Patent Grant
11971701
This application is a National Stage Entry of PCT/JP2019/014242 filed on Mar. 29, 2019, the contents of all of which are incorporated herein by reference, in their entirety.
The present invention relates to an analysis technique using a smell sensor.
A technique for acquiring information relating to gas by measuring the gas with a sensor has been developed. PTL 1 below discloses a technique for determining a kind of sample gas by utilizing a signal (time-series data of a detection value) acquired by measuring the sample gas with a nanomechanical sensor. Specifically, PTL 1 discloses that, since a diffusion time constant of sample gas relative to a receptor of a sensor is determined by a combination of a kind of receptor and a kind of sample gas, the kind of sample gas can be determined based on the diffusion time constant acquired from a signal, and the kind of receptor.
A technique of PTL 1 performs an analysis of a kind of gas using a feature value acquired from output data of a sensor. However, output data (an output waveform) of a sensor that senses smell is essentially a high-order feature value, and performing an analysis with a high degree of accuracy is difficult. When a dynamic characteristic of a smell sensor can be successfully extracted as a feature value converted to a low order, performing an analysis using a smell sensor with a high degree of accuracy becomes easy.
The present invention has been made in view of the problem described above. One object of the present invention is to provide a technique for improving accuracy of an analysis using a smell sensor.
An information processing apparatus according to the present invention includes:
An information processing method performed by a computer according to the present invention includes:
A program according to the present invention causes a computer to execute the above-described information processing method.
According to the present invention, a technique for generating a feature value easy to handle in an analysis using a smell sensor is provided.
The above-described object, the other objects, features, and advantages will become more apparent from a suitable example embodiment described below and the following accompanying drawings.
Example embodiments according to the present invention are described below by use of the drawings. Note that, a similar reference sign is assigned to a similar component in all the drawings, and description is not repeated where appropriate. Further, unless otherwise specially described, each block represents, in each block diagram, not a configuration on a hardware basis but a configuration on a function basis. Moreover, a direction of an arrow in the drawings serves for easy understanding of flow of information, and does not limit a direction of communication (one-way communication/two-way communication) unless otherwise specially described.
<Functional Configuration>
Herein, the sensor 10 is a sensor having a receptor to which a molecule (a smell component) included in gas being a measurement target adheres, as illustrated in
A detection value (the output signal 14) of the sensor 10 changes due to an operation of exposing gas being a measurement target to the sensor 10 (hereinafter, this is also referred to as a “sampling operation”), and an operation of removing gas being a measurement target from the sensor 10 (hereinafter, this is also referred to as a “purge operation”). For example, a non-illustrated pump mechanism sucks in gas being a measurement target (a sampling operation) in a rising period of the input signal 12 (a period in which a signal level is High). Moreover, the non-illustrated pump mechanism removes gas being a measurement target from the sensor 10 by use of impurity gas (air, or the like) or the like (a purge operation) in a falling period of the input signal 12 (a period in which a signal level is Low). A detection value (the output signal 14) of the sensor 10 varies by control of the sampling operation or the purge operation in response to a value of the input signal 12. In other words, it can be said that an input signal for controlling the sampling operation and the purge operation is equivalent to input to the sensor 10 in a system called the sensor 10. In the following description, as needed, the input signal 12 and the output signal 14 are also referred to as U and Y, respectively. Moreover, a value of the input signal 12 at a time t and a value of the output signal 14 at a time t are also referred to as u(t) and y(t), respectively. U becomes a matrix in which u(t)s are enumerated. Y becomes a matrix in which y(t)s are enumerated.
Returning to
The model generating unit 210 learns an input/output relation of the sensor 10 by use of input data of the sensor 10 and output data of the sensor 10, and generates an Auto-Regressive with eXogenous input (ARX) model indicating the input/output relation of the sensor 10. Herein, the input data of the sensor 10 are data that control an input operation (sampling operation/purge operation) of gas including a smell component being a measurement target. As to the example of
Herein, an ARX model of the sensor 10 is represented by an equation (1) below. In the equation (1), y(t) is an output of the sensor 10 at a time t, u(t) is an input to the sensor 10 at a time t, ai (with underline) is an autoregressive coefficient, and bi (with underline) is an exogenous input coefficient. The model generating unit 210 can learn (generate) an input/output relation of the sensor 10 as an ARX model indicated by the equation (1) below, from, for example, the input signal 12 and the output signal 14 as illustrated in
Next, the feature value computation unit 220 generates a feature value indicating a characteristic of a sensor by use of the ARX model generated by the model generating unit 210. First, the feature value computation unit 220 performs Z-Transform on the ARX model generated by the model generating unit 210. Further, the feature value computation unit 220 computes a transfer function of a first-order lag system by subjecting, to partial fraction decomposition, a result of subjecting the ARX model to Z-Transform.
The feature value computation unit 220 first acquires an equation (2) below by subjecting, to Z-Transform, the ARX model indicated by the equation (1). In the equation (2) below, Y(z)/U(z) is a ratio of Z-Transform of an output Y to an input U of a sensor (i.e., a transfer function in a Z-area).
Further, the feature value computation unit 220 acquires a following equation by subjecting a right side of the equation (2) to partial fraction decomposition.
In the equation (3), ai indicates a feature value relating to a desorption rate of a smell component i, and bi (with tilde) indicates a feature value relating to an adsorption rate of the smell component i. Note that, bi with tilde is also referred to briefly as “bi” in the following description. The feature value computation unit 220 acquires a pair of ai and bi as a first-order lag transfer function feature value, as indicated in an equation (4) below.
{ai,{tilde over (b)}i}i=1m [Mathematical 4]
WHERE ai:=(1−Δtβi), {tilde over (b)}i:=Δtγiαiρi,
Note that, in the equation (4) above, αi indicates an adsorption rate of the smell component i, βi indicates a desorption rate of the smell component i, γi indicates a proportionality constant of the number of molecules adhering to a sensor receptor relating to the smell component i and a sensor output generated thereby, ρi indicates density of the smell component i, and Δt indicates a time interval in a discrete-time system. The first-order lag transfer function feature value represented by the equation (4) above can be utilized as a feature value representing a combination of a sensory membrane being set in the sensor 10 and the smell component i. Moreover, dynamics of the sensor 10 is physically interpretable for each smell component i from the above-described relational expression of ai and bi.
A first-order lag transfer function feature value acquired by the present example embodiment is a feature value of an order lower than output data of the sensor 10. Accuracy of a discrimination analysis or a regression analysis can be improved by using the first-order lag transfer function feature value converted to a low order in this way.
<Hardware Configuration of Information Processing Apparatus 20>
Each functional configuration unit of the information processing apparatus 20 may be achieved by hardware (example: a hard-wired electronic circuit, or the like) that achieves each functional configuration unit, or may be achieved by a combination of hardware and software (example: a combination of an electronic circuit and a program controlling the electronic circuit, or the like). A case where each functional configuration unit of the information processing apparatus 20 is achieved by a combination of hardware and software is further described below.
The computer 1000 includes a bus 1020, a processor 1040, a memory 1060, a storage device 1080, an input/output interface 1100, and a network interface 1120. The bus 1020 is a data transmission path through which the processor 1040, the memory 1060, the storage device 1080, the input/output interface 1100, and the network interface 1120 transmit/receive data to/from each other. However, a method of mutually connecting the processor 1040 and the like is not limited to bus connection.
The processor 1040 includes various types of processors such as a central processing unit (CPU), a graphics processing unit (GPU), and a field-programmable gate array (FPGA). The memory 1060 is a main storage apparatus achieved by use of a random access memory (RAM) or the like. The storage device 1080 is an auxiliary storage apparatus achieved by use of a hard disk, a solid state drive (SSD), a memory card, a read only memory (ROM), or the like.
The input/output interface 1100 is an interface for connecting the computer 1000 and an input/output device. For example, an input apparatus such as a keyboard or a touch panel, and an output apparatus such as a display device or a speaker are connected to the input/output interface 1100.
The network interface 1120 is an interface for connecting the computer 1000 to a communication network. The communication network is, for example, a local area network (LAN) or a wide area network (WAN). A method of connecting the network interface 1120 to the communication network may be wireless connection or may be wired connection.
The storage device 1080 stores a program module that achieves each functional configuration unit (the model generating unit 210, the feature value computation unit 220, and the like) of the information processing apparatus 20. The processor 1040 reads each of the program modules onto the memory 1060, executes the read program module, and thereby achieves a function being associated with each of the program modules.
<Flow of Processing>
First, the model generating unit 210 acquires input/output data for each sensory membrane (S102). For example, the model generating unit 210 acquires input data U to the sensor 10 and output data YK of the sensory membrane K as input/output data of the sensory membrane K. Moreover, the model generating unit 210 acquires input data U to the sensor 10 and output data YL of the sensory membrane L as input/output data of the sensory membrane L. Then, the model generating unit 210 generates an ARX model for each sensory membrane, based on input/output data for each sensory membrane (S104). For example, the model generating unit 210 generates an ARX model regarding the sensory membrane K, based on the input data U to the sensor 10 and the output data YK of the sensory membrane K. Moreover, the model generating unit 210 generates an ARX model regarding the sensory membrane L, based on the input data U to the sensor 10 and the output data YL of the sensory membrane L. Then, the feature value computation unit 220 performs Z-Transform on the ARX model generated for each sensory membrane (S106). Then, the feature value computation unit 220 subjects, to partial fraction decomposition, each result of subjecting each ARX model to Z-Transform, and computes a first-order lag transfer function feature value for each sensory membrane (S108).
A model generating unit 210 may be configured in such a way as to extract a plurality of pieces of partial input data and a plurality of pieces of partial output data by use of a plurality of windows, and generate a plurality of ARX models.
The model generating unit 210 generates an ARX model by use of one pair of partial input data and partial output data extracted for each window as the above-described input data and output data. In other words, the model generating unit 210 generates a plurality of ARX models by use of a plurality of pieces of partial input data and a plurality of pieces of partial output data extracted by use of a plurality of windows.
Then, a feature value computation unit 220 computes a plurality of first-order lag transfer function feature values by subjecting the plurality of ARX models to Z-Transform and partial fraction decomposition. Then, the feature value computation unit 220 executes machine learning by use of the plurality of computed first-order lag transfer function feature values as learning data, and determines a first-order lag transfer function feature value of the sensor 10. Moreover, the feature value computation unit 220 may be configured in such a way as to determine a first-order lag transfer function feature value of the sensor 10 after performing statistical processing such as abnormal value removal on the plurality of computed first-order lag transfer function feature values.
The configuration according to the present modification example allows acquisition of a first-order lag transfer function feature value having a higher degree of accuracy as compared with a case of computing a first-order lag transfer function feature value (a feature value indicating a dynamic characteristic of the sensor 10) by use of one ARX model.
In the present example embodiment, one example of application of a first-order lag transfer function feature value is described. When two or more sensory membranes differing in kind from each other are set in a sensor 10, an information processing apparatus 20 can generate a feature value being robust against a change of a measurement environment by use of a first-order lag transfer function feature value for each sensory membrane.
First, a model generating unit 210 acquires input/output data for each sensory membrane, in response to measurement of the sample gas including a smell component i (S202). Then, the model generating unit 210 generates an ARX model for each sensory membrane, based on input/output data for each sensory membrane (S204). Then, the feature value computation unit 220 subjects the ARX model for each sensory membrane to Z-Transform (S206). Further, the feature value computation unit 220 computes a first-order lag transfer function feature value (ai, bi) for each sensory membrane by subjecting, to partial fraction decomposition, a result of subjecting the ARX model to Z-Transform (S208). The processing in S202 to S208 is similar to the processing in S102 to S108 in
Herein, an output of the sensor 10 can vary in response to not only a kind of smell component being a measurement target, but also an environment (e.g., temperature, humidity, or the like at measurement) in which the smell component is measured. Note that, an output of the sensor 10 varies depending on a change of stress generated in a support member by a smell component adhering to a sensory membrane set in the sensor 10, as described by use of
Note that, the above-described assumption has an ideal that a parameter specific for each kind of sensory membrane is always a constant value regardless of a measurement environment. However, in reality, it can hardly be said that a parameter specific for each membrane kind is not at all affected by a change of a measurement environment. Thus, the information processing apparatus 20 may be configured in such a way as to perform the following processing.
First, the sensor 10 measures sample gas including the smell component i under environments differing from each other a plurality of times. For each single measurement, the model generating unit 210 generates each of an ARX model regarding the sensory membrane K and an ARX model regarding the sensory membrane L, and the feature value computation unit 220 computes, based on the ARX models of the sensory membranes K and L, the first-order lag transfer function feature values “biK” and “biL” related to an adsorption rate of the smell component i, respectively. Moreover, the feature value computation unit 220 acquires the ratio “biK/biL” of the first-order lag transfer function feature values “biK” and “biL” for each single measurement. Then, the feature value computation unit 220 determines, based on a plurality of “biK/biL” acquired by a plurality of times of measurements, a range of a measurement environment in which “biK/biL” becomes constant (an inclination is 0). Then, the feature value computation unit 220 stores, in a storage area such as a storage device 1080, information indicating the determined range of the measurement environment, in association with information indicating a kind of smell component and information indicating a kind of sensory membrane (a combination of sensory membranes).
For example, it is assumed that a result as illustrated in
In the example of
The configuration according to the present example embodiment allows generation of a database that accumulates a condition (a combination of a kind of smell component, a kind of sensory membrane, and a measurement environment) in which a feature value being robust against a change of a measurement environment can be acquired. The database can be utilized, for example, as described in a third example embodiment.
<Functional Configuration>
As one example, the output unit 230 can output information indicating a recommended measurement environment for each kind of smell component, with input of information indicating a configuration of a sensor 10 (a kind of sensory membrane being set in the sensor 10). As a specific example, it is assumed that the output unit 230 acquires input information indicating a combination of a sensory membrane K and a sensory membrane L in a state where information as illustrated in
In the present example, when a configuration of the sensor 10 is determined, information indicating a recommended measurement environment is output for each kind of smell component, by inputting information indicating the configuration of the sensor 10. With such information, a user of the sensor 10 can easily determine how to utilize the sensor 10 having the determined configuration (under what environment and for what smell component measurement is performed).
As another example, the output unit 230 can output information indicating a recommended configuration of the sensor 10 (a combination of sensory membranes) for each kind of smell component, with input of information indicating a measurement environment (an environment in which the sensor 10 is placed) of the smell component. As a specific example, it is assumed that the output unit 230 acquires input information indicating that temperature of a measurement environment is in a range of T1 to T2 in a state where information as illustrated in
In the present example, when a measurement environment (an environment in which the sensor 10 is placed) is determined, information indicating a recommended configuration of the sensor 10 is output for each kind of smell component, by inputting information on the measurement environment. With such information, a user can be notified of what configuration of a sensor may be used for what smell in the determined measurement environment to enable stable measurement.
As another example, the output unit 230 can output information indicating a recommended configuration of the sensor 10 (a combination of sensory membranes) and a recommended measurement environment (a range of temperature or humidity, or the like), with input of information indicating a kind of smell component being a measurement target. As a specific example, it is assumed that the output unit 230 acquires input information indicating the “smell component i” as information on a smell component being a measurement target in a state where information as illustrated in
In the present example, when a smell component to be a measurement target is determined, information indicating a configuration of the sensor 10 and a measurement environment that are suited to measurement of the smell component is output, by inputting information on the smell component. With such information, a user becomes able to easily determine “what configuration of a smell sensor to prepare and under what environment the smell sensor is operated in order to accurately perform a discrimination analysis of a smell component being a target”.
As another example, the output unit 230 can output information indicating a recommended measurement environment (a range of temperature or humidity, or the like), with input of information indicating a configuration of the sensor 10 (a combination of sensory membranes) and a kind of smell component being a measurement target. As a specific example, it is assumed that the output unit 230 acquires input information indicating a combination of the sensory membrane K and the sensory membrane L and the smell component i in a state where information as illustrated in
In the present example, when a configuration of a smell sensor and a smell component to be measured with the smell sensor are determined, information indicating a recommended measurement environment is output, by inputting information on the configuration of the smell sensor and the smell component. With such information, a user of the sensor 10 can easily recognize an environment in which measurement can be performed with stable accuracy.
As another example, the output unit 230 can output information indicating a kind of smell component recommended as a measurement target, with input of information indicating a configuration of the sensor 10 (a combination of sensory membranes), and information indicating a measurement environment (an environment in which the sensor 10 is placed) of a smell component. As a specific example, it is assumed that the output unit 230 acquires input information indicating a combination of the sensory membrane K and the sensory membrane L and a temperature range of T1 to T2 in a state where information as illustrated in
In the present example, when a configuration of a smell sensor and an environment in which the smell sensor is placed are already known, information indicating a smell component recommended as a measurement target is output, by inputting information on the configuration of the smell sensor and the environment. With such information, a user of the sensor 10 can easily recognize a smell component suited to measurement.
As another example, the output unit 230 can output information indicating a recommended configuration of the sensor 10 (a combination of sensory membranes), with input of information indicating a kind of smell component being a measurement target and a measurement environment (an environment in which the sensor 10 is placed) of the smell component. As a specific example, it is assumed that the output unit 230 acquires input information indicating the smell component i and a temperature range of T1 to T2 in a state where information as illustrated in
In the present example, when a smell component to be a measurement target and an environment in which the smell component is measured are already known, information indicating a recommended configuration of the sensor 10 (a combination of sensory membranes) is output, by inputting information on the smell component and the environment. With such information, even an inexperienced person can easily determine a sensory membrane to be set in the sensor 10.
Moreover, when information indicating a plurality of smell components being measurement targets is acquired as an input together with information indicating a measurement environment, the output unit 230 can also determine a priority order of a configuration of the sensor 10, based on information on the database acquired in the third example embodiment. Specifically, it is assumed that, in a range of a measurement environment indicated by input information, there exist a combination A of sensory membranes in which a ratio of first-order lag transfer function feature values becomes constant (an inclination is 0) with regard to all the smell components indicated by the input information, and a combination B of sensory membranes in which a ratio of first-order lag transfer function feature values does not become constant (an inclination is not 0) with regard to at least some of the smell components. In this case, the output unit 230 gives a higher priority to the combination A that enables an analysis with stable accuracy with regard to all the smell components than the combination B. Then, the output unit 230 outputs information indicating a priority for each combination to a display apparatus or the like connected to the information processing apparatus 20 (example:
In the present example embodiment, another example of application of a first-order lag transfer function feature value is described. When two or more sensory membranes of the same kind are set in a sensor 10, an information processing apparatus 20 can inspect performance of each of the sensory membranes by use of a first-order lag transfer function feature value for each sensory membrane.
<Functional Configuration>
<Flow of Processing>
First, a model generating unit 210 acquires input/output data for each sensory membrane, in response to measurement of the sample gas including a smell component i (S302). Then, the model generating unit 210 generates an ARX model for each sensory membrane, based on input/output data for each sensory membrane (S304). Then, the feature value computation unit 220 subjects the ARX model for each sensory membrane to Z-Transform (S306). Further, the feature value computation unit 220 computes a first-order lag transfer function feature value (ai, bi) for each sensory membrane by subjecting, to partial fraction decomposition, a result of subjecting the ARX model to Z-Transform (S308). The processing in S302 to S308 is similar to the processing in S102 to S108 in
Then, the feature value computation unit 220 computes a ratio (biK/biL) of the first-order lag transfer function feature values bis related to an adsorption rate of the smell component i, with regard to the sensory membrane K and the sensory membrane L (S310). Herein, when the sensory membrane K and the sensory membrane L have the same performance, a first-order lag transfer function feature value of each of the sensory membranes becomes equal. In this case, a value of biK/biL becomes 1. Thus, the product judgement unit 240 judges whether biK/biL satisfies a predetermined reference (biK/biL becomes 1 or a value close to 1) (S312). When biK/biL satisfies the predetermined reference (S312: YES), the product judgement unit 240 judges that the sensory membrane K being an inspection target is an “acceptable product having performance equal to the sensory membrane L being a reference product” (S314). On the other hand, when biK/biL does not satisfy the predetermined reference (S312: NO), the product judgement unit 240 judges that the sensory membrane K being an inspection target is a “rejected product that does not have performance equal to the sensory membrane L being a reference product” (S316).
In consequence, the present example embodiment enables to determine whether performance of a sensory membrane is good/poor, by using a first-order lag transfer function feature value acquired by the method described in the first example embodiment.
In the present example embodiment, another example of application of a first-order lag transfer function feature value is described. When two or more sensory membranes of the same kind are set in a sensor 10, an information processing apparatus 20 can correct an individual difference (a margin of error of output performance) between the sensory membranes by use of a first-order lag transfer function feature value for each sensory membrane.
<Functional Configuration>
<Flow of Processing>
First, a model generating unit 210 acquires input/output data for each sensory membrane, in response to measurement of the sample gas including a smell component i (S402). Then, the model generating unit 210 generates an ARX model for each sensory membrane, based on input/output data for each sensory membrane (S404). Then, the feature value computation unit 220 subjects the ARX model for each sensory membrane to Z-Transform (S406). Further, the feature value computation unit 220 computes a first-order lag transfer function feature value (ai, bi) for each sensory membrane by subjecting, to partial fraction decomposition, a result of subjecting the ARX model to Z-Transform (S408). The processing in S402 to S408 is similar to the processing in S102 to S108 in
Then, the feature value computation unit 220 computes a ratio (biK/biL) of the first-order lag transfer function feature values bis related to an adsorption rate of the smell component i, with regard to the sensory membrane K and the sensory membrane L (S410). The individual difference correction unit 250 determines a correction coefficient for correcting an output value of either one of the sensory membrane K and the sensory membrane L, based on the ratio (biK/biL) between a first-order lag transfer function feature value biK and a first-order lag transfer function feature value biL acquired by the processing in S410 (S412). Specifically, the individual difference correction unit 250 determines a reciprocal number of biK/biL as a correction coefficient for the output value of the sensory membrane K, and stores the determined correction coefficient in a memory 1060 or the like. Alternatively, the individual difference correction unit 250 may determine biK/biL as a correction coefficient for the output value of the sensory membrane L, and stores the determined correction coefficient in the memory 1060 or the like. Then, the individual difference correction unit 250 corrects the output value of the sensory membrane K or the sensory membrane L by use of the correction coefficient determined in S412 (S414).
In the present example embodiment, an individual difference for each sensory membrane is corrected by way of software. Variation in accuracy of an analysis of the sensor 10 can be prevented by equalizing performance of each sensory membrane.
While the example embodiments of the present invention have been described above with reference to the drawings, the present invention should not be limited to the example embodiments and interpreted accordingly, and various modifications, improvements, and the like can be made based on knowledge of a person skilled in the art without departing from the spirit of the present invention. Various inventions can be formed by a suitable combination of a plurality of components disclosed in the example embodiments. For example, some of all the components indicated in the example embodiments may be deleted, or components in differing example embodiments may be suitably combined. Moreover, in each example embodiment, an order of illustrated processes (steps) can be altered to the extent consistent with contents.
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/JP2019/014242 | 3/29/2019 | WO |
| Publishing Document | Publishing Date | Country | Kind |
|---|---|---|---|
| WO2020/202338 | 10/8/2020 | WO | A |
| Number | Name | Date | Kind |
|---|---|---|---|
| 20190317066 | Imamura et al. | Oct 2019 | A1 |
| Number | Date | Country |
|---|---|---|
| 2006-120049 | May 2006 | JP |
| 2010-019107 | Jan 2010 | JP |
| 2017-083188 | May 2017 | JP |
| 2017-156254 | Sep 2017 | JP |
| 2018-087722 | Jun 2018 | JP |
| Entry |
|---|
| International Search Report for PCT Application No. PCT/JP2019/014242, dated Jun. 18, 2019. |
| Number | Date | Country | |
|---|---|---|---|
| 20220221839 A1 | Jul 2022 | US |
