WO2020045720A1

WO2020045720A1 - Device and method for detecting malfunction in multimodal sensor by using artificial neural network

Info

Publication number: WO2020045720A1
Application number: PCT/KR2018/010654
Authority: WO
Inventors: 이병옥
Original assignee: 엠텍비젼 주식회사
Priority date: 2018-08-30
Filing date: 2018-09-12
Publication date: 2020-03-05
Also published as: KR20200027088A

Abstract

The present invention relates to a device for detecting a malfunction in a multimodal sensor by using an artificial neural network, the device comprising: a multimodal sensor unit comprising a plurality of sensors for measuring an analyte in a measurement area and generating an output signal; and an analysis unit for receiving the output signal through the multimodal sensor unit, extracting a characteristic (pattern) of the output signal according to at least one of a new (normal) state, an obsolescence state, and a fault state of the plurality of sensors, and analyzing a state of the plurality of sensors by using an artificial neural network that is trained by learning about the extracted characteristic (pattern) of the output signal.

Description

Multimodal Sensor Anomaly Detection Apparatus and Method Using Artificial Neural Network

The present invention relates to a multimodal sensor abnormality detection apparatus and method using an artificial neural network.

In general, when a failure occurs in an industrial facility, large losses such as large-scale power outages due to system accidents and exposure to toxic gases are inevitable, and large losses are inevitable. Therefore, monitoring of industrial facilities is very important in terms of preventing accidents in advance.

As a monitoring technology for these industrial facilities, it is possible to visually check / monitor abnormalities by moving by car or foot, or to monitor and install CCTV. Recently, multimodal sensors composed of several sensors have been installed in industrial facilities. The method of monitoring the analyte is mainly used.

The prior art of analyte monitoring using such a multi-modal sensor is Korea Patent Publication No. 10-1079846.

However, the conventional technology uses a multi-modal sensor to monitor the situation of an industrial facility in real time, thereby improving safety and efficient maintenance and preventing accidents. However, the multi-modal sensor cannot detect abnormalities. If one of the sensors is abnormal, there is a problem that can not detect the correct situation.

Japanese Laid-Open Patent Publication No. 1998-019614 is a conventional technique of a multi-sensor system for diagnosing such a problem and a diagnostic method and apparatus.

However, the prior art diagnoses an abnormality by comparing the output variable of the sensor, but because the output variable is different for each sensor, there are many output variables to compare, so it is difficult to combine the variables in the case of the non-skilled person, and when the abnormality is determined using the wrong combination, There is a problem.

The technical problem to be solved by the present invention, in real-time monitoring of the situation of the industrial facilities using a multi-modal sensor, to detect the abnormality of the multi-modal sensor in advance so that the failure of the multi-modal sensor can be replaced in advance before failure It is.

Another technical problem to be solved by the present invention is to accurately and quickly detect an abnormality of a multimodal sensor using an artificial neural network learned about various output variables of a multimodal sensor composed of a plurality of identical sensors or heterogeneous sensors. will be.

In order to solve the above technical problem, according to a preferred aspect of the present invention, a multi-modal sensor unit consisting of a plurality of sensors for measuring the analyte in the measurement area and generating an output signal; And receiving the output signal through the multi-modal sensor unit, extracting a feature (pattern) of the output signal according to at least one of a new (normal) state, an old state, and a failure state of the plurality of sensors, and extracting the extracted signal. It can provide a multi-modal sensor abnormality detection apparatus using an artificial neural network, including; an analysis unit for analyzing the state of the plurality of sensors using the learned artificial neural network for the characteristics (patterns) of the output signal.

Here, the characteristic (pattern) of the output signal is a variable for learning the artificial neural network, the variable is measured according to at least one of the new (normal) state, the old state and the failure state of the plurality of sensors At least one of distance, the analyte concentration, voltage, resistance, current, time, electrical conductivity change amount, pressure change amount, absorption intensity, relative absorbance, and wavelength may be used.

The apparatus may further include a temperature / humidity measurement unit for measuring temperature and humidity in the measurement area, wherein the analysis unit further receives the temperature and humidity measured by the temperature / humidity measurement unit. Extracts a feature (pattern) of the output signal according to at least one of a (normal) state, the aging state, the fault state, the temperature, and the humidity, and learns about a feature (pattern) of the extracted output signal. An artificial neural network may be used to analyze the state of the plurality of sensors.

Here, the characteristic (pattern) of the output signal is a variable for learning the artificial neural network, wherein the variable is one of the new (normal) state, the old state, the failure state, the temperature and the humidity of the plurality of sensors At least one of the measurement distance, the analyte concentration, the voltage, the resistance, the current, the time, the electrical conductivity change amount, the pressure change amount, the absorption intensity, the relative absorbance and the wavelength according to at least one.

Here, the plurality of sensors of the multi-modal sensor unit may be two or more identical sensors or heterogeneous sensors.

According to another preferred aspect of the present invention, the abnormal detection method of the multi-modal sensor abnormality detection apparatus using an artificial neural network, the multi-modal sensor unit consisting of a plurality of sensors measuring the analyte in the measurement area and generates an output signal; An analysis unit receiving the output signal from the multi-modal sensor unit and extracting a feature (pattern) of the output signal according to at least one of a new (normal) state, an old state, and a failure state of the plurality of sensors; And analyzing the state of the plurality of sensors by using the artificial neural network learned about the feature (pattern) of the extracted output signal by the analysis unit. Can be.

According to another preferred aspect of the present invention, in the abnormal detection method of the multi-modal sensor abnormality detection apparatus using an artificial neural network, the multi-modal sensor unit consisting of a plurality of sensors measuring the analyte in the detection area and generates an output signal ; Detecting, by a temperature / humidity measurement unit, temperature and humidity in the measurement area; The analysis unit receives the output signal, the measured temperature and the humidity from the multi-modal sensor unit and the temperature / humidity measuring unit, and the new (normal) state, the old state, the fault state, the temperature, and the like. Extracting a feature (pattern) of the output signal according to at least one of the humidity; And analyzing the state of the plurality of sensors by using the artificial neural network learned about the feature (pattern) of the extracted output signal by the analysis unit. Can be.

The present invention, in real-time monitoring of industrial facilities using a multi-modal sensor, detects the abnormality of the multi-modal sensor in advance and can replace the sensor in advance of the failure of the multi-modal sensor in advance before failure to improve safety and maintenance It is effective to prevent accidents by making them more efficient.

In addition, it is possible to detect abnormalities of the multi-modal sensor accurately and quickly by using the neural network learned about various output variables of the multi-modal sensor composed of a plurality of same sensors or heterogeneous sensors. It is effective to determine the point of management and later inspection.

1 is a block diagram of a multi-modal sensor abnormality detection device using an artificial neural network according to an embodiment of the present invention.

2A and 2B are characteristics (patterns) of output signals according to new (normal) states, aging states, and failure states of a plurality of sensors analyzed by a multimodal sensor abnormality detection apparatus using an artificial neural network according to an embodiment of the present invention; ) Is a graph showing the measurement distance.

3A and 3B are analysis of the characteristics (patterns) of the output signal according to the new (normal) state, the aging state of the plurality of sensors analyzed by the multi-modal sensor abnormality detection apparatus using an artificial neural network according to an embodiment of the present invention This graph shows water concentration and measurement distance.

4A and 4B are characteristics of an output signal according to a new (normal) state, an aging state, a temperature, and a humidity of a plurality of sensors analyzed by a multimodal sensor abnormality detection apparatus using an artificial neural network according to an embodiment of the present invention ( A graph showing the analyte concentration in the pattern).

5A and 5B are characteristics (patterns) of output signals according to a new (normal) state, an aging state, and a failure state of a plurality of sensors analyzed by a multimodal sensor abnormality detection apparatus using an artificial neural network according to an embodiment of the present invention; ) Is a graph showing the analyte concentration according to voltage and voltage.

6A and 6B are characteristics (patterns) of output signals according to a new (normal) state, an aging state, and a failure state of a plurality of sensors analyzed by a multimodal sensor abnormality detection apparatus using an artificial neural network according to an embodiment of the present invention; ) Is a graph showing the current according to the analyte concentration and the analyte concentration.

7A and 7B are characteristics (patterns) of output signals according to new (normal) states, aging states, and failure states of a plurality of sensors analyzed by a multimodal sensor abnormality detection apparatus using an artificial neural network according to an embodiment of the present invention; ) Is a graph showing resistance and analyte concentration according to resistance.

8A and 8B are characteristics (patterns) of output signals according to new (normal) states, aging states, and failure states of a plurality of sensors analyzed by a multimodal sensor abnormality detection apparatus using an artificial neural network according to an embodiment of the present invention; ) Is a graph showing absorption intensity according to wavelength and wavelength.

9A and 9B are characteristics (patterns) of output signals according to new (normal) states, aging states, and failure states of a plurality of sensors analyzed by a multimodal sensor abnormality detection apparatus using an artificial neural network according to an embodiment of the present invention; ) Is a graph showing the electric conductivity change and the pressure change.

FIG. 10 shows wavelengths and relative absorption among characteristics (patterns) of output signals according to the new (normal) state and the aging state of a plurality of sensors analyzed by a multimodal sensor abnormality detection apparatus using an artificial neural network according to an embodiment of the present invention. It is a graph which shows a figure.

11 is a flowchart illustrating a multimodal sensor abnormality detection method using an artificial neural network according to another embodiment of the present invention.

12 is a flowchart illustrating a multimodal sensor abnormality detection method using an artificial neural network according to another embodiment of the present invention.

As the inventive concept allows for various changes and numerous embodiments, particular embodiments will be illustrated in the drawings and described in detail in the written description. However, this is not intended to limit the present invention to specific embodiments, it should be understood to include all changes, equivalents, and substitutes included in the spirit and scope of the present invention.

Terms including ordinal numbers such as first and second may be used to describe various components, but the components are not limited by the terms. These terms are only used to distinguish one component from another.

When a component is said to be 'connected' or 'connected' to another component, it may be directly connected to or connected to that other component, but other components may be present in between. It should be understood that. On the other hand, when a component is said to be 'directly connected' or 'directly connected' to another component, it should be understood that there is no other component in between.

The terminology used herein is for the purpose of describing particular example embodiments only and is not intended to be limiting of the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In the present application, the term 'comprises' or 'having' is intended to indicate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, and one or more other features. It is to be understood that the present invention does not exclude the possibility of the presence or the addition of numbers, steps, operations, components, components, or a combination thereof.

Referring to FIG. 1, the apparatus 100 for detecting abnormality of multimodal sensors using an artificial neural network includes a multimodal sensor unit 110, an analysis unit 120, and an output unit 130.

The multi-modal sensor unit 110 is composed of a plurality of sensors to measure the analyte in the measurement area and generate an output signal.

A plurality of sensors of the multi-modal sensor unit 110 may be composed of two or more identical sensors or heterogeneous sensors.

The analysis unit 120 receives an output signal through the multi-modal sensor unit 110 connected through a wired or wireless network, and outputs a signal according to at least one of a new (normal) state, an old state, and a fault state of a plurality of sensors. (Pattern) is extracted, and the state of the plurality of sensors is analyzed using the artificial neural network learned about the characteristics (pattern) of the extracted output signal.

The characteristic (pattern) of the output signal extracted by the analysis unit 120 is a variable for learning the artificial neural network, and the variable is measured according to at least one of a new (normal) state, an old state, and a fault state of a plurality of sensors. At least one of distance, analyte concentration, voltage, resistance, current, time, electrical conductivity change, pressure change, absorption intensity, relative absorbance, and wavelength, and the artificial neural network is learned by classification learning, deep learning, etc. Can be.

The output unit 130 outputs output signal characteristics (patterns) of the analytes analyzed by the analyzer 120 and the states of the plurality of sensors.

Multimodal sensor abnormality detection device 100 using an artificial neural network may further include a temperature / humidity measuring unit 140.

The temperature / humidity measurement unit 140 measures the temperature and humidity in the measurement area and provides the measured temperature and humidity to the analysis unit 130 connected to the wired / wireless network.

The analyzer 130 further receives the temperature and humidity measured by the temperature / humidity measurement unit 140, according to at least one of the new (normal) state, aging state, failure state, temperature, and humidity of the plurality of sensors. A feature (pattern) of the output signal may be extracted, and the state of the plurality of sensors may be analyzed using an artificial neural network learned about the extracted feature (pattern) of the output signal.

At this time, the parameters for learning the artificial neural network are measured distance, analyte concentration, voltage, resistance, current, time according to at least one of the new (normal) state, aging state, failure state, temperature and humidity of the plurality of sensors, It may be at least one of an electrical conductivity change amount, a pressure change amount, an absorption intensity, a relative absorption degree, and a wavelength. In the case of measuring the gas concentration using a heterogeneous sensor, since the concentration of the gas to be detected is generally determined by determining the composition ratio of the elements constituting the gas, in the case of a gas having a similar element composition, Since the characteristics (patterns) of the output signals are correlated with each other, the correlation of the characteristics (patterns) can be used as variables in the output signals.

2A and 2B are characteristics (patterns) of output signals according to new (normal) states, aging states, and failure states of a plurality of sensors analyzed by a multimodal sensor abnormality detection apparatus using an artificial neural network according to an embodiment of the present invention; ) Is a graph showing the measurement distance. 2A is a graph illustrating a measurement distance among characteristics (patterns) of output signals according to a new (normal) state sensor, an aging state, and a failure state of a plurality of sensors of the multimodal sensor unit 110, and FIG. 2B is a multimodal sensor A plurality of sensors of the unit 110 are graphs showing time and measurement distance among the characteristics (patterns) of the output signal according to the new (normal) state sensor, the deterioration state, and the failure state.

Referring to FIG. 2A, when a plurality of sensors of the multi-modal sensor unit 110 use the same sensor under conditions having the same analyte concentration, temperature, and humidity, the measurement distance should be the same. Since the analysis unit 120 receives the output signal of the sensor, the analysis unit 120 extracts the measurement distance feature (pattern) of the output signal to learn and analyzes which sensor is a new (normal) state or an old state or a failure state.

Referring to FIG. 2B, in the case of the new (normal) state sensor, the analysis unit 120 analyzes that the measurement distance for measuring the analyte is the same as time passes, but as the aging state progresses, the performance of the sensor decreases. The measurement distance to measure the analyte will be less, and if the deterioration state is considerably continued, it is possible to analyze that the measurement value does not come out at any moment, so that the analysis unit 120 has some progress in the deterioration state of a certain sensor. To predict when they will reach their end of life.

3A and 3B are analysis of the characteristics (patterns) of the output signal according to the new (normal) state, the aging state of the plurality of sensors analyzed by the multi-modal sensor abnormality detection apparatus using an artificial neural network according to an embodiment of the present invention This graph shows water concentration and measurement distance. FIG. 3A illustrates a case in which a plurality of sensors are exposed when the analyte is exposed to a measurement distance that each sensor can detect when the installation location is different under the same analyte concentration, temperature, and humidity when the same sensor is used. The graph shows a window clustering the output signal characteristics (patterns) according to the analyte concentration and the measurement distance when all of them are in a new (normal) state, and FIG. 3B shows the same analyte concentration, When the analyte is exposed to the measuring distance that each sensor can detect when the installation location is different under the condition of temperature and humidity, the sensor output according to the analyte concentration and the measuring distance to the sensor which is new (normal) and aging A graph that shows a window of clustering signal features (patterns).

Referring to FIG. 3A, since the plurality of sensors of the multi-modal sensor unit 110 are the same sensor, the performance is the same when the sensor is in a new (normal) state. As a result, the analysis unit 120 analyzes signals measured by the plurality of sensors. The clustered windows are the same size and overlap each other.

Referring to FIG. 3B, since the plurality of sensors of the multi-modal sensor unit 110 are the same sensor, the performance should be the same. However, as the aging state progresses, the performance of the sensor decreases, so that the measurement distance and the analyte concentration may be measured. The value is different, and based on the size of the clustered window analyzed by the analyzer 120 analyzing the signals measured by the plurality of sensors, it is predicted to what extent the deteriorated state of which sensor has progressed and when the end of life.

4A and 4B are characteristics of an output signal according to a new (normal) state, an aging state, a temperature, and a humidity of a plurality of sensors analyzed by a multimodal sensor abnormality detection apparatus using an artificial neural network according to an embodiment of the present invention ( A graph showing the analyte concentration in the pattern). 4A is a graph showing analyte concentrations according to temperature among output signal characteristics (patterns) according to the degree of deterioration of a plurality of sensors and when a plurality of sensors are in a new (normal) state under conditions having the same analyte concentration. 4B is a graph showing analyte concentration according to humidity among output signal characteristics (patterns) according to the degree of deterioration of a plurality of sensors and when a plurality of sensors are in a new (normal) state under conditions having the same analyte concentration.

Referring to Figure 4a, the concentration of the analyte is affected by the temperature, the degradation of the performance of the sensor as the aging state progresses, the effect on temperature is different from the analysis unit based on the new (normal) state sensor and the old state sensor 120 predicts how old the sensor is and how long it has reached its end of life.

Referring to Figure 4b, the concentration of the analyte is affected by the humidity, the performance of the sensor is degraded as the aging state progresses, the effect of the humidity on the analysis unit based on the difference between the new (normal) state sensor and the old state sensor 120 predicts how old the sensor is and how long it has reached its end of life.

5A and 5B are characteristics (patterns) of output signals according to a new (normal) state, an aging state, and a failure state of a plurality of sensors analyzed by a multimodal sensor abnormality detection apparatus using an artificial neural network according to an embodiment of the present invention; ) Is a graph showing the analyte concentration according to voltage and voltage. 5A is a graph illustrating voltages according to a new (normal) state sensor, an aging state, and a failure state of the plurality of sensors of the multi-modal sensor unit 110. FIG. 5B illustrates a plurality of sensors of the multi-modal sensor unit 110. It is a graph showing the analyte concentration according to the new (normal) state sensor, the aging state and the voltage.

Referring to FIG. 5A, when a plurality of sensors of the multimodal sensor unit 110 use the same sensor under conditions having the same analyte concentration, temperature, and humidity, the voltage applied to the sensor becomes less as the aging becomes less. 130 receives the output signal of the sensor and extracts and learns a voltage characteristic (pattern) of the output signal, and analyzes which sensor is new (normal), old, or faulty.

Referring to FIG. 5B, the concentration of the analyte is influenced by the voltage, and as the aging state progresses, the performance of the sensor decreases and the voltage applied to the sensor decreases. Predict how far it has progressed and when it will reach its end of life.

6A and 6B are characteristics (patterns) of output signals according to a new (normal) state, an aging state, and a failure state of a plurality of sensors analyzed by a multimodal sensor abnormality detection apparatus using an artificial neural network according to an embodiment of the present invention; ) Is a graph showing the current according to the analyte concentration and the analyte concentration. FIG. 6A is a graph illustrating analyte concentration according to a new (normal) state sensor, an aging state, and a failure state of a plurality of sensors of the multimodal sensor unit 110. FIG. 6B is a plurality of sensors of the multimodal sensor unit 110. FIG. The sensor is a graph showing the analyte concentration according to the new (normal) state sensor, aging state and current.

Referring to FIG. 6A, when the same sensor is used in a plurality of sensors of the multimodal sensor unit 110 under the same analyte concentration, temperature, and humidity, the analyte concentration should be the same because the performance is the same. As the performance decreases with age, the analysis unit 120 receives the output signal of the sensor and extracts and analyzes the analyte concentration characteristic (pattern) of the output signal to determine which sensor is new (normal), old, or faulty. Analyze

Referring to FIG. 6B, the amount of current generated according to the analyte concentration is changed. As the aging state progresses, the performance of the sensor decreases, so that the measured analyte concentration is small, and thus the analysis unit 120 is small. Predicts how old the sensor is and how old it is.

7A and 7B are characteristics (patterns) of output signals according to new (normal) states, aging states, and failure states of a plurality of sensors analyzed by a multimodal sensor abnormality detection apparatus using an artificial neural network according to an embodiment of the present invention; ) Is a graph showing resistance and analyte concentration according to resistance. FIG. 7A is a graph illustrating resistance of a plurality of sensors of the multi-modal sensor unit 110 according to a new (normal) state sensor, an aging state, and a failure state. FIG. 7B illustrates a plurality of sensors of the multi-modal sensor unit 110. It is a graph showing the analyte concentration according to the new (normal) state sensor, the aging state and the resistance.

Referring to FIG. 7A, when a plurality of sensors of the multi-modal sensor unit 110 use the same sensor under conditions having the same analyte concentration, temperature, and humidity, the resistance of the sensor increases as the age of the analysis unit 130 increases. ) Receives the output signal of the sensor and extracts the resistance characteristic (pattern) of the output signal and learns it to analyze which sensor is new (normal), old, or faulty.

Referring to FIG. 7B, the concentration of the analyte is affected by the resistance. As the aging state progresses, the performance of the sensor decreases and the resistance of the sensor increases, so that the analysis unit 120 determines the degree of deterioration of a certain sensor. Progress is made and predicts when the end of life will be reached.

8A and 8B are characteristics (patterns) of output signals according to new (normal) states, aging states, and failure states of a plurality of sensors analyzed by a multimodal sensor abnormality detection apparatus using an artificial neural network according to an embodiment of the present invention; ) Is a graph showing absorption intensity according to wavelength and wavelength. 8A is a graph illustrating wavelengths of a plurality of sensors of the multi-modal sensor unit 110 according to a new (normal) state sensor, an aging state, and a failure state, and FIG. 8B illustrates a plurality of sensors of the multi-modal sensor unit 110. It is a graph showing the absorption intensity according to the new (normal) state sensor, the aging state and the wavelength.

Referring to FIG. 8A, when a plurality of sensors of the multi-modal sensor unit 110 have the same wavelength under a condition having the same analyte concentration, temperature, and humidity, the output of the light source sensor is lowered as the age becomes older. Since this becomes small, the analysis unit 130 receives the output signal of the sensor, extracts and learns the wavelength characteristic (pattern) of the output signal, and analyzes which sensor is a new (normal) state, an old state or a failure state.

Referring to FIG. 8B, based on the degree of deterioration of the new (normal) state sensor and the sensor, and the absorption intensity of the sensor being lowered, the analyzer 120 proceeds to a certain degree of deterioration of a certain sensor. Predict when the end of life.

9A and 9B are characteristics (patterns) of output signals according to new (normal) states, aging states, and failure states of a plurality of sensors analyzed by a multimodal sensor abnormality detection apparatus using an artificial neural network according to an embodiment of the present invention; ) Is a graph showing the electric conductivity change and the pressure change. FIG. 9A is a graph illustrating changes in conductivity of the plurality of sensors of the multimodal sensor unit 110 according to a new (normal) state sensor, an aging state, and a failure state, and FIG. 9B is a diagram of a plurality of sensors of the multimodal sensor unit 110. The sensor shows a graph of the pressure change according to the new (normal) state sensor, the aging state and the failure state.

Referring to FIG. 9A, when a plurality of sensors of the multi-modal sensor unit 110 use the same semiconductor gas sensor under conditions having the same analyte concentration, temperature, and humidity, the semiconductor gas sensor is a gas on the ceramic semiconductor surface. If the same sensor is used to measure the amount of change in electrical conductivity that occurs when the contact is made, the amount of change in electrical conductivity should be the same because the performance is the same, but as the sensor gets older, the analysis unit 120 outputs the sensor output. It receives the signal and extracts the electrical conductivity variation feature (pattern) and learns it, and analyzes which sensor is new (normal), old or broken.

Referring to FIG. 9B, a plurality of sensors of the multi-modal sensor unit 110 under the same analyte concentration, temperature, and humidity may be used to emit light of gas molecules by using a characteristic in which gas molecules absorb light of a specific wavelength, usually infrared rays. When the same sensor is used to measure the pressure change generated when it expands due to thermal excitation when it is absorbed by a precise microphone, the pressure change should be the same because the performance is the same, but the performance decreases as the sensor gets older. Therefore, the analysis unit 120 receives the output signal of the sensor and extracts the pressure variation feature (pattern) of the output signal to learn and analyzes which sensor is in a new (normal) state or an old state or a failure state.

FIG. 10 shows wavelengths and relative absorption among characteristics (patterns) of output signals according to the new (normal) state and the aging state of a plurality of sensors analyzed by the multimodal sensor abnormality detection apparatus using an artificial neural network according to an embodiment of the present invention. It is a graph which shows a figure. 10 (a) is a novel (normal) when a plurality of sensors of the multi-modal sensor unit 110 is a heterogeneous sensor as a carbon monoxide (CO) sensor and a carbon dioxide (CO ₂ ) sensor under conditions having the same analyte concentration, temperature, and humidity. 10) is a graph showing the wavelength and the relative absorbance when the state, Figure 10 (b) is a plurality of sensors of the multi-modal sensor unit 110 and the carbon monoxide (CO) sensor under the conditions having the same analyte concentration, temperature, humidity This is a graph showing the wavelength and the relative absorbance when the carbon dioxide (CO ₂ ) sensor is a heterogeneous sensor in the aging state.

Referring to FIGS. 10 (a) and 10 (b), the relative absorption of the carbon dioxide (CO ₂ ) sensor is analyzed because the deteriorated state of FIG. 10 (b) is lower than the new (normal) state of FIG. 10 (a). The unit 120 analyzes that the carbon dioxide (CO ₂ ) sensor is in an aging state.

Referring to FIG. 11, in operation S1110, the multimodal sensor unit 110 including a plurality of sensors measures analytes in a measurement area and generates an output signal.

In operation S1120, the analysis unit 120 receives the output signal from the multi-modal sensor unit 110, and thus the characteristics (pattern) of the output signal according to at least one of a new (normal) state, an old state, and a failure state of the plurality of sensors. Extract

In step S1130, the analysis unit 120 analyzes the state of the plurality of sensors using the learned neural network for the feature (pattern) of the extracted output signal. Here, the analysis unit 120 may transmit the analyzed result to the output unit 130 and output the result analyzed by the output unit 130 in the analysis unit 120.

In step S1210 the multi-modal sensor unit 110 composed of a plurality of sensors measures the analyte in the detection area and generates an output signal.

In step S1220 the temperature / humidity measuring unit 140 detects the temperature and humidity in the measurement area.

In step S1230, the analysis unit 120 receives the output signal and the measured temperature and humidity from the multi-modal sensor unit 110 and the temperature / humidity measurement unit 140, and the temperature, humidity, and new (normal) state of the plurality of sensors. Extracts a feature (pattern) of the output signal according to at least one of a deterioration state and a failure state.

In step S1240, the analysis unit 120 analyzes the state of the plurality of sensors using the learned artificial neural network for the feature (pattern) of the extracted output signal. Here, the analysis unit 120 may transmit the analyzed result to the output unit 130 and output the result analyzed by the output unit 130 in the analysis unit 120.

Although embodiments according to the present invention have been described above, these are merely exemplary, and those skilled in the art will understand that various modifications and equivalent embodiments of the present invention are possible therefrom. . Therefore, the true technical protection scope of the present invention will be defined by the following claims.

Claims

A multimodal sensor unit configured to measure analytes in the measurement area and generate an output signal; And

The output signal is input through the multi-modal sensor unit to extract a feature (pattern) of the output signal according to at least one of a new (normal) state, an old state, and a failure state of the plurality of sensors, and the extracted output. It includes; an analysis unit for analyzing the state of the plurality of sensors using the artificial neural network learned about the characteristics (pattern) of the signal

Multimodal sensor abnormality detection device using an artificial neural network, characterized in that.
The method of claim 1,

The characteristic (pattern) of the output signal is a variable for learning the artificial neural network, wherein the variable is a measurement distance according to at least one of the new (normal) state, the aging state and the fault state of the plurality of sensors, At least one of the analyte concentration, voltage, resistance, current, time, electrical conductivity change amount, pressure change amount, absorption intensity, relative absorbance and wavelength

Multimodal sensor abnormality detection device using an artificial neural network, characterized in that.
The method of claim 1,

Further comprising: a temperature / humidity measuring unit for measuring the temperature and humidity in the measurement area,

The analyzer may further receive the temperature and humidity measured by the temperature / humidity measurement unit to at least one of the new (normal) state, the old state, the failure state, the temperature, and the humidity of the plurality of sensors. Extracting a feature (pattern) of the output signal and analyzing the state of the plurality of sensors using an artificial neural network learned about the extracted output signal feature (pattern)

Multimodal sensor abnormality detection device using an artificial neural network, characterized in that.
The method of claim 3,

The characteristic (pattern) of the output signal is a variable for learning the artificial neural network, wherein the variable is at least one of the new (normal) state, the aging state, the failure state, the temperature, and the humidity of the plurality of sensors. At least one of measurement distance, analyte concentration, voltage, resistance, current, time, electrical conductivity change, pressure variation, absorption intensity, relative absorbance, and wavelength according to one

Multimodal sensor abnormality detection device using an artificial neural network, characterized in that.
The method of claim 1,

The plurality of sensors of the multi-modal sensor unit is two or more of the same sensor or heterogeneous sensor

Multimodal sensor abnormality detection device using an artificial neural network, characterized in that.
In the abnormal detection method of the multi-modal sensor abnormality detection device using an artificial neural network,

Measuring an analyte in the measurement area and generating an output signal by the multimodal sensor unit including a plurality of sensors;

An analysis unit receiving the output signal from the multi-modal sensor unit and extracting a feature (pattern) of the output signal according to at least one of a new (normal) state, an old state, and a failure state of the plurality of sensors; And

And analyzing the state of the plurality of sensors by using the artificial neural network learned about the feature (pattern) of the extracted output signal by the analysis unit.

Multimodal sensor abnormality detection method using an artificial neural network characterized in that.
The method of claim 6,

The characteristic (pattern) of the output signal is a variable for learning the artificial neural network, wherein the variable is a measurement distance according to at least one of the new (normal) state, the aging state and the failure state of the plurality of sensors, At least one of the analyte concentration, voltage, resistance, current, time, electrical conductivity change amount, pressure change amount, absorption intensity, relative absorbance and wavelength

Multimodal sensor abnormality detection device using an artificial neural network, characterized in that.
In the abnormal detection method of the multi-modal sensor abnormality detection device using an artificial neural network,

Measuring an analyte in the detection area and generating an output signal by the multimodal sensor unit including a plurality of sensors;

Detecting, by a temperature / humidity measurement unit, temperature and humidity in the measurement area;

The analysis unit receives the output signal, the measured temperature and the humidity from the multi-modal sensor unit and the temperature / humidity measuring unit and at least one of a new (normal) state, an old state, and a fault state of the plurality of sensors. Extracting a feature (pattern) of the output signal according to the present invention; And

And analyzing the state of the plurality of sensors by using the artificial neural network learned about the feature (pattern) of the extracted output signal by the analysis unit.

Multimodal sensor abnormality detection method using an artificial neural network characterized in that.
The method of claim 8,

The characteristic (pattern) of the output signal is a variable for learning the artificial neural network, wherein the variable is at least one of the new (normal) state, the aging state, the failure state, the temperature, and the humidity of the plurality of sensors. At least one of measurement distance, analyte concentration, voltage, resistance, current, time, electrical conductivity change, pressure variation, absorption intensity, relative absorbance, and wavelength according to one

Multimodal sensor abnormality detection device using an artificial neural network, characterized in that.