Gas sensing device

Abstract

The present invention relates to a gas sensing device. In the present invention, a processor (20), a first gas sensor (30), and a second gas sensor (32) are installed on a substrate (10). The first gas sensor (30) and the second gas sensor (32) are positioned between the two ends of the processor (20) and the corresponding edges of the substrate (10), respectively. A temperature / humidity sensor (34) is installed on the substrate (10), and a through channel (12) is formed through the substrate (10) between the temperature / humidity sensor (34) and the processor (20). The first gas sensor (30) and the second gas sensor (32) detect the types and concentrations of a plurality of gases, respectively, and the processor (20) detects a mixed gas and analyzes the types and concentrations of each gas of the detected mixed gas using a deep learning model. At this time, the learning model uses a first-stage classification model and a second-stage regression model.

Description

gas detection device

[0001] The present invention relates to a gas detection device that detects multiple types of harmful gases in real time.

[0002] As air pollution worsens and concerns about air quality grow, the need for indoor air quality management and maintenance is growing. Emissions of various harmful gases are increasing, and their types are becoming more diverse, not only in industrial settings but also in everyday environments, including homes.

[0003] Accordingly, technology for monitoring indoor environmental information in real time has developed, and the demand for gas sensors to detect various harmful gases that cause phenomena such as sick house syndrome has recently increased significantly.

[0004] Generally, gas sensors detect hazardous gases in the air and determine their composition and concentration. Depending on their measurement method, gas sensors can be categorized into various types, including semiconductor, catalytic, and optical. Among these, semiconductor gas sensors are actively being researched due to their simple manufacturing process and low-cost use of metal oxides.

[0005] Conventional techniques have been proposed for detecting hazardous gases using gas sensors. For example, Korean Patent Publication No. 10-2021-0104390 detects contamination by outputting the ionic conductivity of the gas sensor electrolyte. However, since ionic conductivity varies with temperature, the output value is increased or decreased whenever the temperature changes, and the output value is corrected. If the corrected value exceeds a threshold, toxic gas is detected. However, although multiple gas sensors are used to detect multiple gases, the actual gas concentration cannot be determined because only the sensor current value is measured. In addition, the use of electrolytes is disadvantageous for durability, and the electrical conductivity varies with temperature, which increases the error rate. Furthermore, because the large gas sensor is used, the device size increases in proportion to the number of gases detected when detecting multiple gases.

[0006] As another example, Korean Patent No. 10-1674048 discloses a device that detects hazardous chemicals and has network-based connectivity, enabling rapid transmission in the event of an accident. The device comprises a preprocessing unit for concentrating hazardous gases and maintaining a constant temperature and humidity, an environmental sensor unit for collecting external environmental information, a communication unit for communicating with external devices, a sensor array unit for detecting various hazardous gases, and a display, input unit, and control unit, most of which are installed within a single case.

[0007] However, since a heater is required here to maintain temperature and discharge the adsorbed gas, a barrier is required to prevent the heater's heat from affecting the gas measurement device, making the device larger. The use of a heater also increases power consumption, and the need to discharge the adsorbed gas means that the sensor response is not real-time. Furthermore, the inclusion of a hazardous gas concentration unit means that the sensor must concentrate the gas to detect it, making it difficult to detect at low concentrations and in real time.

[0008] As another example, Patent Publication No. 10-2022-0142030 discloses a method for detecting VOC gas using a gas sensor and quickly classifying VOC gas in a transient state using a recurrent neural network.

[0009] However, it is necessary to predict not only the type but also the concentration of indoor harmful gases in real time, to compensate for the sensitivity value that varies depending on temperature and humidity to obtain a constant output value, and to minimize the size of the entire device to minimize the space required for installation.

[0010]

[0011] The present invention provides a gas detection device in which the entire device is miniaturized.

[0012] The present invention provides a gas detection device that detects gas using a gas sensor and accurately provides the type and concentration of the gas.

[0013] The present invention is to enable real-time detection of the type and concentration of gas while miniaturizing the entire device.

[0014] The present invention analyzes the type and concentration of gas by training a deep learning model based on pre-prepared learning data, and provides the concentration and type of gas.

[0015] The present invention compensates for sensitivity values ​​that vary depending on temperature / humidity to always produce a constant output value.

[0016]

[0017] The gas sensor used in the gas detection device of the present invention can detect the type and concentration of multiple gases in real time.

[0018] In the present invention, learning data on the sensing sensitivity of a sample gas is acquired, a deep learning model is trained with the learning data, and the learning model is trained with the sensing sensitivity of the detected target gas to analyze the type and concentration of the target gas.

[0019] In the present invention, a first gas sensor, a second gas sensor, and a temperature / humidity sensor can be installed at predetermined angular intervals at both ends based on a processor that analyzes the type and concentration of a target gas.

[0020] The first gas sensor, the second gas sensor, and the temperature / humidity sensor may be positioned between the edge of the processor and the edge of the corresponding substrate, respectively.

[0021] The present invention may include a substrate, a processor installed on the substrate and analyzing detected data, a first gas sensor installed between the processor and detecting different types of target gases, and a second gas sensor installed between the processor and detecting different types of target gases, wherein the first gas sensor may be positioned between one edge of the processor and a corresponding edge of the substrate, and the second gas sensor may be positioned between the other edge of the processor and a corresponding edge of the substrate.

[0022] The above processor can analyze the type and concentration of the detected target gas.

[0023] The above processor obtains learning data on the sensing sensitivity of the sample gas, trains a deep learning learning model with the learning data, and trains the learning model with the sensing sensitivity of the detected target gas to analyze the type and concentration of the target gas. The learning model may include a classification model of the first stage and a regression model of the second stage.

[0024] The above classification model may be a Fully Connected Neuronal Network (FCN) and the above regression model may be a Multiple Linear Regression (MLR).

[0025] The above FCN can receive the sensing sensitivity of the detected target gas as input and output the type of the target gas, and the above MLR can receive the type of the target gas output from the FCN and the sensing sensitivity of the detected target gas as input and output the concentration of the target gas.

[0026] The above processor can obtain the sensing sensitivity of the plurality of sample gases, generate distribution data of the sensing sensitivity, and generate normal distribution data based on the average and standard deviation of the distribution data, thereby obtaining the normal distribution data as the learning data.

[0027] Either of the first gas sensor and the second gas sensor can detect ethanol (C2H6O), formaldehyde (HCHO), toluene (C7H8), or hydrogen (H2).

[0028] A temperature / humidity sensor may be installed between one edge of the substrate and the edge of the processor.

[0029] A through channel may be formed through the substrate between the temperature / humidity sensor and the processor.

[0030] The above through channel can be formed to surround the temperature / humidity sensor except for some sections.

[0031] The above temperature / humidity sensor, the first gas sensor, and the second gas sensor may be arranged at 90° intervals between the processor and the corresponding edge of the substrate based on the processor.

[0032] A connector for signal connection with the outside is installed on one edge of the above substrate, and the connector can be positioned offset to one side from the edge.

[0033] An ESD diode for static electricity removal can be used on the above substrate.

[0034] The substrate further includes a cover that shields the surface on which the processor is installed from the outside, and a processor shielding portion is formed in the cover at a position corresponding to the position of the processor, and a first sensor window, a second sensor window, and a third sensor window may be formed between an edge of the processor shielding portion and an edge of the cover, respectively.

[0035] A partition grid can be formed in the third sensor window.

[0036]

[0037] The gas detection device according to the present invention has at least one of the following effects.

[0038] The present invention provides a miniaturized gas detection device capable of detecting the type and concentration of multiple hazardous gases in real time. This device comprises multiple MEMS gas sensors positioned on a substrate with other components interposed therebetween, thereby securing the necessary distance between the gas sensors. Furthermore, a through-channel is formed in the substrate to shield the temperature / humidity sensors from the thermal effects of surrounding components. Consequently, the distance between components mounted on the substrate can be minimized, thereby miniaturizing the overall device.

[0039] The present invention applies a deep learning model, thereby providing the type and concentration of detected gas in real time.

[0040] In the present invention, by applying a deep learning model, the components and concentrations of each gas can be analyzed for mixed gases mixed at various ratios.

[0041] Additionally, the present invention compensates for sensitivity values ​​that vary with temperature and humidity. To achieve this, absolute humidity values ​​were used to calibrate sensor sensitivity. This approach allowed for similar sensitivity values ​​based on absolute humidity differences, and applied to deep learning algorithms, significantly reducing error rates.

[0042]

[0043] Figure 1 is a block diagram showing the configuration of a preferred embodiment of a gas detection device of the present invention.

[0044] Figure 2 is a plan view showing an important part of an embodiment of the present invention.

[0045] FIG. 3 is a plan view showing a cover shielding the substrate illustrated in FIG. 2 in an embodiment of the present invention.

[0046] Figure 4 is a configuration diagram of a gas sensor constituting a gas detection device according to an embodiment of the present invention.

[0047] Figure 5 is a configuration diagram of a device that uses sample gas to obtain learning data used in an embodiment of the present invention.

[0048] Figures 6a to 6c are diagrams explaining the process of obtaining normal distribution data using learning data of sample gas according to the present embodiment.

[0049] Figure 7 is an example diagram of the process of deep learning used in an embodiment of the present invention.

[0050] Figure 8 is an example diagram of a deep learning model used in an embodiment of the present invention.

[0051] Figure 9 is a flow chart showing a gas analysis method used in an embodiment of the present invention.

[0052] Figure 10 is an example diagram of a process for a gas analysis experiment used in an embodiment of the present invention.

[0053] Figure 11 is a drawing showing the experimental results showing the discrimination power for each gas according to the experiment of Figure 10.

[0054] Figure 12 is a drawing showing the experimental results showing the concentration error rate for each gas according to the experiment of Figure 10.

[0055]

[0056] Hereinafter, some embodiments of the present disclosure will be described in detail with reference to exemplary drawings. When designating components in each drawing, it should be noted that, where possible, identical components are given the same reference numerals, even if they appear in different drawings. Furthermore, when describing embodiments of the present disclosure, detailed descriptions of related known structures or functions will be omitted if they are deemed to hinder understanding of the embodiments of the present disclosure.

[0057] Figure 1 is a block diagram showing the configuration of a gas detection device according to an embodiment of the present invention. The gas detection device according to an embodiment of the present invention may include an analysis unit (200), a detection unit (300), a communication unit (400), and a management unit (500).

[0058] The above analysis unit (200) is a unit that analyzes the detected gas-related information and can also process a deep learning algorithm. The analysis unit (200) may include a processor (20) to be described below. The analysis unit (200) may analyze the type and concentration of gas using the sensing sensitivity of the gas detected by the detection unit (300) to be described below. The analysis unit (200) trains a deep learning learning model using training data for pre-prepared sample gases and trains the learning model using the sensing sensitivity of the target gas to be analyzed, thereby analyzing the type and concentration of each gas of the target gas. Accordingly, the analysis unit (200) may include a processor or microcomputer (micom) that can execute software or a program for analyzing the type and concentration of gas by training a deep learning learning model. Alternatively, it may include a PC equipped with such a processor or microcomputer.

[0059] The above detection unit (300) may include a first gas sensor (30), a second gas sensor (32), and a temperature / humidity sensor (34) to be described below. The detection unit (300) may detect information about a gas to be detected, as well as temperature and humidity around the device. Here, the first gas sensor (30) and the second gas sensor (32) may each detect different types of mixed gases. That is, the first gas sensor (30) may detect various types of gases, and the second gas sensor (32) may also detect various types of gases. Of course, the types of gases detected by the first gas sensor (30) and the second gas sensor (32) may all be different.

[0060] The above communication unit (400) is a configuration for communication between the device of the present invention and the outside. For example, it may include a level shifter (40), a connector (42), and first and second external connection pins (44, 46).

[0061] The above management unit (500) may include an ESD diode (50) and a regulator (42).

[0062] The various components constituting the embodiment of the present invention are mainly installed on a substrate (10). The substrate (10) may be a general printed circuit board. The printed circuit board may have a vertical dimension of approximately 19 mm to 24 mm based on the drawing, and a horizontal dimension of approximately 33 mm to 40 mm.

[0063] The above substrate (10) has a circuit pattern (not shown) formed inside, and both surfaces are made of a surface layer of insulating material. A portion of the circuit pattern exposed serves as an electrode, and various components are mounted on the electrode so as to be electrically connected.

[0064] A through channel (12) is formed by penetrating the substrate (10). The through channel (12) is formed along a position where the circuit pattern is not formed. In the illustrated embodiment, it has a 'C' shape. A temperature / humidity sensor (34) to be described below is installed in an area surrounded by the through channel (12). The through channel (12) can serve to block the transfer of heat generated from components mounted on the substrate (10). That is, it prevents heat generated from surrounding components from being transferred to the temperature / humidity sensor (34), thereby preventing the operation of the temperature / humidity sensor (34) from being affected.

[0065] A processor (20) is mounted on the substrate (10). The processor (20) can execute software or a program for analyzing the type and concentration of gas by training a deep learning learning model. In addition, the processor (20) can control the process of storing or transmitting various data to the outside. The processor (20) can be arranged so that its center is on the vertical center line based on the drawing. That is, the distance from the top and bottom (based on the drawing) of the processor (20) to the edge of the substrate (10) can be the same. The processor (20) is electrically connected to a circuit pattern by soldering a plurality of leads (not given drawing symbols) to electrodes of the substrate (10).

[0066] The substrate (10) has a first gas sensor (30) and a second gas sensor (32). The first gas sensor (30) may be, for example, located at the top (based on the drawing) of the processor (20), and the second gas sensor (32) may be located at the bottom (based on the drawing) of the processor (20). That is, the first sensor (30) and the second sensor (32) are positioned with the processor (20) interposed therebetween. That is, the first gas sensor (30) and the second gas sensor (32) may be respectively positioned in an area corresponding to the lengthwise ends of the processor (20) and the edge of the substrate (10). The specific configurations of the first and second gas sensors (30, 32) are described below.

[0067] A temperature / humidity sensor (34) may be installed in the area surrounded by the above-mentioned through channel (12). The position where the temperature / humidity sensor (34) is installed is between the edge of the processor (20) with the lead and one edge of the substrate (10). Therefore, the first gas sensor (30), the temperature / humidity sensor (34), and the second gas sensor (32) are arranged at an interval of approximately 90° between the edges of the substrate (10) corresponding to the processor (20). The temperature / humidity sensor (34) detects the temperature and humidity of the surroundings and provides them to the processor (20).

[0068] A level shifter (40) is installed on the opposite side of the temperature / humidity sensor (34) based on the above processor (20). The level shifter (40) is for communication with the product in which the device is used and is for 12C communication.

[0069] A connector (42) is provided on one side of the substrate (10) for signal connection between the substrate (10) and the outside (a product in which the device is used). The connector (42) is electrically connected to the circuit pattern of the substrate (10), and a separate external connector (not shown) is coupled to perform signal connection. In the illustrated embodiment, the connector (42) is located on one edge of the substrate (10). The connector (42) may be located on the edge opposite to the side where the temperature / humidity sensor (34) is installed.

[0070] The above connector (42) can be mounted at a position offset from one edge of the substrate (10), as shown in FIG. 2. In FIG. 2, it is installed relatively offset from the left edge toward the top with respect to the drawing. This can also serve as a reference for the position when mounting the substrate (10) on a product.

[0071] On both sides of the width direction of the connector (42), there may be a first external connection pin (44) and a second external connection pin (46). The first external connection pins (44) are a set of four, and can have the function of downloading a program used in the processor (20). The second external connection pins (46) are a set of two, and can be used to check sensor information.

[0072] The above substrate (10) has an ESD diode (50). There may be multiple ESD diodes (50). In the illustrated embodiment, there are three ESD diodes (50). The ESD diode (50) is for protection against static electricity. It is to prevent the sensor from being shocked and becoming inoperable due to static electricity generated by a worker's contact during assembly work.

[0073] The above substrate (10) may also use a regulator (52). The regulator (52) converts an externally supplied voltage into a voltage for driving the sensors (30, 32, 34). For example, if the voltage used in the product is 5 V and the voltage used in the sensors (30, 32, 34) is 3.3 V, the regulator (52) lowers the externally supplied voltage of 5 V to 3.3 V.

[0074] Meanwhile, in the embodiment of the present invention, a cover (60) may be used. The cover (60) may serve to shield the portion of the substrate (10) where various components, such as the processor (20), are installed from the outside. The cover (60) may serve to prevent static electricity.

[0075] The cover (60) can be injection-molded with an insulating material. The cover (60) has a cover body (61) having an edge shape corresponding to the surface shape of one side of the substrate (10) to form a skeleton. The cover body (61) has a processor shield (62). The processor shield (62) is located at a position corresponding to the position of the processor (20) on the substrate (10). The processor shield (62) protrudes compared to other parts of the cover body (61). This is designed in consideration of the degree to which the processor (20) protrudes from the substrate (10).

[0076] The processor shield (62) is formed with a plurality of through holes (62'). The through holes (62') serve to discharge heat generated from the processor (20) to the outside. The number and size of the through holes (62') are determined so that the processor shield (62) can maintain sufficient rigidity to protect the processor (20) while also allowing smooth heat dissipation.

[0077] Adjacent to the processor shield (62) are a first sensor window (63) and a second sensor window (63'). The first and second sensor windows (63, 63') allow the first gas sensor (30) and the second gas sensor (32) to be exposed to the outside to facilitate gas detection. To this end, the first sensor window (63) and the second sensor window (63') are formed at positions corresponding to the first gas sensor (30) and the second gas sensor (32) on the substrate (10). In the present embodiment, the first sensor window (63) and the second sensor window (63') are rectangular. However, any shape may be used as long as the first gas sensor (30) and the second gas sensor (32) can be accurately exposed to the outside.

[0078] The cover (60) has a third sensor window (64). The third sensor window (64) is located at a position corresponding to the temperature / humidity sensor (34) on the substrate (10). The third sensor window (64) is connected to the outside so that the temperature / humidity sensor (34) can accurately detect temperature and humidity. A partition grid (64') is formed on the third sensor window (64). The partition grid (64') serves to protect the temperature / humidity sensor (34) while allowing the temperature and humidity of the temperature / humidity sensor (34) to be detected smoothly through the third sensor window (64).

[0079] One edge of the cover (60) may have a substrate hooking lever (66) for coupling with the substrate (10). Two substrate hooking levers (66) may be provided side by side. In the present embodiment, the substrate hooking lever (66) is formed on one edge of the cover body (61) adjacent to the third sensor window (64). The edges perpendicular to both ends of the edge where the substrate hooking lever (66) is formed may be inserted into and hooked to a specific structure of the product together with the edge of the substrate (10).

[0080] According to Fig. 4, the first and second gas sensors (30, 32) may include a sensor substrate (110). The sensor substrate (110) may be a silicon substrate used in a general semiconductor process or a ceramic substrate such as Al2O3, ZrO2, or MgO having good properties at high temperatures.

[0081] A first insulating film (120) may be formed on the upper portion of the sensor substrate (110). The first insulating film (120) may be composed of a single or multiple silicon oxide films or silicon nitride films. A heater (130) may be formed on the upper portion of the first insulating film (120).

[0082] The heater (130) may serve to increase the ambient temperature to improve the detection characteristics of the gas. The heater (130) may be made of a metal or a conductive metal oxide such as Pt, Pd, W, or Au, and may be formed in various patterns consisting of lines. A second insulating film (140) may be formed on the upper portion of the heater (130). The second insulating film (140) may be composed of a single or multiple silicon oxide films or silicon nitride films.

[0083] A reference electrode (150) and a plurality of detection electrodes (160) may be formed on the upper portion of the second insulating film (140). No detection film may be formed on the upper portion of the reference electrode (150), and detection films (170) made of different detection materials may be formed on the upper portions of the detection electrodes (160).

[0084] In the present invention, the detection electrode (160) is composed of four detection electrodes (161 to 164). Of course, it is self-evident that the number of detection electrodes (160) can be changed. The detection film (170) is composed of a detection material that reacts with gas, and can cause a change in the electrical characteristics of the detection electrode (160) when reacting with gas.

[0085] Since a detection film is not formed on the above reference electrode (150), it does not react with gas. Electrical connection means for transmitting a gas detection signal to the analysis unit (200) may be provided at both ends of the reference electrode (150) and the detection electrode (160). The electrical connection means may be soldered to the substrate.

[0086] The reference electrode (150) and the sensing electrode (160) may be made of, for example, a metal such as Pt, Pd, W, or Au. The sensing electrode (160) may be disposed on the upper surface of the heater (130). Accordingly, the sensing electrode (160) may be heated by the heat provided from the heater (130). The gas sensing sensitivity of the sensing electrode (160) may be controlled by appropriately adjusting the heat provided from the heater (130). As shown in the drawing, a second insulating film (140) may be disposed between the heater (130) and the electrodes (150, 160) to electrically insulate the heater (130) and the electrodes (150). The heater (130) may also be connected to an external circuit by a heater electrode pad (not shown) and a bonding wire (not shown). Meanwhile, in another embodiment, the heater (130) may be placed at a position corresponding to the sensing electrode (160) on the lower surface of the substrate (110).

[0087] The first gas sensor (30) and the second gas sensor (32) used in the present invention can detect gases such as ethanol (C2H6O), hydrogen (H2), toluene (C7H8), and formaldehyde (HCHO). That is, the first sensing film (171) formed on the upper portion of the first sensing electrode (161) can react with ethanol, and the second sensing film (172) formed on the upper portion of the second sensing electrode (162) can react with hydrogen. The third sensing film (173) formed on the upper portion of the third sensing electrode (163) can react with toluene, and the fourth sensing film (174) formed on the upper portion of the fourth sensing electrode (164) can react with hydrogen.

[0088] As the first to fourth sensing films (171 to 174) and each gas react, the change in resistance of the first to fourth sensing electrodes (161 to 164) can be transmitted as sensing sensitivity to the analysis unit (200) via the bonding wire. In another embodiment, in order to analyze gases other than the above gases, a sensing film made of a sensing material that reacts to other gases can be formed on top of the sensing electrode (160).

[0089] The resistance of the reference electrode (150) can also be transmitted to the analysis unit (200) as a reference signal. The analysis unit (200) can analyze the type and concentration of the target gas to be analyzed by using the size of the sensing sensitivity of the detection electrode (160) compared to the reference signal of the reference electrode (150).

[0090] Meanwhile, a temperature sensor (not shown) for measuring the temperature of the heater (130) may be additionally provided. A thermistor may be used as the temperature sensor.

[0091] The first gas sensor (30) and the second gas sensor (32) configured in this way cause oxidation upon reaction with gas in the sensing film (170) formed on the surface of the sensing electrode (160), and the resistance decreases due to movement of electrons inside the sensing electrode (170). The analysis unit (200) can detect the target gas based on this change in resistance for each sensing electrode (170).

[0092] The number of channels of the above detection electrode (160) and the detection material of the detection film (170) can be determined according to the type of target gas to be detected. Different detection materials are used to detect different types of target gases.

[0093] Figure 5 is a configuration diagram of a device for obtaining learning data using a sample gas to be used in an embodiment of the present invention.

[0094] Referring to FIG. 5, the learning data acquisition device (100) of the present invention may be equipped with a gas chamber (101). A gas sensor (30, 32) may be placed in the internal space of the gas chamber (101).

[0095] A supply pipe (102) through which sample gas is supplied from the outside may be formed on one side of the gas chamber (101), and an exhaust pipe (103) through which sample gas is exhausted to the outside may be formed on the opposite side.

[0096] In the process of the sample gas supplied into the space through the above supply pipe (102) being exhausted through the exhaust pipe (103), the gas sensor (30, 32) can detect the sample gas.

[0097] In the embodiment of the present invention, the gas sensor (30, 32) may be configured with four channels of detection electrodes that detect ethanol (C2H6O), hydrogen (H2), toluene (C7H8), and formaldehyde (HCHO), respectively.

[0098] The above gas sensor (30, 32) can be electrically connected to a data processing unit (106). The data processing unit (106) can receive sensing sensitivity from the gas sensor (30, 32) and obtain training data for a deep learning learning model based on the sensing sensitivity for each sample gas.

[0099] Figures 6a to 6c are drawings for explaining the process of constructing a learning data DB of a sample gas according to an embodiment of the present invention.

[0100] In the present invention, for example, 72 gas sensors (30, 32) are arranged in the internal space of a gas chamber (101) to construct a learning data database, and sample gases are supplied to the internal space. The sample gases are, for example, ethanol, hydrogen, toluene, and formaldehyde, and the gas sensor (100) is arranged with four channels of detection electrodes (170) that detect each of these sample gases.

[0101] In the data processing unit (106), learning data for a deep learning model is acquired based on the sensing sensitivity of each detection electrode of 72 gas sensors (30, 32) and a database (DB) is constructed.

[0102] Below, the DB construction process is described in detail with reference to FIGS. 6a to 6c.

[0103] As an example, the sensing sensitivity of 72 first gases is shown in Fig. 6a. That is, the sensing sensitivity of the first gas detected by 72 first sensing electrodes (161).

[0104] The sensing sensitivity data of the first gas is converted into distribution data by sensing sensitivity for the first gas, as shown in Fig. 6b.

[0105] At this time, the distribution data by sensing sensitivity in Fig. 6b does not show distribution data for all sensing sensitivities. That is, if such distribution data is used as learning data, errors may occur during gas analysis through learning.

[0106] Therefore, in order to display distribution data for all sensing sensitivities, normal distribution data is generated based on the mean and standard deviation of the distribution data in Fig. 6b. This is done by normalizing the distribution data as in Fig. 6c to generate normal distribution data, thereby ensuring that distribution data for all sensing sensitivities exist.

[0107] Figure 6c shows the result of converting the distribution data by sensing sensitivity into normal distribution data. The data processing unit (106) sets the normal distribution data as training data for a deep learning learning model and constructs a training data DB.

[0108] Here, in order to build a DB with high-quality learning data, it is necessary to consider the dispersion between each gas sensor (30, 32), the dispersion between each detection electrode (170), and the dispersion according to measurement error, and for this purpose, the largest possible amount of sensing sensitivity values ​​are required.

[0109] The 72 sensing sensitivity values ​​presented in the present invention are provided as examples for convenience of explanation. It is better to build a learning data database based on a larger number of sensing sensitivity values. Preferably, more than 700,000 learning data points are constructed. Since it is difficult to obtain sensing sensitivity in all continuous areas with only 700,000 learning data points, virtual learning data points are generated by normalizing the distribution data based on the mean and standard deviation of the distribution data points. This enables accurate gas analysis without errors even for untrained data points.

[0110] Figures 6a to 6c illustrate examples of constructing a learning data database for the sensing sensitivity of a first gas detected by a first sensing electrode. Therefore, in order to construct a learning data database for other second, third, and fourth gases, a learning data database for the second, third, and fourth gases is also required through the same process as Figures 6a to 6c.

[0111] Figure 7 illustrates an example of a deep learning process according to an embodiment of the present invention. According to this, a two-stage learning model is applied to analyze gas in the present invention.

[0112] These learning models may be generated by training a deep learning algorithm using pre-prepared training data. The training data may be the sensing sensitivity data generated as shown in Figures 5 and 6a-6c.

[0113] The learning model in the first stage may be a classification model. The learning model in the second stage may be a regression model.

[0114] The classification model in the first stage inputs the sensing sensitivity of multiple gases, trains the learning model with the sensing sensitivity data, and outputs the type of each gas.

[0115] The regression model in the second stage takes the sensing sensitivity and gas type of multiple gases as inputs, learns the sensing sensitivity and gas type data, and outputs the concentration of each gas.

[0116] Figure 8 is an exemplary diagram of a deep learning model according to an embodiment of the present invention. In the present invention, the learning model may include a first-stage classification model and a second-stage regression model.

[0117] The classification model is a fully connected neural network (FCN).

[0118] The regression model is multiple linear regression (MLR).

[0119] As a learning method, supervised learning is used so that FCN predicts the type of gas and MLR predicts the concentration of gas.

[0120] In the example of the drawing, the FCN receives the gas sensing sensitivity of four channels (c1, c2, c3, c4). That is, for example, the sensing sensitivity of hydrogen, ethanol, toluene, and formaldehyde is input as the input layer, and it uses hidden layer 1 with 12 nodes and hidden layer 2 with 24 nodes, and the activation function of the hidden layer uses ReLU (Recited Linear Unit). In the output layer, Softmax is used to normalize the output value to a probability value between 0 and 1 to perform supervised learning.

[0121] MLR receives the gas type output from FCN and the gas sensing sensitivity of the four channels (c1, c2, c3, c4) as input layers, uses the same hidden layers 1 and 2 as FCN, and also uses ReLU (Recited Linear Unit) as the activation function of the hidden layer. In the output layer, supervised learning is performed so that linear concentration values ​​can be output without an activation function.

[0122] In the present disclosure, in the case of a mixed gas containing different types of gases, when the concentration is predicted using a regression model and the gas is determined to be a mixed gas, the regression model can be designed to output the concentration separately for each gas.

[0123] Figure 9 is a flow chart showing a gas analysis method performed in an embodiment of the present invention.

[0124] Referring to FIG. 9, in the gas analysis method according to the present invention, learning data for a deep learning model is acquired for a sample gas (S101).

[0125] The above learning data uses normal distribution data generated by detecting a sample gas from a gas sensor for sample gas, converting the sensing sensitivity data of the sample gas into distribution data by sensing sensitivity, and normalizing the distribution data based on the mean and standard deviation of the distribution data.

[0126] A plurality of different types of target gases are detected by the gas sensor (30, 32) (S102).

[0127] Deep learning is performed by applying the sensing sensitivity of the target gas to the learning model (S103).

[0128] In the case of deep learning, a deep learning model is trained using training data for sample gas, and then when the target gas is detected, the trained deep learning model is trained using the sensing sensitivity of the target gas.

[0129] In the analysis unit (300), the type and concentration of the target gas are analyzed using the learning results of the deep learning model (S104).

[0130] In the present invention, deep learning uses FCN in the first stage and MLR in the second stage.

[0131] FCN receives the sensing sensitivity of the target gas as input and outputs the type of target gas, and MLR receives the type of target gas and the sensing sensitivity of the target gas output from FCN and outputs the concentration of the target gas.

[0132] In the present invention, the target gases are ethanol, hydrogen, toluene, and formaldehyde.

[0133] Figure 10 shows an experimental process for gas analysis according to the present invention.

[0134] In this experiment, four types of gases (ethanol, hydrogen, toluene, formaldehyde) and air were supplied into the gas chamber (101). Gas sensors (30, 32) are placed in the gas chamber (101).

[0135] The supply amount of each gas was finely controlled by a mass flow controller (MFC) (33) and supplied in an appropriate amount according to the measured concentration.

[0136] The output value of the gas sensor (30, 32) was confirmed through a PC (or processor (20)) connected to the gas sensor (30, 32).

[0137] In this experiment, 72 gas sensors were placed within a gas chamber and the sensing sensitivities of a total of 144 sensors were measured twice each. The experimental results are shown in Figures 11 and 12.

[0138] As shown in Figure 11, the discrimination power (accuracy) of the four gases was 94% for hydrogen, 97% for ethanol, 97% for toluene, and 82% for formaldehyde.

[0139] As shown in Figure 12, the concentration error rates of the four gases were found to be within approximately ±30% for hydrogen, ethanol, and toluene, and within approximately ±50% for formaldehyde.

[0140] As described above, the gas detection device according to the present disclosure can accurately measure harmful gases generated in real-time, and thus can be installed and used in conjunction with air conditioners, air purifiers, and the like. This allows for appropriate control of air conditioners and air purifiers based on the type and concentration of the detected harmful gases, thereby providing a pleasant environment for users.

[0141] Although all components constituting the embodiments of the present disclosure have been described as being combined or operating in combination, the present disclosure is not necessarily limited to such embodiments. That is, within the scope of the present disclosure, all components may be selectively combined and operated one or more times. Furthermore, terms such as "include," "comprise," or "have" described above, unless specifically stated otherwise, imply that the corresponding component may be inherent. Therefore, they should be interpreted to include other components rather than excluding other components.

Claims

1. Substrate and, A processor installed on the above substrate for analyzing the detected data, It includes a first gas sensor that detects different types of target gases and a second gas sensor that detects different types of target gases, which are installed with the above processor in between. A gas detection device wherein the first gas sensor is positioned between one edge of the processor and a corresponding edge of the substrate, and the second gas sensor is positioned between the other edge of the processor and a corresponding edge of the substrate.

2. In paragraph 1, the processor is a gas detection device that analyzes the type and concentration of the detected target gas.

3. In the second paragraph, the processor obtains learning data on the sensing sensitivity of the sample gas, trains a deep learning learning model with the learning data, and trains the learning model with the sensing sensitivity of the detected target gas to analyze the type and concentration of the target gas, wherein the learning model includes a classification model of the first stage and a regression model of the second stage.

4. A gas detection device in the third paragraph, wherein the classification model is a Fully Connected Neuronal Network (FCN) and the regression model is a Multiple Linear Regression (MLR).

5. In the fourth paragraph, the FCN is a gas detection device that receives the sensing sensitivity of the detected target gas and outputs the type of the target gas, and the MLR receives the type of the target gas output from the FCN and the sensing sensitivity of the detected target gas and outputs the concentration of the target gas.

6. A gas detection device according to claim 1, wherein the processor obtains the sensing sensitivities of the plurality of sample gases, generates distribution data of the sensing sensitivities, and generates normal distribution data based on the average and standard deviation of the distribution data, thereby obtaining the normal distribution data as the learning data.

7. In the first paragraph, a gas detection device wherein one of the first gas sensor and the second gas sensor detects ethanol (C2H6O), formaldehyde (HCHO), toluene (C7H8), or hydrogen (H2).

8. A gas detection device according to any one of claims 1 to 7, wherein a temperature / humidity sensor is installed between one edge of the substrate and an edge of the processor.

9. A gas detection device in accordance with claim 8, wherein a through channel is formed through the substrate between the temperature / humidity sensor and the processor.

10. A gas detection device in accordance with claim 9, wherein the through channel is formed to surround the temperature / humidity sensor except for some sections.

11. In the 8th paragraph, the temperature / humidity sensor, the first gas sensor, and the second gas sensor are a gas detection device arranged at 90° intervals between the processor and the corresponding edge of the substrate based on the processor.

12. A gas detection device in accordance with claim 1, wherein a connector for signal connection with the outside is installed on one edge of the substrate, and the connector is positioned offset from the edge.

13. A gas detection device according to claim 1, wherein an ESD diode for removing static electricity is used on the substrate.

14. A gas detection device in accordance with claim 1, wherein a cover is further provided to shield a surface of the substrate on which the processor is installed from the outside, wherein a processor shielding portion is formed in a position corresponding to the position of the processor in the cover, and a first sensor window, a second sensor window, and a third sensor window are respectively formed between an edge of the processor shielding portion and an edge of the cover.

15. A gas detection device in accordance with claim 14, wherein a partition grid is formed in the third sensor window.

Images