JP2018120488A - Event classification apparatus, event classification program, and fault/defect determination apparatus - Google Patents

Abstract

PROBLEM TO BE SOLVED: To classify a large amount of data at a high speed with high accuracy.SOLUTION: An event classification apparatus includes: chaos scale calculation means 20-1 to 20-n which divide data obtained from sensors 11-1 to 11-N by predetermined time unit in time series, to calculate chaos scale information from the divided data; and pattern mining classification means 40 which performs pattern mining classification on the chaos scale information calculated by the chaos scale calculation means 20-1 to 20-N, to obtain a classification result based on the data.SELECTED DRAWING: Figure 1

Description

  The present invention relates to an event classification device, an event classification program, and a failure / failure determination device configured using the event classification device.

  In the field of product manufacturing, etc., the distribution and characteristics of data obtained from sensors provided in the manufacturing apparatus sometimes fluctuate day by day, and the rules and thresholds for determining failure or failure of the apparatus fluctuate. The data from such sensors tends to have a large number of sensors and increase the amount of data. Therefore, if a determination is made taking into account such fluctuations, it takes time and cost, and it is difficult to make a determination in real time.

  Therefore, instead of using all the data obtained from the sensor, determination is performed on the thinned data. However, when the determination is performed using the thinned data, it is natural that the accuracy is lowered, and in some cases, erroneous determination may occur.

  As a conventional failure prediction system, the one described in Patent Document 1 is known. The system described in Patent Document 1 includes a failure prediction device and a learning server, and the failure prediction device stores time-series sensor data until the failure occurs when a failure occurs in the target device. The learning model is transmitted to the learning server and the learning server predicts the failure of the target device by learning using the time-series sensor data until the failure occurs as the teacher data. Then, the generated learning model is transmitted to the failure prediction apparatus, and the learning model is operated in the failure prediction apparatus.

  The learning model of the failure prediction device operates by predicting the probability of failure based on data obtained from the sensor of the target device. Thereby, the failure of the target device can be detected in advance, and the maintenance quality of the cloud or network can be improved.

  Further, from the viewpoint of data analysis widely, the system described in Patent Document 2 can be cited as a conventional system. This system is realized by a computer processor executing a program stored in a memory. By executing this program, the number of repetitions in each time-series data of the time-series pattern and the appearance frequency, which is the number of time-series data in which the number of repetitions of the time-series pattern is a predetermined number or more, are counted.

  As described above, a time series pattern is extracted that has a predetermined number of repetitions and is equal to or greater than the predetermined number of time series data and equal to or greater than the predetermined number of repetitions. Furthermore, checkpoints are provided at predetermined intervals in each time series data, and the upper limit value of the number of repetitions of the time series pattern in each time series data is calculated at each check point, and the appearance frequency is calculated using this upper limit value. To limit the range of the time-series pattern and time-series data for counting the number of repetitions.

  Furthermore, Patent Document 3 discloses a time-series data abnormality monitoring device that detects a transition of a monitoring target and a sign thereof by a trajectory parallel measure method based on chaos theory. The orbital parallel measure method determines whether a trajectory constructed by embedding time series data in a high-dimensional space is deterministically regular or probabilistic. Specifically, it is performed by calculating the orbit parallel measure as an evaluation value. It is assumed that if the evaluation value is close to 0, it has deterministic regularity, and if it is close to 0.5, the probabilistic property is large.

  Non-Patent Document 1 discloses a new Deep Learning technique for classifying time-series data with high accuracy. The technology of this non-patent document 1 is that (1) time-series data obtained by a sensor is converted into a graphic based on chaos theory, (2) the graphic is converted into a unique vector expression by quantification based on topology, and (3 ) The vector expression obtained above is learned by a convolutional neural network.

Japanese Patent Application Laid-Open No. 2006-173782 Japanese Unexamined Patent Publication No. 2011-123652 JP, 2006-57649, A

Fujitsu Laboratories Ltd., “Developing New Deep Learning Technology to Analyze Time Series Data Highly”, [online], February 16, 2016, PRESS RELEASE, January 23, 2017, Internet <URL: http://pr.fujitsu.com/jp/news/2016/02/16.html>

  However, since the thing of patent document 1 collects sensor data from a plurality of sensors as learning data, generates a learning model, and operates it, it requires a lot of time and cannot perform high-speed classification.

  Further, although Patent Document 2 attempts to improve efficiency by limiting the range of time series patterns and time series data, there is a limit that this method is not applicable to all event classifications.

  The thing of patent document 3 and the thing described in nonpatent literature 1 have the problem that arithmetic processing is complicated and it is not suitable for the thing which performs a high-speed classification.

  The present invention has been made in view of the current state of event classification processing as described above, and an object thereof is to provide an event classification apparatus capable of performing classification based on sensor data from a plurality of sensors at high speed and with high accuracy. It is. Another object of the present invention is to provide an event classification program for realizing the event classification apparatus as described above, and a failure / failure determination apparatus configured using the event classification apparatus.

  The event classification apparatus according to the present invention categorizes data obtained from a plurality of sensors for each predetermined time unit in a time series, and calculates chaos scale information from the divided data, and the chaos scale calculation means includes: Pattern mining classification is performed on the calculated chaos scale information, and pattern mining classification means for obtaining a classification result based on the data is provided.

  In the event classification device according to the present invention, the chaos measure calculating means calculates one series of chaos measure information from time series data obtained from one sensor.

  In the event classification device according to the present invention, the chaos measure calculating means calculates one series of chaos measure information from time series data obtained from N (an integer of 2 or more) sensors.

  In the event classification device according to the present invention, the chaos measure calculating means calculates one piece of chaos measure information from n + 1 (n is an integer of 1 or more) data obtained in time series from each sensor. To do.

  In the event classification device according to the present invention, pattern mining classification model generation means for performing pattern mining learning based on the chaos scale information calculated by the chaos scale calculation means and generating a pattern mining classification model used by the pattern mining classification means; It is characterized by comprising.

  In the event classification device according to the present invention, the pattern mining classification uses classification based on a random forest.

  The event classification program according to the present invention is a chaos measure calculation means for dividing data obtained from a plurality of sensors into predetermined time units in a time series, and calculating chaos measure information from the divided data. The chaos scale information calculated by the means is subjected to pattern mining classification and functions as pattern mining classification means for obtaining a classification result based on the data.

  The event classification program according to the present invention is characterized in that the computer serves as the chaos measure calculating means so as to function to calculate one series of chaos measure information from time series data obtained from one sensor.

  In the event classification program according to the present invention, the computer functions as the chaos measure calculation means so as to calculate one series of chaos measure information from time series data obtained from N (an integer of 2 or more) sensors. It is characterized by.

  In the event classification program according to the present invention, one piece of chaos measure information is calculated from n + 1 (n is an integer of 1 or more) data obtained in time series from each sensor using the computer as the chaos measure calculation means. It is characterized by functioning.

  In the event classification program according to the present invention, the computer further performs pattern mining learning based on the chaos scale information calculated by the chaos scale calculation means, and generates a pattern mining classification model used by the pattern mining classification means. It functions as a classification model generation means.

  In the event classification program according to the present invention, a random forest classification is used in the pattern mining classification.

  The failure / defect determination device according to the present invention is the sensor according to any one of claims 1 to 6, wherein the sensor is provided at a required position of the manufacturing apparatus. And using the event classification device as determination means for determining failure or failure based on data obtained from the sensor.

  According to the present invention, before obtaining a classification result from data obtained from a plurality of sensors, the data obtained from the plurality of sensors is divided in time series for each predetermined time unit, and chaos scale information is calculated from the divided data. As a result, the data is compressed, so that classification based on sensor data from a plurality of sensors can be performed quickly and accurately.

The block diagram which shows 1st Embodiment of the event classification device which concerns on this invention. The block diagram which shows embodiment of the pattern mining classification model production | generation means which is the principal part of the event classification apparatus which concerns on this invention. The flowchart for demonstrating operation | movement of the event classification device which concerns on this invention. The flowchart for demonstrating operation | movement of the event classification device which concerns on this invention. The block diagram which shows 2nd Embodiment of the event classification device which concerns on this invention. The block diagram which shows 3rd Embodiment of the event classification device which concerns on this invention. The block diagram which shows 4th Embodiment of the event classification device which concerns on this invention.

  Embodiments of an event classification apparatus according to the present invention will be described below with reference to the accompanying drawings. In the drawings, the same components are denoted by the same reference numerals, and redundant description is omitted. FIG. 1 shows a block diagram of a first embodiment of an event classification apparatus according to the present invention. In this embodiment, sensors 11-1 to 11-N are provided at required portions of the target apparatus 100 such as a manufacturing apparatus (N: an integer of 2 or more). Each of the sensors 11-1 to 11-N has a configuration for detecting temperature and humidity in the manufacturing process, the speed of the device, the current flowing through the site of the device, the voltage at the detection point, and the like.

  Chaos scale calculating means 20-1 to 20-N are connected to the sensors 11-1 to 11-N. The chaos scale calculating means 20-1 to 20-N classifies data obtained from the corresponding sensors 11-1 to 11-N in a time series for each predetermined time unit, and calculates chaos scale information from the divided data. It is.

  The configurations of the chaos scale calculating means 20-1 to 20-N are the same. Therefore, the detailed configuration of the chaos scale calculation unit 20-1 will be described as a representative, and the detailed configuration of each system in the other chaos scale calculation units 20-2 to 20-N will be omitted. It is assumed that data (sensor log) continuously arrives from the sensor 11-1 and time-series data having a data length of n + 1 is given as shown in (Expression 1).

Then, the section I including the above _xt is divided into m sections. Here, if the divided section is X _i (i = 1, 2,..., M), the section I can be expressed by the following (Expression 2).

Is then calculated as _{x t} ∈X _i become a probability distribution _{p (i), x t ∈X} i, x t + 1 ∈X j become probability distribution p (i, j) a. The result is the following (Equation 3).

Next, the chaos scale information H _CD is obtained by the following (formula 4) using the above-described p (i) and p (i, j).

It should be noted that the maximum value of the chaos scale information _{H CD} is a logm.

A pattern mining classification means 40 is arranged at the subsequent stage of the chaos scale calculation means 20-1 to 20-N. The chaos scale information obtained as described above by the chaos scale calculating means 20-1 to 20-N is defined as H _{CD-1 to} H _CD-N . The pattern mining classification means 40 performs pattern mining classification on the H _{CD-1 to} H _CD-N obtained by the chaos scale calculation means 20-1 to 20-N, and the chaos scale calculation means 20-1 to 20- A classification result based on the N classification results is obtained. In this pattern mining classification, for example, classification by random forest can be used.

FIG. 2 shows a detailed configuration diagram of the pattern mining classification means 40. The pattern mining classification unit 40 includes a data set generation unit 41A, a pattern mining classification model generation unit 41, and a pattern mining determination unit 42. A data set generation unit 41A that generates a learning data set from the calculation results of the chaos scale calculation units 20-1 to 20-N is provided in the preceding stage of the pattern mining classification model generation unit 41. Here, in order to classify the chaos scale information H _{CD-1 to} H _CD-N calculated in the calculation of chaos scale information into a and b, the data sets of the chaos scale information to be classified into a and the chaos scale information to be classified into b. Is generated.

  In the chaos scale information (data) shown in the data set generation unit 41A, the vertical direction corresponds to the sensor. Here, an example in which the number N of sensors is four is shown. In the chaos scale information (data) shown in the data set generation unit 41A, the horizontal direction corresponds to time. In the above example, one piece of chaos scale information is calculated for the data length n + 1. For this reason, here, it is shown that five pieces of chaos scale information are calculated over time.

  The data set generated by the data set generation unit 41A is given to the pattern mining classification model generation means 41. The pattern mining classification model generation means 41 performs pattern mining learning using the data set and generates a pattern mining classification model. This pattern mining classification model is a classification model used by the pattern mining classification means 40 for classification of data given from the chaos scale calculation means 20-1 to 20-N.

The pattern mining determination unit 42 is arranged after the chaos scale calculation means 20-1 to 20-N, and the plurality of sensors 11-1 to 11-n calculated by the chaos scale calculation means 20-1 to 20-N. Pattern mining classification is performed on the chaos scale information H _{CD-1 to} H _CD-N obtained for each data. That is, the pattern mining determination unit 42 performs pattern mining classification using the pattern mining classification model generated by the pattern mining classification model generation means 41, and obtains a result. The result of this pattern mining classification can be sent to a display device or printing device (not shown) for output.

  The above event classification device can be constituted by a server computer or the like that takes the sensors 11-1 to 11-N into the input port, digitizes them, and performs computer processing. Each means executes the event classification program by the server computer Can be realized.

  3 and 4 show flowcharts corresponding to the event classification program. The operation of the event classification device will be described with reference to this flowchart. FIG. 3 shows a process for generating a pattern mining classification model. When the process is started, data from the sensors 11-1 to 11-N are taken in, the data for each sensor is divided into time units in predetermined time units, and chaos scale information is calculated from the divided data (S11). ).

  Next, the chaos scale information for teachers is labeled with events (S12). The event here may be the temperature and humidity in the manufacturing process, the speed of the device, the current flowing through the site of the device, the voltage at the detection point, and the like as the sensing targets of the sensors 11-1 to 11-N.

  Next, for each label, a data set is generated from the chaos scale information, and pattern mining learning is performed using the generated data set (S13). And the said pattern mining learning is performed and the pattern mining classification model is produced | generated as a result (S14).

  FIG. 4 shows the calculation processing of the chaos scale information to be classified and the actual classification processing using the pattern mining classification model generated by the program of the flowchart of FIG. When the process starts, data from the sensors 11-1 to 11-N is taken in, the data for each sensor is divided into time units in a predetermined time unit, and chaos scale information is calculated from the divided data (S21). ).

  Next, classification is performed by performing classification using the pattern mining classification model by integrating the N series of chaos scale information (S22). In step S22, as shown in FIG. 2, N-series chaos scale information is handled as one, and a process of classifying it into one of a and b is performed for each predetermined time unit.

  As described above, in this embodiment, chaos scale information is calculated for each series, and the chaos scale information for each series is collected into one and classified using the pattern mining classification model. By calculating, data for each series can be compressed, and the processing speed up to classification can be increased.

  Next, a second embodiment of the event classification device according to the present invention will be described. FIG. 5 shows a block diagram of a second embodiment of the event classification apparatus according to the present invention. Also in this embodiment, the sensors 11-1 to 11-N are provided at required portions of the target apparatus 100 such as a manufacturing apparatus. The outputs of the sensors 11-1 and 11-2 are given to the bivariate chaos scale calculating means 30-1, the outputs of the sensors 11-3 and 11-4 are given to the bivariate chaos scale calculating means 30-2, The outputs of the sensors 11- (N-1) and 11-N are given to the bivariate chaos scale calculation means 30-s.

  The configuration of the chaos scale calculation means 30-1 to 30-s is the same. Therefore, the detailed configuration of the chaos scale calculation unit 30-1 will be described as a representative, and the detailed configuration of each system in the other chaos scale calculation units 30-2 to 30-s will be omitted. It is assumed that data (sensor log) arrives continuously from the sensors 11-1 and 11-2, and time-series data having a data length of n + 1 is given as shown in (Expression 5).

In the above, x _t and y _t are data at the same time when the values of t are the same. Therefore, the section I _x including x _t is divided into m (x) sections, and the section I _y including y _t is divided into m (y) sections. Here, if the divided sections are X _i (i = 1, 2,..., M (x)) and Y _i (i = 1, 2,..., M (y)), the above section I Can be expressed by the following (formula 6).

Next, at time t in the _{x t ∈X i, y t ∈Y} j become joint probability distribution p (i, j) and the time t and time t + 1 for _{x t ∈X i, y t ∈Y} j, x t + 1 A simultaneous probability distribution p (i, j, u, v) that satisfies εX _u , y _{t + 1} εY _v is calculated. The result is the following (Equation 7).

Next, the chaos scale information H _CD is obtained by the following (Equation 8) using the above p (i, j) and p (i, j, u, v).

The maximum value of the chaotic scale information _{H CD} by the above calculation is {logm (x) × m ( y)}.

A pattern mining classification unit 40B is arranged following the chaos scale calculation unit 30-1 to 30-s. The chaos scale information obtained as described above by the chaos scale calculating means 30-1 to 30-s is defined as H _{CD-1 to} H _CD-s . The pattern mining classification means 40B performs pattern mining classification on the H _{CD-1 to} H _CD-s obtained by the chaos scale calculation means 30-1 to 30-s, and the chaos scale calculation means 30-1 to 30- A classification result based on the classification result of s is obtained. In this pattern mining classification, for example, classification by random forest can be used. Since the pattern mining classification means 40B in the second embodiment performs the same processing as the pattern mining classification means 40 of the first embodiment except that the input is an s series, detailed description thereof is omitted.

  The event classification apparatus according to the second embodiment described above can be configured by a server computer or the like that takes the sensors 11-1 to 11-N into an input port, digitizes them, and performs computer processing. This can be realized by the server computer executing the event classification program. Since the flowchart corresponding to the event classification program according to the second embodiment is basically the same as that of the first embodiment, the description thereof is omitted.

  Next, a third embodiment of the event classification device according to the present invention will be described. FIG. 6 shows a block diagram of a third embodiment of the event classification apparatus according to the present invention. Also in this embodiment, the sensors 11-1 to 11-N are provided at required portions of the target apparatus 100 such as a manufacturing apparatus. The outputs of the sensors 11-1, 11-2, and 11-3 are given to the trivariate chaos scale calculation means 50-1, and the outputs of the sensors 11-4, 11-5, and 11-6 are given to the trivariate chaos scale calculation means. 50-2, and the outputs of the sensors 11- (N-2), 11- (N-1), and 11-N are given to the trivariate chaos scale calculating means 50-r.

  The configurations of the chaos scale calculating means 50-1 to 50-r are the same. Therefore, the detailed configuration of the chaos scale calculation unit 50-1 will be described as a representative, and the detailed configuration of each system in the other chaos scale calculation units 50-2 to 50-r will be omitted. It is assumed that data (sensor log) arrives continuously from the sensors 11-1, 11-2, and 11-3, and time-series data with a data length of n + 1 is given as shown in (Equation 9).

In the above, x _t and y _t and z _t is when the value of t is the same is the data of the same time. Therefore, the section I _x including x _t is divided into m (x) sections, the section I _y including y _t is divided into m (y) sections, and a section including z _t is included. Divide I _z into m (z) intervals. Here, X _i (i = 1, 2,..., M (x)), Y _i (i = 1, 2,..., M (y)) and Z _i (i = 1, 2,..., M (z)), the sections I _x , I _y , and I _z can be expressed by the following (Equation 10).

Then, _{x t} ∈X _i at time _{t, y t ∈Y j, z} t ∈Z k become the joint probability distribution p (i, j, k) and the time t and time t + 1 for _{x t} ∈X _{i, y t ∈Y j, z t ∈Z} k, x t + 1 ∈X u, y t + 1 ∈Y v, z t + 1 ∈Z w become joint probability distribution p (i, j, k, u, v, w) Is calculated. The result is the following (Equation 11).

Next, the chaos scale information H _CD is obtained by the following (formula 12) using the above-described p (i, j) and p (i, j, u, v).

The maximum value of the chaotic scale information _{H CD} by the above calculation is {logm (x) × m ( y)}.

A pattern mining classification unit 40C is arranged at the subsequent stage of the chaos scale calculation units 50-1 to 50-r. The chaos scale information obtained as described above by the chaos scale calculating means 50-1 to 50-r is defined as H _{CD-1 to} H _CD-r . The pattern mining classification means 40C performs pattern mining classification on the H _{CD-1 to} H _CD-r obtained by the chaos scale calculation means 50-1 to 50-r, and the chaos scale calculation means 50-1 to 50- A classification result based on the classification result of r is obtained. In this pattern mining classification, for example, classification by random forest can be used. The pattern mining classification means 40C in the third embodiment performs the same processing as the pattern mining classification means 40 in the first embodiment except that the input is an r series, and therefore detailed description thereof is omitted.

  The event classification apparatus according to the third embodiment described above can be configured by a server computer or the like that takes the sensors 11-1 to 11-N into the input ports, digitizes them, and performs computer processing. This can be realized by the server computer executing the event classification program. Since the flowchart corresponding to the event classification program according to the third embodiment is basically the same as that of the first embodiment, the description thereof is omitted.

  Next, a fourth embodiment of the event classification device according to the present invention will be described. FIG. 7 shows a block diagram of a fourth embodiment of the event classification apparatus according to the present invention. Also in this embodiment, the sensors 11-1 to 11-N are provided at required portions of the target apparatus 100 such as a manufacturing apparatus. The outputs of the sensors 11-1 to 11-N are given to a chaos scale calculating means 60 of N (N ≧ 4) variables. In addition, since this embodiment is a description continuing from the first embodiment to the third embodiment, and is for the case of four or more variables, N is set to four or more. The explanation is valid when N is 1 or more. Therefore, in the case of 1 to 3 variables, it goes without saying that the processing according to the present embodiment may be adopted.

  Data (sensor log) continuously arrives at the chaos scale calculating means 60 from the sensors 11-1 to 11-N, and as shown in (Equation 13), the data length is set for each sensor 11-1 to 11-N. Assume that n + 1 time-series data is given.

Then i (i = 1,2, ···, N) and divides the interval I _i that contains th time-series data to the m (i) number of intervals, the interval I _i is the following (Equation 14 ).

Here, when a section number including the i-th sequence value x _{i, j} at a certain time t (t = 1, 2,..., N + 1) is represented as L (i), the sequence value x _{i, j} The section of (Expression 14) has the following relationship (Expression 15).

  Here, L (i) has the following relationship (Formula 16).

  Considering that L (i) has the relationship of (Equation 16), the following transformation of (Equation 17) is considered for all sequences.

  From the above (Expression 17), it can be seen that u has the following relationship (Expression 18).

Said u is because the value at a certain time t, which all the time t = 1, 2, · · ·, for n + 1 thinking, generate {u _1, u _2, · · ·, u _{n + 1}} sequence To do. By such an operation, the N-variate data has been replaced with the univariate integer value series shown in (Equation 19).

Thereafter, processing may be performed in the same manner as in the case of univariate. _That, u t = i and (1 ≦ i ≦ M) and comprising a probability distribution _{p (i), u t =} i and _{u t + 1 = j (1} ≦ i, j ≦ M) and comprising a probability distribution p (i, j ) Is calculated by the following (formula 20).

Next, the chaos scale information H _CD is obtained by the following (formula 21) using the above-described p (i) and p (i, j).

The maximum value of the chaotic scale information H _CD by the calculation is the following (Equation 22).

A pattern mining classification unit 40D is arranged at the subsequent stage of the chaos scale calculation unit 60. The chaos degree calculation means 60 the chaos degree information obtained as described above to H _CD. Pattern Mining classifying unit 40D, for H _CD obtained by chaos degree calculation means 60 performs pattern mining classification to thereby obtain a classification result based on the classification result of the chaos degree calculation means 60. In this pattern mining classification, for example, classification by random forest can be used. Since the pattern mining classification means 40D in the fourth embodiment performs the same processing as the pattern mining classification means 40 of the first embodiment except that the input is one series, detailed description thereof is omitted.

  The event classification apparatus according to the fourth embodiment described above can be configured by a server computer or the like that takes the sensors 11-1 to 11-N into an input port, digitizes them, and performs computer processing. This can be realized by the server computer executing the event classification program. Since the flowchart corresponding to the event classification program according to the fourth embodiment is basically the same as that of the first embodiment, the description thereof is omitted.

  Although the embodiment of the event classification device has been described above, it can be a failure / failure determination device. In this case, the sensors 11-1 to 11-N are sensors provided at required positions of the manufacturing apparatus, the other configurations are the same as those of the event classification apparatus of the embodiment, and the sensors 11-1 Based on the data obtained from ˜11-N, the event classification device is used as a determination means for determining failure or failure of the manufacturing device. Accordingly, it is possible to provide a failure / failure determination device that performs failure / failure determination of the manufacturing apparatus (or a product manufactured by the manufacturing apparatus) with less error by analysis using a large amount of data.

  In each embodiment, the same variable chaos scale calculation means is used. For example, when the number of sensors is six (in the case of six variables), the chaos scale of the first embodiment (univariate) is used. One calculation means, one chaos scale calculation means of the second embodiment (bivariate), one chaos scale calculation means of the third embodiment (trivariate), and so on. You may mix and use a scale calculation means. Moreover, it is also possible to select an appropriate number of chaos scale calculation means with different variables, and to receive and classify the data of a predetermined number of sensors in total.

11-1 to 11-N Sensors 20-1 to 20-N Chaos Scale Calculation Units 30-1 to 30-s Chaos Scale Calculation Units 40, 40A, 40B, 40C. 40D Pattern mining classification means 41 Pattern mining classification model generation means 41A Data set generation section 42 Pattern mining determination sections 50-1 to 50-r Chaos scale calculation means 60 Chaos scale calculation means 100 Target device

Claims (13)

Chaos scale calculation means for dividing data obtained from a plurality of sensors into time series for each predetermined time unit, and calculating chaos scale information from the divided data;
Pattern mining classification means for performing pattern mining classification on the chaos scale information calculated by the chaos scale calculation means, and obtaining a classification result based on the data; and
An event classification apparatus comprising:   2. The event classification apparatus according to claim 1, wherein the chaos measure calculating means calculates one series of chaos measure information from time series data obtained from one sensor.   2. The event classification apparatus according to claim 1, wherein the chaos measure calculating means calculates one series of chaos measure information from time series data obtained from N (an integer of 2 or more) sensors.   4. The chaos scale calculation means calculates one piece of chaos scale information from n + 1 (n: integer greater than or equal to 1) data obtained in time series from each sensor. The event classification apparatus according to item 1. Pattern mining classification model generation means for performing pattern mining learning based on chaos scale information calculated by the chaos scale calculation means, and generating a pattern mining classification model used by the pattern mining classification means;
The event classification device according to any one of claims 1 to 4, further comprising:   6. The event classification apparatus according to claim 1, wherein classification by random forest is used in the pattern mining classification. A chaos scale calculating means for dividing data obtained from a plurality of sensors into a predetermined time unit in time series and calculating chaos scale information from the divided data;
Pattern mining classification means for performing pattern mining classification on the chaos scale information calculated by the chaos scale calculation means, and obtaining a classification result based on the data,
Event classification program characterized by functioning as   8. The event classification program according to claim 7, wherein the computer functions as the chaos measure calculating means so as to calculate one series of chaos measure information from time series data obtained from one sensor.   8. The computer according to claim 7, wherein the computer functions as the chaos scale calculation unit so as to calculate one series of chaos scale information from time series data obtained from N (an integer of 2 or more) sensors. Event classification program.   5. The computer according to claim 1, wherein the chaos scale calculating means functions to calculate one piece of chaos scale information from n + 1 (n: integer greater than or equal to 1) data obtained in time series from each sensor. Item 10. The event classification program according to any one of Items 7 to 9. Said computer further
Pattern mining classification model generation means for performing pattern mining learning based on the chaos scale information calculated by the chaos scale calculation means and generating a pattern mining classification model used by the pattern mining classification means;
The event classification program according to any one of claims 7 to 10, wherein   12. The event classification program according to claim 7, wherein classification by random forest is used in the pattern mining classification. The event classification device according to any one of claims 1 to 6, wherein the sensor according to any one of claims 1 to 6 is a sensor provided at a required position of a manufacturing device,
A failure / failure determination device, wherein the event classification device is used as determination means for determining failure or failure based on data obtained from the sensor.

Images