JP6944188B2 - Smell judgment system and odor judgment program - Google Patents

Description

The present invention relates to an odor determination system and an odor determination program for determining an odor.

As a conventional odor determination system, a system that determines the quality and intensity of an odor by forming a neural network with a plurality of gas sensors as inputs and performing machine learning is known (see, for example, Non-Patent Document 1). .). As for the quality of odor, which can be judged by the odor determination system described in Non-Patent Document 1, ammonia in the body is not detoxified due to excessive drinking or accumulation of fatigue and is released by sweat, or left sweat becomes bacteria. The "sweat odor" generated by decomposition and the odor of diacetyl produced by the decomposition of lactic acid contained in sweat from the sebaceous glands by bacteria on the skin, and the odor of medium-chain fatty acids from the sebaceous glands. "Middle fat odor" generated by combining, palmitooleic acid in the sebaceous glands that secrete sebum to keep the skin moisturized, and lipid peroxide increase with aging, and these are decomposed and oxidized. There is an "aging odor" that occurs in.

Konica Minolta Co., Ltd., "Kunkun body PRODUCT", [online], [Search on September 11, 2017], Internet <URL: https://kunkunbody.konicaminolta.jp/product/>

However, the conventional odor determination system has a problem that the quality of odor that can be determined is low.

Therefore, an object of the present invention is to provide an odor determination system and an odor determination program capable of increasing the odor quality that can be determined as compared with the conventional case.

In the odor determination system of the present invention, a column for separating gas components, a plurality of types of gas sensors for detecting the gas whose components are separated by the column, and a chromatogram based on the detection result of the gas sensor are used for the gas. The quality of odor is determined by using a chromatogram generating means generated for each sensor and an odor quality neural network as a neural network in which the value for each time in the chromatogram is used as an explanatory variable and the quality of odor is at least the objective variable. The odor determining means is provided with an odor determining means for determining, and the odor determining means comprises a plurality of chromatograms composed of the chromatograms based on the detection results of the plurality of types of gas sensors generated by the chromatogram generating means. It is characterized in that the quality of the odor is determined based on the above.

With this configuration, the odor determination system of the present invention determines the odor quality of the gas whose components are separated by the column based on a plurality of chromatograms based on the detection results of each of the plurality of types of gas sensors. , The quality of odor that can be judged can be increased more than before.

In the odor determination system of the present invention, the odor determination means uses an odor intensity neural network as a neural network in which the value for each time in the chromatogram is used as an explanatory variable and the odor intensity is used as an objective variable. The intensity of the odor may be determined based on a plurality of chromatograms.

With this configuration, the odor determination system of the present invention determines the odor quality and intensity using the odor quality neural network and the odor intensity neural network, respectively. Compared with the configuration in which both the odor quality and the intensity are determined by using the network, the accuracy of each determination of the odor quality and the intensity can be improved even with a small amount of training data.

In the odor determination system of the present invention, the odor determination means may input one chromatogram obtained by combining and aggregating any one of the waveforms in the plurality of chromatograms to the neural network for each time zone.

With this configuration, the odor determination system of the present invention uses one chromatogram, which is a combination of waveforms of any one of a plurality of chromatograms for each time zone, as an input of the neural network, and thus all of the plurality of chromatograms. The explanatory variables input to the neural network can be reduced as compared with the configuration in which is used as the input of the neural network, and as a result, the required training data can be reduced. Therefore, the odor determination system of the present invention can improve the accuracy of odor determination even with a small amount of learning data.

In the odor determination system of the present invention, the one chromatogram may be a chromatogram in which the waveform having the highest peak value among the waveforms in the plurality of chromatograms is combined and aggregated for each time zone.

With this configuration, the odor determination system of the present invention is a neural network that aggregates one waveform that has the highest peak value among the waveforms in a plurality of chromatograms for each time zone, that is, a combination of characteristic waveforms. Since it is input to, the accuracy of odor determination can be improved even with a small amount of training data.

In the odor determination system of the present invention, the odor determination means may input the one chromatogram to which the distinction information for distinguishing the waveform for each of the original chromatograms is added to the neural network.

With this configuration, the odor determination system of the present invention adds distinctive information for distinguishing waveforms for each underlying chromatogram to a chromatogram input to a neural network, thereby performing neural features in a plurality of chromatograms. Since it is added to the chromatogram input to the network, the accuracy of odor determination can be improved even with a small amount of training data.

In the odor determination system of the present invention, the odor determination means simplifies the shape of the waveform so that at least one value other than the peak value is zero in at least one of the plurality of waveforms in the chromatogram. The chromatogram may be input to the neural network.

With this configuration, the odor determination system of the present invention neurals a chromatogram that simplifies the shape of the waveform so that at least one of the plurality of waveforms in the chromatogram has at least one value other than the peak value set to zero. Since it is input to the network, among the explanatory variables input to the neural network, non-zero values, that is, valid explanatory variables can be reduced, and as a result, the required training data can be reduced. Therefore, the odor determination system of the present invention can improve the accuracy of odor determination even with a small amount of learning data.

In the odor determination system of the present invention, the odor determination means can change the chromatogram to be input to the neural network based on the plurality of chromatograms according to an instruction, and the plurality of chromatograms and the plurality of chromatograms can be changed. A screen including the chromatogram input to the neural network may be generated at the same time.

With this configuration, the odor determination system of the present invention generates a screen containing a plurality of chromatograms based on the detection results of each of the plurality of types of gas sensors and a chromatogram input to the neural network at the same time. It is possible to improve the convenience when the user instructs to change the chromatogram input to the network.

The odor determination system of the present invention does not have to include a mass spectrometer that detects the mass of the components separated by the column.

With this configuration, the odor determination system of the present invention does not need to be equipped with a large helium gas cylinder and a large vacuum pump, which are components of the mass spectrometer, and thus can be miniaturized. Can be improved. Therefore, the odor determination system of the present invention can be moved to the site where the odor is generated, and can easily determine the odor at the site. Therefore, in the odor determination system of the present invention, when the odor-generating gas is put into the container for transportation, the odor is adsorbed on the container, or the odor-generating gas is transported. Even if the odor components change due to changes over time or chemical changes during the process, the odor-generating gas is transported from the site where the odor is generated to the odor judgment system. By moving the odor determination system itself to the site without being performed, the odor can be determined with high accuracy.

The odor determination program of the present invention is a chromatogram generating means for generating a chromatogram for each gas sensor based on the detection results of a plurality of types of gas sensors that detect a gas whose components are separated by a column for separating gas components. A computer is used as an odor determination means for determining the quality of odor by using an odor quality neural network as a neural network in which the value for each time in the chromatogram is used as an explanatory variable and the quality of odor is at least the objective variable. The odor determining means is based on a plurality of chromatograms composed of the chromatograms generated by the chromatogram generating means based on the detection results of the plurality of types of gas sensors. It is characterized by determining the quality.

With this configuration, the computer that executes the odor determination program of the present invention has an odor quality based on a plurality of chromatograms based on the detection results of each of the plurality of types of gas sensors for the gas whose components are separated by the column. Therefore, the quality of the odor that can be determined can be increased as compared with the conventional case.

The odor determination system and the odor determination program of the present invention can increase the quality of odor that can be determined.

It is a block diagram of the odor determination system which concerns on one Embodiment of this invention. It is a block diagram of the computer shown in FIG. It is a figure which shows an example of the chromatogram generated by the chromatogram generating means shown in FIG. (A) It is a figure which shows an example of the odor quality neural network used in the odor determination system shown in FIG. (B) It is a figure which shows an example of the odor intensity neural network used in the odor determination system shown in FIG. It is a figure which shows the structure of the data for learning data shown in FIG. It is a figure which shows an example of the chromatogram for determination mode 1 generated based on the PID chromatogram, the semiconductor A chromatogram and the semiconductor B chromatogram shown in FIG. It is a figure which shows an example of the chromatogram for determination mode 2 generated based on the PID chromatogram, the semiconductor A chromatogram and the semiconductor B chromatogram shown in FIG. It is a figure which shows an example of the chromatogram for determination mode 3 generated based on the PID chromatogram, the semiconductor A chromatogram and the semiconductor B chromatogram shown in FIG. It is a flowchart of the method of generating the data for learning data using the odor determination system shown in FIG. It is a flowchart of the verification method of the neural network using the odor determination system shown in FIG. It is a figure which shows an example of the verification result screen output in the verification method shown in FIG. 10 about the neural network for determination mode 1. It is a flowchart of the odor determination method using the odor determination system shown in FIG. It is a figure which shows an example of the determination result screen output in the determination method shown in FIG. 12 when the scent of coffee and the determination mode 3 are specified as the odor category and the determination mode, respectively.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

First, the configuration of the odor determination system according to the present embodiment will be described.

FIG. 1 is a configuration diagram of an odor determination system 10 according to the present embodiment.

As shown in FIG. 1, the odor determination system 10 takes in the sample gas injection unit 11 into which the gas to be determined for odor (hereinafter referred to as “sample gas”) is injected and the air as a carrier gas from the surroundings. By the pump 12, the VOC filter 13 that removes volatile organic compounds (VOCs) from the carrier gas delivered by the pump 12, the sample gas injected from the sample gas injection unit 11, and the pump 12. A mixing section 14 in which the carrier gas that has been sent out and passed through the VOC filter 13 is mixed, a column 15 that separates the components of the sample gas that is mixed with the carrier gas in the mixing section 14 and moved by the carrier gas, and a column 15 A PID (Photo Generation Detector) sensor 16, a semiconductor gas sensor 17, a semiconductor gas sensor 18, and a computer 20 for displaying the odor quality and intensity of a sample gas as a gas sensor for detecting a gas whose components are separated by It has. That is, the odor determination system 10 is a system that determines the odor quality and intensity of the sample gas by using gas chromatography.

As the column 15, either a packed column or a capillary column may be adopted, but it is preferable to adopt a capillary column having a high performance of separating gas components.

The semiconductor gas sensor 17 and the semiconductor gas sensor 18 are both semiconductor gas sensors, but they are gas sensors having different gas detection performances, that is, different types of gas that can be detected. The PID sensor 16 differs in gas detection performance from both the semiconductor gas sensor 17 and the semiconductor gas sensor 18. That is, the PID sensor 16, the semiconductor gas sensor 17, and the semiconductor gas sensor 18 are different types of gas sensors. For example, the PID sensor 16 has high VOC gas detection performance, the semiconductor gas sensor 17 has high acetic acid-based gas detection performance, and the semiconductor gas sensor 18 has high sulfur-based gas detection performance.

FIG. 2 is a block diagram of the computer 20.

As shown in FIG. 2, the computer 20 is an operation unit 21 which is an input device such as a keyboard and a mouse into which various operations are input, and a display device such as an LCD (Hard Disk Disk Play) which displays various information. Various types of display unit 22 and communication unit 23, which is a communication device that directly communicates with an external device via a network such as LAN (Local Area Information) or the Internet, or directly by wire or wirelessly without a network. It is provided with a storage unit 24 which is a non-volatile storage device such as a semiconductor memory and an HDD (Hard Disk Drive) for storing the information of the above, and a control unit 25 which controls the entire computer 20. The computer 20 is, for example, a PC (Personal Computer).

The storage unit 24 stores an odor determination program 24a for determining an odor. The odor determination program 24a may be installed in the computer 20 at the manufacturing stage of the computer 20, for example, or external storage such as a USB (Universal Serial Bus) memory, a CD (Compact Disk), or a DVD (Digital Versaille Disk). It may be additionally installed on the computer 20 from the medium, or may be additionally installed on the computer 20 from the network.

The storage unit 24 can store the neural network 24b. Further, the storage unit 24 can store at least one neural network having the same configuration as the neural network 24b, in addition to the neural network 24b. The storage unit 24 has an odor quality for determination mode 1, which will be described later, for each odor category such as sick odor, mouth odor, coffee odor, wine odor, chicken farm odor, waste disposal plant odor, and household odor. An odor neural network, which is a neural network for determining odor, an odor intensity neural network, which is a neural network for determining odor intensity for determination mode 1, and an odor quality for determination mode 2, which will be described later. The neural network, the odor intensity neural network for the determination mode 2, the odor intensity neural network for the determination mode 3 described later, and the odor intensity neural network for the determination mode 3 can be stored.

The storage unit 24 can store data (hereinafter, referred to as “learning data data”) 24c for the training data of the neural network. Further, the storage unit 24 can store at least one learning data data having the same configuration as the learning data data 24c, in addition to the learning data data 24c. Similar to the neural network, the storage unit 24 can store a plurality of training data data for each odor category.

The control unit 25 includes, for example, a CPU (Central Processing Unit), a ROM (Read Only Memory) that stores programs and various data, and a RAM (Random Access Memory) that is used as a work area of the CPU. There is. The CPU executes a program stored in the ROM or the storage unit 24.

The control unit 25 uses a chromatogram generating means 25a that generates a chromatogram based on the detection result of the gas sensor for each gas sensor by executing the odor determination program 24a, and a neural network for odor quality and intensity. The odor determination means 25b for determining is realized.

The chromatogram generating means 25a determines the detection values (AD values) of the PID sensor 16, the semiconductor gas sensor 17, and the semiconductor gas sensor 18 from a specific timing after the injection of the sample gas from the sample gas injection unit 11 is completed. Start importing. Assuming that the timing at which the capture of the detected value is started is 0 seconds, the chromatogram generating means 25a captures the detected value every second for up to 2999 seconds. That is, the chromatogram generating means 25a is based on a chromatogram consisting of 3000 detection values (hereinafter referred to as "PID chromatogram") based on the detection result of the PID sensor 16 and the detection result of the semiconductor gas sensor 17. A chromatogram consisting of 3000 detected values (hereinafter referred to as "semiconductor A chromatogram") and a chromatogram consisting of 3000 detected values based on the detection result of the semiconductor gas sensor 18 (hereinafter referred to as "semiconductor B chromatogram"). ”) And three chromatograms are generated for each measurement of the sample gas.

FIG. 3 is a diagram showing an example of a chromatogram generated by the chromatogram generating means 25a.

In FIG. 3, "PID", "semiconductor A", and "semiconductor B" mean a PID sensor 16, a semiconductor gas sensor 17, and a semiconductor gas sensor 18, respectively.

FIG. 3 shows the PID chromatogram, the semiconductor A chromatogram, and the semiconductor B chromatogram in an overlapping manner.

FIG. 4A is a diagram showing an example of an odor quality neural network 71 used in the odor determination system 10. FIG. 4B is a diagram showing an example of an odor intensity neural network 72 used in the odor determination system 10.

The odor quality neural network 71 shown in FIG. 4A is an example of a neural network for determining the quality of the aroma of coffee. The odor intensity neural network 72 shown in FIG. 4B is an example of a neural network for determining the intensity of the aroma of coffee.

The odor quality neural network 71 is a neural network in which the hourly value in the chromatogram is used as an explanatory variable and the odor quality such as mocha scent, kona scent, and kirimanjaro scent is used as an objective variable. The setting items of the odor neural network 71 are the number of input nodes, that is, the number of input nodes indicating the number of explanatory variables, the number of hidden layers, the number of nodes of each hidden layer, and the number of output nodes. That is, the number of output nodes that indicate the number of objective variables, what to use for the activation function, what to use for optimization, whether to use Supervised learning, and how to use Weight data. The presence / absence of, the presence / absence of use of Dropout, and the number of repetitions of supervised learning indicating the number of repetitions of supervised learning for each learning data. In the odor neural network 71 shown in FIG. 4B, the number of input nodes is 3000, which is the same as the number of values for each time in the chromatogram, and the number of hidden nodes indicating the number of nodes in the first layer of the hidden layer is shown. The number of 1 is 300, the number of hidden nodes 2 indicating the number of nodes in the second layer of the hidden layer is 200, and the number of output nodes is 50 such as mocha scent, kona scent, and Kirimanjaro scent. Yes, the activation function is ReLU, the optimization is Adam, the Batch neuralization is used, the Weight decay is used, the Dropout is used, and the number of teacher learning repetitions is 10,000.

The odor intensity neural network 72 is a neural network in which the value for each time in the chromatogram is used as an explanatory variable, and the odor intensity such as intensity 1 to intensity 6 is used as an objective variable. The setting items of the odor intensity neural network 72 are the same as the setting items of the odor quality neural network 71. In the odor intensity neural network 72 shown in FIG. 4B, the number of input nodes is 3000, which is the same as the number of values for each time in the chromatogram, the number of hidden nodes is 50, and the number of output nodes is. There are 6 intensities 1 to 6, the activation function is ReLU, the optimization is Adam, the Batch neuralization is used, the Weight decay is used, the Dropout is used, and the number of supervised learning repetitions is 3000 times. Is.

FIG. 5 is a diagram showing the configuration of the training data data 24c.

As shown in FIG. 5, the training data data 24c includes the original data 30 generated by the chromatogram generating means 25a by measuring the sample gas and the data for the determination mode 1 (hereinafter referred to as “determination mode 1 data”). ) 40, data for determination mode 2 (hereinafter referred to as “data for determination mode 2”) 50, and data for determination mode 3 (hereinafter referred to as “data for determination mode 3”) 60. I'm out.

The original data 30 includes PID chromatogram data (hereinafter referred to as “PID chromatogram data”) 31 generated by the chromatogram generating means 25a by measuring the same sample gas, and semiconductor A chromatogram data (hereinafter referred to as “PID chromatogram data”). It includes (referred to as "semiconductor A chromatogram data") 32 and data of a semiconductor B chromatogram (hereinafter referred to as "semiconductor B chromatogram data") 33.

The determination mode 1 data 40 is one chromatograph composed of 3000 values generated based on the basic information data 41, the PID chromatogram data 31, the semiconductor A chromatogram data 32, and the semiconductor B chromatogram data 33. Gram (hereinafter referred to as "chromatogram for determination mode 1") data (hereinafter referred to as "chromatogram data for determination mode 1") 42 and data for determining the quality of odor (hereinafter referred to as "quality determination data"). ) 43 and data for determining the intensity of odor (hereinafter referred to as “intensity determination data”) 44 are included.

The basic information data 41 includes a measurement ID 41a, which is identification information for measuring the sample gas, which is the basis for generating the training data 24c, a measurement date 41b indicating the date when this measurement was performed, and the measurement date 41b. It includes a measurement time 41c indicating the time when this measurement is performed, a measurement mode 41d of this measurement, a measurer ID 41e which is identification information of the measurer of this measurement, and a comment 41f by the measurer. The measurer is preferably a qualified odor judge.

The odor determination system 10 can employ several measurement modes in which the measurement method is patterned, such as the total measurement time, the flow rate of injecting the sample gas into the column 15, and the setting of the temperature change of the column 15. The measurement mode 41d is one of the measurement modes adopted by the odor determination system 10 that is executed by measuring the sample gas, which is the basis for generating the training data 24c.

Comment 41f is freely described by the measurer. For example, the comment 41f describes the content felt by the measurer regarding the measurement, such as "smell of mushrooms".

The quality determination data 43 teaches the quality of the odor 43a determined by the measurer, the success / failure information 43b indicating the success / failure of the determination, the correction information 43c indicating the correct quality when the determination fails, and the quality 43a. It includes a teacher data flag 43d that is set when it is used as data and is defeated when the quality 43a is not used as teacher data. When the teacher data flag 43d is set, the combination of the chromatogram data 42 for the determination mode 1 as an explanatory variable and the quality 43a as the teacher data becomes the training data of the odor quality neural network for the determination mode 1. ..

The intensity determination data 44 trains the odor intensity 44a determined by the measurer, the success / failure information 44b indicating the success / failure of the determination, the correction information 44c indicating the correct intensity when the determination fails, and the intensity 44a. It includes a teacher data flag 44d that is set up when used as data and is defeated when the intensity 44a is not used as teacher data. When the teacher data flag 44d is set, the combination of the chromatogram data 42 for the determination mode 1 as an explanatory variable and the intensity 44a as the teacher data becomes the training data of the odor intensity neural network for the determination mode 1. ..

The determination mode 2 data 50 is one chromatograph composed of 3000 values generated based on the basic information data 51, the PID chromatogram data 31, the semiconductor A chromatogram data 32, and the semiconductor B chromatogram data 33. It includes gram data (hereinafter referred to as "chromatogram data for determination mode 2") data (hereinafter referred to as "chromatogram data for determination mode 2") 52, quality determination data 53, and intensity determination data 54. ..

The basic information data 51 is the same information as the basic information data 41.

The quality determination data 53 teaches the quality of the odor 53a determined by the measurer, the success / failure information 53b indicating the success / failure of the determination, the correction information 53c indicating the correct quality when the determination fails, and the quality 53a. It includes a teacher data flag 53d that is set when it is used as data and is defeated when the quality 53a is not used as teacher data. The quality 53a is the same information as the quality 43a. When the teacher data flag 53d is set, the combination of the chromatogram data 52 for the determination mode 2 as an explanatory variable and the quality 53a as the teacher data becomes the training data of the odor quality neural network for the determination mode 2. ..

The intensity determination data 54 trains the odor intensity 54a determined by the measurer, the success / failure information 54b indicating the success / failure of the determination, the correction information 54c indicating the correct intensity when the determination fails, and the intensity 54a. It includes a teacher data flag 54d that is set up when used as data and is defeated when the intensity 54a is not used as teacher data. The strength 54a is the same information as the strength 44a. When the teacher data flag 54d is set, the combination of the chromatogram data 52 for the determination mode 2 as an explanatory variable and the intensity 54a as the teacher data becomes the training data of the odor intensity neural network for the determination mode 2. ..

The determination mode 3 data 60 is one chromatograph composed of 3000 values generated based on the basic information data 61, the PID chromatogram data 31, the semiconductor A chromatogram data 32, and the semiconductor B chromatogram data 33. It includes gram data (hereinafter referred to as "chromatogram data for determination mode 3") data (hereinafter referred to as "chromatogram data for determination mode 3") 62, quality determination data 63, and intensity determination data 64. ..

The basic information data 61 is the same information as the basic information data 41.

The quality determination data 63 teaches the quality of the odor 63a determined by the measurer, the success / failure information 63b indicating the success / failure of the determination, the correction information 63c indicating the correct quality when the determination fails, and the quality 63a. It includes a teacher data flag 63d that is set when it is used as data and is defeated when the quality 63a is not used as teacher data. The quality 63a is the same information as the quality 43a. When the teacher data flag 63d is set, the combination of the chromatogram data 62 for the determination mode 3 as an explanatory variable and the quality 63a as the teacher data becomes the training data of the odor quality neural network for the determination mode 3. ..

The intensity determination data 64 trains the odor intensity 64a determined by the measurer, the success / failure information 64b indicating the success / failure of the determination, the correction information 64c indicating the correct intensity when the determination fails, and the intensity 64a. It includes a teacher data flag 64d that is set when it is used as data and is defeated when the intensity 64a is not used as teacher data. The strength 64a is the same information as the strength 44a. When the teacher data flag 64d is set, the combination of the chromatogram data 62 for the determination mode 3 as an explanatory variable and the intensity 64a as the teacher data becomes the training data of the odor intensity neural network for the determination mode 3. ..

Next, the odor determination mode in the odor determination system 10 will be described.

The odor determination system 10 has a determination mode 1, a determination mode 2, and a determination mode 3 as odor determination modes.

The determination mode 1 is one chromatogram obtained by combining and aggregating the waveforms having the highest peak values among the waveforms in the PID chromatogram, the semiconductor A chromatogram, and the semiconductor B chromatogram for each time zone, that is, the chromatograph for the determination mode 1. This is a mode in which a gram is generated and a chromatogram for determination mode 1 is input to each of the odor quality neural network and the odor intensity neural network.

For example, when the PID chromatogram, the semiconductor A chromatogram, and the semiconductor B chromatogram are the chromatograms shown in FIG. 3, the chromatogram for determination mode 1 is, for example, as shown in FIG. In FIG. 6, “PID”, “semiconductor A”, and “semiconductor B” mean a PID sensor 16, a semiconductor gas sensor 17, and a semiconductor gas sensor 18, respectively. In FIG. 6, the waveform 81a is the waveform in the semiconductor A chromatogram, the waveform 81b is the waveform in the semiconductor B chromatogram, and the waveform 81c, which is a waveform other than the waveform 81a and the waveform 81b, is the waveform in the PID chromatogram.

In the determination mode 2, the waveforms having the highest peak values among the waveforms in the PID chromatogram, the semiconductor A chromatogram, and the semiconductor B chromatogram are combined and aggregated for each time zone, and the underlying PID chromatogram and the semiconductor A chromatogram are aggregated. One chromatogram with distinctive information for distinguishing waveforms for each gram and semiconductor B chromatogram, that is, a chromatogram for judgment mode 2, is generated, and a judgment mode is generated for each of the odor quality neural network and the odor intensity neural network. This is a mode for inputting a chromatogram for 2. That is, the chromatogram for determination mode 2 is a chromatogram in which distinction information for distinguishing waveforms is added to the chromatogram for determination mode 1 for each of the underlying PID chromatogram, semiconductor A chromatogram, and semiconductor B chromatogram. .. Here, the distinction information is the shape of the waveform. That is, in the chromatogram for determination mode 2, the waveform whose base chromatogram is the PID chromatogram is maintained in its original shape, and the waveform whose base chromatogram is the semiconductor A chromatogram is specified. A waveform whose underlying chromatogram is a semiconductor B chromatogram is changed to a pulse wave with a width of a specific width shorter than the width of the waveform whose underlying chromatogram is a semiconductor A chromatogram. Be changed.

For example, when the PID chromatogram, the semiconductor A chromatogram, and the semiconductor B chromatogram are the chromatograms shown in FIG. 3, the chromatogram for determination mode 2 is, for example, as shown in FIG. 7. In FIG. 7, “PID”, “semiconductor A”, and “semiconductor B” mean a PID sensor 16, a semiconductor gas sensor 17, and a semiconductor gas sensor 18, respectively. In FIG. 7, the waveform 82a is obtained by changing the waveform 81a shown in FIG. 6 to a pulse wave having a specific width, and the waveform 82b is a waveform 81b shown in FIG. 6 having a specific width shorter than the width of the waveform 82a. It is changed to the pulse wave of.

In the determination mode 3, the shape of the waveform is set so that all the values other than the peak value are set to zero in all the waveforms whose base chromatogram is the PID chromatogram with respect to the chromatogram for the determination mode 2. This is a mode in which a simplified chromatogram, that is, a chromatogram for determination mode 3 is generated, and a chromatogram for determination mode 3 is input to each of the odor quality neural network and the odor intensity neural network.

For example, when the PID chromatogram, the semiconductor A chromatogram, and the semiconductor B chromatogram are the chromatograms shown in FIG. 3, the chromatogram for determination mode 3 is, for example, as shown in FIG. In FIG. 8, “PID”, “semiconductor A”, and “semiconductor B” mean a PID sensor 16, a semiconductor gas sensor 17, and a semiconductor gas sensor 18, respectively. In FIG. 8, the waveform 82c is obtained by changing the waveform 81c shown in FIG. 7 into a pulse wave having a specific width shorter than the width of the waveform 82b, for example, a pulse wave having a width of one second, which is the minimum width.

Next, a method of generating data for learning data using the odor determination system 10 will be described.

FIG. 9 is a flowchart of a method of generating data for learning data using the odor determination system 10.

As shown in FIG. 9, the measurer puts a certain amount of sample gas into a sample bag for putting the sample gas (S101). For example, the measurer fills a sample bag with a capacity of 1 liter with sample gas.

Next, the measurer sniffs the sample gas put in the sample bag in S101, and determines which of the objective variables defined in the odor neural network in the sample gas category corresponds to the odor quality of the sample gas. It is determined which of the objective variables defined in the odor intensity neural network in the sample gas category corresponds to the odor intensity of the sample gas (S102).

Next, the measurer registers the basic information data, the quality in the quality determination data, and the intensity in the intensity determination data from the operation unit 21 in the learning data data corresponding to the sample gas this time (S103). Here, the measurer registers the odor quality determined in S102 as the quality in the quality determination data, and registers the odor intensity determined in S102 as the intensity in the intensity determination data.

Next, the measurer sets the sample bag containing the sample gas in the sample gas injection unit 11 in S101, and then instructs the computer 20 to perform the measurement according to the predetermined measurement conditions via the operation unit 21 (. S104). Here, the measurer specifies the category of the sample gas this time in the instruction of S104.

Therefore, the chromatogram generating means 25a of the computer 20 executes the measurement according to the predetermined measurement conditions to generate the PID chromatogram, the semiconductor A chromatogram and the semiconductor B chromatogram, and is specified in the instruction in S104. In each category, the training data data corresponding to the sample gas of this time is registered as PID chromatogram data, semiconductor A chromatogram data, and semiconductor B chromatogram data (S105).

Next, the odor determining means 25b of the computer 20 is based on the PID chromatogram, the semiconductor A chromatogram and the semiconductor B chromatogram generated in S105, the chromatogram for the determination mode 1, the chromatogram for the determination mode 2, and the determination mode. Chromatograms for 3 are generated and registered as chromatogram data for determination mode 1, chromatogram data for determination mode 2, and chromatogram data for determination mode 3 in the learning data data corresponding to the sample gas this time, respectively (S106). ).

Next, the odor determining means 25b causes each neural network to learn each chromatogram data by setting all the teacher data flags in the training data data corresponding to the sample gas this time (S107). That is, the odor determination means 25b causes the odor quality neural network for the determination mode 1 to learn the chromatogram data for the determination mode 1, and causes the odor intensity neural network for the determination mode 1 to learn the chromatogram data for the determination mode 1. The odor neural network for judgment mode 2 is trained with the chromatogram data for judgment mode 2, the odor intensity neural network for judgment mode 2 is trained with the chromatogram data for judgment mode 2, and the odor neural network for judgment mode 3 is trained. The network is made to learn the chromatogram data for the determination mode 3, and the odor intensity neural network for the determination mode 3 is made to learn the chromatogram data for the determination mode 3.

The method shown in FIG. 9 ends after the step S107.

By repeating the method shown in FIG. 9 to generate data for learning data of a large number of measurers, the accuracy of determination in each neural network is improved.

Next, a method for verifying the neural network using the odor determination system 10 will be described.

FIG. 10 is a flowchart of a neural network verification method using the odor determination system 10.

As shown in FIG. 10, when the worker desires the verification of the neural network of a specific category, the operator instructs the computer 20 to verify the neural network of the specific category via the operation unit 21 (S121).

When the verification of the neural network of a specific category is instructed, the odor determining means 25b of the computer 20 determines each neural network by inputting a chromatogram for each neural network into each neural network of the specified category. Is executed (S122). That is, the odor determination means 25b causes the odor quality neural network for the determination mode 1 to determine the odor quality based on each of all the chromatogram data for the determination mode 1, and causes the odor intensity neural network for the determination mode 1 to determine the odor quality. The intensity of the odor is judged based on each of the chromatogram data for all the judgment modes 1, and the odor quality neural network for the judgment mode 2 is made to judge the quality of the odor based on each of the chromatogram data for all the judgment modes 2. Is made to judge, and the odor intensity neural network for judgment mode 2 is made to judge the odor intensity based on each of the chromatogram data for all judgment modes 2, and the odor quality neural network for judgment mode 3 is made to judge all judgment modes. The quality of the odor is determined based on each of the chromatogram data for 3 and the odor intensity neural network for the determination mode 3 is made to determine the intensity of the odor based on each of the chromatogram data for all the determination modes 3.

Next, the odor determination means 25b collates the determination result in S122 with the determination result by the measurer associated with the chromatogram for each neural network, determines the success or failure of the determination in S122, and determines. The result is registered in the success / failure information (S123). For example, when the odor determination means 25b causes the odor quality neural network for determination mode 1 to determine the quality of odor based on the chromatogram data for specific determination mode 1, the result of this determination and this determination mode When the quality of the quality judgment data associated with the chromatogram data for 1 matches, "correct answer" is registered in the success / failure information of the quality judgment data associated with the chromatogram data for judgment mode 1. If they do not match, "incorrect answer" is registered in this success / failure information. When the odor determination means 25b registers "incorrect answer" in the success / failure information, the quality determined based on the chromatogram data for the determination mode 1 may be registered in the correction information corresponding to the success / failure information.

Next, the odor determining means 25b outputs the verification result to the display unit 22 based on the determination result in S123 (S124).

The method shown in FIG. 10 ends after the step S124.

FIG. 11 is a diagram showing an example of a verification result screen output in S124 with respect to the neural network for the determination mode 1.

According to the verification result screen shown in FIG. 11, regarding the odor quality, 187 of the 200 determination mode 1 chromatogram data are correct and 13 are incorrect in the determination of the odor neural network. It can be seen that the correct answer rate is 93.5%. Similarly, regarding the intensity of odor, in the judgment of the odor neural network, 145 of the 200 judgment mode 1 chromatogram data are correct answers, 55 are incorrect answers, and the correct answer rate is 72.5. It turns out that it is%. Further, on the verification result screen shown in FIG. 11, a list of measurement IDs of the chromatogram data for determination mode 1, which was an incorrect answer, is shown for each of the odor quality and intensity.

The verification result shown in FIG. 11 relates to the neural network for the determination mode 1. The operator also outputs the verification result regarding the neural network for the determination mode 2 and the verification result regarding the neural network for the determination mode 3 to the display unit 22 by instructing the computer 20 via the operation unit 21. be able to.

The operator can perform the verification of any category of neural network by the method shown in FIG.

When the operator determines that the correct answer rate of the judgment is low as a result of the verification of the neural network by the method shown in FIG. 10, the setting items of the neural network, that is, the number of input nodes, the number of hidden layers, and each layer of the hidden layer Number of nodes, number of output nodes, what to use for activation function, what to use for optimization, with or without supervised learning, with or without weight decay, with Dropout After changing the presence / absence and the number of repetitions of supervised learning, the verification of the neural network can be executed again by the method shown in FIG.

In addition, the worker can remove at least one of the quality and intensity of the training data from the training data by defeating the teacher data flag in the training data. For example, if the worker determines that the result of the judgment by a specific measurer indicated by the measurer ID in the basic information data is clearly not appropriate, the worker can exclude the result of the judgment by the measurer from the teacher data. .. In addition, when the worker judges that the result of the judgment by the measurement of a specific period indicated by the measurement date in the basic information data is clearly not appropriate, the result of the judgment by the measurement of this period is excluded from the teacher data. be able to.

Next, an odor determination method using the odor determination system 10 will be described.

FIG. 12 is a flowchart of an odor determination method using the odor determination system 10.

As shown in FIG. 12, the operator puts a certain amount of sample gas into a sample bag for putting the sample gas (S141).

Next, the operator sets the sample bag containing the sample gas in S141 in the sample gas injection unit 11, and then instructs the computer 20 via the operation unit 21 to make a determination according to the predetermined measurement conditions. S142). Here, the operator specifies the category of the sample gas this time and one of the determination modes 1 to 3 in the instruction of S142.

Therefore, the chromatogram generating means 25a of the computer 20 executes the measurement according to the predetermined measurement conditions to generate the PID chromatogram, the semiconductor A chromatogram, and the semiconductor B chromatogram (S143).

Next, the odor determination means 25b of the computer 20 is based on the PID chromatogram, the semiconductor A chromatogram, and the semiconductor B chromatogram generated in S143, the chromatogram for determination mode 1, the chromatogram for determination mode 2, and the determination mode. A chromatogram for 3 is generated (S144).

Next, the odor determination means 25b was generated in S144 for each of the odor quality neural network and the odor intensity neural network corresponding to the determination mode specified in the instruction in S142 in the category specified in the instruction in S142. By inputting the chromatogram corresponding to the judgment mode specified in the instruction in S142 among the chromatogram for judgment mode 1, the chromatogram for judgment mode 2, and the chromatogram for judgment mode 3, the neural network is made to execute the judgment. (S145).

Next, the odor determining means 25b outputs the result of the determination in S145 to the display unit 22 (S146).

The method shown in FIG. 12 ends after the step S146.

FIG. 13 is a diagram showing an example of a determination result screen output in S146 when the aroma of coffee and the determination mode 3 are specified as the odor category and the determination mode, respectively.

In FIG. 13, “PID”, “semiconductor A”, and “semiconductor B” mean a PID sensor 16, a semiconductor gas sensor 17, and a semiconductor gas sensor 18, respectively.

According to the determination result screen shown in FIG. 13, the odor quality and intensity of the sample gas are Kirimanjaro and intensity 3, respectively. The determination result screen shown in FIG. 13 includes the PID chromatogram, the semiconductor A chromatogram, and the semiconductor B chromatogram generated in S143 as all sensor waveforms. Further, the determination result screen shown in FIG. 13 includes the chromatogram input to the neural network in S145 as the determination waveform.

When the judgment result screen is displayed, the operator instructs the change from the current judgment mode to any of the judgment modes 1 to 3 other than the current judgment mode. , The chromatogram input to the neural network can be changed by the odor determining means 25b. When the odor determination means 25b is instructed to change the determination mode, the chromatogram for determination mode 1 generated in S144 is applied to each of the odor quality neural network and the odor intensity neural network corresponding to the specified determination mode. , The neural network is made to execute the determination by inputting the chromatogram corresponding to the designated determination mode among the chromatogram for determination mode 2 and the chromatogram for determination mode 3. Then, the odor determination means 25b outputs the determination result to the display unit 22 in the same manner as the determination result screen shown in FIG.

As described above, the odor determination system 10 has a PID chromatogram based on the detection results of the PID sensor 16, the semiconductor gas sensor 17, and the semiconductor gas sensor 18 for the sample gas whose components have been separated by the column 15. Since the odor quality is determined based on the semiconductor A chromatogram and the semiconductor B chromatogram, the determinable odor quality can be increased as compared with the conventional case.

Since the odor determination system 10 determines the quality and intensity of the odor by using the odor quality neural network and the odor intensity neural network, respectively, the odor is determined by using the neural network having both the odor quality and the intensity as objective variables. Compared with the configuration for determining both the quality and intensity of the odor, the accuracy of each determination of the quality and intensity of the odor can be improved even with a small amount of training data.

The odor determination system 10 may be configured to determine both the odor quality and the intensity by using a neural network having both the odor quality and the intensity as objective variables.

The odor determination system 10 is one chromatogram obtained by combining and aggregating any one waveform of a PID chromatogram, a semiconductor A chromatogram, and a semiconductor B chromatogram for each time zone, that is, a chromatogram for determination mode 1, a determination mode. Since the chromatogram for 2 or the chromatogram for determination mode 3 is input to the neural network, the explanatory variables input to the neural network can be reduced, and as a result, the required training data can be reduced. Therefore, the odor determination system 10 can improve the accuracy of odor determination even with a small amount of learning data.

The odor determination system 10 may have a configuration in which all of the PID chromatogram, the semiconductor A chromatogram, and the semiconductor B chromatogram are input to the neural network. For example, the odor determination system 10 simplifies the shape of the waveform so that at least one value other than the peak value becomes zero, and all of the PID chromatogram, the semiconductor A chromatogram, and the semiconductor B chromatogram of the neural network A determination mode for input may be provided.

When the determination mode is any of the determination mode 1, the determination mode 2, and the determination mode 3, the odor determination system 10 is the most among the waveforms in the PID chromatogram, the semiconductor A chromatogram, and the semiconductor B chromatogram for each time zone. Since a waveform having a high peak value, that is, one chromatogram obtained by combining and aggregating characteristic waveforms is input to the neural network, the accuracy of odor determination can be improved even with a small amount of training data.

The odor determination system 10 is one chromatogram obtained by combining and aggregating any one waveform of a PID chromatogram, a semiconductor A chromatogram, and a semiconductor B chromatogram for each time zone, and is a PID chromatogram for each time zone. It may be configured to input a chromatogram to the neural network, which is not one chromatogram obtained by combining and aggregating the waveforms having the highest peak values among the waveforms in the gram, the semiconductor A chromatogram and the semiconductor B chromatogram. For example, the odor determination system 10 combines and aggregates the waveforms of any one of the PID chromatogram, the semiconductor A chromatogram, and the semiconductor B chromatogram for each time zone, and aggregates the underlying PID chromatogram, the semiconductor A chromatogram, and the semiconductor A chromatogram. A determination mode may be provided in which one chromatogram to which the distinction information for distinguishing the waveform is added for each semiconductor B chromatogram is used as an input of the neural network. Further, the odor determination system 10 aggregates the waveforms of any one of the PID chromatogram, the semiconductor A chromatogram and the semiconductor B chromatogram in combination for each time zone, and aggregates the underlying PID chromatogram, the semiconductor A chromatogram and the semiconductor A chromatogram. One chromatogram that simplifies the shape of the waveform is used as the input of the neural network so that at least one value other than the peak value is set to zero by adding distinction information that distinguishes the waveform for each semiconductor B chromatogram. A determination mode may be provided.

When the determination mode is either the determination mode 2 or the determination mode 3, the odor determination system 10 inputs the distinction information for distinguishing the waveform for each of the underlying chromatograms, that is, the shape of the waveform into the neural network. By adding to the chromatogram, that is, the chromatogram for determination mode 2 and the chromatogram for determination mode 3, the features in the PID chromatogram, the semiconductor A chromatogram and the semiconductor B chromatogram can be added to the chromatogram for determination mode 2 and the chromatogram for determination mode 2. Since it is added to the chromatogram for No. 3, it is possible to improve the accuracy of odor determination even with a small amount of training data.

The odor determination system 10 may have a configuration in which the distinction information for distinguishing the waveform for each underlying chromatogram is not added to the chromatogram input to the neural network. For example, the odor determination system 10 combines and aggregates one of the waveforms of the PID chromatogram, the semiconductor A chromatogram, and the semiconductor B chromatogram for each time zone, and sets at least one value other than the peak value to zero. In addition, a determination mode may be provided in which one chromatogram that simplifies the shape of the waveform is used as an input of the neural network.

The odor determination system 10 may adopt a method other than the method shown in the present embodiment as a method for adding the distinction information. For example, the odor determination system 10 may adopt other shapes such as a triangular wave and a sine wave in addition to the pulse wave described above as the shape of the waveform.

The odor determination system 10 sets at least one value other than the peak value to zero in at least one of the plurality of waveforms in the chromatogram when the determination mode is either the determination mode 2 or the determination mode 3. By inputting a chromatogram that simplifies the shape of the waveform into the neural network, it is possible to reduce non-zero values, that is, valid explanatory variables, among the explanatory variables input to the neural network, and as a result, it is necessary. Training data can be reduced. Therefore, the odor determination system 10 can improve the accuracy of odor determination even with a small amount of learning data.

In particular, when the determination mode is the determination mode 3, the odor determination system 10 provides the retention time and peak value of the waveform and the distinction information for distinguishing the waveform for each underlying chromatogram in the chromatogram for the determination mode 3. Since the characteristics are narrowed down only to the pulse width of, the accuracy of odor determination can be improved even with a small amount of training data.

Since the odor determination system 10 can improve the accuracy of odor determination even with a small amount of learning data, it can produce a valuable and small amount of odor, for example, when a rare odor existing in the natural world is collected. Even when collected, the necessary learning data can be generated, and as a result, a valuable and small amount of odor can be determined with high accuracy.

The odor determination system 10 includes a PID chromatogram, a semiconductor A chromatogram, and a semiconductor B chromatogram based on the detection results of the PID sensor 16, the semiconductor gas sensor 17, and the semiconductor gas sensor 18, and a chromatogram input to the neural network. Since the determination result screen (see FIG. 13) including the above is generated at the same time, it is possible to improve the convenience when the user instructs to change the chromatogram input to the neural network.

In the present embodiment, the odor determination system 10 does not include a mass spectrometer that detects the mass of the components separated by the column 15. However, the odor determination system 10 can be changed to a system for determining the odor quality and intensity of a sample gas by using a gas chromatography-mass spectrometry (GCMS) by providing a mass spectrometer. be. When GCMS is used, the odor determination system 10 can obtain a chromatogram in which gas components are further separated as compared with the case where only gas chromatography is used, so that the accuracy of odor determination can be improved. Can be done.

On the other hand, when the odor determination system 10 is not provided with a mass spectrometer, it is not necessary to provide a large helium gas cylinder and a large vacuum pump, which are components of the mass spectrometer, so that the odor determination system 10 can be miniaturized. The portability can be improved. Therefore, the odor determination system 10 can be moved to the site where the odor is generated, and can easily determine the odor at the site. Therefore, in the odor determination system 10, when the odor-generating gas is put into the container for transportation, the odor is adsorbed on the container, or the odor-generating gas is transported. Even if the odor components change due to changes over time or chemical changes, the odor-generating gas is transported from the site where the odor is generated to the odor determination system 10. By moving the odor determination system 10 itself to the site without having to do so, the odor can be determined with high accuracy.

Since the odor determination system 10 takes in air as a carrier gas from the surroundings and uses it, portability can be improved.

When GCMS is used, the odor determination system 10 generally requires warm-up for about 30 minutes after operating the vacuum pump for 24 hours or more, unlike the case where only gas chromatography is used. Further, the odor determination system 10 requires a longer measurement time when GCMS is used than when only gas chromatography is used. Therefore, when using GCMS, the odor determination system 10 creates a necessary number of learning data for appropriately determining the quality and intensity of odor as compared with the case where only gas chromatography is used. It takes a lot of time to do this. On the other hand, when only gas chromatography is used, the odor determination system 10 can appropriately determine the quality and intensity of odor in a short time as compared with the case where GCMS is used.

In the present embodiment, the odor determination system 10 includes a PID sensor 16, a semiconductor gas sensor 17, and a semiconductor gas sensor 18 as a plurality of types of gas sensors. However, the odor determination system 10 may include a combination other than the combination of the PID sensor 16, the semiconductor gas sensor 17, and the semiconductor gas sensor 18 as a plurality of types of gas sensors. For example, the odor determination system 10 may include a semiconductor sensor array in which 100 or more semiconductor gas sensors are arranged as a plurality of types of gas sensors.

In addition, a part of the functions realized by the computer 20 in this embodiment may be realized by a system external to the computer 20 such as a cloud service. Similarly, at least a part of the data stored in the computer 20 in the present embodiment may be stored by a system external to the computer 20.

10 Smell judgment system 15 Column 16 PID sensor (gas sensor)
17 Semiconductor gas sensor (gas sensor)
18 Semiconductor gas sensor (gas sensor)
24a Odor Judgment Program 25a Chromatogram Generation Means 25b Odor Judgment Means 71 Odor Neural Network 72 Odor Intensity Neural Network 81a-81c, 82a-82c Waveforms

Claims (10)

A column that separates gas components and
Multiple types of gas sensors that detect the gas whose components are separated by the column, and
A chromatogram generating means for generating a chromatogram based on the detection result of the gas sensor for each gas sensor,
It is provided with an odor determination means for determining the quality of odor by using an odor neural network as a neural network in which the value for each time in the chromatogram is used as an explanatory variable and the quality of odor is at least the objective variable.
The odor determination means, produced by the chromatogram generating means consists of the chromatogram based on the respective detection results of the plurality of types of gas sensors, based on a plurality of chromatograms, to determine the odor quality ,
The odor determination means is an odor determination system, characterized in that one chromatogram obtained by combining and aggregating any one of the waveforms in the plurality of chromatograms is input to the neural network for each time zone. A column that separates gas components and
Multiple types of gas sensors that detect the gas whose components are separated by the column, and
A chromatogram generating means for generating a chromatogram based on the detection result of the gas sensor for each gas sensor,
An odor determination means for determining the quality of odor using an odor neural network as a neural network in which the value for each time in the chromatogram is used as an explanatory variable and the quality of odor is at least the objective variable.
With
The odor determining means determines the quality of odor based on a plurality of chromatograms composed of the chromatograms generated by the chromatogram generating means based on the detection results of the plurality of types of gas sensors. ,
The odor determining means can change the chromatogram input to the neural network based on the plurality of chromatograms according to an instruction, and the plurality of chromatograms and the chromatogram input to the neural network are described. An odor determination system characterized by generating a screen containing a chromatogram at the same time. The odor determining means according to claim 2 , wherein one chromatogram obtained by combining and aggregating any one of the waveforms in the plurality of chromatograms is input to the neural network for each time zone. Smell judgment system. The first or third aspect of the present invention, wherein the one chromatogram is a chromatogram obtained by combining and aggregating the waveforms having the highest peak values among the waveforms in the plurality of chromatograms for each time zone. Smell judgment system. 1. Item 4. The odor determination system according to item 4. The odor determining means applies the chromatogram, which simplifies the shape of the waveform, to the neural network so that at least one value other than the peak value becomes zero in at least one of the plurality of waveforms in the chromatogram. The odor determination system according to any one of claims 1 to 5, wherein the input is performed. The odor determining means uses an odor intensity neural network as a neural network in which the value for each time in the chromatogram is used as an explanatory variable and the odor intensity is used as an objective variable, and is based on the plurality of chromatograms. The odor determination system according to any one of claims 1 to 6, wherein the odor intensity is determined. The odor determination system according to any one of claims 1 to 7, wherein the mass spectrometer for detecting the mass of the components separated by the column is not provided. A chromatogram generating means for generating a chromatogram based on the detection results of a plurality of types of gas sensors for detecting a gas whose components are separated by a column for separating gas components, and a chromatogram generating means for each gas sensor.
An odor determination means for determining the quality of odor by using an odor neural network as a neural network in which the value for each time in the chromatogram is used as an explanatory variable and the quality of odor is at least the objective variable is realized in a computer. Let me
The odor determination means, produced by the chromatogram generating means consists of the chromatogram based on the respective detection results of the plurality of types of gas sensors, based on a plurality of chromatograms, to determine the odor quality ,
The odor determination means is an odor determination program characterized in that one chromatogram obtained by combining and aggregating any one of the waveforms in the plurality of chromatograms is input to the neural network for each time zone. A chromatogram generating means for generating a chromatogram based on the detection results of a plurality of types of gas sensors for detecting a gas whose components are separated by a column for separating gas components, and a chromatogram generating means for each gas sensor.
An odor determination means for determining the quality of odor using an odor neural network as a neural network in which the value for each time in the chromatogram is used as an explanatory variable and the quality of odor is at least the objective variable.
To the computer,
The odor determining means determines the quality of odor based on a plurality of chromatograms composed of the chromatograms generated by the chromatogram generating means based on the detection results of the plurality of types of gas sensors. ,
The odor determining means can change the chromatogram input to the neural network based on the plurality of chromatograms according to an instruction, and the plurality of chromatograms and the chromatogram input to the neural network are described. An odor determination program characterized by generating a screen containing a chromatogram at the same time.

Images