JP2019003603A - Artificial intelligence module development method, artificial intelligence module development system, and artificial intelligence module development integration system - Google Patents

Abstract

To provide an artificial intelligence module development method capable of selecting an artificial intelligence module (neural network and the like) most suitable for an analysis purpose by itself and promptly generating a highly-reliable artificial intelligence module by automatic learning with simple operation, an artificial intelligence module development system, and an artificial intelligence module development integration system.SOLUTION: An artificial intelligence module development method comprises the steps of causing an operator to: select an analysis purpose; register the data necessary for the analysis purpose; set a target for achieving the analysis purpose; instruct implementation of the learning tailored to the analysis purpose and the target using the registered data; confirm the artificial intelligence module requested as learning results; and instruct incorporation of the learning results containing the confirmed artificial intelligence module into other device.SELECTED DRAWING: Figure 3

Description

  The present invention relates to an artificial intelligence module development system and an artificial intelligence module development integration system related to, for example, a data analysis system for analyzing data, and more particularly, for example, to be used as an advanced analysis infrastructure for machine learning by analyzing a large amount of data. The present invention relates to a possible artificial intelligence module development method, an artificial intelligence module development system, and an artificial intelligence module development integration system.

  As a result of the progress of computerization of society due to the development of computers, enormous amounts of data are widely and closely related to the activities of manufacturing products by companies or individuals. For this reason, recently, there is a need to accurately sort out desired information from a vast amount of data and use only valid data.

  As an approach for extracting desired information from a vast amount of data, computer analysis, selection, conversion, grouping, range, etc. are implemented using various algorithms, but the range of use is still narrow. In order to improve the effective use of enormous amounts of data, aiming at automatic learning and machine learning based on the program in the computer, artificial intelligence that selects and uses information independently in the system to improve judgment and judgment accuracy It is desirable to develop and utilize it.

  Even in the recent development of systems that repeat machine learning, artificial analysis that can perform analysis, selection, and conversion of data independently, and make judgments and judgments on external events faster, more accurately and accurately than humans can. It must be said that the development of intelligence has not yet been reached.

  For example, Patent Literature 1 discloses a technical idea of a machine learning system for providing a means for comparing machine learning algorithm candidates.

According to this, the machine learning system
・ Machine learning algorithm library,
A data input module for receiving a data set and receiving a selection of machine learning models obtained from a machine learning algorithm library;
An experimental module configured to train and evaluate each machine learning model and generate performance results for each machine learning model;
-Aggregation module configured to aggregate all performance results of machine learning models and create performance comparison statistics, receive data sets, receive machine learning model selections, and Training models are trained and evaluated, performance results are generated for each machine learning model, performance results are aggregated, performance comparison statistics are created, and performance comparison statistics are displayed.

JP 2017-4509 A

  However, according to this disclosed technique, it is difficult to obtain a machine learning model suitable for the analysis purpose because the analysis purpose is not specified when the machine learning model is created. The machine learning model (artificial intelligence module) that fits the purpose of the analysis can be obtained automatically, simply, and at high speed. The problem of difficulty has not been solved.

  The present invention is intended to solve such problems in the prior art, and an artificial intelligence module (such as a neural network) that is most adapted to the purpose of analysis is selected by itself, and an artificial intelligence module having high reliability by automatic learning. It is an object of the present invention to provide an artificial intelligence module development method, an artificial intelligence module development system, and an artificial intelligence module development integrated system that can be quickly generated by simple operations.

In order to solve such a problem, the artificial intelligence module development method according to the present invention,
1) selecting an analysis purpose;
2) registering data necessary for the analysis purpose;
3) setting a goal to achieve the analysis purpose;
4) Instructing execution of learning that matches the analysis purpose and the goal using the registered data;
5) confirming the artificial intelligence module obtained as a result of the learning;
6) Instructing to incorporate the learning result including the confirmed artificial intelligence module into another device.

  In this way, the user of the artificial intelligence module can acquire the desired artificial intelligence module simply by sequentially proceeding with such steps, and can actually use it in another device.

The artificial intelligence module development system according to the present invention includes a central processing unit that performs various calculations, a screen of an output device that outputs information that is a result of the calculation performed by the central processing unit, and the central processing unit. An artificial intelligence module development system comprising a storage device for storing information exchanged with
1) A first screen for prompting the user to input analysis purpose information, which is data defining an analysis purpose, is displayed on the screen, and the center is set so that the input analysis purpose information is stored in the storage device. A first screen generator for operating the arithmetic device;
2) The central processing unit so that a second screen for prompting the user to input analysis-necessary data necessary for the analysis purpose is displayed on the screen and the input analysis-necessary data is stored in the storage device. A second screen generator for operating
3) A third screen for allowing the user to set analysis objective achievement target data that defines the analysis objective achievement target, which is a specific target for achieving the analysis objective, is displayed on the screen and input. A third screen generator for operating the central processing unit to store analysis goal achievement target data in the storage device;
4) A fourth process for accepting an instruction from a user to execute learning suitable for the analysis purpose and the analysis purpose achievement target to the central processing unit using the analysis purpose necessary data stored in the storage device. A fourth screen generating unit that operates the central processing unit so that the screen is displayed on the screen;
5) The artificial intelligence module that is the result of the central processing unit executing the learning is displayed on the screen, and the user receives the confirmation information that is the result of confirming the artificial intelligence module. A fifth screen generator for operating the central processing unit so that the screen of 5 is displayed on the screen;
6) A sixth screen for receiving an instruction from a user to incorporate the learning result including the artificial intelligence module to which the confirmation information is input into another device is displayed on the screen. And a sixth screen generation unit that operates the central processing unit.

  In this way, the user of the artificial intelligence module can obtain the desired artificial intelligence module simply by proceeding sequentially with such guidance on the screen, and can actually use it with other devices. can do.

Further, the artificial intelligence module development system according to the present invention is:
A) a data input interface block for inputting data;
B) A data preprocessing block comprising a data preprocessing module library and a data preprocessing module for performing preprocessing for converting data input by the data input interface block into a form necessary for analysis;
C) Analyzing the data preprocessed by the data preprocessing block,
Pre-learning neural network library,
Pre-learning neural network setting module,
Neural network learning setting library,
A neural network learning execution module;
A neural network learning block comprising a post-learning neural network setting module;
D) a data post-processing block comprising a data post-processing module library and a data post-processing module for post-processing output data output by the neural network learning block;
E) An artificial intelligence module output interface block for outputting an artificial intelligence module, which is a product generated by the data post-processing block, may be provided.

  With such a specific configuration, the artificial intelligence module development method and the artificial intelligence module development apparatus described above are guided to obtain a desired artificial intelligence module.

The artificial intelligence module development system according to the present invention further includes:
The data preprocessing module may select an analysis purpose from an analysis purpose menu and transfer the selected analysis purpose to the pre-learning neural network setting module.

  In this way, it is possible to select a neural network suitable for the purpose of analysis.

The artificial intelligence module development system according to the present invention further includes:
The data preprocessing module preprocesses learning data input via the data input interface block using a selected data preprocessing algorithm, and the preprocessed data is set to the pre-learning neural network setting. It may be characterized by being transferred to a module.

  In this way, the accuracy of the pre-learning neural network selection is improved by using appropriately preprocessed data.

The artificial intelligence module development system according to the present invention further includes:
The pre-learning neural network setting module selects a pre-learning neural network and a learning method suitable for the purpose from the pre-learning neural network library according to the selected analysis purpose, and transfers them to the neural network learning setting module. May be a feature.

  In this way, an appropriate pre-learning neural network and learning method suitable for the purpose of analysis are selected.

The artificial intelligence module development system according to the present invention further includes:
The neural network learning setting module performs learning using the selected pre-learning neural network, the selected learning method, and the preprocessed learning data, and after completion, the learned The neural network, the learning result, and / or the selected data pre-processing algorithm may be transferred to the data post-processing module.

The artificial intelligence module development system according to the present invention further includes:
The data post-processing module may select a data post-processing algorithm including a determination formula from the data post-processing module library and determine the learning result.

  If it does in this way, judgment of a learning result will be performed appropriately.

The artificial intelligence module development system according to the present invention further includes:
The artificial intelligence module may include the selected data pre-processing algorithm, the selected pre-learning neural network, the selected learning method, and / or the selected data post-processing algorithm. Good.

  If it does in this way, the calculated | required artificial intelligence module can be utilized appropriately.

  In addition, a multiple artificial intelligence module development system characterized in that a plurality of artificial intelligence module development systems according to the present invention described so far are connected in series may be used.

  Alternatively, it may be a multiple artificial intelligence module development system in which a plurality of artificial intelligence module development systems according to the present invention described so far are connected in parallel.

  Alternatively, the above-described artificial intelligence module development system according to the present invention may be a multiple artificial intelligence module development system characterized in that a plurality of the artificial intelligence module development systems are connected in series and in parallel.

  In this way, by using a multiple-type artificial intelligence module development system in which a plurality of artificial intelligence modules are connected in series and / or in parallel, an artificial intelligence module superior to a single artificial intelligence module development system can be created. It becomes possible.

  Also, an artificial intelligence module having the artificial intelligence module development system according to the present invention described so far, and a translation middleware for making the artificial intelligence module obtained by the artificial intelligence module development system compatible with other programming languages It may be an aspect of a development integrated system.

  If it does in this way, an artificial intelligence module can be easily utilized with another apparatus.

  ADVANTAGE OF THE INVENTION According to this invention, the artificial intelligence module development method which can select the artificial intelligence module (neural network etc.) most suitable for the analysis objective by itself, and can produce | generate a highly reliable artificial intelligence module by simple operation quickly by simple operation. The artificial intelligence module development system and the artificial intelligence module development integrated system can be provided.

It is a hardware block diagram of the artificial intelligence module development system which concerns on one Embodiment of this invention. It is a functional lineblock diagram of an artificial intelligence module development system concerning one embodiment of the present invention. It is the flowchart seen from the operator side of the artificial intelligence module development system which concerns on one Embodiment of this invention. It is the screen transition figure seen from the operator side of the artificial intelligence module development system which concerns on one Embodiment of this invention. It is the screen transition figure seen from the operator side of the artificial intelligence module development system which concerns on one Embodiment of this invention. It is the screen transition figure seen from the operator side of the artificial intelligence module development system which concerns on one Embodiment of this invention. It is the screen transition figure seen from the operator side of the artificial intelligence module development system which concerns on one Embodiment of this invention. It is the screen transition figure seen from the operator side of the artificial intelligence module development system which concerns on one Embodiment of this invention. It is the screen transition figure seen from the operator side of the artificial intelligence module development system which concerns on one Embodiment of this invention. It is the timing flowchart which showed the process of the artificial intelligence module development system which concerns on one Embodiment of this invention. It is the conceptual diagram which showed the image of the translation environment of the artificial intelligence module development integrated system which concerns on one Embodiment of this invention. It is a conceptual diagram which shows the multiple type | mold artificial intelligence module development system which connected the artificial intelligence module development system in series based on one Example of another aspect of this invention. It is a conceptual diagram which shows the multiple | multiplex type | mold artificial intelligence module development system which connected the artificial intelligence module development system in series according to one Example of another aspect of this invention in series. It is a conceptual diagram which shows the multi-type artificial intelligence module development system which connected the artificial intelligence module development system in parallel based on one Example of another aspect of this invention. It is a conceptual diagram which shows the multi-type artificial intelligence module development system which connected the artificial intelligence module development system in parallel with one Example of another aspect of this invention in parallel. It is a conceptual diagram which shows the multiple | multiplex type | mold artificial intelligence module development system which connected the artificial intelligence module development system in series according to one Example of another aspect of this invention in series and parallel.

  Hereinafter, embodiments for carrying out the present invention will be described with reference to the drawings. In the following, the range necessary for the description for achieving the object of the present invention is schematically shown, and the range necessary for the description of the relevant part of the present invention will be mainly described. According to a known technique.

  FIG. 1 is a hardware configuration diagram of an artificial intelligence module development system 1 according to an embodiment of the present invention. The hardware configuration of the artificial intelligence module development system includes a central processing unit 1001 that performs control and calculation related to the artificial intelligence module development system, a storage device 1002 that stores data and programs related to the artificial intelligence module development system, and an artificial intelligence module development system. Input device 1003, such as keyboard, touch panel, microphone, switch, sensor, etc., input information and signals from artificial intelligence module development system, display, screen, speaker, earphone, printer, etc. Output device 1004, and an interface device 1005 for connecting the artificial intelligence module development system to an external device or a network.

  As such an artificial intelligence module development system, in addition to various servers, PCs, tablets, and the like may be used as long as they include the above-described configuration, and some systems may not be integrated with all functions. May be a system in which is an external device. Furthermore, these functions may be configured by cloud computing.

  FIG. 2 is a functional configuration diagram of the artificial intelligence module development system according to an embodiment of the present invention. Such a function is configured using a hardware configuration as shown in FIG.

  As shown in the figure, the artificial intelligence module development system 1 according to an embodiment of the present invention functionally includes a data input interface block 10, a data preprocessing block 20, a neural network learning block 30, A data post-processing block 40 and an artificial intelligence module output interface block 50 are provided.

  The data input interface block 10 is an interface for inputting learning data D to the system from the outside. In terms of hardware configuration, the central processing unit 1001 controls the interface device 1005 to realize an interface for data input. To do.

  The data preprocessing block 20 includes a data preprocessing module library 21 and a data preprocessing module 22 for performing preprocessing for converting the data input by the data input interface block 10 into a form necessary for analysis. ing.

  Regarding the data preprocessing block 20, these functions are realized by the data preprocessing module library 21 stored in the storage device 1002 and the data preprocessing module 22 executed by the central processing unit 1001 in terms of hardware configuration. Note that the following functional blocks are also realized by the hardware configuration in the same manner, and thus description thereof is omitted.

  The data preprocessing module library 21 holds a list of analysis purposes, for example, abnormality detection by time series data, abnormality detection by vibration, and further, for example, averaging, FFT (Fourier transform), normalization, two Preprocessing means such as valuation and labeling are provided. Note that details of these means are known techniques, and thus detailed description thereof is omitted here.

  The data preprocessing module 22 selects an analysis purpose, further selects a preprocessing means that matches the input data from the data preprocessing module library 21, and performs preprocessing on the input data.

  The neural network learning block 30 is a pre-learning neural network library 31, a pre-learning neural network setting module 32, and a neural network learning method setting library for analyzing and calculating the data preprocessed by the data preprocessing block 20. 33, a neural network learning execution module 34, and a post-learning neural network setting module 35.

  The pre-learning neural network library 31 includes means such as AE (self-encoder), CNN (convolutional neural network), LSTM (deep learning principle processing), and F-CNN (forward number neural network). I have. Note that details of these means are known techniques, and thus detailed description thereof is omitted here.

  The pre-learning neural network setting module 32 selects means from the pre-learning neural network library 31 that matches the analysis purpose and the preprocessed data.

  The neural network learning method setting library 33 includes means such as the number of learning times and the learning time.

  The neural network learning execution module 34 performs learning on the preprocessed input data according to the means selected from the neural network learning method setting library 33.

  The post-learning neural network setting module 35 sets a neural network obtained as a result of learning and transmits it to the data post-processing module 40.

  The data post-processing block 40 includes a data post-processing module library 41 and a data post-processing module 42 for post-processing output data output by the neural network learning block 30.

  The data post-processing module library 41 includes means such as K-means, filtering, threshold setting, labeling, classification, I-FFT (Fast Fourier Transform).

  The data post-processing module 42 performs post-processing on the learned neural network set by the post-learning neural network setting module 35 and transmitted to the data post-processing block 40 by means selected from the data post-processing module library 41. This is an artificial intelligence module that can be output to the outside.

  The artificial intelligence module output interface block 50 is an interface for outputting the artificial intelligence module 51, which is a product generated by the data post-processing block, to the outside.

  The functions to be stored in each library of the artificial intelligence module development system according to the present invention are not limited to those described above, and any logic functions that are theoretically or implemented in the prior art. Needless to say, these are all included in the technical idea of the present application.

  Next, the operation and action of the artificial intelligence module development system according to the present invention configured as described above will be described.

  FIG. 3 is a flowchart seen from the operator side of the artificial intelligence module development system according to an embodiment of the present invention, and FIGS. 4 to 9 illustrate operations of the artificial intelligence module development system according to the embodiment of the present invention. It is the screen transition figure seen from the person side. 4 to 8, the subdivided drawings indicate that the same screen is displayed by a method such as scrolling, but may be divided into different screens.

  Here, a case where the purpose of analysis by the artificial intelligence module is time-series abnormality detection will be described. First, the analysis purpose is selected (step S01 and FIG. 4A). Specifically, it is selected whether the project is a new project or an existing project. If the project is a new project, it is selected whether the project is newly created or imported from the outside. Here, it is assumed that creation of a new project is selected.

  This step will be described with a hardware configuration. A first screen for prompting the user to input analysis purpose information, which is data defining the analysis purpose, is displayed on the screen of the output device 1004 and the input analysis purpose information is input. Is stored in the storage device 1002 and a first screen generation unit that operates the central processing unit 1001 is included.

  Further, it is selected whether the analysis purpose is time-series data, vibration, torque or spectrum among the abnormality detections (FIGS. 4B to 4C). Here, it is assumed that abnormality detection of time-series data, that is, data that changes in time series such as current, voltage, and temperature is selected.

  Analytical purposes include “time series data abnormality detection / predictive maintenance”, “motor vibration abnormality detection / predictive maintenance”, etc., “image recognition”, “abnormality detection / predictive maintenance function corresponding to abnormal noise” It can be extended to "."

  Next, learning data necessary for the purpose of analysis is registered (step S02 and FIG. 5). Specifically, there are CSV format, AI UNIT, and existing data options. CSV format is generally learning data from the outside. AI UNIT is learning data held by another external unit and existing data is This is the existing data accumulated so far. Here, it is assumed that the CSV format is selected (FIGS. 5A to 5B).

  This step will be described in relation to the hardware configuration. In the artificial intelligence module development system of the present invention, the second screen that prompts the user to input the analysis purpose necessary data necessary for the analysis purpose is displayed on the screen of the output device 1004. A second screen generation unit is provided for operating the central processing unit 1001 so as to store the displayed analysis purpose necessary data displayed and stored in the storage device 1002.

  Then, in accordance with the selection result, necessary data is uploaded and displayed on the display (FIG. 5C).

  Next, a target for achieving the analysis purpose is set (step S03 and FIG. 6). More specifically, regarding abnormality detection, whether to perform normal / abnormal distribution or to determine abnormality type is selected. Here, it is assumed that normal / abnormal distribution is selected (FIG. 6A).

  Further, regarding the determination of the abnormal data, the operator selects whether the abnormal data is designated by the operator or whether the abnormal data is automatically determined. Here, it is assumed that automatic determination is selected (FIG. 6B).

  If this step is explained in relation to the hardware configuration, the artificial intelligence module development system of the present invention will provide the analysis objective achievement target data that defines the analysis objective achievement target, which is a specific target for achieving the analysis objective to the user. And a third screen generator for operating the central processing unit 1001 so as to store the inputted analysis goal achievement target data in the storage device 1002 on the screen of the output device 1004. doing.

  Next, execution of learning suitable for the analysis purpose and goal is instructed using the registered data (step S04 and FIG. 7). As for the number of abnormality detection patterns, 100 patterns are designated as initial values, and if there is no change, learning is performed for 100 patterns (FIG. 7A). Note that the number of patterns can be changed by an operator.

  The progress state during learning is shown as in FIG. 7A, and when completed, it is displayed as shown in FIG. 7B.

  When this step is described in relation to the hardware configuration, the artificial intelligence module development system of the present invention uses the analysis purpose necessary data stored in the storage device 1002 to perform the learning that matches the analysis purpose and the analysis purpose achievement goal as a central processing. A fourth screen generation unit that operates the central processing unit 1001 is displayed so that a fourth screen that accepts a user instruction to be executed by the device 1001 is displayed on the screen of the output device 1004.

  Next, when learning is completed, the artificial intelligence module obtained as a result of learning is confirmed (step S05 and FIG. 8). Specifically, a result such as an abnormality detection rate shown for each detection pattern is displayed and confirmed, and an optimum one is selected (FIG. 8A).

  It is also possible to display detailed data and display a histogram divided by abnormal values (FIGS. 8B to 8C). Here, it is also possible to change by automatically dragging and dropping a threshold value (abnormal value is 0.04 in the figure) determined automatically.

  This step will be described in relation to the hardware configuration. In the artificial intelligence module development system of the present invention, the artificial intelligence module, which is the result of execution of learning by the central processing unit 1001, is displayed on the screen of the output device 1002, and the user can A fifth screen generating unit that operates the central processing unit 1001 so that a fifth screen that accepts input of confirmation information that is a result of confirming the intelligence module is displayed on the screen of the output device 1002; doing.

  Finally, it is instructed to incorporate the learning result including the selected artificial intelligence module into another device (step 06 and FIG. 9). Specifically, a device is specified, and learning results that are optimal for the device are installed.

  Explaining this step in relation to the hardware configuration, the artificial intelligence module development system of the present invention uses instructions for incorporating the learning result including the artificial intelligence module into which the confirmation information is input into another device. A sixth screen generation unit that operates the central processing unit 1001 so that a sixth screen to be received from a person is displayed on the screen of the output device 1004.

  Next, an operation on the artificial intelligence module development system side corresponding to the flowchart from the operator side will be described. FIG. 10 is a timing flowchart showing the steps of the artificial intelligence module development system 1 according to an embodiment of the present invention.

  First, as shown in FIG. 4, an input of a signal (T1) instructing that abnormality of time series data is detected in a new project is received.

  Then, the data preprocessing module 22 selects “time series data abnormality detection” from the data preprocessing module library 21.

  Further, the result of the selection is also transmitted to the neural network learning block 30 and used to select an appropriate means from the pre-learning neural network library 31.

  Next, as shown in FIG. 5, upon receiving a signal (T2) for registering the input data D for learning, for example, a signal instructing to register the data necessary for the analysis purpose by selecting the CSV format. The data preprocessing module 22 issues a signal (T3) for requesting data export to the storage unit 1002 (DS1) in the artificial intelligence module development system or the storage unit 64 (DS2) of the external device, and the input data for learning D is imported into the artificial intelligence module development system 1.

  Next, as shown in FIG. 6, as a goal to achieve the analysis purpose, in response to abnormality detection, normal abnormality distribution is performed, and further, a signal (T4) instructing automatic determination of abnormality data is received, The data preprocessing module 22 preprocesses the imported data by the target and the appropriate preprocessing means selected from the data preprocessing module library 21.

  Subsequently, the pre-learning neural network setting module 32 selects an appropriate pre-learning neural network in the pre-learning neural network library 31 according to the preprocessed data and the selected analysis purpose and target.

  Next, as shown in FIG. 7, upon receiving a signal (T5) instructing learning by designating the number of patterns, the neural network learning execution module 34 performs neural network learning on the selected pre-learning neural network. An appropriate learning method is selected from the method setting library 33, learning is executed, and learning results for each pattern are displayed on the display.

  Next, as shown in FIG. 8, the post-learning neural network setting module 35 receives the signal (T6) instructing selection of the optimal learned neural network from the displayed learning result, and the post-learning neural network setting module 35 To decide.

  If necessary, the data post-processing module 42 receives a signal (T7) for confirming the learning result, and determines a judgment formula (eg, whether the learning result is good or bad) from the data post-processing module library 41 for the learned neural network. , Threshold) can also be selected.

  For example, the normal / abnormal judgment criteria can be changed by dragging and dropping the threshold value of the histogram of the learning result on the screen (FIG. 8C).

  Next, as shown in FIG. 9, in response to a signal (T8) instructing to incorporate the learning result including the selected artificial intelligence module into an external device, the external device 60 is designated and the external device 60 is designated. The preprocessing method, the learned neural network, and the judgment formula are configured and incorporated as the artificial intelligence module 51.

  In general, general-purpose microcomputers (MCUs) used in all products cannot read and use the artificial intelligence module 51 selected this time as it is because of language differences. There is a need for middleware that converts (translates) the language used by 51 into a language that can be read by a general-purpose microcomputer.

  As shown in FIG. 10, the translation middleware 61 (also referred to as a translator) can be provided and provided in the external device to be analyzed. In this way, development of the artificial intelligence module including the analysis target device is possible. The integrated system 2 can be realized.

  The artificial intelligence module 51 that has been configured after the post-processing is automatically transported to a translation middleware 61 prepared in advance in the external device 60 to be analyzed and converted into a format that can be operated by a controller 62 (such as a CPU) in the analysis target. It is converted and transported to the controller 62 as a translated artificial intelligence module 65. The translated artificial intelligence module 65 transported to the controller 62 performs the preprocessing on the sensor data 63 input to the analysis target, and performs discrimination / discrimination by the learned neural network. It is possible to make a determination with respect to the determination formula.

  FIG. 11 is a conceptual diagram showing an image of a translation environment of the artificial intelligence module development integrated system 2 according to an embodiment of the present invention. The translation middleware 61 that translates the application 71 and the artificial intelligence module and the operation system 72 are compiled by the compiler 73. Then, it is output together with the translation middleware 61 as software and ported to the general-purpose microcomputer 74 or the like.

  That is, since the artificial intelligence module 51 is created in a dedicated program language, the translation middleware 61 is compiled together with the application program 71 to be evaluated, and ported to the microcomputer 74 (MPU) in advance, so that general-purpose communication or the like is performed. When the artificial intelligence module 51 transmitted in the above is installed in the MPU 74, the language is automatically converted and the artificial intelligence module 51 can be used as a new application.

  Next, another aspect of the artificial intelligence module development system 1 of the present invention will be described.

  FIG. 12 is a conceptual diagram showing a multiple artificial intelligence module development system in which artificial intelligence module development systems are connected in series according to an embodiment of another aspect of the present invention. As shown in the figure, the artificial intelligence module development system according to the present invention has a configuration in which a plurality of artificial intelligence module development systems are connected in series, that is, an artificial intelligence module development system X231 and an artificial intelligence module development system Y232 are connected in series. A concatenated configuration can also be taken and an artificial intelligence module 290 is generated.

  As an example of such serial connection, the artificial intelligence module developed in the front row is an artificial intelligence module related to image recognition, and the artificial intelligence module developed in the back row is an artificial intelligence development module related to the operation of the manipulator (robot arm). Some cases are assumed.

  In this case, after determining what the target object is and what kind of properties it is by image recognition, the result is used to develop an artificial intelligence module related to the operation of the manipulator. It is possible to realize an artificial intelligence module more suitable for

  FIG. 13 is a conceptual diagram showing a multiple artificial intelligence module development system in which a plurality of artificial intelligence module development systems are connected in series according to an embodiment of another aspect of the present invention. As shown in the figure, the artificial intelligence module development system according to the present invention has a configuration in which a plurality of artificial intelligence module development systems are connected in series, that is, an artificial intelligence module development system X2311 and an artificial intelligence module development system Y2321 are connected in series. The connected row 1 and the row 2 in which the artificial intelligence module development system X2312 and the artificial intelligence module development system Y2322 are connected in series may be connected in parallel. In the column 1, the artificial intelligence module development system X2311 starts from the learning data input step 101, the learning data preprocessing step 102, the learning data input step 103 to the pre-learning neural network, the learning step 104, and the learning completed. The process includes a step 105 in which a neural network is generated and output, and a step 106 in which post-processing is performed on the value output from the learned neural network. The artificial intelligence module development system Y2321 adds an artificial data in addition to the learning data input steps 101 to 106. The process includes an output process 107 of the intelligent module development system 23. A characteristic of column 1 is that the data in step 106 of performing post-processing on the values output from the learned neural network in the front row in series is input to the learning data input step 101 in the rear row in series. On the other hand, in column 2, the artificial intelligence module development system X2312 includes learning data input steps 101 to 105, and the artificial intelligence module development system Y2321 includes learning data input steps 101 to output step 107. Is done. A characteristic in column 2 is that the data in step 105, which is generated and output the learned neural network in the front row in series, is input to the learning data input step 101 in the rear row in series.

  Here, the column 1 and the column 2 include the post-processing step 106 after the learned neural network output step 105 in the artificial intelligence module development system X2311 in the column 1, and in the column 2, the artificial intelligence module development system X2312 The difference is that up to the learned neural network output step 105. In the example described above, this is the case where the image recognition result of the front row is processed until post-processing and then used as an input for developing the artificial intelligence module of the manipulator operation of the rear row, and the manipulator operation of the rear row without post-processing. This is equivalent to the case of using as an input for developing an artificial intelligence module.

  Although it may be easier to develop artificial intelligence modules in the rear row if the post-processing is performed, it may be possible to increase the degree of freedom in developing artificial intelligence modules in the rear row without using post-processing. Any one may be selected according to the performance.

  FIG. 14 is a conceptual diagram showing a multiple artificial intelligence module development system in which artificial intelligence module development systems are connected in parallel according to an embodiment of another aspect of the present invention. As shown in the figure, the artificial intelligence module development system according to the present invention has a configuration in which the artificial intelligence module development system is connected in parallel, that is, the artificial intelligence module development system X231 and the artificial intelligence module development system Y232 are connected in parallel. The artificial intelligence module 290 is generated.

  As an example of such parallel connection, an artificial intelligence module developed by the artificial intelligence module development system X231 is related to speech recognition, and is an intelligent module, and an artificial intelligence module developed by the artificial intelligence module development system Y232 is A case that relates to image recognition is assumed.

  In this case, it is possible to develop an artificial intelligence module that determines in parallel whether or not the utterance source is appropriate as an instruction source by image recognition when determining an instruction such as machine operation by voice recognition. And a highly reliable artificial intelligence module can be realized.

  FIG. 15 is a conceptual diagram showing a multiple artificial intelligence module development system in which a plurality of artificial intelligence module development systems are connected in parallel according to an embodiment of another aspect of the present invention. As shown in the figure, the artificial intelligence module development system according to the present invention has a configuration in which column 1, column 2 and column 3 are connected in parallel. In column 1, the artificial intelligence module development system X2313 includes learning data input step 101 to output step 107, and in column 2, the artificial intelligence module development system Y2323 extends to learning data input step 101 to step 105. In column 3, the artificial intelligence module development system Z2333 is configured to include the learning data input steps 101 to 106, and the artificial intelligence module 293 is generated.

  As shown in this figure, the learned neural network of the artificial intelligence module development system 2323 according to column 2 is generated and output, and the data of step 105 is output from the learned neural network of the artificial intelligence module development system 2313 of column 1. The data of the step 106 for performing post-processing on the values output from the learned neural network of the artificial intelligence module development system 2333 related to the column 3 is input to the step 106 for performing post-processing on the values. It is characterized by being input to the output step 107 of the intelligent module development system 2313.

  FIG. 16 is a conceptual diagram showing a multiple artificial intelligence module development system in which a plurality of artificial intelligence module development systems are connected in series and in parallel according to an embodiment of another aspect of the present invention. As shown in the figure, the artificial intelligence module development system according to the present invention includes an array 1 composed of an artificial intelligence module development system X2314 and an artificial intelligence module development system Y2324 connected in series, and a single artificial intelligence module. The artificial intelligence module 294 is generated by connecting the column 2 configured in the development system Z2334 in parallel.

  As described above, as in another aspect of the artificial intelligence module development system shown in FIGS. 12 to 16, in the multiple artificial intelligence module development system, the artificial intelligence modules obtained through the above steps are connected in series / parallel. It is possible to obtain a more precise result by forming one artificial intelligence module as a whole by connecting one or more to each other.

  In addition, in the multiple artificial intelligence module development system described so far, it is also possible to provide the artificial intelligence module development integrated system 2 including the analysis target device in which the translation middleware 61 is included in the external device 60 to be analyzed. It is.

  The artificial intelligence module developed by the artificial intelligence module development system for anomaly detection and the artificial intelligence module development integration system described above as an example for the manufacturing industry has an abundant pre-learning neural network, on-premises The main features are that an artificial intelligence module can be developed with, and that learning and identification / judgment with external devices are seamlessly connected.

  In addition, for example, the artificial intelligence module generated by the artificial intelligence module development system of the present invention is not only used as an artificial intelligence module development tool for anomaly detection and predictive maintenance for the manufacturing industry, but also succeeds the skill of a skilled person to the successor. Can also help.

  Furthermore, such an artificial intelligence module development system is adapted to specific issues, such as factory automation, abnormality detection, predictive maintenance, automatic operation, automatic design, etc. in the manufacturing industry, optimal team prediction, personnel evaluation, Route optimization in map guidance operations such as retirement prediction, withdrawal prediction, skill prediction, drug discovery in the pharmaceutical industry, medical imaging diagnosis, disease prediction, learning support in the education industry, automatic translation, legal case search and classification in the legal community, etc. Arbitrary terrain, fuel optimization, map creation, intrusion detection, unauthorized site diagnosis, unauthorized access inspection, surveillance camera, etc. in the security field, optimal power consumption in the energy field, solar power generation prediction, remote sensing, etc. Stock price prediction, credit, and various finance related to financial securities, such as automatic design, disaster prevention and mitigation (seismic and vibration control) Target prediction, etc., such as production factories and optimal environment prediction in agriculture, defense, sales forecasts in the marketing relationship between user behavior prediction and trade area analysis and the like, it can be applied also to the entertainment applications · Web API, such as in the service industry.

  As described above, the artificial intelligence module development and integration system of the artificial intelligence module development system according to an embodiment of the present invention can perform the development and operation of artificial intelligence capable of realizing the above-described process in an integrated and seamless manner. it can. In other words, it is an integrated development / operation system generated from the user, the object of analysis, and the functional module.

  As described above, the artificial intelligence module development system according to the present invention can be configured by comparing and selecting a neural network for specific data analysis. Since you can select multiple neural networks, you can select the neural network in the most necessary form for the target data analysis from the library, and output the artificial intelligence module by repeating machine learning The analysis accuracy of this is dramatically improved.

  In addition, the artificial intelligence module development system according to the present invention includes a pre-learning neural network, a learning system for learning learning data, a post-learning neural network, and an algorithm for ranking, selecting, setting, etc. of output data from the neural network. And an artificial intelligence module to be processed and judged by the system.

  The artificial intelligence module of the artificial intelligence module development system according to the present invention is configured to perform further necessary post-processing on the output result that is each machine learning effect, perform a determination evaluation, and generate the result as an effective one. Has been. The artificial intelligence module also takes into account the necessary preprocessing of the data. The artificial intelligence module of the artificial intelligence module development system according to the present invention is configured to aggregate all performance results of machine learning and create performance comparison statistics. The accuracy of judgment / judgment / conclusion by the artificial intelligence module can be dramatically improved.

  The artificial intelligence module development system according to the present invention makes use of machine learning technology including deep learning, which is the ultimate form of machine learning, in various industries that have been difficult to solve with conventional technology. The problem can be solved. In addition to image data analysis, various time series numerical data analysis, vibration sensor analysis, acoustic data analysis, video analysis, and various data analysis can be performed. Based on these abundant analysis results, it becomes possible to quickly propose an excellent solution to an unknown problem.

  As described above, the artificial intelligence module development system according to an embodiment of the present invention is realized as an integrated development / operation system capable of seamlessly performing the development and integration of artificial intelligence capable of realizing all the above-described processes. It is what is done.

  Note that the present invention is not limited to the above-described embodiment, and various modifications can be made without departing from the spirit of the present invention. These are all part of this technical idea.

  According to the present invention, an enormous amount of data is arbitrarily extracted, and it is recognized and determined in the algorithm what kind of judgment the data is necessary for making a decision, and a necessary analysis algorithm is selected and automatically selected. Artificial intelligence that can be called an infrastructure that recognizes various social events, sets thresholds, and makes judgments and judgments using artificial intelligence that automatically repeats analysis, automatic learning and various correction processes The entire system can be provided.

  Therefore, the present invention is not limited to the IT industry, but can be used and applied to all material industries and application industries. Thus, the present application brings great benefits to various industries involving particularly large-scale data processing.

DESCRIPTION OF SYMBOLS 1 Artificial intelligence module development system 2 Artificial intelligence module development integrated system 10 Data input interface block 20 Data pre-processing block 21 Data pre-processing module library 22 Data pre-processing module 30 Neural network learning block 31 Pre-learning neural network library 32 Before learning Neural network setting module 33 Neural network learning method setting library 34 Neural network learning execution module 35 Post-learning neural network setting module 40 Data post-processing block 41 Data post-processing module library 42 Data post-processing module 50 Artificial intelligence module output interface block 51 Artificial intelligence module 60 External device 61 Translation Middleware 62 Controller 63 Sensor data 64 Storage unit 65 Translated artificial intelligence module 1001 Control unit 1002 Storage unit 1003 Input unit 1004 Output unit 1005 Interface unit

Claims (13)

1) selecting an analysis purpose;
2) registering data necessary for the analysis purpose;
3) setting a goal to achieve the analysis purpose;
4) Instructing execution of learning that matches the analysis purpose and the goal using the registered data;
5) confirming the artificial intelligence module obtained as a result of the learning;
6) Instructing to incorporate the learning result including the confirmed artificial intelligence module into another device. A central processing unit that performs various operations, a screen of an output device that outputs information that is a result of the operations performed by the central processing unit, and information that is exchanged between the central processing unit In an artificial intelligence module development system equipped with a storage device,
1) A first screen for prompting the user to input analysis purpose information, which is data defining an analysis purpose, is displayed on the screen, and the center is set so that the input analysis purpose information is stored in the storage device. A first screen generator for operating the arithmetic device;
2) The central processing unit so that a second screen for prompting the user to input analysis-necessary data necessary for the analysis purpose is displayed on the screen and the input analysis-necessary data is stored in the storage device. A second screen generator for operating
3) A third screen for allowing the user to set analysis objective achievement target data that defines the analysis objective achievement target, which is a specific target for achieving the analysis objective, is displayed on the screen and input. A third screen generator for operating the central processing unit to store analysis goal achievement target data in the storage device;
4) A fourth process for accepting an instruction from a user to execute learning suitable for the analysis purpose and the analysis purpose achievement target to the central processing unit using the analysis purpose necessary data stored in the storage device. A fourth screen generating unit that operates the central processing unit so that the screen is displayed on the screen;
5) The artificial intelligence module that is the result of the central processing unit executing the learning is displayed on the screen, and the user receives the confirmation information that is the result of confirming the artificial intelligence module. A fifth screen generator for operating the central processing unit so that the screen of 5 is displayed on the screen;
6) A sixth screen for receiving an instruction from a user to incorporate the learning result including the artificial intelligence module to which the confirmation information is input into another device is displayed on the screen. An artificial intelligence module development system, comprising: a sixth screen generation unit for operating the central processing unit. A) a data input interface block for inputting data;
B) A data preprocessing block comprising a data preprocessing module library and a data preprocessing module for performing preprocessing for converting data input by the data input interface block into a form necessary for analysis;
C) Analyzing the data preprocessed by the data preprocessing block,
Pre-learning neural network library,
Pre-learning neural network setting module,
Neural network learning setting library,
A neural network learning execution module;
A neural network learning block comprising a post-learning neural network setting module;
D) a data post-processing block comprising a data post-processing module library and a data post-processing module for post-processing output data output by the neural network learning block;
E) An artificial intelligence module development system comprising: an artificial intelligence module output interface block for outputting an artificial intelligence module which is a product generated by the data post-processing block.   4. The artificial intelligence module development system according to claim 3, wherein the data pre-processing module causes an analysis purpose to be selected from the analysis purpose-specific menu and transfers the selected analysis purpose to the pre-learning neural network setting module.   The data preprocessing module preprocesses learning data input through the data input interface block by a selected data preprocessing algorithm, and sets the preprocessed data to the pre-learning neural network setting 4. The artificial intelligence module development system according to claim 3, wherein the system is transferred to a module.   The pre-learning neural network setting module selects a pre-learning neural network and a learning method suitable for the purpose from the pre-learning neural network library according to the selected analysis purpose, and transfers them to the neural network learning setting module. The artificial intelligence module development system according to claim 3.   The neural network learning setting module performs learning using the selected pre-learning neural network, the selected learning method, and the preprocessed learning data, and after completion, the learned 4. The artificial intelligence module development system according to claim 3, wherein the neural network, the learning result, and / or the selected data pre-processing algorithm are transferred to the data post-processing module.   4. The artificial intelligence module development system according to claim 3, wherein the data post-processing module selects a data post-processing algorithm including a determination formula from the data post-processing module library and determines the learning result.   The artificial intelligence module includes the selected data pre-processing algorithm, the selected pre-learning neural network, the selected learning method, and / or the selected data post-processing algorithm. The artificial intelligence module development system according to any one of claims 3 to 8.   A multi-type artificial intelligence module development system, wherein a plurality of the artificial intelligence module development systems according to claim 2 are connected in series.   A multi-type artificial intelligence module development system, wherein a plurality of the artificial intelligence module development systems according to any one of claims 2 to 9 are connected in parallel.   A multi-type artificial intelligence module development system, wherein a plurality of the artificial intelligence module development systems according to any one of claims 2 to 9 are connected in series and in parallel. The artificial intelligence module development system according to any one of claims 2 to 12,
An artificial intelligence module development integrated system comprising: translation middleware for causing the artificial intelligence module obtained by the artificial intelligence module development system to correspond to another programming language.

Images