JP7230086B2 - Polymorphic data analysis system and polymorphic data analysis program - Google Patents

Description

The present disclosure relates to parsing of polymorphic data.

Multiple types of sensors are used to ascertain the state of equipment. In addition, the time interval of measurement, the number of values obtained by measurement, the types of values obtained by measurement, and the like differ depending on the sensor. Therefore, it is necessary to analyze a plurality of sensor data having different formats.
However, no method has been established for collectively analyzing a plurality of sensor data having different formats. Furthermore, there are no techniques for analyzing data such as numerical or categorical data along with time series data such as sensor data.

Patent Literature 1 discloses that a convolutional neural network performs operations in the order of an input layer, a fully connected layer, a plurality of convolutional layers, and an output layer in order to eliminate dependence on the order of sensor data.
However, Patent Literature 1 does not disclose a technique for collectively analyzing a plurality of sensor data having different formats.

JP 2020-027375 A

The present disclosure aims to enable collective analysis of multi-form data.

The polymorphic data analysis system of the present disclosure comprises:
a format conversion unit that converts each of the plurality of individual numeric strings into a prescribed numeric string having a prescribed format based on format information data indicating the format of each of the plurality of individual numeric strings having formats different from each other;
a normalization unit that converts each of a plurality of specified numeric sequences corresponding to the plurality of individual numeric sequences into a normalized numeric sequence that is a normalized specified numeric sequence;
an output unit that outputs an analysis result obtained by using the plurality of normalized numerical value sequences corresponding to the plurality of defined numerical value sequences as two-dimensional data.

wherein the format information data indicates the number of digits of each numerical value contained in the individual numerical value string as a set number of digits for each individual numerical value string;
The format conversion unit obtains one or more numerical values by dividing the individual numerical value sequence by the set number of digits for each individual numerical value sequence, and obtains one or more numerical values for each individual numerical value sequence based on the specified Generate a numeric string.

Each numerical value contained in the prescribed numerical sequence is a numerical value represented by a prescribed base number,
wherein the format information data indicates, for each individual numeric string, the base of each numeric value contained in the individual numeric string as a set base;
The format conversion unit converts each of the one or more numerical values from the set base number to the specified base number for each individual numerical value sequence.

The prescribed numerical sequence consists of a prescribed number of numerical values,
wherein the format information data indicates the number of numerical values contained in each individual numerical value string as a set number;
The format conversion unit extracts a plurality of set numbers corresponding to the plurality of individual numerical strings from the format information data, calculates a least common multiple of the plurality of set numbers as the prescribed number, and A quotient is calculated by dividing the prescribed number by the set number, and each of the one or more numerical values is increased to the number indicated by the quotient for each individual numerical sequence, thereby dividing the one or more numerical values into the prescribed number. Convert to a numeric value.

The polymorphic data analysis system comprises:
a receiving unit that receives category data indicating a category identifier that identifies a specific category among a plurality of categories;
a digitization unit that obtains the plurality of individual numerical sequences by generating an individual numerical sequence for the specific category and an individual numerical sequence for each category other than the specific category.

A distinct numeric string for the specific category indicates positive values, and a distinct numeric string for each category other than the specific category indicates negative values.

The format information data indicates, for each individual numeric string, a set number of digits, which is the number of digits of each numeric value included in the individual numeric string, and a set base, which is the base of each numeric value included in the individual numeric string,
The normalization unit extracts from the format information data a set number of digits of an individual numeric string corresponding to the specified numeric string and a set base of the individual numeric string corresponding to the specified numeric string, and converts the extracted set base to A value obtained by subtracting 1 from the exponent obtained by subtracting 1 from the power obtained as the exponent of the set number of digits extracted as the base is calculated as a divisor, and the quotient is calculated by dividing each numerical value contained in the prescribed numerical value sequence by the divisor, and the prescribed numerical value sequence normalize the defined sequence of numbers by replacing each number contained in , with the calculated quotient.

The polymorphic data analysis system comprises:
a visualization unit that generates a two-dimensional image using the plurality of normalized numerical sequences as the two-dimensional data;
The output unit displays the two-dimensional image as the analysis result.

The polymorphic data analysis system comprises:
a model application unit that calculates the model using the plurality of normalized numerical sequences as the two-dimensional data that is input to a model of a convolutional neural network;
a visualization unit that generates a computation result graph representing computation results of the model,
The output unit displays the calculation result graph as the analysis result.

The model application unit performs one-dimensional convolution for each normalized numerical value sequence in the computation of the model.

The polymorphic data analysis system comprises:
A model building unit that builds the model using a data set that is a pair of two-dimensional data and a label as learning data,
The model application unit obtains, as the calculation result, label information for the two-dimensional data that is input to the model by calculating the model.

The model construction unit receives one or more knowledge data corresponding to two-dimensional data included in the data set, and determines labels included in the data set based on documents indicated in the one or more knowledge data. do.

The polymorphic data analysis program of the present disclosure comprises:
a format conversion unit that converts each of the plurality of individual numeric strings into a prescribed numeric string having a prescribed format based on format information data indicating the format of each of the plurality of individual numeric strings having formats different from each other;
a normalization unit that converts each of a plurality of specified numeric sequences corresponding to the plurality of individual numeric sequences into a normalized numeric sequence that is a normalized specified numeric sequence;
The computer is caused to function as an output unit that outputs analysis results obtained by using the plurality of normalized numerical value sequences corresponding to the plurality of specified numerical value sequences as two-dimensional data.

According to the present disclosure, polymorphic data can be transformed into two-dimensional data. As a result, it becomes possible to collectively analyze multiform data.

1 is a configuration diagram of a multiform data analysis system 100 according to Embodiment 1. FIG. 4 is a flowchart of a multiform data analysis method according to Embodiment 1; 4 is a diagram showing an example of formal expression of sensor data according to the first embodiment; FIG. FIG. 2 shows multi-format data 191 and format information data 192 according to Embodiment 1. FIG. 4 is a flowchart of step S120 in Embodiment 1; FIG. 4 shows a specific example of step S121 in Embodiment 1; 4 shows a specific example of step S122 in the first embodiment; FIG. 4 shows a specific example of step S123 in Embodiment 1. FIG. 4 shows a specific example of step S130 in Embodiment 1. FIG. Fig. 2 shows a two-dimensional image 193 according to Embodiment 1; FIG. 2 is a configuration diagram of a multiform data analysis system 100 according to Embodiment 2; 4 is a flowchart of a multiform data analysis method according to Embodiment 2; FIG. 10 shows format information data 192 and category data 194 according to the second embodiment; FIG. 11 shows a specific example of step S222 in the second embodiment; FIG. FIG. 11 is a configuration diagram of a multiform data analysis system 100 according to Embodiment 3; 10 is a flow chart of a multiform data analysis method (construction) according to Embodiment 3. FIG. 11 is a flowchart of a multiform data analysis method (application) according to Embodiment 3; The block diagram of the CNN model 195 in Embodiment 3. FIG. FIG. 11 is an explanatory diagram of one-dimensional convolution according to Embodiment 3; FIG. 11 shows computation result graphs (196, 197) in the third embodiment; FIG.

The same or corresponding elements are denoted by the same reference numerals in the embodiments and drawings. Descriptions of elements having the same reference numerals as those described will be omitted or simplified as appropriate. Arrows in the figure mainly indicate the flow of data or the flow of processing.

Embodiment 1.
A multiform data analysis system 100 will be described with reference to FIGS. 1 to 10. FIG.

*** Configuration description ***
The configuration of the multiform data analysis system 100 will be described based on FIG.
The multiform data analysis system 100 is a computer equipped with hardware such as a processor 101 , a memory 102 , an auxiliary storage device 103 , a communication device 104 and an input/output interface 105 . These pieces of hardware are connected to each other via signal lines.
However, the multiform data analysis system 100 may be composed of two or more computers.

The processor 101 is an IC that performs arithmetic processing and controls other hardware. For example, processor 101 is a CPU, DSP or GPU.
IC is an abbreviation for Integrated Circuit.
CPU is an abbreviation for Central Processing Unit.
DSP is an abbreviation for Digital Signal Processor.
GPU is an abbreviation for Graphics Processing Unit.

Memory 102 is a volatile or non-volatile storage device. Memory 102 is also referred to as main storage or main memory. For example, memory 102 is RAM. The data stored in the memory 102 is saved in the auxiliary storage device 103 as required.
RAM is an abbreviation for Random Access Memory.

Auxiliary storage device 103 is a non-volatile storage device. For example, the auxiliary storage device 103 is ROM, HDD or flash memory. Data stored in the auxiliary storage device 103 is loaded into the memory 102 as required.
ROM is an abbreviation for Read Only Memory.
HDD is an abbreviation for Hard Disk Drive.

Communication device 104 is a receiver and transmitter. For example, communication device 104 is a communication chip or NIC. Communication with the multiform data analysis system 100 is accomplished using a communication device 104 .
NIC is an abbreviation for Network Interface Card.

The input/output interface 105 is a port to which an input device and an output device are connected. For example, the input/output interface 105 is a USB terminal, the input device is a keyboard and mouse, and the output device is a display. Input/output of the multiform data analysis system 100 is performed using the input/output interface 105 .
USB is an abbreviation for Universal Serial Bus.

The multiform data analysis system 100 includes elements such as a reception unit 111 , a format conversion unit 112 , a normalization unit 113 , a visualization unit 114 and an output unit 115 . These elements are implemented in software.

Auxiliary storage device 103 stores a multiformat data analysis program for causing the computer to function as reception unit 111 , format conversion unit 112 , normalization unit 113 , visualization unit 114 , and output unit 115 . A polymorphic data analysis program is loaded into memory 102 and executed by processor 101 .
The auxiliary storage device 103 further stores an OS. At least part of the OS is loaded into memory 102 and executed by processor 101 .
The processor 101 executes the multiform data analysis program while executing the OS.
OS is an abbreviation for Operating System.

The input/output data of the multiform data analysis program are stored in the storage unit 190 .
Memory 102 functions as storage unit 190 . However, a storage device such as the auxiliary storage device 103 , a register within the processor 101 and a cache memory within the processor 101 may function as the storage unit 190 instead of or together with the memory 102 .

The multiform data analysis system 100 may include multiple processors in place of the processor 101 .

The polymorphic data analysis program can be recorded (stored) in a computer-readable manner on a non-volatile recording medium such as an optical disc or flash memory.

***Description of operation***
The procedure of operation of the multiform data analysis system 100 corresponds to the multiform data analysis method. Further, the procedure of operation of the multiform data analysis system 100 corresponds to the procedure of processing by the multiform data analysis program.

Based on FIG. 2, the multiform data analysis method will be described.
At step S 110 , the user inputs the multiform data and the form information data into the multiform data analysis system 100 .
The receiving unit 111 receives the polyformat data and the format information data.

Polymorphic data is a plurality of data having formats that are different from each other. Each piece of data included in the multiform data is called "target data".
Each target data indicates a numeric string consisting of one or more numeric values. A numeric string of each target data is called an “individual numeric string”.

Multiple sensor data obtained from multiple sensors is an example of polymorphic data.
Sensor data is time-series data or numerical data, and is an example of target data.
The number of digits of a numerical value, the base number of a numerical value, the number of numerical values, the time interval in a numerical sequence, and the time length in a numerical sequence are examples of elements that specify the data format.

The format information data indicates the format of each of the multiple individual numeric strings.

Note that sensor information data indicating the format of sensor data for each sensor may be stored in the storage unit 190 in advance. In that case, the receiving unit 111 receives sensor data as target data, selects sensor information data corresponding to the received sensor data, and acquires the sensor data format from the selected sensor information data. The acquired format is treated as the format indicated in the format information data.

FIG. 3 shows an example of a formal representation of sensor data.
For example, the format of the sensor data is expressed as shown in FIG. 3, specifying the number of digits, base number, and number.
The base of a number is binary when the base is represented by "bit".
When the base number is expressed in "byte", the base number of the number is hexadecimal. Also, since 1 byte corresponds to 2 characters, "N×2" is the number of digits.
If no number is given, the number is one.
In many sensor data, numerical values in time series are lined up from the left.

FIG. 4 shows polymorphic data 191 and format information data 192 .
The multi-format data 191 is an example of multi-format data, and the format information data 192 is an example of format information data.
Time series A, time series B, and time series C are time series data.
Numerical value A is numerical data.
The time series A, the time series B, the time series C, and the numerical value A are different from each other in terms of the number of digits, number of bases, and number.

Returning to FIG. 2, the description continues from step S120.
In step S120, the format conversion section 112 converts the multi-format data into standard format data based on the format information data.
In other words, the format conversion unit 112 converts each of the plurality of individual numeric strings into a specified numeric string based on the format information data.
A prescribed numeric string is a numeric string having a prescribed format. Specifically, the prescribed numerical value sequence consists of a prescribed number of numerical values. In addition, each numerical value included in the prescribed numerical value sequence is a numerical value represented by a prescribed base number. The prescribed number and the prescribed number will be described later.

The procedure of step S120 will be described based on FIG.
The format information data indicates the set number of digits, the set base number, and the set number for each individual numeric string (target data).
The set number of digits is the number of digits of each numerical value included in the individual numerical value string.
The set base number is the base number of each numerical value contained in the individual numerical value sequence.
The set number is the number of numerical values included in the individual numerical value sequence.

In step S<b>121 , the format conversion unit 112 divides each individual numeric string by the set number of digits. This yields one or more numerical values for each distinct numerical sequence.

A numeric string consisting of one or more numeric values obtained in step S121 is called a "divided numeric string".

FIG. 6 shows a specific example of step S121.
The set number of digits for the time series A is "2". Therefore, the format conversion unit 112 divides the individual numerical value sequence of the time series A into two digits each. This gives 32 numerical values.
The set number of digits for the time series B is "4". Therefore, the format conversion unit 112 divides the individual numerical value sequence of the time series B into four digits each. This gives 16 numbers.
The set number of digits for the time series C is "8". Therefore, the format conversion unit 112 divides the individual numerical value sequence of the time series C into eight digits each. This gives four numbers.
The set number of digits of the numerical value A is "2". Therefore, the format conversion unit 112 divides the individual numerical value string of the numerical value A into two digits each. This gives a single numerical value.

Returning to FIG. 5, the description continues from step S122.
In step S122, the format conversion unit 112 converts each of the one or more numeric values from the set base number to the specified base number for each individual numeric string.
The prescribed base number is a predetermined base number. A specific example of the prescribed base number is a decimal number.

A numeric string consisting of one or more numeric values obtained in step S122 is referred to as a "base numeric string".

FIG. 7 shows a specific example of step S122. The prescribed base number is a decimal number.
The base number set for the time series A is "16". Therefore, the format conversion unit 112 converts each of one or more numerical values in the time series A from hexadecimal numerical values to decimal numerical values. For example, hexadecimal "05" is converted to decimal "5".
The base number set for the time series B is "16". Therefore, the format conversion unit 112 converts each of one or more numerical values in the time series B from hexadecimal numerical values to decimal numerical values. For example, hexadecimal "14FA" is converted to decimal "5370".
The base number set for the time series C is "2". Therefore, the format conversion unit 112 converts each of one or more numerical values in the time series C from binary numerical values to decimal numerical values. For example, binary "00000000" is converted to decimal "0".
The base number set for the numeric value A is "16". Therefore, the format conversion unit 112 converts each of the one or more numerical values of the numerical value A from a hexadecimal numerical value to a decimal numerical value. For example, hexadecimal "08" is converted to decimal "8".

Returning to FIG. 5, the description continues from step S123.
In step S123, the format conversion unit 112 converts one or more numerical values into a specified number of numerical values for each individual numerical value sequence.

Specifically, the format conversion unit 112 operates as follows.
First, the format conversion unit 112 extracts a plurality of setting numbers corresponding to a plurality of individual numeric strings from the format information data.
Next, the format conversion unit 112 calculates the least common multiple of the multiple set numbers. The calculated lowest common multiple is the "specified number".
Next, the format conversion unit 112 calculates a quotient by dividing the specified number by the set number for each individual numerical value sequence.
Then, the format conversion unit 112 increases each of the one or more numerical values to the same number as the number indicated by the quotient for each individual numerical value sequence.

The specified number of numerical values obtained in step S123 is referred to as a "number number sequence". The count numeric string is a defined numeric string.

FIG. 8 shows a specific example of step S123.
The set number of time series A is "32". The set number of time series B is "16". The set number of time series C is "4". The set number of the numerical value A is "1". The lowest common multiple of these set numbers is "32". Therefore, the specified number is "32".
A quotient of "1" is obtained by dividing the specified number "32" by the set number "32" of the time series A. Therefore, the format conversion unit 112 does not increase the number of numerical values in the time series A.
A quotient of “2” is obtained by dividing the specified number “32” by the set number of time series B “16”. Therefore, the format conversion unit 112 increases each numerical value of the time series B to two. For example, the format conversion unit 112 increases the first numerical value "5370" of the time series B to two.
A quotient of "8" is obtained by dividing the prescribed number "32" by the set number "4" of the time series C. Therefore, the format conversion unit 112 increases each numerical value of the time series C to eight. For example, the format conversion unit 112 increases the first numerical value "0" of the time series C to eight.
A quotient of "32" is obtained by dividing the prescribed number "32" by the set number "1" of the numerical value A. Therefore, the format conversion unit 112 increases each numerical value of the numerical value A to 32 pieces. That is, the format conversion unit 112 increases the single numerical value "8" of the numerical value A to 32 numbers.

Returning to FIG. 5, the procedure of step S120 is supplemented.
The order of steps S122 and S123 may be exchanged.
That is, the format conversion unit 112 may convert the divided numerical value sequence (S121) into the number numerical value sequence (S123) and then convert the number numerical value sequence (S123) into the hexadecimal numerical value sequence (S122). In this case, the hexadecimal number string (S122) becomes the prescribed number string.

Returning to FIG. 2, the description continues from step S130.
In step S130, the normalization unit 113 normalizes the prescribed format data.
In other words, the normalization unit 113 converts each of the plurality of specified numeric strings corresponding to the plurality of individual numeric strings into a normalized numeric string.
A normalized numeric sequence is a normalized prescribed numeric sequence.

Specifically, the normalization unit 113 converts each prescribed numerical value sequence into a normalized numerical value sequence as follows.
First, the normalization unit 113 extracts the set number of digits of the individual numeric string corresponding to the specified numeric string and the set base number of the individual numeric string corresponding to the specified numeric string from the format information data.
Next, the normalization unit 113 uses the extracted set base number as a base and the extracted set number of digits as an exponent to calculate a power. Then, the normalization unit 113 calculates a value obtained by subtracting 1 from the calculated exponent. That is, when the set base number is "x" and the set number of digits is "y", the normalization unit 113 calculates "x ^y -1". The calculated value is the maximum value of the numerical values in the prescribed numerical sequence.
Next, the normalization unit 113 uses the calculated maximum value as a divisor, divides each numerical value included in the prescribed numerical sequence by the divisor, and calculates a quotient. The calculated quotient is a value of 0 or more and 1 or less.
Then, the normalization unit 113 replaces each numerical value included in the prescribed numerical sequence with the calculated quotient.

FIG. 9 shows a specific example of step S130.
The set number of digits for the time series A is "2", and the set base number for the time series A is "16". Therefore, the maximum numerical value in the prescribed numerical sequence of time series A is "255 (=16 ² -1)". Therefore, the normalization unit 113 divides each numerical value of the prescribed numerical sequence of the time series A by "255" and replaces it with a quotient. For example, "5" is replaced with "0.02 (=5/255)".
The set number of digits for the time series B is "4", and the set base number for the time series B is "16". Therefore, the maximum numerical value in the prescribed numerical sequence of time series B is "65535 (=16 ⁴ -1)". Therefore, the normalization unit 113 divides each numerical value of the specified numerical sequence of the time series B by "65535" and replaces it with a quotient. For example, "5370" is replaced with "0.08 (=5370/65535)".
The set number of digits for the time series C is "8", and the set base number for the time series C is "2". Therefore, the maximum numerical value in the prescribed numerical sequence of time series C is "255 (=2 ⁸ -1)". Therefore, the normalization unit 113 divides each numerical value of the specified numerical sequence of the time series C by "255" and replaces it with a quotient. For example, "0" is replaced with "0.00 (=0/255)".
The set number of digits of the numerical value A is "2", and the set base number of the numerical value A is "16". Therefore, the maximum numerical value in the prescribed numerical sequence of numerical value A is "255 (=16 ² -1)". Therefore, the normalization unit 113 divides each numerical value in the prescribed numerical sequence of the numerical value A by "255" and replaces it with a quotient. For example, "8" is replaced with "0.03 (=8/255)".

Returning to FIG. 2, the description continues from step S140.
In step S140, the visualization unit 114 generates a two-dimensional image using the normalized data as two-dimensional data.
That is, the visualization unit 114 generates a two-dimensional image using a plurality of normalized numerical value sequences as two-dimensional data.

A two-dimensional image 193 is shown in FIG.
A two-dimensional image 193 is an example of a two-dimensional image generated in step S140.
The two-dimensional image 193 corresponds to an image obtained by visualizing two-dimensional data of “time series x sensor”.
One row of the two-dimensional image 193 represents one normalized numerical value sequence.
One square in the two-dimensional image 193 represents one numerical value in the normalized numerical value sequence.
Each numerical value is distinguished by the density of hatching. The smaller the numerical value, the lower the density of hatching, and the larger the numerical value, the higher the density of hatching.
However, each numerical value may be distinguished by color gradation, may be distinguished by color, pattern or color, or may be distinguished by other methods.

Returning to FIG. 2, step S150 will be described.
In step S150, the output unit 115 displays the two-dimensional image on the display.
A two-dimensional image is an example of an analysis result of polymorphic data. The display is an example of output.
A user can visually grasp the content of the multi-format data by referring to the two-dimensional image. For example, by comparing a two-dimensional image of multi-format data obtained before the occurrence of a failure with a two-dimensional image of multi-format data obtained after the occurrence of a failure, the user can see changes in the multi-format data before and after the occurrence of the failure. It can be grasped visually.

*** Effect of Embodiment 1 ***
According to Embodiment 1, polymorphic data can be converted into two-dimensional data. As a result, it becomes possible to collectively analyze multiform data.
Also, two-dimensional data can be visualized. This allows the user to visually grasp the contents of the multi-format data.

Embodiment 2.
A form in which category data is included in multiform data will be described mainly with reference to FIGS. 11 to 14 for differences from the first embodiment.

*** Configuration description ***
The configuration of the multiform data analysis system 100 will be described based on FIG.
The multiform data analysis system 100 further includes a digitizing section 121 .
The multiform data analysis program further causes the computer to function as the digitizing section 121 .

***Description of operation***
Based on FIG. 12, the multiform data analysis method will be described.
In step S210, accepting unit 111 accepts the polyformat data and the format information data.
Step S210 is the same as step S110 of the first embodiment.

In step S221, the digitization unit 121 determines whether category data is included in the polymorphic data.
Category data is data indicating a category identifier.
A category identifier identifies a particular category of a plurality of categories. For example, category identifiers are represented by letters, symbols, numbers, or combinations thereof.

For example, the digitization unit 121 determines whether the target data is categorical data based on the format indicated in the format information data for each target data included in the multiformat data. Target data with no base number is category data.

If the multiform data includes category data, the process proceeds to step S222.
If the multiform data does not contain category data, the process proceeds to step S230.

FIG. 13 shows the category data 194. As shown in FIG.
Category data 194 is an example of category data.
Possible values for the category identifier are 'A', 'B' or 'C'.
Category data 194 indicates the category identifier "A".

Returning to FIG. 12, step S222 will be described.
In step S222, the digitization unit 121 generates a plurality of individual numerical value string data corresponding to a plurality of categories based on the category data.
Then, the digitization unit 121 replaces the category data included in the polymorphic data with a plurality of individual numeric string data.
Also, the digitization unit 121 updates the format information data for a plurality of individual numeric string data.

Individual numerical value string data is data indicating an individual numerical value string.
A plurality of individual numeric string data corresponding to category data consists of separate numeric string data for a specific category and separate numeric string data for each category other than the specific category.
A specific category is a category identified by a category identifier indicated in the category data.

Each individual numeric string data corresponding to category data is numeric data.
Distinct numeric string data for a particular category indicates a positive value as a distinct numeric string. A positive value means that the corresponding category is the particular category.
The individual numeric string data for each category other than the specific category shows negative values as the individual numeric string. A negative value means that the corresponding category is not the specific category.

FIG. 14 shows a specific example of step S222.
Since there are three categories "A", "B", and "C", the digitizing unit 121 generates three separate numerical value string data for category A, category B, and category C. The set number is "2", the positive value is "1", and the negative value is "0".
The individual numerical string of category A is "1", and the individual numerical string of each of category B and category C is "0".

Returning to FIG. 12, steps S230 to S260 will be described.
Steps S230 to S260 are the same as steps S120 to S150 of the first embodiment.

*** Effect of Embodiment 2 ***
According to the second embodiment, multi-format data containing category data can be converted into two-dimensional data. As a result, it becomes possible to collectively analyze multiform data including category data.

Embodiment 3.
A form of analyzing multiform data using a CNN model will be described mainly with reference to FIGS. 15 to 20 for differences from the first embodiment.

CNN is an abbreviation for convolutional neural network.
A CNN model is a model of a convolutional neural network.

*** Configuration description ***
The configuration of the multiform data analysis system 100 will be described based on FIG.
The multiform data analysis system 100 further comprises a model construction section 131 and a model application section 132 .
The multiform data analysis program further causes the computer to function as a model construction unit 131 and a model application unit 132 .

***Description of operation***
Based on FIG. 16, the multiform data analysis method (construction) will be described.
A polymorphic data analysis method (construction) is a method for constructing a CNN model.

Steps S301 to S303 are the same as steps S110 to S130 of the first embodiment.
As a result, two-dimensional data, which are a plurality of normalized numerical value sequences, are obtained.

In step S304, the model construction unit 131 determines labels for the two-dimensional data.

Specifically, labels for two-dimensional data are determined as follows.
A user inputs knowledge data corresponding to multiform data, that is, knowledge data corresponding to two-dimensional data, into the multiform data analysis system 100 . Knowledge data is document data specified by a user.

For example, the user specifies the knowledge data as follows.
In step S301, the user inputs into the multiform data analysis system 100 the multiform data obtained when a device failure occurs.
In step S304, the user specifies a troubleshooting manual for the failure that has occurred based on basic information about the failed device or diagnostic information for the failure that has occurred. Also, the user specifies the response history for the failure that occurred.
The specified troubleshooting manual and the specified failure handling history become one or more knowledge data corresponding to the two-dimensional data.

The model construction unit 131 receives one or more pieces of knowledge data, analyzes the document indicated by the one or more pieces of knowledge data by natural language analysis or the like, and determines a label for the two-dimensional data based on the analysis result. For example, the model construction unit 131 searches for common terms in the troubleshooting manual and the failure handling history, and determines the found terms as labels for the two-dimensional data.

A set of two-dimensional data and a label is called a "dataset".
Steps S301 through S304 are performed on a plurality of polymorphic data to obtain a plurality of data sets.

In step S305, the model construction unit 131 constructs a CNN model by performing deep learning using a plurality of data sets as learning data.
Specifically, the model construction unit 131 inputs two-dimensional data for each two-dimensional data included in a plurality of data sets, calculates the CNN model, and learns the weight of the feature amount of the two-dimensional data. Then, the model construction unit 131 registers a set of the weight of the feature amount of the two-dimensional data and the label in the CNN model for each two-dimensional data included in the plurality of data sets.
The computation of the CNN model will be described later.

Based on FIG. 17, the multiform data analysis method (application) will be described.
A polymorphic data analysis method (application) is a method for applying a CNN model.

Steps S311 to S313 are the same as steps S110 to S130 of the first embodiment.
As a result, two-dimensional data, which are a plurality of normalized numerical value sequences, are obtained.

In step S314, the model application unit 132 receives the two-dimensional data and calculates the CNN model.

The calculation of the CNN model 195 will be explained based on FIG.
CNN model 195 is an example of a CNN model.

The CNN model 195 has an input layer, multiple convolutional layers, multiple pooling layers, two fully connected layers, and an output layer.
First, the input layer receives and outputs two-dimensional data.
Then the convolutional and pooling layers are iteratively executed. The convolution layer performs one-dimensional convolution for each normalized numeric sequence. The pooling layer performs pooling.
Then two fully connected layers perform regularization and dropout.
The output layer then computes the sigmoid function or softmax function and outputs the result.

Based on FIG. 19, the one-dimensional convolution will be described using the first convolution layer as an example.
Assume that the input layer receives two-dimensional data of "wxh".
The first convolution layer selects kernels in order from the top (left end) to the end (right end) for each row, and performs convolution on multiple numerical values included in the kernels. A kernel is a number corresponding to the kernel size. If the kernel size is 4 (=1×4), the kernel is 4 consecutive numbers.
As a result, two-dimensional data with a reduced number of columns can be obtained. When the kernel size is 4 (=1×4), two-dimensional data of “(w−3)×h” are obtained.

Returning to FIG. 17, the description is continued.
Through step S314, label information for the two-dimensional data is obtained as a calculation result. For example, the calculation result indicates the probability that two-dimensional data corresponds to each label.

In step S315, the visualization unit 114 generates a graph representing the calculation result. The generated graph is called a "computation result graph".

FIG. 20 shows calculation result graph 196 and calculation result graph 197 .
Each of the computation result graph 196 and the computation result graph 197 is an example of a computation result graph.
The calculation result graph 196 is a bar graph showing probabilities for each of the six labels "01" to "06".
The calculation result graph 197 is a pie chart showing the probability for each of the two labels "recoverable" and "recoverable".

Returning to FIG. 17, step S315 will be described.
In step S315, the output unit 115 displays the computation result graph on the display.
The operation result graph is an example of the analysis result of polymorphic data.
The user can visually grasp the contents of the multi-form data by referring to the calculation result graph. For example, the user can visually grasp the failure state of the equipment by referring to the operation result graph of the multi-format data obtained when the failure occurs.

*** Effect of Embodiment 3 ***
According to Embodiment 3, polymorphic data can be converted into two-dimensional data, and a convolutional neural network (CNN) can be applied with the two-dimensional data as input. As a result, it becomes possible to collectively analyze multiform data.
Also, the results obtained by applying CNN can be visualized. This allows the user to visually grasp the contents of the multi-format data.

***Description of Examples***
The third embodiment may be combined with the first embodiment. That is, the visualization unit 114 may generate a two-dimensional image, and the output unit 115 may display the two-dimensional image.
The third embodiment may be combined with the second embodiment. That is, the multiform data analysis system 100 may include the digitizing section 121 .

***Summary of Embodiments***
In Embodiments 1 to 3, two-dimensional conversion of multi-format sensor data has been described.
The multi-format data analysis system 100 converts data in various formats (sensor data, basic information, operation information) such as time-series data with different acquisition intervals and acquisition times, non-time-series numerical data, and category data into two-dimensional data. do. In the two-dimensional data, items such as sensors are shown in the vertical direction, and time series are shown in the horizontal direction. The multiform data analysis system 100 also applies a convolutional neural network (CNN) to the two-dimensional data.
As a result, it is possible to learn data values and time changes with CNN while retaining information on time-series data with different acquisition intervals and acquisition times, and data other than time-series data (category data, numerical data). .
The CNN is used to learn weights of feature amounts of sensor data, basic information, and operational information, and to calculate the probability of a specific label corresponding to the data.

In Embodiment 3, one-dimensional convolution of two-dimensional data has been described.
A convolutional layer of a CNN for image recognition performs a two-dimensional convolution. On the other hand, the convolution layer of the CNN in Embodiment 3 performs one-dimensional convolution for each sensor.
This makes it possible to learn changes in data along the time axis without reflecting the order in which the sensors are arranged in model learning.

In the third embodiment, utilization of knowledge of experts has been described.
When constructing the CNN model, the engineer's insight information is utilized to obtain the correct labels. Specifically, design information (troubleshooting manual, device failure diagnosis information) and failure response history are utilized.
By using the knowledge of experts in the sensor data, basic information, and correct labels for operation information when an elevator (an example of equipment) fails, it is possible to estimate the cause of the failure and the probability of success in restoring remote failures.
A fusion of deep learning and the knowledge of experts realizes an estimation that satisfies all the people (managers, administrators, engineers) involved in the system into which the third embodiment is introduced.

*** Supplement to the embodiment ***
Each embodiment is an example of a preferred form and is not intended to limit the technical scope of the present disclosure. Each embodiment may be implemented partially or in combination with other embodiments. The procedures described using flowcharts and the like may be changed as appropriate.

Each element of multiform data analysis system 100 may be implemented in software, hardware, firmware, or a combination thereof.
A “unit” that is an element of the multiform data analysis system 100 may be read as “processing”, “process”, “circuit” or “circuitry”.

100 Multiform Data Analysis System 101 Processor 102 Memory 103 Auxiliary Storage Device 104 Communication Device 105 Input/Output Interface 111 Reception Unit 112 Format Conversion Unit 113 Normalization Unit 114 Visualization Unit 115 Output Unit 121 Numerical unit 131 model construction unit 132 model application unit 190 storage unit 191 polymorphic data 192 format information data 193 two-dimensional image 194 category data 195 CNN model 196 operation result graph 197 operation result graph .

Claims (6)

a format conversion unit that converts each of the plurality of individual numeric strings into a prescribed numeric string having a prescribed format based on format information data indicating the format of each of the plurality of individual numeric strings having formats different from each other;
a normalization unit that converts each of a plurality of specified numeric sequences corresponding to the plurality of individual numeric sequences into a normalized numeric sequence that is a normalized specified numeric sequence;
an output unit that outputs an analysis result obtained by using the plurality of normalized numerical value sequences corresponding to the plurality of specified numerical value sequences as two-dimensional data;
with
The multiple format data analysis system, wherein the plurality of individual numerical sequences includes an individual numerical sequence composed of one or more numerical values arranged in time series and an individual numerical sequence composed of one or more numerical values other than the time series. The polymorphic data analysis system comprises:
a receiving unit that receives the time-series individual numerical value sequence, the non-time-series individual numerical value sequence, category data indicating a category identifier that identifies a specific category among a plurality of categories, and
a digitizing unit that generates an individual numeric string for the specific category and an individual numeric string for each category other than the specific category;
2. The multiform data analysis of claim 1, wherein said plurality of distinct numerical columns includes said distinct numerical sequence for said specific category and said distinct numerical sequence for each category other than said specific category. system. 3. A multiform data analysis system according to claim 2, wherein said distinct numerical column for said particular category indicates positive values and said distinct numerical column for each category other than said particular category indicates negative values. The format information data indicates, for each individual numeric string, a set number of digits, which is the number of digits of each numeric value included in the individual numeric string, and a set base, which is the base of each numeric value included in the individual numeric string,
The normalization unit extracts from the format information data a set number of digits of an individual numeric string corresponding to the specified numeric string and a set base of the individual numeric string corresponding to the specified numeric string, and converts the extracted set base to A value obtained by subtracting 1 from the exponent obtained by subtracting 1 from the power obtained as the exponent of the set number of digits extracted as the base is calculated as a divisor, and the quotient is calculated by dividing each numerical value contained in the prescribed numerical value sequence by the divisor, and the prescribed numerical value sequence 4. A multiform data analysis system according to any one of claims 1 to 3 , wherein said prescribed numerical sequence is normalized by replacing each numerical value contained in by a calculated quotient. The polymorphic data analysis system comprises:
a visualization unit that generates a two-dimensional image using the plurality of normalized numerical sequences as the two-dimensional data;
5. The multi-format data analysis system according to claim 1, wherein said output unit displays said two -dimensional image as said analysis result. a format conversion unit that converts each of the plurality of individual numeric strings into a prescribed numeric string having a prescribed format based on format information data indicating the format of each of the plurality of individual numeric strings having formats different from each other;
a normalization unit that converts each of a plurality of specified numeric sequences corresponding to the plurality of individual numeric sequences into a normalized numeric sequence that is a normalized specified numeric sequence;
A multiformat data analysis program for causing a computer to function as an output unit that outputs analysis results obtained by using a plurality of normalized numerical sequences corresponding to the plurality of prescribed numerical sequences as two-dimensional data,
A multi-format data analysis program in which the plurality of individual numerical sequences includes an individual numerical sequence composed of one or more numerical values arranged in time series and an individual numerical sequence composed of one or more numerical values other than the time series.

Images