KR101995026B1 - System and Method for State Diagnosis and Cause Analysis - Google Patents

System and Method for State Diagnosis and Cause Analysis Download PDF

Info

Publication number
KR101995026B1
KR101995026B1 KR1020170061210A KR20170061210A KR101995026B1 KR 101995026 B1 KR101995026 B1 KR 101995026B1 KR 1020170061210 A KR1020170061210 A KR 1020170061210A KR 20170061210 A KR20170061210 A KR 20170061210A KR 101995026 B1 KR101995026 B1 KR 101995026B1
Authority
KR
South Korea
Prior art keywords
sensor data
target system
state
sensor
abnormal state
Prior art date
Application number
KR1020170061210A
Other languages
Korean (ko)
Other versions
KR20180126311A (en
Inventor
구형일
Original Assignee
아주대학교산학협력단
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 아주대학교산학협력단 filed Critical 아주대학교산학협력단
Priority to KR1020170061210A priority Critical patent/KR101995026B1/en
Publication of KR20180126311A publication Critical patent/KR20180126311A/en
Application granted granted Critical
Publication of KR101995026B1 publication Critical patent/KR101995026B1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning

Abstract

The present invention relates to a system and method for diagnosing a state and causal analysis of a target system, and comprising: a state diagnosis device for diagnosing a state of the target system based on sensor data received from a plurality of sensors of the target system, and an abnormal state of the target system. (Abnormal State) Diagnosis includes a cause analysis device that calculates the relevance of each sensor to an abnormal state diagnosis result, so that it is possible to accurately diagnose the state of the target system, warn of the risk of an abnormal state, and perform maintenance in advance. To increase the safety of the target system.

Description

System and Method for State Diagnosis and Cause Analysis

The present invention relates to a system and method for diagnosing a condition and a cause analysis system of a target system. More particularly, the present invention applies a machine learning-based method to sensor data obtained from a sensor installed in a target system to be diagnosed. The present invention relates to a system and method for diagnosing and causing a condition of a target system for diagnosing abnormal conditions of the abnormal condition and enabling cause analysis of the abnormal condition.

Neural networks can model general input and output relationships and therefore have a variety of applications. A general situation of predicting (inferring) information of interest via sensor data is shown in FIG. 1. In the case of a human body, a bio signal is input, and thus, it is possible to estimate (estimate) the type of disease or the progress of the bottle, and in the case of semiconductor production, it is possible to grasp the relationship between various sensor conditions and yields. The neural network used here may be a deep neural network (DNN), a convolutional neural network (CNN), or a recurrent neural network (RNN) to process sensor data that changes over time.

The technology using neural networks has been applied to many fields and shows good performance.

However, the conventional neural network is focused on the estimation result and the diagnosis result, so that it is difficult to grasp the causal relationship between the input data and the diagnosis result.

In addition, in order to learn complex relations, a large amount of training data is required at every label, but a high cost is required to acquire the training data, and sometimes there is a problem that it is impossible to obtain the data in a realistic manner. For example, when sensor data (normal operation data) of a target system operating in a normal state is relatively easy to obtain, but sensor data (abnormal operation data) of a target system operating in an abnormal state is impossible to obtain. There is.

Prior Art 1: Korean Patent Registration No. 10-1677358 (January 20, 2006)

An object of the present invention is to diagnose the state of a target system for diagnosing the state of the target system by applying a machine learning (machine learning) -based method to a large amount of sensor data generated from various sensors installed in the target system to be diagnosed. It is to provide a cause analysis system and method.

Another object of the present invention is to provide a system and method for diagnosing a cause of a target system and a cause analysis system for enabling cause analysis of a diagnosis result when an abnormal state of the target system is diagnosed.

It is still another object of the present invention to provide a system and method for diagnosing and causing a condition of a target system capable of determining an abnormal state of the target system even in asymmetric data caused by a large number of normal operation data and a small number of abnormal operation data. .

On the other hand, the technical problem to be achieved by the present invention is not limited to the above-mentioned technical problem, various technical problems can be included within the scope apparent to those skilled in the art from the following description.

According to an aspect of the present invention for solving the above problems, a state diagnosis device for diagnosing a state of the target system based on sensor data received from a plurality of sensors of the target system, the abnormal state of the target system (Abnormal State At the time of diagnosis, a condition diagnosis and cause analysis system of a target system including a cause analysis device that calculates a degree of relevance of each sensor to an abnormal state diagnosis result may be provided.

If the amount of sensor data of the target system operating in the normal state and the sensor data of the target system operating in the abnormal state is asymmetric, the state diagnosis apparatus may detect the sensor data after a predetermined time of the received sensor data. And predict the abnormal state of the target system by comparing the sensor data received from the plurality of sensors after the preset time with the predicted sensor data.

The cause analyzing apparatus calculates a relationship between the sensor data change amount of each sensor of the target system operating in the normal state and the abnormal state signal change amount at the time of diagnosing the abnormal state, thereby determining the relevance of each sensor to the abnormal state diagnosis. You can print

The cause analyzing apparatus may further include a relevance calculator configured to calculate a relevance of each sensor to an abnormal state diagnosis result of a target system including a plurality of sensors, wherein the relevance calculator includes a normal state ( The relationship between the sensor data change amount of each sensor of the target system operating in the normal state and the abnormal state signal change amount at the time of diagnosing the abnormal state can be calculated, and the degree of relevance of each sensor to the abnormal state diagnosis can be output.

In addition, the relevance calculator may output the relevance of each sensor to an abnormal state diagnosis using at least one of a back propagation algorithm and an inverse filtering.

The state diagnosis apparatus may include a predictor configured to predict sensor data after a preset time of sensor data received from a plurality of sensors of a target system, and predict the sensor data received from the plurality of sensors after the preset time. In comparison with the sensor data, a comparison unit for diagnosing an abnormal state of the target system may be included.

According to another aspect of the invention, the step of diagnosing the state of the target system based on the sensor data received from the plurality of sensors of the target system, the cause analysis device, the cause analysis device is an abnormal state (Abnormal State) of the target system In the diagnosis, a method of diagnosing a condition and analyzing a cause of a target system, including calculating a relevance of each sensor with respect to an abnormal state diagnosis result, may be provided.

In this case, when the amount of sensor data of the target system operating in the normal state and the sensor data of the target system operating in the abnormal state is asymmetric, the state diagnosis apparatus may determine whether the sensor data after the preset time of the input sensor data is asymmetric. The abnormal state of the target system may be diagnosed by predicting sensor data and comparing sensor data received from the plurality of sensors after the preset time with the predicted sensor data.

The cause analysis device calculates a relation between sensor data change amount of each sensor of a target system operating in a normal state and an abnormal state signal change amount when diagnosing an abnormal state, and calculates a relevance of each sensor for an abnormal state diagnosis. can do. In addition, the relevance may be calculated using at least one of a back propagation algorithm and inverse filtering.

According to the present invention, by applying a machine learning-based method to a large amount of sensor data generated from various sensors of the target system to be diagnosed, it is possible to accurately diagnose the state of the target system, and to warn and maintain the risk of abnormal state occurrence in advance You can increase the safety of the target system.

In addition, the sensor data of the target system may be used to diagnose an abnormal condition, and may also help cause analysis of a diagnosis result.

In addition, it is possible to increase the maintenance efficiency by quantitatively obtaining the relevance of each sensor with respect to the diagnosis result of the target system and checking the sensor having a high influence on the abnormal state during maintenance.

In addition, the abnormal state of the target system can be diagnosed even in asymmetric data caused by a large number of normal operation data and a small number of abnormal operation data.

On the other hand, the effects of the present invention is not limited to the above-mentioned effects, various effects may be included within the scope apparent to those skilled in the art from the following description.

1 is a diagram illustrating an apparatus for inferring information of interest from sensor data.
2 is a view for explaining the concept of the status diagnosis and cause analysis of the target system according to the present invention.
3 is a diagram illustrating a system for diagnosis and cause analysis of a target system according to the present invention.
4 is a block diagram schematically illustrating a configuration of a state diagnosis apparatus for diagnosing a state of a target system in a situation where the amount of normal operation data and abnormal operation data according to the present invention is asymmetric.
5 is an exemplary diagram for describing an operation of the state diagnosis apparatus illustrated in FIG. 4.
6 is a block diagram schematically showing the configuration of a cause analysis apparatus according to the present invention.
7 is a flowchart illustrating a method for diagnosing a condition and analyzing a cause of a target system according to the present invention.
8 and 9 are exemplary diagrams for explaining a method for diagnosing a condition and analyzing a cause of a target system according to the present invention.

Details of the above-described objects and technical configurations of the present invention and the effects thereof according to the present invention will be more clearly understood by the following detailed description based on the accompanying drawings.

Hereinafter, with reference to the accompanying drawings will be described in detail the 'system diagnosis and cause analysis system and method of the target system' according to the present invention. The described embodiments are provided to enable those skilled in the art to easily understand the technical spirit of the present invention, and the present invention is not limited thereto. In addition, matters represented in the accompanying drawings may be different from the form actually embodied in the schematic drawings in order to easily explain the embodiments of the present invention.

In addition, each component expressed below is only an example for implementing this invention. Thus, other implementations may be used in other implementations of the invention without departing from the spirit and scope of the invention. In addition, each component may be implemented by purely hardware or software configurations, but may also be implemented by a combination of various hardware and software components that perform the same function. In addition, two or more components may be implemented together by one hardware or software.

In addition, the expression "comprising" certain components merely refers to the presence of the components as an 'open' expression, and should not be understood as excluding additional components.

2 is a view for explaining the concept of the status diagnosis and cause analysis of the target system according to the present invention.

Referring to FIG. 2, sensor data X1 (t), X2 (t), Xn-1 (t), and Xn (t) are received from sensors mounted in a target system to be diagnosed. The sensor data is applied to machine learning to diagnose whether the target system is normal. When the target system is diagnosed as an abnormal state, the relevance of each sensor to the diagnosis result is obtained, and cause analysis is performed in order of the most relevant sensor. In other words, the sensor sensor quantifies and expresses how much the input sensor data contributed to the final decision, and visually displays a region of high relevance.

A system for diagnosing a condition and analyzing a cause of the target system will be described with reference to FIG. 3.

3 is a diagram illustrating a system for diagnosis and cause analysis of a target system according to the present invention.

Referring to FIG. 3, the state diagnosis and cause analysis system of the target system includes a state diagnosis device 200 and a cause analysis device 300.

The state diagnosis apparatus 200 receives sensor data from a plurality of sensors 100 installed in a target system to be diagnosed, and diagnoses whether the target system is normal by using the sensor data. Here, the target system is not only a smart device equipped with a sensor, for example, smart devices such as smart TVs, smartphones, tablet PCs, wearable computers, but also sensors, such as cars, drones, ships, toys, etc. A variety of devices equipped with computing devices are referred to generically. The sensor is used to detect data necessary for identifying the operation status information of the target system. The sensor is various types of sensors such as an optical sensor, a sound sensor, an optical sensor, an image sensor, a gyro sensor, a fingerprint sensor, and a motion sensor. It can be different depending on the type of.

The state diagnosis apparatus 200 diagnoses the target system based on the field data, thereby increasing the reliability of the diagnosis result.

Meanwhile, in order for the state diagnosis apparatus 200 to receive sensor data and diagnose an abnormal state of the target system, a large amount of abnormal operation data is required. However, although sensor data (normal operation data) of a target system operating in a normal state is relatively easy to acquire, sensor data (abnormal operation data) of a target system operating in an abnormal state may not be obtained. . In this case, the state diagnosis apparatus 200 should be able to diagnose an abnormal state of the target system using at least labeled sensor data. That is, the state diagnosis apparatus 200 should be able to diagnose the abnormal state of the target system even in a situation where the amount of sensor data is asymmetric with a large number of normal operation data and a few abnormal operation data.

As such, when the amount of sensor data of the target system operating in the normal state and the sensor data of the target system operating in the abnormal state is asymmetric, the state diagnosis apparatus 200 may perform machine learning on the sensor data of the target system operating in the normal state. (Eg, deep learning) is applied to predict sensor data after a preset time, and compare sensor data received from a plurality of sensors 100 after the preset time with the predicted sensor data, thereby causing abnormality of the target system. Abnormal State can be diagnosed. Here, the sensor data may include an image, time-series data that changes with time, an image sequence, three-dimensional data, and multimodal data.

A detailed description of the state diagnosis apparatus 200 for diagnosing an abnormal state in a situation where the amounts of the normal operation data and the abnormal operation data are asymmetric will be given with reference to FIG. 4.

In order to secure the reliability of the diagnosis result, the state diagnosis apparatus 200 needs to have the sensor data used for learning when the operation state of the target system is normal. For this reason, the state diagnosis apparatus 200 obtains a guarantee such as a maintenance record that no failure has occurred in the target system during the period in which the sensor data is input, and treats the input sensor data as being normal.

The cause analyzing apparatus 300 obtains a degree of relevance of each sensor to an abnormal state diagnosis result when the state diagnosis apparatus 200 diagnoses an abnormal state of a target system, and analyzes the cause of the abnormal state. That is, the cause analyzing apparatus 300 calculates a relationship between the sensor data change amount of each sensor of the target system operating in the normal state and the sensor data change amount predicted at the time of diagnosing the abnormal state. Output

The user can proceed with the cause analysis in order of the relevant sensors, and can perform repairs such as replacement of parts based on the relatedness. In the case of healthcare, it can give priority to what the staff needs to review.

A detailed description of the cause analyzing apparatus 300 will be described with reference to FIG. 6.

Meanwhile, although the state diagnosis apparatus 200 and the cause analysis apparatus 300 are implemented as separate apparatuses, the state diagnosis apparatus 200 and the cause analysis apparatus 300 may be implemented as one apparatus.

The system configured as described above may help to diagnose (or predict) the abnormal state of the target system and to analyze the cause. For example, if there is a wearable device that receives a plurality of bio-signals as inputs, the physical abnormal state of the user may be determined from the bio-signal information, and an alarm or a request may be requested when determining the abnormal state. In addition, by providing quantitatively what causes the abnormal state at present, it can be used for abnormal state cause analysis and resolution. Specifically, by providing which sensor is most relevant to the abnormal state, it can be used to analyze and resolve the cause of the abnormal state. In the case of the manufacturing industry as well, it is possible to diagnose and alert the problem of the production line by itself, and to help identify the list of sensors to consider in order to solve the problem.

4 is a block diagram schematically illustrating a configuration of a state diagnosis apparatus for diagnosing a state of a target system in a situation where the amounts of normal operation data and abnormal operation data are asymmetrical, and FIG. 5 is a state diagnosis shown in FIG. 4. An illustration for explaining the operation of the device.

Referring to FIG. 4, the state diagnosis apparatus 200 for diagnosing a state of a target system in a situation where the amounts of normal operation data and abnormal operation data are asymmetric includes a predictor 210 and a comparator 220.

The predictor 210 predicts sensor data after a preset time of the sensor data input from the plurality of sensors of the target system in normal operation. That is, the predictor 210 predicts sensor data after a preset time by applying machine learning to the received sensor data.

As such, the prediction unit 210 learns to predict future sensor data using sensor data received on a machine learning basis. In this case, the predictor 210 may use a structure of an auto-encoder. In addition, the prediction unit 210 may include a convolutional neural network (CNN) for an image, a recurrent neural network (RNN) for time-series data that changes with time, a combination structure of the CNN and the RNN for an image sequence, and three-dimensional data (eg, In the case of MRI data) and multi-modal data, a structure combining a CNN and a long short term memory (LSTM) based on the baseline CNN may be used.

When the sensor data is received from the plurality of sensors of the target system after the preset time, the comparator 220 compares the sensor data after the preset time with the sensor data predicted by the predictor 210 to determine an abnormality of the target system. Diagnose the condition.

That is, the comparator 220 compares the difference between the predicted value predicted by the predictor 210 and the actual value received from the sensor, and diagnoses an abnormal state of the target system.

A method of diagnosing a state of a target system using the RNN by the state diagnosis apparatus 200 having such a configuration will be described with reference to FIG. 5. Referring to FIG. 5, the prediction unit 210

Figure 112017047071371-pat00001
If is input as sensor data, sensor data after (t + Δ) using RNN (
Figure 112017047071371-pat00002
Figure 112017047071371-pat00003
Predict and output

The comparator 220 receives the sensor data from the sensor after (t + Δ).

Figure 112017047071371-pat00004
) Is received,
Figure 112017047071371-pat00005
Wow
Figure 112017047071371-pat00006
Compare As a result of the comparison, predicted sensor data
Figure 112017047071371-pat00007
Wow
Figure 112017047071371-pat00008
If the difference is greater than or equal to a predetermined value, the comparator 220 determines that the target system is in an abnormal state and notifies the abnormal state diagnosis. The state diagnosis apparatus 200 configured as described above may diagnose an abnormal state of the target system even in a situation where sensor data is asymmetric with a large number of normal operation data and a small number of abnormal operation data.

The state diagnosis apparatus 200 may further include a learning DB (not shown) in which sensor data learned by the predictor 210 is stored. The learning DB stores sensor identification information, sensor data received from each sensor, sensor data predicted by the sensor data, and the like.

In addition, the state diagnosis apparatus 200 may further include a diagnosis result DB (not shown) in which information on the diagnosis result of the comparator 220 is stored. The diagnostic result DB stores sensor identification information, sensor data received from each sensor, sensor data predicted by the predicting unit 210, diagnostic result, and the like, and the diagnostic result includes a normal state or an abnormal state of the target system.

In addition, the state diagnosis apparatus 200 may further include an alarm unit (not shown) that notifies the abnormal state when the diagnosis result of the comparison unit 220, the target system is an abnormal state. The alarm unit may notify the abnormal state of the target system by means of sound, text, video, and message.

In addition, the state diagnosis apparatus 200 may further include a display unit (not shown) for outputting a diagnosis result of the comparator 220.

In addition, the state diagnosis apparatus 200 may further include a controller (not shown) for controlling operations of various components of the state diagnosis apparatus 200. The control unit may include at least one arithmetic unit, wherein the arithmetic unit is a general-purpose central processing unit (CPU), programmable device elements (CPLD, FPGA) implemented for a specific purpose, and application-specific semiconductor processing unit (ASIC) Or a microcontroller chip.

Each configuration in the state diagnosis apparatus 200 described above may be implemented in the form of a software module or a hardware module executed by a processor, or may be implemented in the form of a combination of a software module and a hardware module.

As such, a software module executed by a processor, a hardware module, or a combination of software modules and hardware modules may be implemented as a hardware system (eg, a computer system).

On the other hand, the state diagnosis device 200 implemented as described above may be used as a stand-alone.

6 is a block diagram schematically showing the configuration of a cause analysis apparatus according to the present invention.

Referring to FIG. 6, the cause analyzing apparatus 300 includes a relevance calculator 320.

The relevance calculator 320 calculates a relevance of each sensor with respect to an abnormal state diagnosis result of the target system provided with a plurality of sensors. The relevance calculation unit 320 calculates a relationship between the change amount of sensor data of each sensor of the target system operating in the normal state and the change amount of the abnormal state signal when the abnormal state is diagnosed, and outputs the relevance of each sensor with respect to the abnormal state diagnosis result. do. The abnormal state signal may be at least one of an abnormal state diagnosis result of the target system or sensor data predicted with respect to a plurality of sensors of the target system operating in the abnormal state. In addition, the abnormal state signal may be a kind of sensor, or may be a result obtained through another test as in the case of diagnosis.

In other words, the relevance calculation unit 320 grasps the relational expression of the change amount of the output and the change amount of the input for each sensor input based on the previously learned network. Here, the input may be sensor data according to time obtained from a plurality of sensors operating in the normal state of the target system, and the output may be an abnormal state signal. The neural network obtained through the learning process can be viewed as a "function of the relationship between the input and output", and the relevance calculator 320 can express the correlation between the output and the input value using this function relationship. Here, the correlation can be obtained using, for example, differential, inverse filtering, or the like.

Specifically, the sensor data X (t) input from the target system operating in the normal state is

Figure 112017047071371-pat00009
, The abnormal state signal
Figure 112017047071371-pat00010
If, the degree of relevance calculation unit 320 is an abnormal state signal change amount (
Figure 112017047071371-pat00011
) And sensor data change amount for each sensor
Figure 112017047071371-pat00012
) Is calculated as in Equation 1.

Figure 112017047071371-pat00013

Equation 1 obtains the relevance for each sensor and arranges the sensors in the order of high relevance. That is, the relevance calculation unit 320 changes the sensor data change amount of the sensor 1 (

Figure 112017047071371-pat00014
) And the degree of change in sensor data using the abnormal state signal change amount
Figure 112017047071371-pat00015
) And the degree of change in the abnormal state signal to obtain the relevance of the sensor 2. In this manner, the relevance calculator 320 obtains relevance for each of the n sensors, and arranges the sensors in ascending order. Here, the relevance may be a value that can be compared in size.

It can be determined that the sensor having the largest value of relevance is highly related to the abnormal state of the current target system.

Equation 1 is

Figure 112017047071371-pat00016
As the input
Figure 112017047071371-pat00017
If the model for predicting is previously learned, it can be learned using this, and the relevance can be calculated mathematically using a differential relation such as Equation 2.

Figure 112017047071371-pat00018

The relevance calculation formula of Equation 2 is a time interval and

Figure 112017047071371-pat00019
Normalized according to the range of, to calculate the relevance of each sensor.

The user can analyze the cause in the order of the relevant sensors and repair work such as replacement of parts based on the relatedness. In the case of healthcare, it can give priority to what the staff needs to review.

In addition, the relevance calculator 320 may output a relevance of each input sensor data to the final decision as a quantified value using a back propagation algorithm, inverse filtering, or the like.

Meanwhile, the cause analyzing apparatus 300 receives sensor data from a plurality of sensors of a target system operating in a normal state, diagnoses an abnormal state of the target system using the received sensor data, and diagnoses the result. It may further include an update unit 310 for updating in real time. In this case, the updater 310 may update the diagnosis result of the state diagnosis device for diagnosing the abnormal state in real time.

In addition, the cause analyzing apparatus 300 may further include a controller (not shown) for controlling operations of various components of the cause analyzing apparatus 300. The control unit may include at least one arithmetic unit, wherein the arithmetic unit is a general-purpose central processing unit (CPU), programmable device elements (CPLD, FPGA) implemented for a specific purpose, and application-specific semiconductor processing unit (ASIC) Or a microcontroller chip.

The cause analysis apparatus 300 configured as described above may quantitatively evaluate the degree of influence on the output for each sensor data so as to help cause analysis of an abnormal state of the target system. In addition, the cause analysis apparatus 300 may help in selecting the priority to be reviewed so that the cause analysis may be performed in order of the sensors which contributed by quantitatively obtaining the relevance of each sensor to the diagnosis result.

Each component in the cause analysis apparatus 300 described above may be implemented in the form of a software module or a hardware module executed by a processor, or may be implemented in a combination of a software module and a hardware module.

As such, a software module executed by a processor, a hardware module, or a combination of software modules and hardware modules may be implemented as a hardware system (eg, a computer system).

The cause analysis apparatus 300 may use a neural network structure of a single layer, or may use a complex network structure as the complexity and the system allow.

7 is a flowchart illustrating a method for diagnosing a cause and analyzing a cause of a target system according to the present invention, and FIGS. 8 and 9 are exemplary diagrams for describing a method for diagnosing a state and a cause of a target system according to the present invention.

Referring to FIG. 7, when sensor data is received from a plurality of sensors installed in a target system (S702), the state diagnosis apparatus diagnoses a state of the target system using the sensor data (S704). In this case, when the amount of normal operation data and abnormal operation data is asymmetric, the state diagnosis apparatus predicts sensor data after a preset time of the sensor data received from the plurality of sensors of the target system, and after the preset time, the plurality of sensors When the sensor data is received from the sensor data, the sensor data after a preset time is compared with the predicted sensor data to diagnose an abnormal state of the target system.

When the target system is diagnosed as an abnormal state, the state diagnosis apparatus transmits the diagnosis result to the cause analyzing apparatus to analyze the cause of the abnormal state (S706). In this case, the diagnosis result may include an abnormal state signal and sensor data for each sensor.

When the cause analysis device receives a diagnosis result from the state diagnosis device, the cause analysis device calculates an association degree of each sensor with respect to the abnormal state diagnosis result (S708). That is, the cause analyzing apparatus calculates the relationship between the sensor data change amount and the abnormal state signal change amount of each sensor of the target system operating in the normal state, and outputs the degree of relevance of each sensor to the abnormal state diagnosis result.

For example, when sensor data 1, sensor data 2,..., And sensor data n are input as shown in FIG. 8A, the state diagnosis device applies machine learning to the sensor data to generate a normal, check required, or abnormal state. Output the diagnosis result of the target system. In this case, the state diagnosis apparatus uses a machine learning method such as a convolutional neural network (CNN) for an image, a recurrent neural network (RNN) for a time-series data that changes with time, or a combination structure of a CNN and an RNN. Diagnosis can be made.

When diagnosed as in (a), the cause analysis device uses a back propagation algorithm, inverse filtering, and the like to analyze the cause of the 'need to check' or abnormal condition occurrence, as shown in (b). The degree of relevance of each sensor data can be output as a quantified value.

Next, when time series data is input, a state diagnosis and cause analysis of the target system will be described with reference to FIG. 9. When the time series data is input, the state diagnosis apparatus diagnoses the state of the target system by using the RNN structure for inputting the time series data as shown in (a). In this case, the method of learning the RNN may use BPTT (Back propagation through time).

The cause analyzing apparatus may analyze the cause of the abnormal state of the target system by using the RNN structure as shown in (b). When time series data is input, the cause analyzing apparatus may analyze not only a sensor type which is highly related to the diagnosis result but also what time interval influenced the diagnosis result.

Meanwhile, a method of diagnosing a condition and analyzing a cause of a target system may be written by a program, and codes and code segments constituting the program may be easily inferred by a programmer in the art. In addition, a program relating to a method of diagnosing a condition and analyzing a cause of a target system may be stored in a readable media that can be read by an electronic device, and read and executed by the electronic device.

As such, those skilled in the art will recognize that the present invention can be implemented in other specific embodiments without changing the technical spirit or essential features thereof. Therefore, it should be understood that the embodiments described above are merely exemplary and are not limitative in scope. In addition, the flowcharts shown in the drawings are merely sequential orders illustrated to achieve the most desirable results in practicing the present invention, and other additional steps may be provided or some steps may be omitted. .

The technical features and implementations described herein may be embodied in digital electronic circuitry, implemented in computer software, firmware, or hardware, including the structures and structural equivalents described herein, or a combination of one or more of these. It can be implemented. An implementation that implements the technical features described herein is also a module relating to computer program instructions encoded on a program storage medium of tangible type for controlling or by the operation of a computer program product, ie a processing system. It may be implemented.

On the other hand, the term "system" and "device" in the present specification include all the devices, devices and machines for processing data, including, for example, a processor, a computer, or a multiprocessor or a computer. The processing system includes, in addition to hardware, all code that forms an execution environment for a computer program on demand, such as code constituting processor firmware, a protocol stack, a database management system, an operating system, or a combination of one or more thereof. can do. Computer programs, known as programs, software, software applications, scripts, or code, may be written in any form of programming language, including compiled or interpreted languages, or a priori, procedural languages. It may be implemented in any form, including other units suitable for use in a routine or computer environment.

Elements embodying the technical features of the present invention included in the block diagrams and flowcharts shown in the accompanying drawings herein indicate logical boundaries between the elements. However, according to an embodiment of the software or hardware, the illustrated configuration and its functions are executed in the form of stand-alone software modules, monolithic software structures, codes, services, and combinations thereof, and can execute stored program codes, instructions, and the like. All such embodiments should also be considered to be within the scope of the present invention, as the functions may be implemented by being stored in a computer executable processor.

Accordingly, although the accompanying drawings and descriptions thereof illustrate technical features of the present invention, they should not be inferred simply unless the specific arrangement of software for implementing such technical features is clearly stated. That is, there may be various embodiments described above, and such embodiments may be modified in part while having the same technical features as the present invention, which should also be regarded as falling within the scope of the present invention.

In addition, although flowcharts depict operations in the drawings in a particular order, they are shown to obtain the most desirable results, which must be performed in the specific order shown or in the sequential order shown or all illustrated actions must be executed. It should not be understood to be. In certain cases, multitasking and parallel processing may be advantageous. In addition, the separation of the various system components of the embodiments described above should not be understood as requiring such separation in all embodiments, and the described program components and systems are generally integrated together into a single software product or may be combined into multiple software products. It should be understood that it can be packaged.

As such, this specification is not intended to limit the invention by the specific terms presented. Thus, although the present invention has been described in detail with reference to the embodiments described above, those skilled in the art to which the present invention pertains without departing from the scope of the invention modifications, changes and Modifications can be made. The scope of the present invention is shown by the following claims rather than the detailed description, and all changes or modifications derived from the meaning and scope of the claims and their equivalent concepts are included in the scope of the present invention. Should be.

100 sensor 200 status diagnosis device
210: prediction unit 220: comparison unit
300: cause analysis device 310: update unit
320: relevance calculation unit

Claims (10)

A state diagnosis device for diagnosing a state of the target system based on sensor data received from a plurality of sensors of a target system; And
When the state diagnosis device diagnoses a state in which the state of the target system is changed from a normal state to an abnormal state, a cause analysis for quantitatively calculating a relevance of each sensor involved in the abnormal state A device;
The cause analyzing apparatus includes (i) an amount of change according to time of sensor data when the plurality of sensors operate in the normal state, and (ii) sensor data or prediction when the plurality of sensors operate in the abnormal state. A condition diagnosis and cause analysis system of a target system, characterized in that the relationship between the amount of change in time of one sensor data is calculated, and the degree of relevance of each sensor that has caused an abnormal state of the target system is output in large order. The method of claim 1,
If the amount of sensor data of the target system operating in the normal state and the sensor data of the target system operating in the abnormal state are asymmetric, the state diagnosis apparatus may determine whether the sensor data after the preset time of the input sensor data is asymmetric. Predicting sensor data, comparing sensor data received from the plurality of sensors after the preset time with the predicted sensor data, and diagnosing an abnormal state of the target system; and Cause Analysis System. delete The method of claim 1,
The cause analysis apparatus may include a relevance calculation unit configured to calculate a relevance of each sensor with respect to an abnormal state diagnosis result of a target system provided with the plurality of sensors.
The relevance calculator calculates a relation between sensor data change amount of each sensor of the target system operating in the normal state and change amount of abnormal state signal at the time of diagnosing the abnormal state, thereby relevance of each sensor to the abnormal state diagnosis. System for diagnosis and cause analysis of the target system, characterized in that for outputting. The method of claim 4, wherein
The relevance calculator may be configured to output a relevance of each sensor to an abnormal condition diagnosis using at least one of a back propagation algorithm and inverse filtering. The method of claim 1,
The state diagnosis apparatus may include a predictor configured to predict sensor data after a predetermined time of sensor data received from a plurality of sensors of the target system; And
And a comparator for diagnosing an abnormal state of the target system by comparing sensor data received from the plurality of sensors after the preset time with the predicted sensor data. Diagnostic and cause analysis system. Diagnosing a state of the target system based on sensor data received from a plurality of sensors of the target system by a state diagnosis apparatus; And
When the state diagnosis apparatus diagnoses a state in which the state of the target system is changed from a normal state to an abnormal state, the cause analyzing apparatus quantitatively determines the relevance of each sensor involved in the abnormal state. Comprising; including;
The calculating of the degree of relevance of each sensor includes (i) an amount of change over time of sensor data when the plurality of sensors operate in the normal state and (ii) when the plurality of sensors operate in the abnormal state. Calculating the relationship between the amount of change in sensor data or the predicted sensor data over time, and outputting the relevance of each sensor that has caused an abnormal state of the target system in large order. Analytical Method. The method of claim 7, wherein
If the amount of sensor data of the target system operating in the normal state and the sensor data of the target system operating in the abnormal state is asymmetric, the state diagnosis apparatus may detect the sensor data after a predetermined time of the received sensor data. And diagnosing an abnormal state of the target system by comparing sensor data received from the plurality of sensors after the preset time with the predicted sensor data, and diagnosing a cause of the target system. Way. delete The method of claim 7, wherein
The relevance is calculated using at least one of a back propagation algorithm and inverse filtering.
KR1020170061210A 2017-05-17 2017-05-17 System and Method for State Diagnosis and Cause Analysis KR101995026B1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
KR1020170061210A KR101995026B1 (en) 2017-05-17 2017-05-17 System and Method for State Diagnosis and Cause Analysis

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
KR1020170061210A KR101995026B1 (en) 2017-05-17 2017-05-17 System and Method for State Diagnosis and Cause Analysis

Publications (2)

Publication Number Publication Date
KR20180126311A KR20180126311A (en) 2018-11-27
KR101995026B1 true KR101995026B1 (en) 2019-09-30

Family

ID=64603379

Family Applications (1)

Application Number Title Priority Date Filing Date
KR1020170061210A KR101995026B1 (en) 2017-05-17 2017-05-17 System and Method for State Diagnosis and Cause Analysis

Country Status (1)

Country Link
KR (1) KR101995026B1 (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102250498B1 (en) * 2019-11-07 2021-05-11 한양대학교 산학협력단 Architecture and electronc device implementing the architecture for safety and survivability wearable system based on artificial intelligence and internet of things, and operating method thereof
KR102497586B1 (en) * 2020-07-24 2023-02-10 한국전자통신연구원 Method and system for diagnosing abnormality of a plurality of apparatus based on sequential learning
KR102347168B1 (en) * 2021-08-04 2022-01-04 백우민 Sensor signal aggregating devcie for aggregating multi-sensor data

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2012150820A (en) * 2001-04-10 2012-08-09 Smartsignal Corp Diagnostic systems and methods for predictive condition monitoring
KR101713985B1 (en) * 2016-09-02 2017-03-09 에이블맥스(주) Method and apparatus for prediction maintenance

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR101677358B1 (en) 2015-01-28 2016-11-18 주식회사 엑센솔루션 System and method of failure prognostic as to die casting facilities using neural network algorithm

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2012150820A (en) * 2001-04-10 2012-08-09 Smartsignal Corp Diagnostic systems and methods for predictive condition monitoring
KR101713985B1 (en) * 2016-09-02 2017-03-09 에이블맥스(주) Method and apparatus for prediction maintenance

Also Published As

Publication number Publication date
KR20180126311A (en) 2018-11-27

Similar Documents

Publication Publication Date Title
Schlegel et al. Towards a rigorous evaluation of XAI methods on time series
JP7012871B2 (en) Devices and methods for controlling the system
Little et al. Using and understanding cross-validation strategies. Perspectives on Saeb et al.
US10977568B2 (en) Information processing apparatus, diagnosis method, and program
US10089203B2 (en) Artificial intelligence based health management of host system
KR101711936B1 (en) Generalized pattern recognition for fault diagnosis in machine condition monitoring
JP6585482B2 (en) Device diagnostic apparatus and system and method
JP2017120649A (en) Machine learning method and machine learning device for learning fault condition, and fault prediction device and fault prediction system including machine learning device
KR101995026B1 (en) System and Method for State Diagnosis and Cause Analysis
US20190139643A1 (en) Facilitating medical diagnostics with a prediction model
US20210133018A1 (en) A unifying semi-supervised approach for machine condition monitoring and fault diagnosis
JP5944241B2 (en) Method, monitoring system, and computer program product for monitoring health of monitored system using associative memory method
US10311404B1 (en) Software product development defect and issue prediction and diagnosis
WO2013188326A1 (en) Discriminative hidden kalman filters for classification of streaming sensor data in condition monitoring
JP2018180759A (en) System analysis system and system analysis method
WO2022009010A1 (en) Model fidelity monitoring and regeneration for manufacturing process decision support
del Amo et al. Hybrid recommendations and dynamic authoring for AR knowledge capture and re-use in diagnosis applications
KR20150007913A (en) Failure Prediction Device
WO2021240300A1 (en) Machine learning model error detection
RU2559401C1 (en) Diagnostic of aircraft service systems operating conditions
US20220391724A1 (en) Unsupervised Anomaly Detection With Self-Trained Classification
US20230091197A1 (en) Automatic sensor trace validation using machine learning
CN110858063B (en) Device and method for monitoring mechanical condition of robot
Ayerdi et al. Metamorphic Runtime Monitoring of Autonomous Driving Systems
US20200104234A1 (en) Event log processing

Legal Events

Date Code Title Description
E902 Notification of reason for refusal
E701 Decision to grant or registration of patent right
GRNT Written decision to grant