google Patents

Abstract

To accurately detect only the abnormal state by using an unsupervised abnormality detection technique even with a process in which the normal pattern is to change.SOLUTION: A data identification device which identifies whether or not input data are singular data comprises: a feature quantity extracting unit 11 which extracts a feature quantity by using a first neural network from the input data; a data reconstruction unit 12 which generates reconstruction data by using a second neural network from the feature quantity; a parameter determination unit 14 which determines parameters of the first neural network and the second neural network by an error inverse propagation method on the basis of an error between the reconstruction data and the input data; a data evaluation unit 13 which finds an evaluation value of the error between the reconstruction data and the input data; a determination unit 15 which determines whether or not the input data are the singular data by comparing the evaluation value with a preset threshold value; and a control unit 17 which effects the operation of two modes of a learning mode and a data identification mode.SELECTED DRAWING: Figure 2

Description

   The present invention relates to an abnormality detection apparatus, an abnormality detection method, and an abnormality detection program for identifying whether data is normal data or abnormal data with respect to data constantly acquired from a production process or data acquired in the past. .

  There is known a machine learning method for identifying whether or not data included in a certain data group is normal data, and is roughly divided into two: supervised abnormality detection method and unsupervised abnormality detection method. As a supervised abnormality detection method, a method as described in Patent Document 1 is known. However, in order to obtain sufficient discrimination accuracy in supervised abnormality detection, it is necessary to prepare normal data and abnormal data in the same degree in advance.

  However, when most of the products obtained from the production facilities of the product are non-defective products and the incidence rate of defective products is low, most of the data obtained from the production facilities are normal data obtained from non-defective products and anomalies obtained from defective products There may be less data. In such a case, when the supervised abnormality detection method is used, since the abnormal data can not be obtained sufficiently, the characteristics of the abnormal data can not be sufficiently captured, and the identification accuracy with respect to the abnormal data is lowered. There is a problem.

  The unsupervised abnormality detection technique is an abnormality detection method using normal data as learning data, and therefore has the advantage that abnormal data can be identified as a state different from normal data without using abnormal data for learning. Patent Document 2 proposes an anomaly detection method using an unsupervised anomaly detection technique.

JP 2014-26455 A JP, 2017-97718, A

However, when using the above-mentioned unsupervised abnormality detection technology in the production process of a product etc., data acquired by changes over time of environmental conditions change, data acquired by updating of production equipment, peripheral equipment, etc. The unsupervised anomaly detection technology alone can not cope with this change, and an accurate judgment may not be made.

  The present invention has been made in view of the above-mentioned point, and an object of the present invention is to perform abnormality detection accurately using an unsupervised abnormality detection method even under a condition where environmental conditions change with time.

In order to solve the above problems, the present invention is a data identification device for identifying whether or not input data is unique data in a data group,
Feature quantity extraction means for extracting a feature quantity from the input data using a first neural network;
Data reconstruction means for generating reconstruction data of the same size as the input data from the feature quantities using a second neural network;
Parameter determination means for determining a parameter of the first neural network and a parameter of the second neural network by an error back propagation method from an error between the reconstructed data and the input data;
Data evaluation means for evaluating the magnitude of an error between the reconstructed data and the input data;
An identification unit that compares the evaluation value calculated by the evaluation unit with a preset threshold value to identify whether the input data is unique data;
The feature quantity extraction means, the learning mode for operating the data reconstruction means and the parameter determination means, and the data identification mode for operating the feature quantity extraction means, the data reconstruction means, the data evaluation means and the identification means And control means for operating the two modes independently or simultaneously.

In the above data identification apparatus of the present invention, the data group is derived from an identification object which is continuously transported,
The data identification apparatus further comprises: data acquisition means for acquiring data on the identification target; and data storage means for accumulating data acquired by the data acquisition means.
When the identification object is in the transport state, the control means inputs the data acquired by the data acquisition means to the feature amount extraction means to operate the data identification mode, and the identification object is not in the transport state It is preferable that the learning mode is operated by inputting data accumulated by the data accumulation unit to the feature quantity extraction unit when the identification target is in the transport state.

  In the data identification device of the present invention, preferably, the identification target is a conveyed product, a resin, or a yarn conveyed by a conveyance belt.

Further, in order to solve the above-mentioned problems, according to the present invention, by causing the first neural network and the second neural network to learn data acquired and accumulated for a certain period, the data currently being acquired is among the data groups. Data identification method for identifying whether or not the data is unique data, and the following steps 1 and 2 are performed.
(Step 1)
Feature 1 is extracted from the data acquired and accumulated for a fixed period using a first neural network,
Generating reconstructed data of the same size as the accumulated data using the feature 1 to the second neural network;
Parameters of the first neural network and parameters of the second neural network are determined by an error back propagation method from an error between the reconstructed data and the accumulated data.
(Step 2)
Feature amount 2 is extracted from the data being acquired using the first neural network in which the parameter determined in the step 1 is set;
Generating reconstructed data of the same size as the data being acquired from the feature amount 2 using the second neural network in which the parameter determined in the step 1 is set;
From the magnitude of the error between the reconstructed data and the data being acquired, it is determined whether the data being acquired is unique data.

The data identification method of the present invention is preferably performed by any of the following methods.
-A method in which the step 1 is performed only once at first, and then the step 2 is continuously performed.
-The data acquired and accumulated for the fixed period used in the step 1 is a method of repeating the step 2 continuously and the step 1 and the step 1 of the data used for the fixed period. Data obtained in the process of step 2 performed continuously with step 1 performed immediately before step 1), and set in the first neural network and the second neural network in the step 2. Are the parameters determined by Step 1 performed immediately before Step 2.
· Data obtained by repeating the step 1 in a predetermined cycle and continuously performing the step 2 while the step 1 is not performed, the data acquired and accumulated for the predetermined period used in the step 1 Is the data obtained in the process of step 2 performed continuously between the step 1 and the step 1 performed immediately before the step 1, and in the step 2, the first neural network and The parameters set in the second neural network are the parameters determined by the process 1 performed immediately before the process 2.

  According to the present invention, it is determined whether the data acquired from the object is normal or abnormal using the unsupervised abnormality detection method, and the acquired data changes due to the time-lapse change of the environmental condition, etc., or the production apparatus Even in a system that affects data acquired by updating of a peripheral device or the like, by providing a function of relearning at an appropriate cycle, it is possible to carry out determination corresponding to a change.

FIG. 1 is a diagram showing a data identification apparatus of the present invention. FIG. 2 is a schematic block diagram of the data identification apparatus of the present invention. FIG. 3 is a diagram showing an example of a first neural network used for the feature quantity extraction unit 11. As shown in FIG. FIG. 4 is a view showing an example of a second neural network used for the data reconstruction unit 12. FIG. 5 is a view showing an example of a neural network in which the first neural network used in the feature amount extraction unit 11 and the second neural network used in the data reconstruction unit 12 are combined. FIG. 6 is an operation flow diagram when the data identification apparatus of the present invention operates in the learning mode. FIG. 7 is an operation flow diagram when the data identification device of the present invention operates in the identification mode. FIG. 8 is a diagram showing one implementation method of the present invention. FIG. 9 is identification data of each time acquired in one of the implementation methods of the present invention. FIG. 10 is a diagram showing the time change of the evaluation value output when the data identification method A is used in one of the implementation methods of the present invention. FIG. 11 is a diagram showing the time change of the evaluation value output when the data identification method B is used in one of the implementation methods of the present invention. FIG. 12 is a diagram showing the time change of the evaluation value output when the data identification method C is used in one of the implementation methods of the present invention.

  Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings.

<Device configuration>
FIG. 1 shows an example of an embodiment of a data identification apparatus according to the present invention. The data identification apparatus includes an arithmetic unit 1, a data acquisition unit 2, and a display unit 3. The computing device 1 is a PC (Personal Computer) composed of a memory composed of a central processing unit (CPU) and a random access memory (RAM), and a recording device such as a hard disk drive (HDD) or a solid state drive (SSD). Etc. The program or data is called from the recording device to the memory, and the CPU carries out the operation. The recording apparatus is not limited to the illustrated configuration, and an HDD may be connected to the outside of the computing apparatus 1, or CD, DVD, Blu-ray (registered trademark), or the like may be used. The data acquisition device 2 is a sensor represented by a camera or a microphone, and is electrically connected to the arithmetic device 1. The display device 3 is a display or the like, is electrically connected to the arithmetic device 1, and can be configured to display the determination result of the identification target.

  FIG. 2 specifically shows the outline of the internal configuration of the data identification apparatus of FIG. The arithmetic device 1 includes a data acquisition unit 10, a feature extraction unit 11, a data reconstruction unit 12, a data evaluation unit 13, a parameter determination unit 14, a determination unit 15, a result output unit 16, and a control unit 17. . Data acquisition unit 10, feature amount extraction unit 11, data reconstruction unit 12 data evaluation unit 13, parameter determination unit 14, determination unit 15, result output unit 16, and control unit 17 are either one stored in operation device 1 or It is a plurality of programs or arithmetic circuits.

<Functional configuration>
The data acquisition unit 10 has a function of acquiring identification target data through the data acquisition device 2 and temporarily recording the data in a memory or the like. Further, it is preferable that the data acquisition unit 10 have a function of storing identification target data in a recording device inside or outside the arithmetic device.

  The feature quantity extraction unit 11 extracts a low-dimensional feature quantity indicating identification target data using the first neural network. The data acquired by the sensor to be identified can be treated as a multi-dimensional vector, and the values of neurons in the first neural network are interpreted as one vector. Specifically, a neural network as shown in FIG. 3 can be considered. As a restriction to the first neural network used for the feature amount extraction unit 11, the number of neurons in the input layer needs to be matched with the dimension n of the identification target data. Furthermore, it is necessary to make the number m of feature quantities to be extracted smaller than n. There is no restriction on the number of layers of the first neural network or the number of neurons used for the middle layer, but it is desirable to increase the number according to the number of dimensions of identification target data. In addition, when the identification target data is an image or a moving image, the correlation between adjacent neurons of the acquired data is important, so use of a convolution layer in which the connection between neurons is restricted. This makes it possible to extract more important features.

  The data reconstruction unit 12 reconstructs n-dimensional data identical to the identification target data from the feature amounts extracted from the feature amount extraction unit 11 using the second neural network. Specifically, a neural network configured as shown in FIG. 4 is used. A limitation on the second neural network used in the data reconstruction unit 12 is that the number of neurons in the output layer is the same as the identification target data. The number of layers in the second neural network and the number of neurons used in the middle layer are not limited, but when the data to be identified is an image or a moving image, the correlation between adjacent neurons in the acquired data is important. By using a deconvolution layer in which connections between neurons are restricted, it is possible to reconstruct data with higher accuracy. The feature amount extraction unit 11 and the data reconstruction unit 12 can also be used as one neural network as shown in FIG.

  The data evaluation unit 13 evaluates the difference between the identification target data and the reconstruction data obtained by the data reconstruction unit 12. As a method of evaluating the difference between the two data, there are a SSD (Sum of Squared Difference), a SAD (Sum of Absolute Difference), and the like. Each index is an index that is 0 when the identification target data matches the reconstructed data, and indicates that the larger the numerical value, the larger the difference between the data.

  The parameter determination unit 14 determines the parameters of the neural network used for the feature amount extraction unit 11 and the data reconstruction unit 12 based on the result of the data evaluation unit 13. Here, the parameters of the neural network are weights and biases. The weight is a numerical value indicating the coupling strength between each neuron of the neural network, and the bias is a constant independent of the value of the neuron to be added when transmitting information between the neurons.

  From the error between the identification target data calculated by the data evaluation unit 13 and the reconstruction data, the parameter is corrected by the error back propagation method. In this case, identification target data is acquired, and the same processing is repeated using a plurality of accumulated data. When learning with a neural network, it is known that using the mini-batch stochastic gradient descent method is effective, and a plurality of data to be identified are combined to create a mini-batch to calculate an error at the same time, It is effective to perform parameter correction using those total values. In the case of performing error back propagation learning, it is preferable to determine final parameters while gradually repeating parameter correction.

  The determination unit 15 compares the value of the error calculated by the error data evaluation unit 13 with a preset threshold value, and determines whether the identification target data is normal or abnormal. Data having a large error is data that has a large difference between the original identification target data and the reconstruction data, and therefore, it is determined that the data has not sufficiently learned in the past, that is, abnormal data. Data with a small error is judged to be sufficiently learned data. The threshold value is a value obtained by adding a constant multiple of the standard deviation to the mean value or median value of normal data. When it is known that there is abnormal data in the identification target data or the accumulated data, the minimum of the error calculated by the data evaluation unit 13 from the abnormal data using the feature amount extraction unit 11 and the data reconstruction unit 12 You can set the value, etc.

  The result output unit 16 notifies or displays the result of the determination unit 15. When the determination unit 15 determines that the data is abnormal, the result output unit 16 notifies the outside by displaying identification target data on the screen, outputting an abnormal signal from the arithmetic device, or the like.

  The control unit 17 controls the above-described units according to the state to operate the data identification device of the present invention. Control is performed in two control modes, a learning mode (step 1) and a discrimination mode (step 2). Control the processing procedure or operation order of the data acquisition unit 10, feature amount extraction unit 11, data reconstruction unit 12, data evaluation unit 13, parameter determination unit 14, determination unit 15, result output unit 16 according to each control mode Do.

<Operation of Data Identification Device>
The operation flow when the data identification apparatus of the present invention operates in the learning mode (step 1) is shown in FIG. 6, and the operation flow when operating in the identification mode (step 2) is shown in FIG.

  In order to identify data using the data identification device of the present invention, first, after operating in the learning mode, it operates in the identification mode.

The operation flow in the case of operating in the learning mode (step 1) will be described using FIG. Before entering the processing step in the learning mode, first, in step S10, accumulated data is divided into predetermined value units (2 to 64). The divided units are called mini-batches. Let the number of mini-batches be N- _b , and let the default value for the number of mini-batches. The parameter modification of the first neural network it is necessary to repeated, setting the prescribed value N _e in advance repeat count. First, before the value of the repeat count counter n _e showing a mini-batch counter n _b and the current repetition count indicating the mini-batch number currently being processed is initialized, starts step S11 follows, adds 1 to the values Do. Next, in step S11, the identification target data group is acquired from the data acquisition unit 10. In step S12, the feature amount extraction unit 11 is used to extract a feature amount 1 group representing the identification target data group. In step S13, the second neural network of the data reconstruction unit 12 generates a reconstruction data group from the obtained feature quantity 1 group. Next, in step S14, the data evaluation unit 13 calculates an error between the identification target data and the reconstruction data. In step S15, when the errors of all the identification target data included in the mini-batch have been calculated, the first neural network of the feature extraction unit 11 and the data recalculation are performed by the error back propagation method from the error whose average value is calculated. The parameters of the second neural network of the configuration unit 12 are corrected to determine the parameters.
Repeat until the series of processes mini-batch number counter n _b being processed is the mini-batch number specified value N _b, it repeats the process of step S11~ step S15 for all mini-batch. When the number of mini-batches exceeds the specified value, it is judged whether the value of the repeat counter n _e is more than the specified value for the repeat count, and if N _e is less than the repeat count, 1 is added to the value of n _e After the value of N _b is initialized, the processing of steps S11 to S15 is repeated again. This process is repeated and the process ends when the value of n _e becomes N _e or more. Hereinafter, this series of flows will be referred to as learning.

  The control unit extracts important parameters representing the identification target data from the identification target data by learning the neural network by using the input layer as the identification target data and the target output as the identification target data, and is close to the identification target data It will be possible to obtain reconstruction data. When reconstructed data close to the identification target data can be obtained, the value of the error calculated for the learned data decreases.

  Next, data identification is performed by operating in the identification mode (step 2). An operation flow in the case of operating in the identification mode will be described with reference to FIG. First, in step S20, the data acquisition unit 10 acquires data to be identification target data. Next, in step S21, the feature amount 2 is extracted from the identification target data acquired in step S20 by the first neural network of the feature amount extraction unit 11. In step S22, the second neural network of the data reconstruction unit 12 is extracted. Create reconstruction data according to. In step S23, the difference between the reconstructed data created by the data evaluation unit 13 and the identification target data is evaluated. In step S24, the determination unit 15 compares the error calculated in step S23 with the threshold value to perform determination. In step S25, the result output unit 16 outputs the determination result obtained in step S24. This series of flows is called identification. Data acquired in step S20 may be real-time data acquired by a camera, a microphone or the like, or data acquired in advance may be read, or identification processing can be performed on them.

<Data identification method>
The data identification method performed by the data identification device of the present invention can be performed by the following three methods.

・ Data identification method A
The first data identification method is operated in a learning mode using learning data acquired in advance, and the parameter determination unit 14 determines parameters of the feature quantity extraction unit 11 and the data reconstruction unit 12. Set (Step 1). Thereafter, it is to operate in the identification mode using the determined parameters (step 2). At this time, identification target data identified by the identification mode may be subjected to data identification with respect to data sequentially acquired by the data acquisition unit 10 or may be acquired in advance and stored inside or outside the arithmetic device 1 Data identification may be performed by acquiring the identification target data group being processed.

  This data identification method is suitable for data whose identification target data is always acquired under the same condition or data acquired in the past.

・ Data identification method B
The second data identification method operates the learning mode and the identification mode in parallel. At this time, the evaluation result calculated by the data evaluation unit 13 is transmitted to the parameter determination unit 14 and the determination unit 15, and the determination and the parameter correction can be performed simultaneously. The corrected parameters can be updated before carrying out a series of flows from the next data acquisition, but after learning a certain amount of data to be identified, the feature quantity extraction unit 11 and the data reconstruction unit 12 Update the parameters used. This data identification method is effective for a system in which a production process is always in operation. In this data identification method, since the learning mode and the identification mode are operated in parallel, it is possible to cope with the state change more favorably than the data identification method A. When the learning mode and the identification mode operate in parallel, the computational load on the arithmetic device increases, and when affecting the operation in the identification mode, two arithmetic devices are prepared, and in each, the learning mode and the identification mode are paralleled. Operation is also possible. In this case, since the operation in the learning mode does not affect the identification mode, the operation stability in the identification mode can be secured.

・ Data identification method C
In the third data identification method, the control unit 17 operates in the identification mode at the time of operation of the production process, particularly when the production process repeatedly operates and stops when the identification target data is acquired from the production process. This is a method of recording identification target data in a recording device inside or outside the arithmetic device 1 and operating the learning mode when the production process is stopped. When using the parameter obtained in the learning mode for the identification mode, it is desirable to prepare test data, confirm in advance the evaluation value obtained, and after that, update the parameters of the neural network used in the identification mode after judgment. This data identification method is preferable because the computing device 1 can be efficiently utilized when the production process repeats stopping and operation. In this data identification method, identification target data acquired when operating in the identification mode is stored, accumulated data is created, and when operating in the learning mode, the accumulated data is used to carry out learning, It is possible to reflect the latest production status at the next operation. As a result, even if temporal or sudden changes occur in the production process, the change state can be reflected in the subsequent production.

  Hereinafter, the present invention will be specifically described by way of examples. However, the present invention is not limited to this.

Example 1
In this embodiment, the data identification method A is used. In the present embodiment, with the configuration shown in FIG. 8, the camera corresponding to the data acquisition device 2 acquires an image of the transported product, and the product appearance inspection is performed using the data identification program stored in the PC as the arithmetic device 1. The result is displayed on the display 3 as a display. Here, using the identification device learned using accumulated data up to time T1, operation is performed in the identification mode with respect to the identification target data acquired from the product to be continuously produced from time T1 to time T6 The product to be transported was identified as normal or abnormal.

  At this time, a slight decrease in the amount of light occurred due to the deterioration of the illumination. The captured image changes as shown in FIG. 9 at the time from T1 to T6 due to the change of the imaging condition caused by the decrease of the light amount, and the time change of the evaluation value becomes as shown in FIG. Moreover, before the time T1, the deterioration of the illumination did not occur. Since the present identification device learns the image before time T1, even if the product is normal, the value of the evaluation value calculated by evaluation unit 13 from the change in the imaging condition occurring in the captured image is time T1 in FIG. It will gradually rise from time T3 to time T3. As shown in the figure, when the condition change for acquiring the identification target data continues, the difference between the accumulated data used for learning and the acquired identification target data becomes large, and the evaluation value continues to increase. For this reason, although the evaluation value has become equal to or higher than the threshold after time T5, since the judgment was normal until the aging T5, the change in the imaging condition is moderate or the time for continuous production is short In some cases, the data identification method A can sufficiently identify whether the product is normal or abnormal.

Example 2
In this embodiment, the data identification method B is used. Here, a product is continuously produced at times T1 to T6 using the configuration of FIG. 8, an image of the product to be transported is acquired, and an appearance inspection of the product is performed. In this embodiment, PC1 and PC2 are used as the arithmetic unit 1. As in the first embodiment, the PC 1 operates on the identification target data acquired from the camera in the identification mode, and the PC 2 sequentially uses the identification target data group stored in the PC 1 as accumulated data, and the time set in advance is determined. In the learning mode.

  Here, since the data identification method B is used, learning is performed at predetermined times T2 and T4. Further, as in the first embodiment, since a slight decrease in the amount of light occurs due to deterioration of the illumination after time T1, the identification target data obtained is as shown in FIG. 9, and the time change of the evaluation value is as in FIG. .

  When the appearance inspection of the product was started from time T1, as shown in FIG. 11, the evaluation value continued to increase as in Example 1 from time T1 to time T3. Here, at time T2, learning is performed using accumulated data from time T1 to T2. By learning with the accumulated data of the latest time in the present identification device, it was possible to reduce the difference between the accumulated data used for the learning and the illumination condition in the acquired identification target data. At this time, by updating the parameters of the feature amount extraction unit 11 and the data reconstruction unit 12 at time T3 with the parameters obtained by learning, the value of the evaluation value is reduced compared to before time T3 after time T3. It was possible. For this reason, it is possible to suppress an increase in the evaluation value after time T3. Similarly, by learning again at time T4 and updating the parameters of the feature quantity extraction unit 12 and the data reconstruction unit 13 at time T5, it becomes possible to suppress the increase in the evaluation value. As a result, at time T1 to T6, the evaluation value did not become equal to or higher than the threshold value, and the determination could be performed accurately.

Example 3
In this embodiment, the data identification method C is used. Here, similarly to the first embodiment and the second embodiment, the image of the product conveyed by the configuration of FIG. 8 is acquired, and the appearance inspection of the product is performed. Especially in this example, the products were produced intermittently. Specifically, production is performed from time T1 to time T2, time T3 to time T4, and time T5 to time T6, and the product is transported, and production is performed from time T2 to time T3 and time T4 to time T5. Not, product transport was stopped. In the present embodiment, the learning mode is operated during the product conveyance stop time.

  Here, since a slight decrease in the amount of light occurs due to the deterioration of the illumination after time T1, the identification target data obtained is as shown in FIG. 9, and the time change of evaluation is as shown in FIG. When the appearance inspection of the product was started from time T1, the evaluation value continued to rise between time T1 and time T2. However, by using the learning mode of the data identification apparatus for accumulated data acquired between time T1 and time T2 between time T2 and time T3, accumulated data used for learning and an identification target to be acquired As the difference in the lighting conditions in the data became smaller, it was possible to reduce the value of the evaluation value. For this reason, it is possible to suppress an increase in the evaluation value after time T3. Similarly, by performing learning also from time T4 to time T5, the determination could be performed accurately even after time T5.

  From the above, data identification method A is effective when it does not change due to accumulated data etc. However, the data acquired as in Example 1 changes with time, and it is time to continue examination without learning. On the other hand, when the degree of temporal change of data is large, there may be a problem in identification accuracy. For this reason, it is preferable to use the data identification method B and the data identification method C when the acquired data changes with time, or when the acquired data is affected by the update of a production apparatus or a peripheral device. In particular, when the production facility performs continuous production, it is necessary to operate in the identification mode at all times, so it is necessary to operate in the learning mode in parallel with the identification mode. Therefore, it is considered preferable to use the data identification method B used in the second embodiment. When the production equipment is batch production, it is preferable to use the data identification method C as in the third embodiment because the computing device can be effectively used by performing learning when the device is stopped.

  INDUSTRIAL APPLICABILITY The present invention is useful for detecting an abnormality in a transported product, a resin, or a yarn transported by a transport belt in a production process.

Reference Signs List 1 arithmetic device 2 data acquisition device 3 display device 10 data acquisition unit 11 feature amount extraction unit 12 data reconstruction unit 13 data evaluation unit 14 parameter determination unit 15 determination unit 16 result output unit 17 control unit

Claims (7)

A data identification device for identifying whether input data is unique data in a data group, comprising:
Feature quantity extraction means for extracting a feature quantity from the input data using a first neural network;
Data reconstruction means for generating reconstruction data of the same size as the input data from the feature quantities using a second neural network;
Parameter determination means for determining a parameter of the first neural network and a parameter of the second neural network by an error back propagation method from an error between the reconstructed data and the input data;
Data evaluation means for evaluating the magnitude of an error between the reconstructed data and the input data;
An identification unit that compares the evaluation value calculated by the evaluation unit with a preset threshold value to identify whether the input data is unique data;
The feature quantity extraction means, the learning mode for operating the data reconstruction means and the parameter determination means, and the data identification mode for operating the feature quantity extraction means, the data reconstruction means, the data evaluation means and the identification means Control means for operating the two modes independently or simultaneously
Data identification device equipped with The data group is derived from a continuously conveyed identification object,
The data identification device
Data acquisition means for acquiring data relating to the identification object;
Data storage means for storing data acquired by the data acquisition means;
And further
The control means
When the identification target is in the transport state, the data acquired by the data acquisition unit is input to the feature extraction unit, and the data identification mode is operated.
When the identification target is not in the transport state, data stored by the data storage unit is input to the feature amount extraction unit when the identification target is in the transport state, and the learning mode is operated.
The data identification device of claim 1.   3. The data identification apparatus according to claim 2, wherein the object to be identified is a conveyed product, a resin, or a yarn conveyed by a conveyance belt. A data identification method for identifying whether or not currently acquired data is unique data in a data group by causing the first neural network and the second neural network to learn data acquired and accumulated for a fixed period of time. And a data identification method for performing the following steps 1 and 2.
(Step 1)
Feature 1 is extracted from the data acquired and accumulated for a fixed period using a first neural network,
Generating reconstructed data of the same size as the accumulated data using the feature 1 to the second neural network;
Parameters of the first neural network and parameters of the second neural network are determined by an error back propagation method from an error between the reconstructed data and the accumulated data.
(Step 2)
Feature amount 2 is extracted from the data being acquired using the first neural network in which the parameter determined in the step 1 is set;
Generating reconstructed data of the same size as the data being acquired from the feature amount 2 using the second neural network in which the parameter determined in the step 1 is set;
From the magnitude of the error between the reconstructed data and the data being acquired, it is determined whether the data being acquired is unique data.   The data identification method according to claim 4, wherein said step 1 is performed only once at first, and then said step 2 is continuously performed. 5. The data identification method according to claim 4, wherein said step 1 is repeatedly performed in a predetermined cycle while said step 2 is continuously performed,
The data acquired and accumulated during the fixed period used in the step 1 is obtained in the process of the step 2 continuously performed between the step 1 and the step 1 performed immediately before the step 1 Data, and
The data identification method, wherein the parameters set in the first neural network and the second neural network in the step 2 are the parameters determined in the step 1 performed immediately before the step 2 respectively. 5. The data identification method according to claim 4, wherein said step 1 is repeatedly performed in a predetermined cycle, and said step 2 is continuously performed while said step 1 is not performed,
The data acquired and accumulated during the fixed period used in the step 1 is obtained in the process of the step 2 continuously performed between the step 1 and the step 1 performed immediately before the step 1 Data, and
The data identification method, wherein the parameters set in the first neural network and the second neural network in the step 2 are the parameters determined in the step 1 performed immediately before the step 2 respectively.

Images