Gas detection device

Abstract

The present application is a gas detection device. In the present application, a substrate (10) is provided with a processor (20), a first gas sensor (30), and a second gas sensor (32). The first gas sensor (30) and the second gas sensor (32) are located between edges corresponding to both ends of the processor (20) and the substrate (10), respectively. The substrate (10) is provided with a temperature / humidity sensor (34), and a through channel (12) is formed between the temperature / humidity sensor (34) and the processor (20) and penetrates the substrate (10). The first gas sensor (30) and the second gas sensor (32) respectively detect the types and concentrations of multiple gases, and the processor (20) analyzes the types and concentrations of each gas in the detected mixed gas using a detection and deep learning learning model for mixed gas. At this time, the learning model uses a first-stage classification model and a second-stage regression model.

Description

Technical Field

[0001] This invention relates to a gas detection device for real-time detection of multiple harmful gases. Background Technology

[0002] As air pollution intensifies, public concern about air quality is growing, leading to an increasing demand for indoor air quality management and maintenance. Not only in industrial settings, but also in daily life environments, including homes, the emission of various harmful gases is increasing, and their types are becoming more diverse.

[0003] As a result, technologies for real-time monitoring of indoor environmental information have been developed, and in recent years, the demand for gas sensors used to detect various harmful gases that cause phenomena such as new home syndrome has gradually increased.

[0004] Gas sensors are typically used to detect harmful gases in the air and to determine their composition and concentration. Gas sensors can be categorized into various types based on their measurement methods, such as semiconductor, catalytic, and optical sensors. Among these, semiconductor gas sensors are receiving significant research attention due to their simple manufacturing process and the use of inexpensive metal oxides.

[0005] As existing technology, techniques for detecting harmful gases using gas sensors have been proposed. For example, Korean Patent Publication 10-2021-0104390 detects pollution by measuring the ionic conductivity of the electrolyte in the gas sensor. However, since the ionic conductivity varies with temperature, the output value increases or decreases with temperature changes. By correcting the output value, a toxic gas is detected when the corrected value reaches a threshold. However, although multiple gas sensors are used to detect various gases, the actual gas concentration cannot be determined by measuring only the sensor current value. The use of an electrolyte also compromises durability, and the temperature-dependent conductivity leads to a higher error rate. Furthermore, the use of large-sized gas sensors results in a problem where the device size increases proportionally to the number of gases being detected.

[0006] As another example, Korean Patent No. 10-1674048 discloses a device that has the function of detecting harmful chemical substances and the function of linkage via a network, so that it can spread rapidly in the event of an accident. The device consists of a pretreatment unit for concentrating harmful gases and maintaining a predetermined temperature and humidity, an environmental sensor unit for collecting information about the external environment, a communication unit for communication with external devices, a sensor array unit for detecting multiple harmful gases, a display, an input unit, and a control unit, and discloses that most of these are housed in a single housing.

[0007] However, the presence of a heater for maintaining temperature and removing adsorbed gases necessitates a barrier to prevent the heater's heat from causing adverse effects, resulting in a larger gas measuring device. The use of a heater also increases power consumption, and the sensor response cannot be real-time due to the need to remove adsorbed gases. Furthermore, the inclusion of a hazardous gas concentration section means the sensor needs to concentrate the gas for detection, leading to difficulties in detecting low concentrations and real-time monitoring.

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

[0009] However, it is necessary not only to predict the types of harmful gases in the room in real time, but also to predict their concentrations in real time. It is also necessary to correct the sensitivity values ​​that change with temperature and humidity to obtain constant output values, and to minimize the size of the entire device to minimize the space required for installation. Summary of the Invention

[0010] The problem that the invention aims to solve

[0011] The purpose of this invention is to provide a miniaturized gas detection device.

[0012] The purpose of this invention is to provide a gas detection device that uses a gas sensor to detect gas in order to accurately provide the type and concentration of the gas.

[0013] The purpose of this invention is to provide a gas detection device that can achieve miniaturization of the overall device while still being able to detect the type and concentration of gas in real time.

[0014] The purpose of this invention is to enable the training of a deep learning model based on pre-prepared training data, thereby analyzing the types and concentrations of gases to provide information on gas concentration and type.

[0015] This invention corrects the sensitivity value that changes with temperature / humidity to ensure that a constant output value is always output.

[0016] Technical solutions to the problem

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

[0018] This invention obtains training data on the sensing sensitivity of a sample gas, uses the training data to train a deep learning model, and uses the sensing sensitivity of the detected target gas to train the learning model, so as to analyze the type and concentration of the target gas.

[0019] In this invention, a processor for analyzing the type and concentration of a target gas is used as a reference, and its two ends can be provided with a first gas sensor, a second gas sensor, and a temperature / humidity sensor at a predetermined angular interval.

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

[0021] The present invention may include: a substrate; a processor disposed on the substrate for analyzing detected data; and a first gas sensor and a second gas sensor disposed between the processor, wherein the first gas sensor detects target gases of different types and the second gas sensor detects target gases of different types; the first gas sensor may be located between one edge of the processor and the corresponding edge of the substrate, and the second gas sensor may be located between the other edge of the processor and the corresponding edge of the substrate.

[0022] The processor can analyze the type and concentration of each of the detected target gases.

[0023] The processor can acquire training data on the sensing sensitivity of the sample gas, and can use the training data to train a deep learning model. The learning model can be trained by using the sensing sensitivity of the detected target gas to analyze the type and concentration of the target gas. The learning model may include a first-stage classification model and a second-stage regression model.

[0024] The classification model can be a fully connected neural network (FCN), and the regression model can be multiple linear regression analysis (MLR).

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

[0026] The processor can acquire the sensing sensitivity of a plurality of sample gases, generate distribution data of the sensing sensitivity, and generate normal distribution data based on the mean and standard deviation of the distribution data, so as to acquire the normal distribution data as the training data.

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

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

[0029] A through-channel can be formed between the temperature / humidity sensor and the processor, and the through-channel extends through the substrate.

[0030] The through-channel can be formed to surround all but one portion of the temperature / humidity sensor.

[0031] The temperature / humidity sensor, the first gas sensor, and the second gas sensor can be arranged at 90° intervals between the corresponding edges of the processor and the substrate.

[0032] A connector for connecting to an external signal may be provided on one side edge of the substrate, and the connector is located on one side of the edge.

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

[0034] The gas detection device may also have a cover that isolates the surface of the substrate in which the processor is disposed from the outside. A processor shielding portion is formed in the cover at a position corresponding to the position of the processor. A first sensor window, a second sensor window, and a third sensor window may be formed between the edge of the processor shielding portion and the edge of the cover, respectively.

[0035] The third sensor window can be divided into grids.

[0036] Invention Effects

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

[0038] This invention provides a miniaturized gas detection device for real-time detection of the type and concentration of various harmful gases. Multiple MEMS gas sensors are arranged with spacing between them and other components mounted on a substrate, ensuring necessary distance between the gas sensors. Furthermore, through-channels are formed in the substrate to prevent the temperature / humidity sensors from being affected by heat from surrounding components. Therefore, the distance between components mounted on the substrate can be minimized, thereby enabling the overall device to be miniaturized.

[0039] This invention uses a deep learning model, which enables it to provide the type and concentration of detected gases in real time.

[0040] This invention uses a deep learning model to analyze the composition and concentration of each gas in a gas mixture at various ratios.

[0041] Furthermore, this invention corrects for sensitivity changes based on temperature / humidity variations. To this end, the sensor sensitivity is calibrated using absolute humidity values. This process brings the sensitivities arising from differences in absolute humidity closer together, and when applied to deep learning algorithms, it significantly reduces the error rate. Attached Figure Description

[0042] Figure 1 This is a block diagram illustrating the configuration of a preferred embodiment of the gas detection device of the present invention.

[0043] Figure 2 This is a top view illustrating an important part of an embodiment of the present invention.

[0044] Figure 3 This illustrates the shielding in an embodiment of the present invention. Figure 2 The top view of the cover of the substrate shown.

[0045] Figure 4 This is a structural diagram of the gas sensor constituting the gas detection device according to an embodiment of the present invention.

[0046] Figure 5 This is a schematic diagram of a device using sample gas to obtain training data in an embodiment of the present invention.

[0047] Figures 6a to 6c This is a diagram illustrating the process of obtaining normally distributed data using the training data of the sample gas in this embodiment.

[0048] Figure 7 This is an example diagram of the deep learning process used in the embodiments of the present invention.

[0049] Figure 8 This is an example diagram of the deep learning model used in the embodiments of the present invention.

[0050] Figure 9 This is a flowchart illustrating the gas analysis method used in an embodiment of the present invention.

[0051] Figure 10 This is an example diagram of the process used in the gas analysis experiment in the embodiments of the present invention.

[0052] Figure 11 It shows the basis Figure 10 The experimental results of the differentiating forces of each gas are shown in the figure.

[0053] Figure 12 It shows the basis Figure 10The experimental results are shown in the figure, which represents the error rate of each gas concentration obtained from the experiment. Detailed Implementation

[0054] Hereinafter, some embodiments of the present disclosure will be described in detail with reference to the exemplary accompanying drawings. When affixing reference numerals to the constituent elements of the various drawings, it should be noted that the same constituent elements are affixed with the same reference numerals as much as possible, even if they are shown in different drawings. Furthermore, in describing embodiments of the present disclosure, detailed descriptions of related well-known structures or functions that are deemed to hinder understanding of the embodiments of the present disclosure are omitted.

[0055] Figure 1 This is a block diagram illustrating 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.

[0056] The analysis unit 200 is responsible for analyzing information related to the detected gas and can also process deep learning algorithms. The analysis unit 200 may include the processor 20 described later. The analysis unit 200 can analyze the type and concentration of the gas using the sensing sensitivity of the gas detected by the detection unit 300 described later. The analysis unit 200 trains a deep learning model using pre-prepared training data of 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 in the target gas. Accordingly, the analysis unit 200 may include a processor or microcontroller capable of running software or programs for analyzing the type and concentration of gases by training a deep learning model. Alternatively, it may include a PC having such a processor or microcontroller.

[0057] The detection unit 300 may include a first gas sensor 30, a second gas sensor 32, and a temperature / humidity sensor 34, described later. The detection unit 300 can detect information about the target gas and the temperature and humidity around the device. The first gas sensor 30 and the second gas sensor 32 can each detect different types of mixed gases. That is, the first gas sensor 30 can detect multiple types of gases, and the second gas sensor 32 can also detect multiple types of gases. Of course, the types of gases detected by each of the first gas sensors 30 and the second gas sensor 32 may be completely different.

[0058] The communication unit 400 is a configuration for the device of the present invention to communicate with the outside. For example, it includes a level converter 40, a connector 42, a first external connection pin 44, and a second external connection pin 46.

[0059] The management unit 500 may include an ESD diode 50 and a regulator 42.

[0060] The various components constituting the embodiments of the present invention are mainly disposed on the substrate 10. The substrate 10 can be a common printed circuit board. The printed circuit board can have a value of approximately 19 mm to 24 mm in the vertical direction and approximately 33 mm to 40 mm in the horizontal direction, with reference to the accompanying drawings.

[0061] The substrate 10 has a circuit pattern (not shown) formed inside, and its two side surfaces are formed as a surface layer of insulating material. A portion of the exposed part of the circuit pattern forms an electrode, on which various components can be mounted to achieve electrical connection.

[0062] A through-channel 12 is formed through the substrate 10. The through-channel 12 is formed along a position where the circuit pattern is not formed. In the illustrated embodiment, it is formed in a "C" shape. A temperature / humidity sensor 34, described later, is disposed in the area surrounded by the through-channel 12. The through-channel 12 functions to block the transfer of heat generated by the components mounted on the substrate 10. That is, heat generated by surrounding components is not transferred to the temperature / humidity sensor 34, so that the operation of the temperature / humidity sensor 34 is not affected.

[0063] The substrate 10 is equipped with a processor 20. The processor 20 runs software or programs for analyzing the type and concentration of gases by training a deep learning model. Additionally, it can control various data storage or external transmission processes. Referring to the accompanying drawings, the processor 20 can be configured with its center located on a vertical centerline. That is, the distance from the top and bottom ends of the processor 20 (referring to the accompanying drawings) to the edge of the substrate 10 can be the same. A plurality of pins of the processor 20 (not labeled in the drawings) are electrically connected to a circuit pattern by soldering electrodes to the substrate 10.

[0064] The substrate 10 is provided with a first gas sensor 30 and a second gas sensor 32. The first gas sensor 30 may be disposed at the upper end of the processor 20 (based on the accompanying drawings), and the second gas sensor 32 may be disposed at the lower end of the processor 20 (based on the accompanying drawings). That is, the first sensor 30 and the second sensor 32 are disposed across the processor 20. Specifically, the first gas sensor 30 and the second gas sensor 32 can be respectively configured in the regions between the two ends of the processor 20 in the longitudinal direction and the edge of the substrate 10. The specific configuration of the first gas sensor 30 and the second gas sensor 32 will be described later.

[0065] A temperature / humidity sensor 34 can be disposed in the area surrounded by the through channel 12. The temperature / humidity sensor 34 can be positioned between the edge where the pins of the processor 20 are located 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 approximately 90° intervals between the edges of the substrate 10 corresponding to the processor 20. The temperature / humidity sensor 34 detects the ambient temperature and humidity and provides this information to the processor 20.

[0066] A level converter 40 is provided on the opposite side of the temperature / humidity sensor 34, with reference to the processor 20. The level converter 40 is used for communication with products using this device, for I2C communication.

[0067] For signal connection between the substrate 10 and an external device (a product using this device), a connector 42 is located on one side of the substrate 10. The connector 42 is electrically connected to the circuit pattern of the substrate 10 and can be coupled with an additional external connector (not shown) to perform signal connection. In the illustrated embodiment, the connector 42 is located at one edge of the substrate 10. The connector 42 may also be located at the edge opposite to the side where the temperature / humidity sensor 34 is located.

[0068] like Figure 2 As shown, the connector 42 can be mounted on one side of the substrate 10 at a position offset to one side. Figure 2 In the accompanying drawing, the connector 42 is positioned on the left edge at a location relatively offset towards the upper end. This can also serve as a positioning reference when mounting the substrate 10 onto the product.

[0069] First external connecting pins 44 and second external connecting pins 46 can be provided on both sides of the connector 42 in the width direction. The first external connecting pins 44 are in groups of four and can be used to download the program used in the processor 20. The second external connecting pins 46 are in groups of two and can be used to confirm sensor information.

[0070] The substrate 10 is provided with ESD diodes 50. Multiple ESD diodes 50 may be provided. In the illustrated embodiment, three ESD diodes 50 are provided. The ESD diodes 50 are used to provide electrostatic discharge protection. Their purpose is to prevent the sensor from being damaged by static electricity generated during assembly operations due to contact with personnel, thus preventing it from malfunctioning.

[0071] The substrate 10 may also use a regulator 52. The regulator 52 is used to convert the voltage supplied from the outside into a voltage used to drive the sensors 30, 32, and 34. For example, if the product uses a voltage of 5V and the sensors 30, 32, and 34 use a voltage of 3.3V, then the regulator 52 functions to reduce the 5V voltage supplied from the outside to 3.3V.

[0072] On the other hand, embodiments of the present invention may use a cover 60. The cover 60 serves to isolate the portion of the substrate 10 in which various components, such as the processor 20, are disposed from the outside. The cover 60 also performs a function to prevent static electricity.

[0073] The cover 60 can be injection molded from an insulating material. A cover body 61, having an edge shape corresponding to the shape of one side surface of the substrate 10, forms the skeleton of the cover 60. The cover body 61 is provided with a processor shielding portion 62. The processor shielding portion 62 is located at a position corresponding to the position of the processor 20 in the substrate 10. The processor shielding portion 62 protrudes beyond the rest of the cover body 61. This is designed with consideration of the degree to which the processor 20 protrudes from the substrate 10.

[0074] The processor shielding portion 62 has a plurality of through holes 62'. The through holes 62' serve to dissipate the heat generated by the processor 20 to the outside. The number or size of the through holes 62' can be determined so that the processor shielding portion 62 can smoothly dissipate heat while maintaining sufficient rigidity to protect the processor 20.

[0075] A first sensor window 63 and a second sensor window 63' are disposed adjacent to the processor shielding portion 62. The first sensor window 63 and the second sensor window 63' allow the first gas sensor 30 and the second gas sensor 32 to be exposed to the outside for smooth gas detection. Therefore, 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 in the substrate 10. In this embodiment, the first sensor window 63 and the second sensor window 63' are formed as quadrilaterals. However, they can be of any shape as long as the first gas sensor 30 and the second gas sensor 32 can be accurately exposed to the outside.

[0076] The cover 60 is provided with a third sensor window 64. The third sensor window 64 is located at a position corresponding to the temperature / humidity sensor 34 in the substrate 10. The third sensor window 64 communicates with the outside to enable the temperature / humidity sensor 34 to accurately detect temperature and humidity. The third sensor window 64 forms a dividing grid 64'. The dividing grid 64' serves to facilitate the smooth detection of temperature and humidity by the temperature / humidity sensor 34 through the third sensor window 64 while protecting the temperature / humidity sensor 34.

[0077] One edge of the cover 60 may be provided with a substrate mounting lever 66 for engaging with the substrate 10. Two substrate mounting levers 66 may be arranged side-by-side. In this embodiment, the substrate mounting lever 66 is formed on the edge of the cover body 61 adjacent to the third sensor window 64. The edges orthogonal to the two ends of the edge where the substrate mounting lever 66 is formed can be inserted together with the edge of the substrate 10 and hung on a specific structure of the product.

[0078] according to Figure 4 The first gas sensor 30 and the second gas sensor 32 may include a sensor substrate 110. The sensor substrate 110 may utilize a silicon substrate used in conventional semiconductor engineering or a ceramic substrate such as Al2O3, ZrO2, or MgO that has good properties at high temperatures.

[0079] A first insulating film 120 may be formed on the upper part 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 part of the first insulating film 120.

[0080] The heater 130 serves to raise the ambient temperature to improve the gas detection characteristics. The heater 130 can be made of metals such as Pt, Pd, W, or Au, or conductive metal oxides, and can be formed into various patterns composed of lines. A second insulating film 140 can be formed on the upper part of the heater 130. The second insulating film 140 can be composed of a single or multiple silicon oxide films or silicon nitride films.

[0081] A reference electrode 150 and a plurality of detection electrodes 160 may be formed on the upper part of the second insulating film 140. No detection film is formed on the upper part of the reference electrode 150, but a detection film 170 made of different detection materials may be formed on the upper part of the detection electrodes 160.

[0082] The detection electrode 160 of the present invention consists of four detection electrodes 161 to 164. Of course, the number of detection electrodes 160 can be varied. The detection membrane 170 can be made of a detection material that reacts with gas, and when it reacts with gas, it can produce a change in the electrical properties of the detection electrode 160.

[0083] The reference electrode 150 does not have a detection film formed on it, and therefore does not react with the gas. Electrical connection means for transmitting gas detection signals to the analysis unit 200 can be provided at both ends of the reference electrode 150 and the detection electrode 160. These electrical connection means can be soldered to a substrate.

[0084] The reference electrode 150 and the detection electrode 160 can be metals such as Pt, Pd, W, or Au. The detection electrode 160 can be disposed on the top surface of the heater 130. Accordingly, the detection electrode 160 can be heated by heat supplied from the heater 130. The gas sensing sensitivity in the detection electrode 160 is controlled by appropriately adjusting the heat supplied from the heater 130. As shown, a second insulating film 140 can be disposed between the heater 130 and the electrodes 150 and 160 to electrically insulate the heater 130 from the electrode 150. The heater 130 can also be connected to an external circuit using heater electrode pads (not shown) and bonding wires (not shown). Alternatively, in another embodiment, the heater 130 can also be disposed on the bottom surface of the substrate 110 at a position corresponding to the detection electrode 160.

[0085] The first gas sensor 30 and the second gas sensor 32 used in this invention can detect gases such as ethanol (C2H6O), hydrogen (H2), toluene (C7H8), and formaldehyde (HCHO). Specifically, the first detection membrane 171 formed on the upper part of the first detection electrode 161 can react with ethanol, and the second detection membrane 172 formed on the upper part of the second detection electrode 162 can react with hydrogen. The third detection membrane 173 formed on the upper part of the third detection electrode 163 can react with toluene, and the fourth detection membrane 174 formed on the upper part of the fourth detection electrode 164 can react with hydrogen.

[0086] As the first to fourth detection films 171-174 react with each gas, the resistance changes of the first to fourth detection electrodes 161-164 can be transmitted to the analysis unit 200 as sensing sensitivity via the bonding wires. In another embodiment, in order to analyze gases other than those mentioned above, a detection film made of a detection material that reacts with other gases may be formed on the upper part of the detection electrode 160.

[0087] 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 by comparing the sensing sensitivity of the detection electrode 160 with the reference signal from the reference electrode 150.

[0088] Alternatively, a temperature sensor (not shown) can be added to measure the temperature of the heater 130. The temperature sensor can be a thermistor.

[0089] The first gas sensor 30 and the second gas sensor 32 configured in this way undergo oxidation when the detection film 170 formed on the surface of the detection electrode 160 reacts with the gas. As a result, the resistance decreases due to the migration of electrons inside the detection electrode 170. The analysis unit 200 can detect the target gas based on this resistance change of each detection electrode 170.

[0090] The number of channels in the detection electrode 160 and the detection material of the detection membrane 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.

[0091] Figure 5 This is a schematic diagram of an apparatus for acquiring training data using the sample gas used in the embodiments of the present invention.

[0092] Reference Figure 5 The training data acquisition device 100 of the present invention may be provided with a gas chamber 101. The internal space of the gas chamber 101 may be configured with gas sensors 30 and 32.

[0093] A supply pipe 102 for supplying sample gas from the outside can be formed on one side of the gas chamber 101, and an exhaust pipe 103 for discharging sample gas to the outside can be formed on the opposite side.

[0094] During the process of the sample gas supplied into the space through the supply pipe 102 being discharged through the exhaust pipe 103, the gas sensors 30 and 32 can detect the sample gas.

[0095] In this embodiment of the invention, the gas sensors 30 and 32 may be configured with detection electrodes for four channels respectively detecting ethanol (C2H6O), hydrogen (H2), toluene (C7H8), and formaldehyde (HCHO).

[0096] The gas sensors 30 and 32 can be electrically connected to the data processing unit 106. The data processing unit 106 receives the sensing sensitivity from the gas sensors 30 and 32, and obtains training data for the deep learning model based on the sensing sensitivity of each sample gas.

[0097] Figures 6a to 6cThis is a diagram illustrating the process of training data DB for component sample gas in an embodiment of the present invention.

[0098] To train the construction of the database, this invention, for example, sets up 72 gas sensors 30, 32 inside the gas chamber 101 and supplies sample gas to the internal space. The sample gas may be, for example, ethanol, hydrogen, toluene, or formaldehyde. The gas sensors 100 are equipped with detection electrodes 170 in four channels for detecting these sample gases respectively.

[0099] The data processing unit 106 acquires training data for the deep learning model based on the sensing sensitivity of each detection electrode in the 72 gas sensors 30, 32, and constructs a database (DB).

[0100] The following is for reference Figures 6a to 6c This section details the DB construction process.

[0101] Figure 6a As an example, the sensing sensitivity of 72 first gases is shown. That is, the sensing sensitivity of the first gases detected by 72 first detection electrodes 161.

[0102] The sensing sensitivity data of the first gas, as follows Figure 6b As shown, this is converted into distribution data for different sensing sensitivities of the first gas.

[0103] at this time, Figure 6b The distribution data for different sensing sensitivities does not show the distribution data for all sensing sensitivities. That is, if such distribution data is used as training data, errors may occur when performing gas analysis using the trained data.

[0104] Therefore, in order to show the distribution data of all sensing sensitivities, based on Figure 6b The mean and standard deviation of the distributed data generate normally distributed data. This is as follows: Figure 6c As shown, normally distributed data is generated by normalizing the distributed data, so that the distribution data of all sensing sensitivities are present.

[0105] Figure 6c This is the result of converting distribution data with different sensing sensitivities into normally distributed data. The data processing unit 106 sets the normally distributed data as training data for a deep learning model and constructs a training data database (DB).

[0106] In order to build the database using high-quality training data, it is necessary to consider the distribution among the gas sensors 30 and 32, the distribution among the detection electrodes 170, and the distribution caused by measurement errors. For this purpose, a large number of sensing sensitivity values ​​are required.

[0107] The 72 sensing sensitivity values ​​proposed in this invention are for ease of explanation; it is preferable to construct a training dataset (DB) based on more sensing sensitivity values. Preferably, more than 700,000 training data points are constructed. However, since it is difficult to obtain the sensing sensitivity of all continuous regions using only 700,000 training data points, the distribution data is normalized based on the mean and standard deviation to generate virtual training data. Therefore, accurate gas analysis can be achieved even with untrained data.

[0108] Figures 6a to 6c This is an example of constructing a database (DB) of training data on the sensing sensitivity of the first gas detected by the first detection electrode. Therefore, in order to construct training data DBs for other second, third, and fourth gases, it is also necessary to use data from the second, third, and fourth gases... Figures 6a to 6c The same process is used to build the training data database.

[0109] Figure 7 An example diagram illustrating the deep learning process of an embodiment of the present invention. Accordingly, in this invention, a two-stage learning model is employed for gas analysis.

[0110] These learning models can be generated by training deep learning algorithms using pre-prepared training data. The training data can be, for example... Figure 5 or Figures 6a-6c The sensor sensitivity data generated as shown.

[0111] The first-stage learning model can be a classification model. The second-stage learning model can be a regression model.

[0112] The first-stage classification model takes into account the sensing sensitivities of multiple gases, uses the sensing sensitivity data to train the learning model, and outputs the type of each gas.

[0113] The second-stage regression model takes into account the sensing sensitivity and types of multiple gases, and uses the sensing sensitivity and gas type data for training to output the concentration of each gas.

[0114] Figure 8 This is an example diagram of a deep learning model according to an embodiment of the present invention. In this invention, the learning model may include a first-stage classification model and a second-stage regression model.

[0115] The classification model can be a fully connected neural network (FCN).

[0116] The regression model is multiple linear regression (MLR).

[0117] As a learning method, supervised learning enables FCN to predict the types of gases and MLR to predict the concentrations of gases.

[0118] In one embodiment of the attached figure, the FCN receives the gas sensing sensitivities of four channels (c1, c2, c3, c4). That is, for example, the sensing sensitivities of hydrogen, ethanol, toluene, and formaldehyde are used as input to the input layer. A hidden layer 1 with 12 nodes and a hidden layer 2 with 24 nodes are used, with ReLU (Recitized Linear Unit) as the activation function. In the output layer, Softmax (normalized exponential function) is used to normalize the output values ​​to probability values ​​between 0 and 1 for supervised learning.

[0119] The MLR inputs the gas species from the FCN output and the gas sensing sensitivities of the four channels (c1, c2, c3, c4) to the input layer, using the same hidden layers 1 and 2 as the FCN, with ReLU (Recitified Linear Unit) activation function. Supervised learning is performed at the output layer to enable the output of linear concentration values ​​without an activation function.

[0120] This disclosure allows for the design of a regression model that, in the case of a mixture of gases composed of different types of gases, when using the regression model to predict concentrations, outputs the concentrations of each gas separately if the mixture is determined to be a mixed gas.

[0121] Figure 9 This is a flowchart illustrating a gas analysis method performed according to an embodiment of the present invention.

[0122] Reference Figure 9 The gas analysis method of the present invention can obtain training data for a deep learning model for a sample gas (S101).

[0123] The training data is obtained by detecting sample gas using a gas sensor, converting the sample gas sensing sensitivity data into distribution data with different sensing sensitivities, normalizing the distribution data based on the mean and standard deviation, and using the normally distributed data generated in this way.

[0124] Gas sensors 30 and 32 detect multiple target gases of different types (S102).

[0125] The sensing sensitivity of the target gas is applied to the learning model and deep learning is performed (S103).

[0126] In the case of deep learning, the deep learning model is trained using training data for the 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.

[0127] The analysis unit 300 uses the training results of the deep learning model to analyze the type and concentration of the target gas (S104).

[0128] In this invention, deep learning utilizes FCN in the first stage and MLR in the second stage.

[0129] The FCN receives the sensing sensitivity of the target gas and outputs the type of target gas. The MLR receives the type of target gas and the sensing sensitivity of the target gas output from the FCN and outputs the concentration of the target gas.

[0130] The target gases in this invention are ethanol, hydrogen, toluene, and formaldehyde.

[0131] Figure 10 The experimental procedure for gas analysis according to the present invention is shown.

[0132] In this experiment, four gases (ethanol, hydrogen, toluene, and formaldehyde) and air are supplied to the interior of the gas chamber 101. Gas sensors 30 and 32 are installed in the gas chamber 101.

[0133] Each gas is finely regulated by a mass flow controller (MFC) 33 to supply the appropriate amount that meets the measured concentration.

[0134] The output values ​​of gas sensors 30 and 32 are confirmed by a PC (or processor 20) connected to gas sensors 30 and 32.

[0135] In this experiment, 72 gas sensors were configured in the gas chamber, and each gas sensor was measured twice, for a total of 144 sensing sensitivities were measured. Figure 11 and Figure 12 The experimental results are shown.

[0136] like Figure 11 As shown, the distinguishing power (accuracy) of the four gases is as follows: hydrogen 94%, ethanol 97%, toluene 97%, and formaldehyde 82%.

[0137] like Figure 12 As shown, the concentration error rates for the four gases are: hydrogen, ethanol, and toluene within approximately ±30%, and formaldehyde within approximately ±50%.

[0138] As described above, the gas detection device disclosed herein can accurately detect harmful gases generated in daily life in real time, and therefore can be used in conjunction with air conditioners, air purifiers, etc. Thus, by linking the detection of the type and concentration of harmful gases, the air conditioner and air purifier can be appropriately controlled to provide a comfortable environment for the user.

[0139] While the foregoing has described all the constituent elements of the embodiments constituting this disclosure as being combined into one or through combination, this disclosure is not limited to such embodiments. That is, all the constituent elements may be selectively combined into more than one to perform operations, provided they are within the scope of the purpose of this disclosure. Furthermore, unless otherwise specifically stated to the contrary, the terms "comprising," "constituting," or "having" used above indicate that the constituent element may be included, and should be interpreted as not excluding other constituent elements, but rather including other constituent elements.

Claims

1. A gas detection device, wherein, include: substrate; A processor, disposed on the substrate, analyzes the detected data; as well as A first gas sensor and a second gas sensor are configured to be separated from the processor, wherein the first gas sensor detects different types of target gases from each other, and the second gas sensor detects different types of target gases from each other. The first gas sensor is located between one edge of the processor and the edge of the substrate corresponding to one edge of the processor, and the second gas sensor is located between the other edge of the processor and the edge of the substrate corresponding to the other edge of the processor.

2. The gas detection device according to claim 1, wherein, The processor analyzes the type and concentration of each of the detected target gases.

3. The gas detection device according to claim 2, wherein, The processor acquires training data on the sensing sensitivity of the sample gas, uses the training data to train a deep learning model, and trains the learning model by using the sensing sensitivity of the detected target gas to analyze the type and concentration of the target gas. The learning model includes a first-stage classification model and a second-stage regression model.

4. The gas detection device according to claim 3, wherein, The classification model is a fully connected neural network, and the regression model is multiple linear regression analysis.

5. The gas detection device according to claim 4, wherein, The fully connected neural network receives the sensing sensitivity of the detected target gas and outputs the type of the target gas. The multiple linear regression analysis receives the type of target gas output from the fully connected neural network and the sensing sensitivity of the detected target gas, and outputs the concentration of the target gas.

6. The gas detection device according to claim 1, wherein, The processor acquires the sensing sensitivity of a plurality of sample gases, generates distribution data of the sensing sensitivity, and generates normal distribution data based on the mean and standard deviation of the distribution data, so as to acquire the normal distribution data as the training data.

7. The gas detection device according to claim 1, wherein, Either the first gas sensor or the second gas sensor can detect ethanol, formaldehyde, toluene, or hydrogen.

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

9. The gas detection device according to claim 8, wherein, A through-channel is formed between the temperature / humidity sensor and the processor, and the through-channel extends through the substrate.

10. The gas detection device according to claim 9, wherein, The through-channel is formed to surround the temperature / humidity sensor except for a portion of the intervals.

11. The gas detection device according to claim 8, wherein, The temperature / humidity sensor, the first gas sensor, and the second gas sensor are arranged at a 90° interval between corresponding edges of the processor and the substrate, with the processor as a reference.

12. The gas detection device according to claim 1, wherein, A connector for connecting to an external signal is provided on one side edge of the substrate, and the connector is located off-center on the edge.

13. The gas detection device according to claim 1, wherein, An ESD diode for removing static electricity is provided on the substrate.

14. The gas detection device according to claim 1, wherein, The gas detection device also has a cover that isolates the surface of the substrate in which the processor is disposed from the outside. A processor shielding portion is formed in the cover at a position corresponding to the position of the processor. A first sensor window, a second sensor window, and a third sensor window are formed between the edge of the processor shielding portion and the edge of the cover, respectively.

15. The gas detection device according to claim 14, wherein, The third sensor window has a grid pattern.

Images