Detecting device, estimating system, detecting method, control program for detecting device, and recording medium

Abstract

According to the present invention, abnormalities and changes in the sensitivity of a detecting device are easily detected. The detecting device includes: a sensor unit capable of outputting time-series data of a detected value corresponding to the concentration of a component of interest; an identifying unit; and an abnormality determining unit. The identifying unit: inputs second time-series data, which are output when the detecting device at a second time point detects the component of interest included in a gas of interest, into an identification model that has machine-learned first time-series data output when the sensor unit detects the component of interest contained in a standard gas at a first time point at which the detecting device is in a first state; and outputs difference information. The abnormality determining unit determines a second state, which is the state of the detecting device at the second time point, on the basis of the difference information.

Description

Detection device, estimation system, detection method, control program for detection device, and recording medium

[0001] The present disclosure relates to a detection device, an estimation system, a detection method, and the like that detect an abnormality in a sensor.

[0002] After shipment, devices equipped with gas sensors may experience some abnormalities or changes in sensitivity, etc. For example, in the case of electrochemical gas sensors, it is known that the sensitivity can change due to changes in the amount of electrolyte inside.

[0003] Conventionally, when inspecting whether an abnormality, sensitivity change, etc. has occurred in an apparatus equipped with a gas sensor, this is done by exposing the gas sensor equipped with the apparatus to a standard gas whose composition and concentration are known. Whether an abnormality, sensitivity change, etc. has occurred in the gas sensor equipped with the apparatus can be determined based on the comparison result between the composition and concentration of the standard gas and the detection result output from the apparatus. For example, Patent Document 1 discloses a technology for correcting a calibration curve using a standard gas and determining an abnormality in the gas sensor.

[0004] Japanese Patent Application Publication No. 2020-173143

[0005] In order to solve the above problem, a detection device according to one aspect of the present disclosure comprises: an identification unit that inputs second time series data output by the detection device at a second time point after the first time point when the detection device detects the target component contained in the target gas when the detection device detects the target component contained in the target gas, into an identification model that has been machine-learned using first time series data output by the detection device, which has a sensor unit capable of outputting time series data of detection values ​​corresponding to the concentration of the target component contained in the target gas, at a first time point when the detection device is in a first state; and an identification unit that outputs difference information regarding the difference between the first time series data and the second time series data; and an abnormality determination unit that determines a second state, which is the state of the detection device at the second time point, based on the magnitude of the difference indicated by the difference information.

[0006] An estimation system according to one aspect of the present disclosure includes a detection device having a sensor unit, and an estimation device communicatively connected to the detection device, wherein the sensor unit is capable of outputting time series data of detection values ​​corresponding to the concentration of a target component contained in a target gas; an identification unit that inputs second time series data output by the detection device at a second time point after the first time point when it detects the target component contained in the target gas into an identification model that has been machine-learned using first time series data output by the sensor unit when the detection device detects the target component contained in a standard gas at a first time point when the detection device is in a first state, and outputs difference information regarding the difference between the first time series data and the second time series data; and an abnormality determination unit that determines the state of the detection device at the second time point based on the magnitude of the difference indicated by the difference information.

[0007] A detection method according to one aspect of the present disclosure includes an identification step of inputting second time series data output by a detection device at a second time point after the first time point when the detection device detects the target component contained in the target gas, into an identification model that has been machine-learned using first time series data output by the detection device, which has a sensor unit capable of outputting time series data of detection values ​​corresponding to the concentration of the target component contained in the target gas, when the sensor unit detects the target component contained in the target gas at a first time point when the detection device is in a first state, and outputting difference information regarding the difference between the first time series data and the second time series data; and a determination step of determining a second state, which is the state of the detection device at the second time point, based on the magnitude of the difference indicated by the difference information.

[0008] An estimation system according to one aspect of the present disclosure includes a sensor unit including a plurality of sensor elements capable of outputting an output signal corresponding to the concentrations of a first gas component and a second gas component different from the first gas component contained in a target gas; a supply mechanism capable of supplying the target gas to the sensor unit; an estimation unit that estimates the concentrations of the first gas component and the second gas component contained in the target gas based on the output signals; and a determination unit that outputs a determination result of estimation accuracy by the estimation unit based on the output signals, wherein the plurality of sensor elements include sensor elements that differ in at least one of detection sensitivity to the first gas component and detection sensitivity to the second gas component. The determination unit uses an estimation model that has been machine-learned to learn feature information including: (1) information created based on a normal signal, which is the output signal output by the sensor unit when all of the sensor elements included in the plurality of sensor elements are operating normally; (2) information created based on an abnormal signal, which is the output signal output by the sensor unit including at least one sensor element that is not operating normally; or (3) information created to imitate the abnormal signal based on the normal signal; and outputs information regarding the estimation accuracy of the estimation result based on the output signal at the time of interest from the feature information in the output signal output by the sensor unit at the time of interest.

[0009] An estimation system according to one aspect of the present disclosure includes a sensor unit having a sensor element capable of outputting an output signal corresponding to the concentration of a target component contained in a target gas, a supply mechanism capable of supplying the target gas to the sensor unit, an estimation unit that estimates an estimated value of the concentration of the target component contained in the target gas based on the output signal, and a judgment unit that determines the accuracy of the estimation of the concentration of the target component by the estimation unit based on the estimated value, and if the estimated value is a negative value, the judgment unit determines that the accuracy of the estimation of the concentration of the target component by the estimation unit has decreased.

[0010] an estimation system according to one aspect of the present disclosure, comprising: a detection device including a sensor unit including a plurality of sensor elements capable of outputting an output signal corresponding to the concentration of target components contained in a target gas, the target components including a first gas component and a second gas component different from the first gas component; a supply mechanism capable of supplying the target gas to the sensor unit; an estimation unit that estimates the concentration of the first gas component and the concentration of the second gas component contained in the target gas based on the output signal; an identification unit that inputs second time series data output by the detection device at a second time point after the first time point into an identification model that has been machine-learned using first time series data output when the sensor unit detects the target component contained in a standard gas at a first time point when the detection device is in a first state, and outputs discrepancy information regarding a difference between the first time series data and the second time series data; a first determination unit that determines a second state, which is the state of the detection device at the second time point, based on the magnitude of the difference indicated by the discrepancy information; a second determination unit that outputs a determination result of the estimation accuracy by the estimation unit based on the output signal; the second determination unit uses an estimation model that has been machine-learned to acquire feature information from the output signal output by the sensor unit at a time point of interest, the feature information including: (1) information created based on a normal signal that is the output signal output by the sensor unit when all of the sensor elements included in the plurality of sensor elements are operating normally; (2) information created based on an abnormal signal that is the output signal output by the sensor unit when at least one of the sensor elements is not operating normally; or (3) information created to imitate the abnormal signal based on the normal signal,The third determination unit outputs information regarding the estimation accuracy of the estimation result based on the output signal at the time point of interest, and if the estimation value is a negative value, the third determination unit determines that the estimation accuracy of the concentration of the target component by the estimation unit has decreased, and the fourth determination unit determines whether the accuracy of the concentration estimation in the detection device is normal or not based on the determination results of the first determination unit, the second determination unit, and the third determination unit.

[0011] The detection device and estimation system according to each aspect of the present disclosure may be realized by a computer. In this case, the control program for the detection device that causes the computer to operate as each part (software element) of the detection device to realize the detection device, and the computer-readable recording medium on which the control program is recorded, also fall within the scope of the present disclosure.

[0012] FIG. 1 is a diagram illustrating an example of a schematic configuration of an estimation system according to an embodiment of the present disclosure. FIG. 2 is a diagram illustrating an example of a function included in a detection device according to an aspect of the present disclosure. FIG. 3 is a functional block diagram illustrating an example of a configuration of a detection device. FIG. 4 is a diagram illustrating an example of a configuration of an identification model. FIG. 5 is a diagram illustrating a comparative example of output data output from an identification model when second waveform data corresponding to second detection data output from a sensor unit of a detection device operating normally is used as input data to be input to the identification model, and input data. FIG. 6 is a diagram illustrating a comparative example of output data output from an identification model when second waveform data corresponding to second detection data output from a sensor unit of a detection device not operating normally is used as input data to be input to the identification model, and input data. FIG. 7 is a diagram illustrating characteristics of waveform data corresponding to second detection data output from the sensor unit when a leak occurs in the pump unit. FIG. 8 is a diagram illustrating a comparative example of output data output from an identification model when second waveform data corresponding to second detection data output from the sensor unit when a leak occurs in the pump unit is used as input data to be input to the identification model, and input data. 17 is a diagram showing a comparative example between output data output from the identification model when second waveform data corresponding to second detection data output from the sensor unit when an abnormality occurs in the valve unit is used as input data to be input to the identification model. FIG. 18 is a flowchart showing an example of the flow of processing performed by the detection device. FIG. 19 is a graph showing the relationship between the median of cumulative difference values ​​calculated from difference information and a value indicating the accuracy of concentration estimation of a target component. FIG. 19 is a block diagram showing an example of the configuration of a detection device according to a fifth embodiment of the present disclosure. FIG. 19 is a diagram showing an example of a graph plotting output values ​​of three sensor elements. FIG. 20 is a diagram showing another example of a graph plotting output values ​​of three sensor elements. FIG. 21 is a flowchart showing an example of the flow of processing performed by the detection device shown in FIG. 13. FIG. 22 is a block diagram showing an example of the configuration of a detection device according to a sixth embodiment of the present disclosure. FIG. 23 is a graph showing the relationship between the estimated value of concentration and estimation accuracy. FIG. 24 is a flowchart showing an example of the flow of processing performed by the detection device shown in FIG.FIG. 13 is a block diagram showing the configuration of a detection device according to a seventh embodiment of the present disclosure.

[0013] It is not easy to provide a standard gas to all users of the device. Furthermore, it is troublesome for users to perform tests using a standard gas to check whether the device is operating normally. According to the present disclosure, it is possible to easily detect abnormalities and sensitivity changes in a device equipped with a gas sensor without performing tests using a standard gas.

[0014] First Embodiment Hereinafter, one embodiment of the present disclosure will be described in detail.

[0015] (Configuration of estimation system 100) First, a schematic configuration of an estimation system 100 according to an embodiment of the present disclosure will be described with reference to Fig. 1. Fig. 1 is a schematic diagram showing an example of a schematic configuration of an estimation system 100 according to an embodiment of the present disclosure. For convenience of explanation, each drawing referred to in this specification is a schematic diagram showing only some components in a simplified form to explain the embodiment. Therefore, the estimation system 100 may include any component not shown in each drawing referred to in this specification.

[0016] The estimation system 100 includes a detection device 1, an estimation device 5, and an electronic device 9. FIG. 1 illustrates the estimation system 100 in which one detection device 1 is installed at each of a plurality of installation locations Q, and an electronic device 9 communicatively connected to the detection device 1 is present at each installation location Q; however, the estimation system 100 is not limited to this configuration. For example, in the estimation system 100, a plurality of detection devices 1 may be installed at each installation location Q. For example, in the estimation system 100, a single electronic device 9 may be communicatively connected to the detection devices 1 installed at each of the plurality of installation locations Q. The detection device 1 and the electronic device 9 may be connected via wireless communication or wired communication. The detection device 1 and the electronic device 9 may each be communicatively connected to the estimation device 5 via a communication network 7.

[0017] The detection device 1 includes a sensor unit 30 capable of outputting time-series data (hereinafter, detection data) of detection values ​​corresponding to the concentration of a detection target component (hereinafter, target component) contained in a gas. The detection device 1 may be configured to output waveform data corresponding to the detection data, the waveform data being obtained by plotting the detection data on two-dimensional coordinates. The sensor unit 30 may include multiple sensor elements 31. When the sensor unit 30 includes multiple sensor elements 31, each sensor element 31 may be capable of outputting detection data corresponding to the concentration of a different target component. When the sensor unit 30 includes multiple sensor elements 31, each sensor element 31 may have a different detection sensitivity for the same target component. The detection device 1 transmits the detection data to the electronic device 9. The detection device 1 is a device used by a user of the estimation system 100 and is installed at a predetermined installation location Q.

[0018] The electronic device 9 transmits the detection data received from the detection device 1 to the estimation device 5, and receives the estimation result output by the estimation device 5 based on the detection data transmitted to the estimation device 5. The electronic device 9 presents the received estimation result to the user by displaying it on a display screen or the like. The electronic device 9 may be a personal computer, smartphone, or the like used by the user who has installed the detection device 1 at the installation location Q.

[0019] In the estimation system 100, the detection device 1 may have a function of estimating the concentration of the target component based on detection data corresponding to the concentration of the target component. In this case, the electronic device 9 may transmit concentration information indicating the concentration of the target component received from the detection device 1 to the estimation device 5, and receive an estimation result output by the estimation device 5 based on the concentration information transmitted to the estimation device 5.

[0020] In the estimation system 100, the detection device 1 may be configured to have some or all of the functions of the electronic device 9. For example, the detection device 1 may be capable of transmitting detection data to the estimation device 5, in which case the electronic device 9 receives an estimation result output by the estimation device 5 based on the detection data transmitted from the detection device 1 to the estimation device 5. Alternatively, the detection device 1 may have all the functions of the electronic device 9, in which case the detection device 1 may be capable of transmitting detection data to the estimation device 5, receiving an estimation result transmitted from the estimation device 5, and presenting the estimation result. For example, the detection device 1 may have some or all of the functions of the estimation device 5.

[0021] The estimation device 5 inputs the detection data received from the electronic device 9 into an estimation model, and transmits the estimation result output from the estimation model to the electronic device 9 that is the sender of the detection data. The estimation device 5 may be a computer or a server device managed by an administrator of the estimation system 100. The estimation model may be a neural network that estimates the air condition at the installation location Q and the user's condition (e.g., health condition) from the concentration of the target component corresponding to the detection data.

[0022] As an example, the estimation system 100 may be a system that detects target components from gas (target gas) containing gas originating from a user in a toilet and estimates the user's health condition. For example, the gas originating from the user may be gas released from the user's stool. When the gas originating from the user is gas released from the user's stool, the detection device 1 may be installed on the toilet bowl. When the detection device 1 is installed on the toilet bowl, the estimation system 100 performs a process in the toilet to detect target components from the target gas containing gas released from the user's stool. Therefore, a user using the estimation system 100 does not need to perform cumbersome tasks such as stool testing, and can simply use the toilet. The gas originating from the user is not limited to gas released from the user's stool, but may be, for example, gas released from the user's exhaled breath, skin, or saliva.

[0023] Furthermore, the detection device 1 does not have to be fixedly installed in one location. For example, the detection device 1 may be portable by the user. Specifically, the user may carry the detection device 1 of the estimation system 100 and attach the detection device 1 to the toilet bowl each time the user uses the toilet. With the above configuration, the user can use the estimation system 100 in any location (for example, while out and about).

[0024] (Functions and configuration of detection device 1) Next, functions of the detection device 1 will be described with reference to Fig. 2. Fig. 2 is a diagram for explaining an example of functions of the detection device 1 according to one embodiment of the present disclosure.

[0025] The detection device 1 that has been confirmed to be operating normally is shipped from the factory P that manufactured the detection device 1 to the user and installed at the installation location Q (for example, a toilet bowl in the user's home). The factory P may be the sender of the detection device 1.

[0026] Here, at the factory, based on the detection data output from the detection device 1 before shipment when a standard gas (standard gas) is supplied to the sensor unit 30, it can be confirmed that the detection device 1 is operating normally and that there are no problems with the detection sensitivity and detection accuracy (first state). The standard gas is a gas known to contain a predetermined concentration of the target component. For example, the standard gas may be a gas formulated to contain a predetermined concentration of the target component. For example, the standard gas may be a gas simulating the target gas containing the target component to be detected by the detection device 1 after shipment at the installation location Q. For example, it is possible to determine whether the detection device 1 is operating normally based on the correspondence between the concentration of the target component contained in the standard gas and the detection data output from the detection device 1.

[0027] For example, if the estimation system 100 is a system that detects target components from target gases, including gases emitted from the user's feces, to estimate the user's health condition, it is important that the detection device 1 operates normally in order to provide the user with accurate and reliable estimation results. However, it is not easy to provide a standard gas to all users. Furthermore, it is troublesome for users to perform tests using a standard gas to confirm whether the detection device 1 is operating normally.

[0028] Therefore, the detection device 1 has a function that can determine whether the device itself is operating normally without performing an inspection using a standard gas. The configuration of the detection device 1 having such a function will be described with reference to Fig. 3. Fig. 3 is a functional block diagram showing an example configuration of the detection device 1. The detection device 1 has a sensor unit 30, a supply mechanism 32, a control unit 10, and a memory unit 20.

[0029] The sensor unit 30 includes a sensor element 31 that can output detection data corresponding to the concentration of a target component. The sensor unit 30 may include one or more sensor elements 31. The sensor unit 30 may also include multiple sensor elements 31 that can detect different target components (i.e., have different specificities).

[0030] The sensor unit 30 may output first detection data (first time-series data) corresponding to the detection results during the period from when the supply of the standard gas to the sensor unit 30 starts to when the discharge of the standard gas from the sensor unit 30 ends. The sensor unit 30 may also output second detection data (second time-series data) corresponding to the detection results during the period from when the supply of the target gas to the sensor unit 30 starts to when the discharge of the target gas from the sensor unit 30 ends.

[0031] The supply mechanism 32 supplies the target gas or standard gas to the sensor unit 30. After supplying the target gas or standard gas to the sensor unit 30, the supply mechanism 32 exhausts the target gas or standard gas from the sensor unit 30. For example, after supplying the target gas or standard gas to the sensor unit 30, the supply mechanism 32 supplies an exhaust gas to the sensor unit 30 for exhausting the target gas or standard gas from the sensor unit 30. For example, by alternately supplying the target gas and the exhaust gas to the sensor unit 30 multiple times, it is possible to obtain multiple detection data for the target gas. The supply mechanism 32 may include a flow path unit 321, a pump unit 322, and a valve unit 323.

[0032] The flow path portion 321 is part of a path through which the gas to be supplied to the sensor portion 30 passes. The flow path portion 321 may be configured by one or more tubular members having a space inside. The valve portion 323 may be one or more valves provided in the flow path portion 321 that switch the destination of the gas. The valve portion 323 may be an electromagnetic valve that operates in response to a control signal from the control unit 10. The flow path portion 321 and the valve portion 323 may be airtightly connected.

[0033] The pump unit 322 may include a first pump that collects gas at the installation location Q, a second pump that supplies gas to the sensor unit 30, and a third pump that exhausts gas from the sensor unit 30. In the pump unit 322, the first pump, the second pump, and the third pump may each be individual pumps. Alternatively, the pump unit 322 may include a pump that has the functions of two or more of the first pump, the second pump, and the third pump. The pump unit 322 may be capable of supplying gas to the sensor unit 30 at a predetermined flow rate in response to a control signal from the control unit 10. The pump unit 322 may be capable of supplying gas to the sensor unit 30 at a predetermined flow rate in response to a control signal from the control unit 10. At the predetermined flow rate, the flow path unit 321 and the pump unit 322 may be airtightly connected.

[0034] The control unit 10 may be, for example, a CPU (Central Processing Unit). The control unit 10 reads a control program, which is software stored in the storage unit 20, expands it in a memory such as a RAM (Random Access Memory), and controls each component of the detection device 1. For simplicity of explanation, the control program is not shown in the storage unit 20 shown in FIG. 3. The control unit 10 may include a mechanism control unit 101, a data acquisition unit 102, an identification unit 103, an abnormality determination unit 104, and a communication control unit 105.

[0035] The mechanism control unit 101 controls the operation of the pump unit 322 and the valve unit 323 of the supply mechanism 32. For example, the mechanism control unit 101 may be able to control the on / off of the pump unit 322. For example, the mechanism control unit 101 may be able to switch the destination of the gas by controlling the valve unit 323. For example, the mechanism control unit 101 may control the operation of the sensor element 31.

[0036] The data acquiring unit 102 acquires detection data from the sensor unit 30. The data acquiring unit 102 acquires first detection data output when the sensor unit 30 detects a target component contained in a standard gas at a first time point when the detection device 1 is operating normally (i.e., in a first state). Here, the first time point may be any time point at which the detection device 1 is confirmed to be operating normally, and may be, for example, a time point before shipment. The data acquiring unit 102 also acquires second detection data output when the detection device 1 detects a target component contained in a target gas at a second time point after the first time point. Here, the second time point may be, for example, any time point after the detection device 1 is installed at the installation location Q.

[0037] The identification unit 103 inputs the second detection data into the identification model 201 that has undergone machine learning of the first detection data, and outputs difference information regarding the difference between the first detection data and the second detection data. The identification unit 103 may input the second detection data into the identification model 201 and output the difference information every time the data acquisition unit 102 acquires the second detection data. Alternatively, the identification unit 103 may input the second detection data into the identification model 201 and output the difference information every time the detection device 1 detects a target component contained in the user's target gas. Alternatively, the identification unit 103 may input the second detection data into the identification model 201 and output the difference information only when the data acquisition unit 102 acquires the second detection data at any timing.

[0038] The discriminant model 201 may be a neural network that has been machine-learned to output first output data that is obtained by restoring the input first detection data. More specifically, the discriminant model 201 may be an autoencoder. The discriminant model 201 is not limited to a neural network. Therefore, when learning the discriminant model 201, k-nearest neighbor, k-means clustering, a support vector machine, a decision tree, etc. may be used.

[0039] The following description will be given taking an example where the discriminative model 201 is an autoencoder. The configuration of the discriminative model 201 will be described using FIG. 4. FIG. 4 is a diagram illustrating an example of the configuration of the discriminative model 201. The discriminative model 201 may be a neural network having an input layer L1, an intermediate layer L2, and an output layer L3. When detection data x1 to xt acquired by the data acquisition unit 102 are input to the input layer L1, output data y1 to yt are output from the output layer L3. Here, the detection data x1 to xt may be detection data acquired from each of the sensor elements 31 included in the detection device 1. The output data y1 to yt are data reconstructed by the discriminative model 201 from the detection data x1 to xt. While FIG. 4 shows an example with three intermediate layers, the number of intermediate layers is not limited to this.

[0040] When the discriminative model 201 is an autoencoder, the discriminative model 201 can be generated by training using the same training data for the input layer L1 and the output layer L3. For example, first detection data is used as the training data. When the trained discriminative model 201 receives first detection data as input to the input layer L1, it can output output data from the output layer L3 that reconstructs the first detection data (i.e., that matches the first detection data highly). When detection data acquired from the sensor unit 30 when the detection device 1 is not operating normally is input to the input layer L1 of such a discriminative model 201, the detection data cannot be accurately reconstructed, resulting in a low degree of match between the input detection data and the output data. For example, when the discriminative model 201 is a k-nearest neighbor method, the discriminative model 201 can be generated by training using training data that includes the first detection data and information indicating whether the first detection data is normal.

[0041] The identification unit 103 outputs discrepancy information based on a comparison result obtained by comparing second output data output from the identification model 201 when second detection data is input to the identification model 201 with the second detection data input to the identification model 201. The discrepancy information may be information extracted based on the degree of agreement or difference between the second detection data input to the identification model 201 and the second output data output from the identification model 201. Because the identification model 201 has been trained to restore the first output data from the first detection data, the discrepancy information based on the comparison result obtained by comparing the second detection data with the second output data can be interpreted as information regarding the difference between the first detection data and the second detection data. For example, if the degree of agreement between the second detection data and the second output data is high, it can be determined that the difference between the first detection data and the second detection data is small and that the operation of the detection device 1 is normal. On the other hand, if the degree of agreement between the second detection data and the second output data is low, it can be determined that the difference between the first detection data and the second detection data is large and that the operation of the detection device 1 is abnormal.

[0042] First waveform data corresponding to the first detection data may be used as training data for generating the discriminative model 201. When the first waveform data is input to the input layer L1, the trained discriminative model 201 can output, as second output data, waveform data that is obtained by restoring the first waveform data from the output layer L3 and that closely matches the first waveform data. In this case, the discriminative unit 103 may output, as difference information, a diagram in which the first waveform data input to the discriminative model 201 and the second output data (waveform data) output from the discriminative model 201 are plotted in the same coordinate system.

[0043] <Difference Information> FIG. 5 illustrates a comparison between the output data output from the identification model 201 and the input data when second waveform data corresponding to the second detection data output from the sensor unit 30 of the detection device 1 operating normally is used as input data to the identification model 201. The second waveform data may include waveforms corresponding to the detection data detected in the first and second periods. In the example illustrated in FIG. 5 , the first period is the period from the start of supply of the target gas to the sensor unit 30 (T=0 in the figure) to the start of supply of the exhaust gas to the sensor unit 30 (T=t1 in the figure). In the example illustrated in FIG. 5 , the second period is the period from the start of supply of the exhaust gas to the sensor unit 30 to the start of the next supply of the target gas. When the target gas and the exhaust gas are alternately supplied to the sensor unit 30 multiple times, multiple detection data are acquired. In this case, one piece of difference information may be generated for each acquired detection data. When the detection device 1 is operating normally, the trained discrimination model 201 can output output data D2 that closely matches the input second waveform data D1. In the example shown in Fig. 5, the output data D2 output from the trained discrimination model 201 well restores the waveform data of the first period and the waveform data of the second period.

[0044] FIG. 6 illustrates a comparison between output data output from the identification model 201 and input data when second waveform data corresponding to second detection data output from the sensor unit 30 of the detection device 1 that is not operating normally is used as input data to the identification model 201. The second waveform data may include waveforms corresponding to detection data detected during a first period and a second period. In the example illustrated in FIG. 6 , the first period is the period from when the supply of target gas to the sensor unit 30 begins (T=0 in the figure) to when the supply of exhaust gas to the sensor unit 30 begins (T=t2 in the figure). In the example illustrated in FIG. 6 , the second period is the period after the exhaust gas is supplied to the sensor unit 30. When second waveform data D3 from when the detection device 1 is not operating normally is input, the trained identification model 201 outputs output data D4 that has a low degree of agreement with the second waveform data D3. Therefore, the discrepancy information output from the identification unit 103 can be information indicating whether the detection device 1 is operating normally. 6 is an example of waveform data corresponding to detection data output when the detection sensitivity of the sensor element 31 provided in the sensor unit 30 of the detection device 1 is reduced compared to the detection sensitivity at the first time point (i.e., during normal operation). In this case, as shown in FIG. 6, the output data D4 output from the trained identification model 201 cannot restore the waveform data of the first period.

[0045] 3 , the abnormality determination unit 104 outputs a determination result of the state (second state) of the detection device 1 at the second time point based on the magnitude of the difference indicated by the difference information. The abnormality determination unit 104 may determine whether or not an abnormality has occurred in either the sensor unit 30 or the supply mechanism 32 based on the difference information. The abnormality determination process in which the abnormality determination unit 104 determines whether or not the second state of the detection device 1 is abnormal will be described later.

[0046] If the detection device 1 is not operating normally, it is possible that there is an abnormality in the detection sensitivity of the sensor element 31 (for example, a decrease in sensitivity) as well as an abnormality in the supply mechanism 32. Abnormalities in the supply mechanism 32 include, for example, the following: - An abnormality in the flow path portion 321 (for example, leakage due to poor connection or damage) - An abnormality in the pump portion 322 (for example, leakage, malfunction) - An abnormality in the operation of the valve portion 323 (for example, poor switching operation, leakage).

[0047] FIG. 7 illustrates the characteristics of waveform data corresponding to the detection data output from the sensor unit 30 when a leak (e.g., a poor connection with the second pump) occurs in the pump unit 322. When this abnormality occurs, the target gas is not stably supplied to the sensor unit 30. Therefore, the detection value detected gradually decreases from approximately 20 seconds after the start of the supply of the target gas to the sensor unit 30 (T=0 in the figure) to the start of the supply of exhaust gas to the sensor unit 30 (T=t3 in the figure). Thus, the second waveform data D6 corresponding to the second detection data is clearly different from the waveform data D5 in a normal state. However, because the amount of exhaust gas supplied to the sensor unit 30 is sufficient, the presence of a leak in the pump unit 322 does not significantly affect the waveform data corresponding to the detection data in the second period.

[0048] FIG. 8 shows a comparison between the output data output from the discrimination model 201 when the second waveform data is used as input data to the discrimination model 201 and the input data. The second waveform data corresponds to the second detection data output from the sensor unit 30 when a leak occurs in the pump unit 322. When second waveform data D6a output from the sensor unit 30 when a leak occurs in the pump unit 322 is input, the trained discrimination model 201 outputs output data D6b, which has a low degree of agreement with the second waveform data D3. In the second waveform data D6a, the detection value gradually decreases from approximately 50 seconds after the start of supplying the target gas to the sensor unit 30 to the start of supplying the exhaust gas to the sensor unit 30 (T=t3 in the figure), but this decrease is not observed in the output data D6b. As shown in FIG. 8, the output data D6b output from the trained discrimination model 201 fails to restore the waveform of the input data during the first period. When a leak occurs in the pump unit 322, the difference between the input data and the output data is large in the first period and small in the second period.

[0049] FIG. 9 is a diagram illustrating the characteristics of waveform data corresponding to detection data output from the sensor unit 30 when an abnormality (e.g., a switching operation failure) occurs in the valve unit 323. When this abnormality occurs, the target gas is hardly supplied to the sensor unit 30, resulting in an unstable state. Therefore, the second waveform data D8 corresponding to the detection data detected during the first and second periods is significantly distorted, unlike the waveform data D7 during normal operation. In the example shown in FIG. 9, the first period is the period from the start of supply of the target gas to the sensor unit 30 (T=0 in the figure) to the start of supply of exhaust gas to the sensor unit 30 (T=t4 in the figure). In the example shown in FIG. 9, the second period is the period after the exhaust gas is supplied to the sensor unit 30.

[0050] FIG. 10 shows a comparative example of the output data output from the discrimination model 201 when the second waveform data is used as input data to the discrimination model 201, and the input data. The second waveform data corresponds to the second detection data output from the sensor unit 30 when an abnormality occurs in the valve unit 323. When second waveform data D8a output from the sensor unit 30 when an abnormality occurs in the valve unit 323 is input, the trained discrimination model 201 outputs output data D8b that has a low degree of agreement with the second waveform data D8a. As shown in FIG. 10, the output data D8b output from the trained discrimination model 201 fails to restore the waveform of the input data in the first and second periods. When an abnormality occurs in the valve unit 323, the magnitude of the difference between the input data and the output data is large in both the first and second periods.

[0051] As such, the detection data output from the sensor unit 30 often reflects the location of an abnormality in the detection device 1 (see FIGS. 7 and 9 ). When input data corresponding to detection data acquired when the detection device 1 is normal is input to the identification model 201, the identification model 201 can output output data that is a restored version of the input data. However, when input data corresponding to detection data acquired when the detection device 1 is abnormal is input, the identification model 201 outputs output data that has a low degree of match with the input data (i.e., cannot be restored). Furthermore, the location where the output data differs from the input data (i.e., the identification model 201 cannot restore) reflects the location of an abnormality in the detection device 1. Therefore, it is possible to determine an abnormality occurring in the detection device 1 based on the difference information output by the identification unit 103.

[0052] Here, the identification unit 103 may be configured to output difference information including first difference information corresponding to the first period and second difference information for the second period. Alternatively, the identification unit 103 may be configured to output partial difference information for each of a plurality of partial periods included in the first period and a plurality of partial periods included in the second period as difference information. With this configuration, the abnormality determination unit 104 can determine whether an abnormality has occurred in either the sensor unit 30 or the supply mechanism 32 of the detection device 1 based on the difference information output by the identification unit 103. Furthermore, with this configuration, the abnormality determination unit 104 can accurately identify the location of the abnormality occurring in the detection device 1 based on the difference information output by the identification unit 103.

[0053] When the concentration of the target component contained in the target gas fluctuates, it may be inaccurate to determine that the detection device 1 is not operating normally based on the magnitude of the difference indicated by one piece of discrepancy information. Therefore, the abnormality determination unit 104 may determine whether the detection device 1 is operating normally based on a result of aggregating the magnitudes of the differences indicated by the multiple pieces of discrepancy information output from the identification unit 103. This improves the accuracy of determining whether the detection device 1 is operating normally.

[0054] <Abnormality Determination Process> The abnormality determination unit 104 may use a cumulative difference value obtained by aggregating the magnitude of the difference indicated by each of the plurality of discrepancy information related to the target gas to determine whether the operation of the detection device 1 is normal. The cumulative difference value may be the sum of difference values ​​calculated for each partial period as the difference between the input data and the output data in the discrepancy information.

[0055] The abnormality determination unit 104 may determine that an abnormality has occurred in the detection device 1 at the second time point by comparing a predetermined threshold value with the frequency distribution of the cumulative difference values ​​at the second time point. Here, the predetermined threshold value is determined in advance (for example, at the first time point) based on the result of analyzing the correlation between the accuracy of the estimation result output from the estimation device 5 and the magnitude of the cumulative difference value. By employing the threshold value determined in this manner, the abnormality determination unit 104 can determine whether an abnormality has occurred in the detection device 1 that will significantly affect the accuracy of the estimation result output from the estimation device 5.

[0056] Alternatively, the abnormality determination unit 104 may calculate a cumulative difference value from each of the multiple pieces of difference information and determine whether the operation of the detection device 1 is normal or not by using a frequency distribution of the calculated multiple cumulative difference values. In this case, the abnormality determination unit 104 determines whether the operation of the detection device 1 is normal or not at the second time point based on a comparison result obtained by comparing the frequency distribution (approximately close to a normal distribution) of the cumulative difference values ​​at the first time point with the frequency distribution of the cumulative difference values ​​at the second time point. For example, the abnormality determination unit 104 may determine that an abnormality has occurred in the detection device 1 at the second time point when at least either (1) or (2) below is met:

[0057] (1) The minimum value (left end) of the frequency distribution of the cumulative difference values ​​at the second time point is greater than the average value of the frequency distribution of the cumulative difference values ​​at the first time point.

[0058] (2) The difference between the average value of the frequency distribution of the cumulative difference values ​​at the first time point and the average value of the frequency distribution of the cumulative difference values ​​at the second time point is greater than a predetermined multiple (e.g., 1 or 2) of the standard deviation of the frequency distribution of the cumulative difference values ​​at the first time point.

[0059] Alternatively, the anomaly determination unit 104 may use the difference between a value obtained by numerically integrating the input data in the difference information (for example, trapezoidal integration is applicable) and a value obtained by numerically integrating the output data. Applying numerical integration can improve the accuracy of determination by the anomaly determination unit 104. Furthermore, numerical integration can be applied even when the time-varying pattern corresponding to the input data differs from the time-varying pattern of the output data.

[0060] The abnormality determination unit 104 first determines whether or not an abnormality has occurred in the detection device 1 based on the difference information. If it determines that an abnormality has occurred, the abnormality determination unit 104 may be configured to subsequently identify the partial period that formed the basis of the determination result. The abnormality determination unit 104 may also identify the partial period that formed the basis of the determination result based on the second detection data or the second output data. The abnormality determination unit 104 may also identify the partial period that formed the basis of the determination result based on the waveform shape of the second waveform data.

[0061] The abnormality determination unit 104 may output, as the determination result, multiple candidates for the state of the detection device 1 in addition to the determination result regarding whether or not an abnormality has occurred in the detection device 1. For example, the abnormality determination unit 104 may output abnormality location information indicating multiple candidates for the location of the abnormality occurring in the detection device 1. The abnormality location information may be information that ranks the locations where the abnormality is most likely to occur. Furthermore, the abnormality location information may include, in addition to the candidates for the location of the abnormality occurring in the detection device 1, indications regarding changes in the user's health condition and lifestyle habits. In this case, the abnormality determination unit 104 may output abnormality location information such as the examples shown in (A) to (C) below. (A): (First candidate) Abnormality of the sensor element 31, (Second candidate) Change in lifestyle habits. (B): (First candidate) Abnormality of the sensor element 31, (Second candidate) Abnormality of the flow path unit 321, (Third candidate) Change in health condition. (C): (first candidate) malfunction of the valve section 323, (second candidate) malfunction of the pump section 322, (third candidate) none.

[0062] The communication control unit 105 transmits the determination result by the abnormality determination unit 104 to the electronic device 9. The communication control unit 105 may transmit the determination result by the abnormality determination unit 104, as well as discrepancy information on which the determination result is based, to the electronic device 9. The electronic device 9 may display the determination result output from the abnormality determination unit 104 on a display screen.

[0063] (Processing Performed by the Detection Apparatus 1) The flow of processing performed by the detection apparatus 1 will be described with reference to Fig. 11. Fig. 11 is a flowchart showing an example of the flow of processing performed by the detection apparatus 1.

[0064] First, the identification unit 103 inputs second detection data output when the detection device detects a target component contained in a target gas at a second time point that is after the first time point to the identification model 201 (step S1: input step). The identification model 201 may be a neural network that has been machine-learned using the first detection data output when the sensor unit of the detection device 1 detects a target component contained in a standard gas at the first time point when the detection device 1 is operating normally.

[0065] Next, the identifying unit 103 outputs difference information regarding the difference between the first detection data and the second detection data (step S2: identifying step).

[0066] Next, the abnormality determination unit 104 determines the state of the detection device at the second time point based on the magnitude of the difference indicated by the difference information (step S3: determination step).

[0067] According to this configuration, the detection device 1 can determine whether or not the device itself is operating normally without carrying out an inspection using a standard gas.

[0068] [Embodiment 2] Another embodiment of the present disclosure will be described below. For convenience of explanation, the same reference numerals will be used to designate components having the same functions as those described in the above embodiment, and the description thereof will not be repeated.

[0069] The abnormality determination unit 104 may be configured to determine the time point at which an abnormality began to occur in the state of the detection device 1, based on the plurality of pieces of difference information output from the identification unit 103 at multiple different time points. In other words, the abnormality determination unit 104 may be configured to identify the time point at which an abnormality began to occur in the state of the detection device 1, based on the plurality of pieces of difference information output from the identification unit 103 at multiple different time points.

[0070] Here, the discrimination unit 103 inputs each of the plurality of second detection data into the discrimination model 201 and generates difference information I Ｔ１ , I Ｔ２ , and I Ｔ３Take the case where the second detection data is output as an example. The second detection data is data acquired by the data acquisition unit 102 at time T1, time T2 after time T1, and time T3 after time T2. Time points T1, T2, and T3 may be second detection data acquired every time a predetermined time (e.g., 10 seconds) has elapsed. Alternatively, time points T1, T2, and T3 may be second detection data acquired every time the detection device 1 detects a target component contained in a target gas of a user.

[0071] The abnormality determination unit 104 receives the difference information I Ｔ１ , I Ｔ２ , and I Ｔ３ If there is a time when the magnitude of the difference indicated by each of the difference information I exceeds a predetermined threshold H, it is possible to identify that time as the time when an abnormality occurred in the state of the detection device 1. For example, Ｔ１ and I Ｔ２ The magnitude of the difference indicated by is smaller than the threshold H, and the difference information I Ｔ３ is equal to or greater than the threshold value H, the abnormality determination unit 104 determines that the time when the abnormality occurred in the detection device 1 is the period between time points T2 and T3. Here, the threshold value H may be a value that can be set in advance based on the difference information that is output when typical second detection data acquired when an abnormality occurs in the state of the detection device 1 is input to the identification model 201.

[0072] In this way, the abnormality determination unit 104 can identify the period during which the accuracy of the estimation result output from the estimation device 5 decreased by identifying the time when the abnormality occurred in the detection device 1. Furthermore, by analyzing what events (e.g., earthquakes, power outages, etc.) occurred during the period identified as the time when the abnormality occurred in the detection device 1, it is also possible to identify the event that caused the abnormality that occurred in the detection device 1.

[0073] [Embodiment 3] Another embodiment of the present disclosure will be described below. For convenience of explanation, the same reference numerals will be used to designate components having the same functions as those described in the above embodiment, and the description thereof will not be repeated.

[0074] In the above description, when the identification model 201 is an autoencoder, the first detection data output by the sensor unit 30 at a first time point when the detection device 1 is operating normally is used for machine learning of the identification model 201. However, the present invention is not limited to this configuration. For example, the first detection data output by the sensor unit 30 at a first time point when the detection device 1 is not operating normally may be used for machine learning of the identification model 201. In other words, the first detection data may be data output from the sensor unit 30 of a detection device 1 that is operating normally, or may be data output from the sensor unit 30 of a detection device 1 that is not operating normally.

[0075] In this case, if input data corresponding to detection data acquired when the detection device 1 is not normal is input to the trained discrimination model 201, the trained discrimination model 201 can output output data that is a restored version of the input data. On the other hand, if input data corresponding to detection data acquired when the detection device 1 is normal is input, the trained discrimination model 201 outputs output data that has a low match rate with the input data (i.e., cannot be restored). The discrimination unit 103 inputs the second detection data to the discrimination model 201 that has undergone machine learning of the first detection data, and outputs difference information regarding the difference between the first detection data and the second detection data.

[0076] The abnormality determination unit 104 can determine a second state, which is the state of the detection device 1 at the second time point, based on the difference information output by the identification unit 103. For example, if the matching rate between the second detection data and the second output data is high, the difference between the first detection data and the second detection data is small, and therefore the abnormality determination unit 104 determines that the state of the detection device 1 at the second time point is abnormal. On the other hand, if the matching rate between the second detection data and the second output data is low, the difference between the first detection data and the second detection data is large, and therefore the abnormality determination unit 104 can determine that the operation of the detection device 1 at the second time point is not abnormal (normal).

[0077] [Embodiment 4] In the first embodiment, when the identification model 201 is an autoencoder, the abnormality determination unit 104 determines whether the operation of the detection device 1 at the second time point is normal by comparing a predetermined threshold value with the frequency distribution of the cumulative difference value at the second time point, but this is not limited to this.

[0078] 12 shows the median of the cumulative difference values ​​calculated from the difference information output by the discrimination unit 103 and the value (R ２ 12 is a graph showing the relationship between the median of the cumulative difference values ​​and the value (R ２ ) there is a correlation between

[0079] Therefore, the abnormality determination unit 104 may determine whether the estimation accuracy of the concentration of the target component in the detection device 1 is normal, using a cumulative difference value calculated based on the difference information acquired from the identification unit 103. In this case, the storage unit 20 may store a trained estimation model that has been machine-learned using a clustering method. The clustering method used to train the trained estimation model may be, for example, a random forest or a support vector machine.

[0080] Below, a method for training a trained estimation model and a method for determining estimation accuracy using the trained estimation model are described. Hereinafter, abnormal estimation accuracy of the concentration estimation of a target component in the detection device 1 will be expressed as "abnormal estimation accuracy" or "degraded estimation accuracy." Also, below, a sensor element 31 that is unable to normally output a signal corresponding to the concentration of a target component due to reduced detection sensitivity or the like will be expressed as a "degraded sensor element 31" or an "abnormal sensor element 31." First, second detection data (normal data) output from the sensor unit 30 of the detection device 1 operating normally and second detection data (degraded data) output from the sensor unit 30 with a degraded sensor element 31 are acquired. If the sensor unit 30 includes multiple types of sensor elements 31, multiple pieces of degradation data may be generated that differ from each other in at least one of the type of degraded sensor element 31 and the degree of degradation.

[0081] The method of degrading the sensor element 31 may be, for example, a method of exposing the sensor element 31 to a gas containing a target component in an amount that the sensor element 31 is expected to be exposed to over a predetermined number of years. Alternatively, for example, data generated from normal data may be used as degradation data. For example, the degradation data may be obtained by multiplying the normal data by a constant. For example, data obtained by multiplying the normal data by 0.9 and data obtained by multiplying the normal data by 0.8 may be used as degradation data. Alternatively, data output at the time when a sensor element of the same type as the sensor element 31 has deteriorated may be used as degradation data.

[0082] Next, the normal data and the deteriorated data are input to an estimation model used for concentration estimation in the detection device 1, and the concentration of the target component is estimated. Based on the estimation result, a value indicating the estimation accuracy of the concentration estimation of the target component (R ２ ) is calculated. ２ Based on the value of R, it is evaluated whether the accuracy of concentration estimation in the detection device 1 has deteriorated. ２ For the second detection data for which a result of ≧0.9 was obtained, the estimation accuracy was evaluated as normal, and R ２ It may be considered that the estimation accuracy has deteriorated for the second detection data for which a result of <0.9 was obtained.

[0083] A learning dataset is created by using the median values ​​of the second detection data, including the normal data and the degraded data, as input data for learning, and the evaluation results of the estimation accuracy for each second detection data as output data for learning. Using the dataset created as described above, machine learning is performed using a random forest or support vector machine technique to create a trained estimation model for evaluating the estimation accuracy.

[0084] The abnormality determination unit 104 may use the trained estimation model to determine whether the estimation accuracy of the target component in the detection device 1 is normal. Specifically, when the abnormality determination unit 104 acquires multiple pieces of difference information from the identification unit 103, it calculates a cumulative difference value from each of the multiple pieces of difference information. Next, the abnormality determination unit 104 further calculates a median of the calculated cumulative difference values. The abnormality determination unit 104 inputs the calculated median of the cumulative difference values ​​into the trained estimation model and acquires a determination result indicating whether the estimation accuracy of the target component in the detection device 1 is normal or has deteriorated.

[0085] [Embodiment 5] Fig. 13 is a block diagram showing an example of the configuration of a detection device 1A according to embodiment 5 of the present disclosure. As shown in Fig. 13, the detection device 1A differs from the detection device 1 according to embodiment 1 in that it includes a sensor unit 30A instead of the sensor unit 30, a control unit 10A instead of the control unit 10, and a storage unit 20A instead of the storage unit 20. The detection device 1A according to this embodiment may be applied to the estimation system 100 shown in Fig. 1. The estimation system 100 shown in Fig. 1 may include the detection device 1A instead of or in addition to the detection device 1.

[0086] The detection device 1A according to this embodiment uses a plurality of sensor elements 31 having detection sensitivity for the target component in order to estimate the concentration of the target component. In the following description, a sensor element 31 in a state where no abnormality such as a decrease in detection sensitivity occurs, or in a state where an abnormality occurs but not to the extent that it affects the estimation accuracy of the detection device 1A, will be referred to as a "normal sensor element 31." On the other hand, a sensor element 31 in which an abnormality has occurred to the extent that it affects the estimation accuracy of the detection device 1A will be referred to as a "degraded sensor element 31."

[0087] 14 and 15 are diagrams showing examples of graphs plotting output values ​​of three sensor elements 31 used in the detection device 1A. FIG. 14 shows a graph in which all three sensor elements 31 are normal, while FIG. 15 shows a graph in which one of the three sensor elements 31 is degraded. As shown in FIG. 14 , when multiple sensor elements 31 capable of detecting the same target component are used for concentration estimation, the ratio of output values ​​among the multiple sensor elements 31 is approximately constant. If all multiple sensor elements 31 are able to detect the target component normally, the ratio does not change significantly even if the shape (waveform) of the graph when the output values ​​are plotted changes due to a change in the concentration of the target component. In this case, the detection device 1A can accurately estimate the concentration of the target component based on the output values ​​from each of the multiple sensor elements 31, and it can be said that the detection device 1A is operating normally in this state.

[0088] On the other hand, if at least one of the multiple sensor elements 31 deteriorates, more specifically, if an abnormality such as a decrease in detection sensitivity occurs in at least one of the multiple sensor elements 31, the accuracy of estimating the concentration of the target component may decrease. As shown in Fig. 15 , when one sensor element 31 deteriorates, the value of the output signal output from that sensor element 31 decreases. If the sensor element 31 deteriorates, estimation will be performed using the output value of the deteriorated sensor element 31, and the accuracy of estimating the concentration of the target component will decrease. In such a case, it can be said that the detection device 1A is not operating normally.

[0089] As described above, when at least one of the multiple sensor elements 31 deteriorates, the output value of that sensor element 31 decreases. This may result in a decrease in the estimation accuracy of the target component in the detection device 1A. Here, as shown in FIG. 15 , when a sensor element 31 deteriorates, the waveform obtained from the output signal of that sensor element 31 becomes distorted, or the ratio of the output value of that sensor element 31 to the output values ​​of the other sensor elements 31 becomes significantly different from that in normal conditions. Therefore, in the detection device 1A according to this embodiment, an estimation model that has learned the relationship between the output values ​​of the multiple sensor elements 31 is used to determine whether the estimation accuracy of the target component in the detection device 1A is normal.

[0090] The detection device 1A will be described in detail below. As shown in FIG. 13 , the sensor unit 30A includes a plurality of sensor elements 31. The sensor elements 31 are capable of detecting a first gas component and a second gas component contained in the target gas. The sensor elements 31 are also capable of outputting output signals corresponding to the concentrations of the first gas component and the second gas component. When the target gas is supplied to the sensor unit 30A, the sensor unit 30A outputs, to the control unit 10A, each of the output signals output by the plurality of sensor elements 31 corresponding to the concentrations of the first gas component and the second gas component contained in the target gas.

[0091] The first gas component and the second gas component may be hydrogen sulfide or MMC (Methyl Mercaptan), respectively. For example, the sensor element 31 detects H ２ S and MMC as a second gas component are detected, and H ２ For example, the sensor element 31 may be the same type of sensor as the sensor element 31 according to the first embodiment.

[0092] The multiple sensor elements 31 differ from one another in at least one of their detection sensitivities to the first gas component and the second gas component. Furthermore, the multiple sensor elements 31 may differ from one another in their detection methods. For example, each of the multiple sensor elements 31 may be an electrochemical sensor, a semiconductor sensor, or a combustion sensor. The following describes, as an example, a configuration in which the sensor unit 30A may include three types of sensor elements 31 (sensor element 31A, sensor element 31B, and sensor element 31C). However, the number of sensor elements 31 included in the sensor unit 30A is not limited to this. For example, the sensor unit 30A may include two sensor elements 31, or may include four or more sensor elements 31.

[0093] The storage unit 20A stores a first estimation model 202 and a second estimation model 203 (estimation models). The second estimation model 203 is a trained estimation model used to determine whether the accuracy of the estimation of the concentrations of the first gas component and the second gas component by the estimation unit 106 is normal. Details of the second estimation model 203 will be described later.

[0094] The first estimation model 202 is a trained estimation model used to estimate the concentrations of a first gas component and a second gas component. The first estimation model 202 is a trained estimation model that uses output signals output by each of a plurality of sensor elements 31 supplied with a target gas as input data and the concentrations of the first gas component and the second gas component contained in the target gas as output data. The first estimation model 202 may be trained by machine learning using a dataset including the input data and the output data. In training the first estimation model 202, output signals output by each of the plurality of sensor elements 31 supplied with a target gas containing known concentrations of the first gas component and the second gas component may be used as input data for training. Furthermore, in training the first estimation model 202, the concentrations of the first gas component and the second gas component contained in the target gas may be used as output data for training.

[0095] As shown in FIG. 13 , the control unit 10A includes a mechanism control unit 101, a data acquisition unit 102, a communication control unit 105, an estimation unit 106, and a determination unit 107. The estimation unit 106 estimates the concentrations of a first gas component and a second gas component contained in the target gas based on an output signal from the sensor unit 30A. When the target gas is supplied to the sensor unit 30A, the estimation unit 106 acquires detection data output from the sensor unit 30A. The estimation unit 106 may acquire output signals output from each of the multiple sensor elements 31 as the detection data. The estimation unit 106 inputs the detection data to a first estimation model 202 and estimates the concentrations of the first gas component and the second gas component contained in the target gas. The estimation unit 106 may transmit concentration information indicating the estimated concentrations of the first gas component and the second gas component to the estimation device 5 via the electronic device 9.

[0096] The determination unit 107 determines whether the accuracy of the estimation of the concentrations of the first gas component and the second gas component by the estimation unit 106 is normal or not, using the second estimation model 203. Details of the determination by the determination unit 107 will be described later.

[0097] A learning method for the second estimation model 203 will be described below. Feature information is used for learning the second estimation model 203. The feature information includes at least one of the following pieces of information (1) to (3). If feature information including at least one of the pieces of information (1) to (3) is used for learning, it is possible to create a second estimation model 203 that can estimate whether the estimation accuracy of the estimation unit 106 is normal. The feature information may include one piece of information from (1) to (3), or may include two or more pieces of information. When the feature information includes a combination of two or more pieces of information from (1) to (3), the estimation accuracy is improved compared to when the feature information includes only one of the pieces of information from (1) to (3). (1) Information created based on a normal signal, which is an output signal output by a sensor unit 30A in which all sensor elements 31 included in the plurality of sensor elements 31 are operating normally. (2) Information created based on an abnormal signal, which is an output signal output by a sensor unit 30A including at least one sensor element 31 that is not operating normally. (3) Information created based on a normal signal, simulating an abnormal signal.

[0098] First, verification data including first data and second data is generated using a gas containing a target component whose concentration is known and a detection device 1A equipped with a plurality of sensor elements 31. The first data is data including a normal signal, which is an output signal obtained from a sensor unit 30A equipped with a plurality of normal sensor elements 31. The first data may be obtained by supplying the gas to a detection device 1A in which all of the sensor elements 31 are normal.

[0099] The second data is data including an abnormal signal, which is an output signal obtained from a sensor unit 30A including at least one deteriorated sensor element 31. The second data may be obtained by supplying the above-mentioned gas to a detection device 1A in which at least one sensor element 31 is deteriorated. Alternatively, data obtained by pseudo-deteriorating the value of a normal signal obtained from a sensor unit 30A in which all sensor elements 31 are normal may be used as the second data created to imitate an abnormal signal. For example, data obtained by multiplying at least a portion of the output values ​​obtained from multiple normal sensor elements 31 by a constant may be used as the second data.

[0100] As a specific example, 64 types of data obtained by multiplying the output signal values ​​of normal sensor elements 31A, 31B, and 31C by 1, 0.9, 0.8, and 0.7, respectively, may be used as the verification data. The multiplication factors used to create the verification data are not limited to those described above. Furthermore, the number of data patterns included in the verification data is not limited to those described above. The detection data may be created using output signals obtained from multiple sensor elements 31 included in the detection device 1A that is actually used, or output signals obtained from multiple other sensor elements 31 of the same type as the multiple sensor elements 31.

[0101] Next, each data included in the verification data is input to the first estimation model 202, and the concentrations of the first gas component and the second gas component are estimated. Then, the estimated concentrations are used to calculate the R ２ Calculate the coefficient of determination (R ２ It can be said that R is a value indicating the estimation accuracy of the concentration of the target component in the detection device 1. For example, R ２ If R is ≥ 0.9, it is determined that the estimation accuracy of the concentration of the target component in the detection device 1 is normal, and a label indicating "normal" is attached to the corresponding data. ２ If R is less than 0.9, it is determined that the estimation accuracy of the target component concentration estimation in the detection device 1 has deteriorated, and a label indicating "deterioration" is attached to the corresponding data. ２ The threshold value of R may be set appropriately. ２ ≧0.8 or R ２If R is ≧0.7, it may be determined that the estimation accuracy of the concentration of the target component in the detection device 1 is normal. ２ <0.8 or R ２ < 0.7, it may be determined that the estimation accuracy of the concentration of the target component in the detection device 1 has deteriorated. ２ However, the estimation accuracy is not limited to this. For example, the estimation accuracy of the target component concentration estimation may be determined using an estimation error, RMSE, MSE, AUC, MCC (Matthews correlation coefficient), or the like.

[0102] Feature information is created from the verification data created as described above. The feature information may be at least one of the peak values ​​of the output signals included in the verification data and the output ratio between the sensor elements 31 using the peak values. The output ratio value may be, for example, the output value of sensor element 31A / the output value of sensor element 31B, the output value of sensor element 31A / the output value of sensor element 31C, or the output value of sensor element 31B / the output value of sensor element 31C.

[0103] The second estimation model 203 is trained using the created feature information as explanatory variables for training and the result of the determination of the estimation accuracy based on the validation data as the objective variable. A label corresponding to each data item included in the validation data, i.e., information indicating whether the estimation accuracy of the target component concentration estimation by the detection device 1 is normal or deteriorated, is used as the objective variable for training. In this way, a dataset of explanatory variables and objective variables is created based on each data item included in the validation data, and the second estimation model is trained by machine learning using the created dataset.

[0104] The k-nearest neighbor method may be used to train the second estimation model 203. This identifies an optimal k value based on a training dataset, and determines parameters for performing estimation. For example, if the peak value of the output signal from the sensor element 31 when measuring MMC is used as an explanatory variable, the value of k may be 5. By performing machine learning in this manner, the second estimation model 203 is created, which uses at least one of the peak value of each sensor element 31 and the output ratio between the sensor elements 31 as an explanatory variable, and information indicating whether the estimation accuracy by the estimation unit 106 is normal as a target variable.

[0105] The determination unit 107 acquires data including an output signal from the sensor unit 30A. The time point at which the sensor unit 30A outputs an output signal is referred to as a time point of interest. Upon acquiring the output signal, the determination unit 107 creates feature information based on the output signal. Specifically, the determination unit 107 creates, as feature information, information indicating the peak value of the output signal of each sensor element 31 and a value indicating the output ratio between the sensor elements 31 based on the output signal. The determination unit 107 inputs at least any of the created feature information to the second estimation model 203. This allows the determination unit 107 to acquire information indicating the estimation accuracy of the estimation unit 106 as the estimation result by the second estimation model 203. The determination unit 107 may output the acquired information to the electronic device 9 as information indicating the determination result of the estimation accuracy of the estimation unit 106.

[0106] If the sensor elements 31 are not deteriorated, the estimation unit 106 can perform estimation normally using the first estimation model 202. In this case, the determination unit 107 outputs a determination result indicating that the estimation accuracy by the estimation unit 106 is normal. If any of the sensor elements 31 has deteriorated and its output value has decreased, the estimation accuracy by the estimation unit 106 will deteriorate. In this case, the determination unit 107 outputs a determination result indicating that the estimation accuracy by the estimation unit 106 has decreased.

[0107] As described in the first embodiment, the supply mechanism 32 supplies the target gas to the sensor unit 30A, and then supplies the sensor unit 30A with exhaust gas for exhausting the target gas from the sensor unit 30A. This causes the target gas to be exhausted from the sensor unit 30A. Here, the determination unit 107 may make a determination using an output signal output from the sensor unit 30A at the timing when the supply mechanism 32 switches the gas supplied to the sensor unit 30A from the target gas to the exhaust gas.

[0108] As described above, when the sensor elements 31 deteriorate, the estimation accuracy by the estimation unit 106 decreases. Furthermore, the second estimation model 203 is trained using the relationship between the output values ​​among the multiple sensor elements 31 and information indicating the estimation accuracy by the estimation unit 106. Therefore, when the second estimation model 203 outputs an estimation result indicating that the estimation accuracy by the estimation unit 106 has decreased, it is considered that the relationship between the output values ​​among the multiple sensor elements 31 is abnormal. In this case, it is considered that there is a high possibility that at least one of the multiple sensor elements 31 has deteriorated, causing an abnormality in detection accuracy.

[0109] Therefore, the determination unit 107 may output information indicating whether the state of the multiple sensor elements 31 included in the sensor unit 30A is normal or deteriorated, as information regarding the estimation accuracy of the estimation unit 106. Specifically, when the determination unit 107 determines that the estimation accuracy by the estimation unit 106 has deteriorated as a result of estimation using the second estimation model 203, the determination unit 107 may determine that at least one of the multiple sensor elements 31 has deteriorated. On the other hand, when the determination unit 107 determines that the estimation accuracy by the estimation unit 106 is normal as a result of estimation using the second estimation model 203, the determination unit 107 may determine that the multiple sensor elements 31 are normal.

[0110] The flow of processing performed by the detection device 1A will be described with reference to Fig. 16. Fig. 16 is a flowchart showing an example of the flow of processing performed by the detection device 1A.

[0111] First, the target gas is supplied to the sensor unit 30A by the operation of the supply mechanism 32. As a result, each of the sensor elements 31 included in the sensor unit 30A outputs an output signal corresponding to the concentrations of the first gas component and the second gas component contained in the target gas. The determination unit 107 acquires the output signal from each of the sensor elements 31.

[0112] When the determination unit 107 acquires an output signal, it inputs at least one of the peak values ​​of each acquired output signal and the ratio of the peak values ​​between multiple sensor elements 31 to the second estimation model 203 (S11: input step).

[0113] The second estimation model 203 performs estimation based on the input data and outputs information indicating whether the estimation accuracy by the estimation unit 106 is normal. The determination unit 107 determines whether the estimation accuracy by the estimation unit 106 is normal based on the information output by the second estimation model 203 (S12: determination step). The determination unit 107 outputs information indicating the determination result to the electronic device 9, etc. (S13: output step).

[0114] Sixth Embodiment Fig. 17 is a block diagram showing an example of the configuration of a detection device 1B according to a sixth embodiment of the present disclosure. Hereinafter, the detection device 1B according to the sixth embodiment of the present disclosure will be described with reference to Fig. 17. As shown in Fig. 17, the detection device 1B includes a control unit 10B and a storage unit 20B instead of the control unit 10 and the storage unit 20. The detection device 1B according to this embodiment may be applied to the estimation system 100 shown in Fig. 1. The estimation system 100 shown in Fig. 1 may include the detection device 1B instead of or in addition to the detection device 1.

[0115] When estimating the concentration of a target component using a target gas containing an unknown concentration of the target component, it is unclear whether the estimated concentration (estimated value) is a correct value (correct value). Therefore, it has been difficult to determine whether the accuracy of the concentration estimation is normal or has deteriorated using the estimated value. Here, the inventors have found that when the estimation accuracy deteriorates due to deterioration of the sensor element 31, the estimated value based on the output value of the sensor element 31 also tends to deteriorate.

[0116] 18 is a graph showing the relationship between the estimated concentration value and the estimation accuracy. In FIG. 18, the vertical axis shows the estimated value based on the output value from the sensor element 31 when the target gas containing the target component at a concentration of 0 ppm is supplied to the sensor element 31, and the horizontal axis shows the R calculated based on the output value. ２ When the sensor element 31 deteriorates, the value of R ２ 18, it was found that there is a correlation between deterioration of the sensor element 31 and a decrease in the concentration estimate based on the output value of the sensor element 31. It was also found that when a low concentration of a target component is detected by a deteriorated sensor element 31, the estimate based on the output signal of the sensor element 31 tends to be a negative value.

[0117] The inventors have discovered that it is possible to determine whether the sensor element 31 has deteriorated based on the tendency of correlation between the degree of deterioration in estimation accuracy and the degree of deterioration in the estimated value. For example, if an estimated value obtained when an estimation is performed using a target gas containing a low concentration of the target component is significantly different from an estimated value estimated when the sensor element 31 is normal, it can be considered that the estimation accuracy of the detection device 1B including the sensor element 31 has deteriorated. Furthermore, this method is performed using an estimated value obtained from the output value of the sensor element 31. Therefore, if it is determined based on this method that the estimation accuracy of the detection device 1B has deteriorated, it can be considered that the sensor element 31 included in the detection device 1B has deteriorated.

[0118] The detection device 1B will be described in detail below. The detection device 1B according to the present disclosure can determine whether the accuracy of the concentration estimation of the target component in the detection device 1B is normal based on an estimated value of a target gas containing a low concentration of the target component. Furthermore, the detection device 1B can also determine whether the sensor element 31 is normal or degraded (abnormal) based on the result of this determination. In this embodiment, the target gas containing a low concentration of the target component includes a target gas in which the concentration of the target component is 0 ppm. The detection device 1B performs a determination measurement at a predetermined timing. The determination measurement is a measurement for determining whether the sensor element 31 is normal or degraded. A target gas with a low concentration of the target component is used for the determination measurement. The detection device 1B may perform the determination measurement at a predetermined time, for example, once every 24 hours. When the detection device 1B is installed in a toilet room, the detection device 1B may acquire outside air outside the toilet room as the target gas and perform the determination measurement. The target component that can be detected by the sensor element 31 included in the detection device 1B may be, for example, hydrogen, hydrogen sulfide, MMC, or a mixed gas of hydrogen sulfide and MMC. The sensor element 31 may be an electrochemical sensor, a semiconductor sensor, or a combustion sensor.

[0119] 17, the storage unit 20B stores past data 204. The past data 204 includes estimated data indicating estimated values ​​that are estimated based on output values ​​of the sensor element 31 acquired during past measurements for determination. The past data 204 includes estimated data that are estimated during n past measurements for determination. The value of n will be described later.

[0120] 17 , the control unit 10B includes a mechanism control unit 101, a data acquisition unit 102, a communication control unit 105, an estimation unit 106, and a determination unit 107B. Upon receiving an estimated value indicating the concentration of the target component from the estimation unit 106, the determination unit 107B determines whether the sensor element 31 that detected the target component is normal or deteriorated. If the detection device 1B includes multiple sensor elements 31, the determination unit 107B may perform a determination for each combination of the type of target component and the type of sensor element 31.

[0121] The determination unit 107B determines the deterioration of the sensor element 31 in three stages. When the determination unit 107B receives information indicating the estimated value from the estimation unit 106, the determination unit 107B performs a first determination. In the first determination, the determination unit 107B determines whether the estimated value is a negative value. If the estimated value is a negative value, the determination unit 107B further determines whether the estimated value is equal to or less than a first threshold value. The first threshold value is a value that indicates that the sensor element 31 is deteriorated when the estimated value is equal to or less than the first threshold value. If the estimated value is a negative value and equal to or less than the first threshold value, the determination unit 107B determines that the sensor element 31 is deteriorated. Alternatively, if the estimated value is a negative value, the determination unit 107B may determine that the sensor element 31 is deteriorated without considering the first threshold value. Alternatively, the determination unit 107B may determine that the sensor element 31 is deteriorated when the estimated value is a negative value and the absolute value of the estimated value is equal to or less than the first threshold value occurs a predetermined number of times or more. The number of times may be the same as the number used in the third determination described below.

[0122] If the estimated value exceeds the first threshold, the determination unit 107B performs a second determination. In the second determination, the determination unit 107B determines whether the estimated value is equal to or less than a second threshold. The second threshold is a value at which it is considered that the sensor element 31 is likely to be deteriorated if the estimated value is equal to or less than the second threshold. The second threshold may be a value smaller than the first threshold. Furthermore, the second threshold may be set for each combination of the type of target component and the type of sensor element 31. If the estimated value exceeds the second threshold, it is likely that the sensor element 31 is normal. Therefore, if the estimated value exceeds the second threshold in the second determination, the determination unit 107B determines that the sensor element 31 is normal.

[0123] If the estimated value is equal to or less than the second threshold in the second determination, the determination unit 107B performs a third determination. In the third determination, the determination unit 107B performs the determination using the estimated value acquired this time and the estimated value included in the past data 204 stored in the storage unit 20B. As described above, the storage unit 20B stores past data 204 including n estimated values. The determination unit 107B newly stores the estimated value acquired this time in the storage unit 20B, and deletes the estimated data indicating the oldest estimated value included in the past data 204. As a result, the storage unit 20B is in a state where past data 204 including estimated data from n measurements is stored. The determination unit 107B uses the n estimated values ​​including the most recent estimated data stored in the storage unit 20B.

[0124] The determination unit 107B determines whether a pattern in which the output signal of the sensor element 31 is a negative value and the absolute value of the estimated value is equal to or less than a second threshold value, which is smaller than the first threshold value, occurs a predetermined number of times or more. Specifically, in the third determination, the determination unit 107B determines whether an estimated value equal to or less than the second threshold value has been estimated m times during n measurements. The determination unit 107B determines whether m or more pieces of estimated data indicating values ​​equal to or less than the second threshold value are included in the estimated data indicating n estimated values ​​stored in the storage unit 20B. If m or more pieces of estimated data indicating values ​​equal to or less than the second threshold value are included, the determination unit 107B determines that the sensor element 31 is degraded. On the other hand, if fewer than m pieces of estimated data indicating values ​​equal to or less than the second threshold value are included, the determination unit 107B determines that the sensor element 31 is normal.

[0125] A method for setting the first threshold and the second threshold will now be described. First, a target gas containing a target component at a known concentration is supplied to a normal sensor element 31. The output value of the sensor element 31 thus obtained is multiplied by a constant to create verification data in which the output value of the sensor element 31 is pseudo-degraded. Specifically, a plurality of pieces of estimated data obtained by multiplying the output value of the sensor element 31 by a constant may be used as the verification data. For example, estimated data may be created by multiplying the output value of the sensor element 31 by 1, 0.9, 0.8, and 0.7, and these pieces of estimated data may be used as the verification data.

[0126] Next, the estimation unit 106 estimates the concentration of the target component based on the verification data. ２ Calculate R ２ Among the estimated data having a value >0.9, the estimated data having the smallest estimated value is identified, and a value obtained by adding a safety margin of about 10% to the estimated value of the identified estimated data is set as the first threshold value.

[0127] As shown in FIG. ２ If ≦0.9, the estimate is R ２ >0.9, which tends to be lower than the lower bound of the estimate. ２ When the value of R is 0.9 or less, the estimation accuracy of the detection device 1B is insufficient. ２ 18. The dashed line in FIG. 18 indicates the first threshold value obtained by adding a value of 10% as a safety margin to the lower limit of the estimated value when R ２ 18 shows the value obtained by adding a 10% value as a safety margin to the lower limit of the estimated value when R ２ When the value is ≦0.9, the estimated value tends to be lower than the value indicated by the dashed line. Therefore, by setting this value as the first threshold, it is possible to determine whether the sensor element 31 is normal or not based on the first threshold and the estimated value.

[0128] The second threshold value used in the second determination will be described. First, in the same way as when setting the first threshold value, R ２ Calculate R ２ Root mean square error (RMSE) is calculated using the estimated data when the RMSE is >0.9, and a value equal to or less than the calculated RMSE is set as the second threshold.

[0129] Next, the values ​​of n and m used in the third determination will be described. In this embodiment, the values ​​of n, m, and the second threshold may be set for each combination of the type of target component and the type of sensor element 31. First, multiple measurements are performed using the sensor element 31. Then, based on the type of target component, the type of sensor element 31, and the second threshold, the combination of n and m that maximizes the TPR (True Positive Rate) and TNR (True Negative Rate) values ​​is calculated. The combination of n and m calculated in this way may be set as the values ​​of n and m used when making a determination based on the combination of the type of target component and the type of sensor element 31.

[0130] If the calculation results in a combination of a type of target component with low TPR and TNR values ​​and a type of sensor element 31, the determination result based on the estimated value for that combination may be used as reference information with low accuracy. For example, if a combination of a type of target component with a type of sensor element 31 with either a TPR or TNR value of 70% or less exists, the determination result based on the estimated value for that combination may be used as reference information.

[0131] The first threshold, the value of n, the value of m, and the second threshold set as described above may be stored in the storage unit 20B in association with information indicating a combination of the type of target component and the type of sensor element 31. The determination unit 107B may identify the first threshold, the second threshold, the value of n, and the value of m to be used for determination based on the combination of the type of target component and the type of sensor element 31.

[0132] Generally, because there are individual differences among the sensor elements 31 used for concentration estimation, even when multiple sensor elements 31 of the same type are used to detect the same concentration of a target component, the output value may vary from sensor element to sensor element. Furthermore, even when a single sensor element 31 is used to detect the same concentration of a target component, the output value may vary from measurement to measurement. For example, when a measurement is performed using a normal sensor element 31, the estimated value may be below the second threshold. Therefore, when only a single measurement result is referenced, the sensor element 31 may be determined to be deteriorated even if it is actually normal. Therefore, in the third determination, the sensor element 31 is determined based on multiple pieces of information indicating concentrations estimated based on the output values ​​of the sensor element 31 in n measurements. This improves the reliability of the determination result by the determination unit 107B.

[0133] The flow of processing performed by the detection device 1B will be described with reference to Fig. 19. Fig. 19 is a flowchart showing an example of the flow of processing performed by the detection device 1B.

[0134] First, the supply mechanism 32 operates to supply the target gas to the sensor unit 30. As a result, the sensor element 31 included in the sensor unit 30 outputs an output signal corresponding to the concentration of the target component contained in the target gas. The estimation unit 106 acquires the output signal. The estimation unit 106 also inputs the acquired output signal into the first estimation model 202 to estimate the concentration of the target component (S21: estimation step). The estimation unit 106 outputs an estimated value indicating the estimated concentration of the target component to the determination unit 107B.

[0135] When the determination unit 107B acquires information indicating the estimated value estimated by the estimation unit 106, the determination unit 107B refers to the storage unit 20B based on the combination of the sensor element 31 and the target component to acquire information indicating a first threshold value. Then, the determination unit 107B determines whether the acquired estimated value is a negative value and is equal to or less than the first threshold value (S22: determination step). If the determination result in S22 is YES, the determination unit 107B determines that the sensor element 31 (sensor) has deteriorated (S27: determination step).

[0136] If the result of S22 is NO, the determination unit 107B refers to the storage unit 20B based on the combination of the sensor element 31 and the target component, and acquires information indicating the second threshold value. Then, the determination unit 107B determines whether the estimated value is equal to or less than the second threshold value (S23: determination step).

[0137] If the answer is YES in S23, after S25, the determination unit 107B refers to the storage unit 20B and acquires the values ​​of n and m corresponding to the combination of the sensor element 31 and the target component. The determination unit 107B then erases the nth estimated data stored in the past data 204 of the storage unit 20B (S24). The determination unit 107B then stores the currently acquired estimated data in the storage unit 20B as new past data 204 (S25). The determination unit 107B then refers to the nth estimated data included in the past data 204 and determines whether the number of data items equal to or less than the second threshold is m or more (S26: determination step). If the answer is YES in S26, the determination unit 107B determines that the sensor element 31 has deteriorated.

[0138] If the answer is NO in S23, the determination unit 107B erases the n-th previous estimated data stored in the past data 204 of the storage unit 20B (S28).The determination unit 107B then stores the currently acquired estimated data in the storage unit 20B as new past data 204 (S29).After S29 or if the answer is NO in S26, the determination unit 107B determines that the sensor element 31 is normal (S30: determination step).

[0139] [Embodiment 7] Figure 20 is a block diagram showing the configuration of a detection device 1C according to embodiment 7 of the present disclosure. The detection device 1C according to this embodiment can determine whether the concentration estimation accuracy is normal or deteriorated by using multiple techniques. If the detection device 1C determines that the concentration estimation accuracy has deteriorated or that the sensor element has deteriorated using at least one of the multiple techniques, the detection device 1C may determine that an abnormality has occurred in the concentration estimation accuracy of the detection device 1C. Furthermore, if the detection device 1C determines that an abnormality has occurred in the concentration estimation accuracy of the detection device 1C, the detection device 1C may further determine the extent of the abnormality. For example, the detection device 1C may determine whether the abnormality is so severe that the estimation model for concentration estimation needs to be corrected, the sensor element 31 needs to be replaced, or the entire substrate storing the sensor element 31 and the estimation model needs to be replaced.

[0140] 20 , the detection device 1C includes a control unit 10C, a memory unit 20C, a sensor unit 30A, and a supply mechanism 32. The control unit 10C includes a mechanism control unit 101, a data acquisition unit 102, an identification unit 103, a communication control unit 105, an estimation unit 106, a first determination unit 108, and a second determination unit 109 (fourth determination unit). The memory unit 20C stores an identification model 201, a first estimation model 202, a second estimation model 203 (estimation model), and past data 204.

[0141] The first determination unit 108 includes an abnormality determination unit 104 (first determination unit), a determination unit 107 (second determination unit), and a determination unit 107B (third determination unit). Hereinafter, the determination method performed by the abnormality determination unit 104 of the detection device 1 according to any one of the first to fourth embodiments will be referred to as the first method. The determination method performed by the determination unit 107 of the detection device 1A according to the fifth embodiment will be referred to as the second method. The determination method performed by the determination unit 107B of the detection device 1B according to the sixth embodiment will be referred to as the third method. The first determination unit 108 can perform determination regarding the accuracy of concentration estimation according to each of the above-described embodiments. The first determination unit 108 outputs the determination results based on each method to the second determination unit 109.

[0142] The abnormality determination unit 104, the determination unit 107, and the determination unit 107B may perform determination only on the sensor element 31 for which determination is possible. For example, the determination unit 107 may determine whether H ２ S and MMC as a second gas component are detected, and H ２ The determination may be made based on the output value of the sensor element 31 that does not detect the

[0143] The second determination unit 109 acquires each of the determination results based on the first method, the second method, and the third method. Then, the second determination unit 109 determines whether or not an abnormality has occurred in the concentration estimation accuracy of the detection device 1C based on each of the determination results. The second determination unit 109 may transmit information indicating the determination result to the electronic device 9.

[0144] As an example, if the determination result based on at least one of the first method, the second method, and the third method indicates a decrease in the concentration estimation accuracy, the second determination unit 109 may determine that an abnormality has occurred in the concentration estimation accuracy of the detection device 1C. This allows a user of the detection device 1C or an administrator of the detection device 1C to recognize that an abnormality has occurred in the concentration estimation accuracy of the detection device 1C by checking the determination result of the second determination unit 109. By recognizing the abnormality in the detection device 1C, the administrator or the like can recognize that at least one of the sensor element 31 of the detection device 1C and the first estimation model 202 for concentration estimation needs to be replaced.

[0145] For example, the first estimation model 202 stored in the detection device 1C may be replaced with another first estimation model 202 capable of estimating the concentration of the target component based on the output value of the degraded sensor element 31. Alternatively, the first estimation model 202 stored in the detection device 1C may be replaced with another first estimation model 202 capable of estimating the concentration of the target component without using the output value of the degraded sensor element 31. As a result, even if the sensor element 31 in the detection device 1C is degraded, the estimation unit 106 can perform estimation using the other first estimation model 202, thereby ensuring the accuracy of the estimation. The detection device 1C may acquire the other first estimation model 202 from a cloud server (not shown) or the like. Alternatively, the storage unit 20C of the detection device 1C may store multiple first estimation models 202. In this case, an appropriate first estimation model 202 may be selected from the first estimation models 202 stored in the storage unit 20C based on an administrator's instruction or the like, and the estimation unit 106 may perform estimation using the selected first estimation model 202.

[0146] Alternatively, the deteriorated sensor element 31 attached to the detection device 1C may be replaced with a normal sensor element 31. This allows the estimation unit 106 to perform estimation with high accuracy by using the output value of the replaced normal sensor element 31. When the sensor element 31 is replaced, the first estimation model 202 corresponding to the replaced sensor element 31 may be stored in the storage unit 20C as an estimation model to be used when the estimation unit 106 performs estimation.

[0147] Furthermore, when the storage unit 20C stores a plurality of first estimation models 202 and there is no first estimation model 202 that can ensure the estimation accuracy, the second determination unit 109 may output information indicating that it is necessary to replace the sensor element 31. In this way, replacement of the sensor element 31 is requested only when it becomes impossible to ensure the estimation accuracy by simply changing the first estimation model 202.

[0148] As described above, in the detection device 1C, the first determination unit 108 can make a determination based on a plurality of methods. For example, the abnormality determination unit 104 makes a determination based on a first method. In the first method, the determination is made based on a change in the waveform shape of each of the sensor elements 31. This allows the abnormality determination unit 104 to determine the deterioration of each of the sensor elements 31 even when deterioration of multiple sensor elements 31 occurs simultaneously. Furthermore, the determination unit 107 makes a determination based on a second method. In the second method, the determination is made based on a change in the relationship between the output values ​​of the multiple sensor elements 31. Furthermore, when H as the first gas component is used, ２ S and MMC as a second gas component are detected, and H ２ If the sensor element 31 that does not detect the temperature is deteriorated, this may have a significant impact on the detection accuracy of the detection device 1C. Here, the second method can determine the accuracy of estimation based on the output value of such a sensor element 31. Furthermore, the determination unit 107B performs determination based on the third method. With the third method, if the sensor element 31 is significantly deteriorated, the deterioration of the sensor element 31 can be accurately determined. Furthermore, with the third method, there is a low possibility of erroneously determining that a sensor element 31 that is not actually deteriorated is deteriorated.

[0149] The second determination unit 109 can determine the estimation accuracy of the concentration in the detection device 1C by referring to the determination results based on the multiple methods described above, each of which has different advantages. In the detection device 1C, the final determination is made based on the determination results using the multiple methods, thereby reducing the possibility that an abnormality in the estimation accuracy will be overlooked.

[0150] When the sensor element 31 that is the subject of the determination is identified, the second determination unit 109 may output information indicating the sensor element 31 that is the subject of the determination together with the determination result. ２ S and MMC as a second gas component are detected, and H ２Here, when the determination unit 107 outputs a determination result indicating that the accuracy of estimating the concentration has decreased, the second determination unit 109 may output information indicating the sensor element 31 that is the target of the determination by the determination unit 107 together with the determination result. This allows the administrator or the like to know which sensor element 31 should be replaced.

[0151] As another example, if the determination result based on the third method indicates a decrease in the concentration estimation accuracy, the second determination unit 109 may determine that an abnormality has occurred in the concentration estimation accuracy of the detection device 1C. Alternatively, if the determination results based on both the first method and the second method indicate a decrease in the concentration estimation accuracy, the second determination unit 109 may determine that an abnormality has occurred in the concentration estimation accuracy of the detection device 1C. In other words, if the determination unit 107B included in the first determination unit 108 determines that the sensor element 31 has deteriorated, the second determination unit 109 may determine that an abnormality has occurred in the concentration estimation accuracy of the detection device 1C. Alternatively, if both the abnormality determination unit 104 and the determination unit 107 included in the first determination unit 108 determine that the concentration estimation accuracy has decreased, the second determination unit 109 may determine that an abnormality has occurred in the concentration estimation accuracy of the detection device 1C.

[0152] [Example of implementation by software] The functions of the detection devices 1, 1A to 1C (hereinafter referred to as "devices") can be realized by a program that causes a computer to function as the device, and a program that causes a computer to function as each control block of the device (particularly each part included in the control unit 10, 10A to 10C).

[0153] In this case, the device includes a computer having at least one control device (e.g., a processor) and at least one storage device (e.g., a memory) as hardware for executing the program. The functions described in each of the above embodiments are realized by executing the program using the control device and storage device.

[0154] The program may be non-transitory and may be recorded on one or more computer-readable recording media. The recording media may or may not be included in the device. In the latter case, the program may be supplied to the device via any wired or wireless transmission medium.

[0155] In addition, some or all of the functions of each of the control blocks can be realized by logic circuits. For example, integrated circuits in which logic circuits that function as each of the control blocks are formed are also included in the scope of the present disclosure. In addition, the functions of each of the control blocks can also be realized by, for example, a quantum computer.

[0156] Furthermore, each process described in each of the above embodiments may be executed by AI (Artificial Intelligence). In this case, the AI ​​may run on the control device or on another device (for example, an edge computer or a cloud server).

[0157] The invention according to the present disclosure has been described above based on the drawings and examples. However, the invention according to the present disclosure is not limited to the above-described embodiments. In other words, the invention according to the present disclosure can be modified in various ways within the scope of the present disclosure, and embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the invention according to the present disclosure. In other words, it should be noted that a person skilled in the art can easily make various modifications or corrections based on the present disclosure. It should also be noted that these modifications or corrections are included in the scope of the present disclosure.

[0158] [Summary] A detection device according to aspect 1 of the present disclosure comprises an identification unit that inputs second time series data output by the detection device at a second time point after the first time point when the detection device detects the target component contained in the target gas, into an identification model that has been machine-learned using first time series data output by the detection device, which has a sensor unit capable of outputting time series data of detection values ​​corresponding to the concentration of a target component contained in a target gas, when the sensor unit detects the target component contained in a standard gas at a first time point when the detection device is in a first state, and outputs discrepancy information regarding the difference between the first time series data and the second time series data; and an abnormality determination unit that determines a second state, which is the state of the detection device at the second time point, based on the magnitude of the difference indicated by the discrepancy information.

[0159] A detection device according to Aspect 2 of the present disclosure may be, in Aspect 1, wherein the first state is a state in which the detection device is operating normally.

[0160] The detection device according to aspect 3 of the present disclosure, in the above-mentioned aspects 1 or 2, may further include a supply mechanism that supplies the target gas or the standard gas to the sensor unit and discharges it from the sensor unit, and the abnormality determination unit may determine whether an abnormality has occurred in the sensor unit or the supply mechanism based on the difference information.

[0161] A detection device according to aspect 4 of the present disclosure is any one of aspects 1 to 3 above, further comprising a supply mechanism that supplies the target gas or the standard gas to the sensor unit and discharges it from the sensor unit, and the sensor unit may output the first time series data corresponding to the detection result during the period from when the supply of the standard gas to the sensor unit begins to when the discharge of the standard gas from the sensor unit ends.

[0162] A detection device according to aspect 5 of the present disclosure, in any one of aspects 1 to 4 above, may further include a supply mechanism that supplies the target gas or the standard gas to the sensor unit and discharges it from the sensor unit, and the sensor unit may output the second time series data corresponding to the detection result during the period from when the supply of the target gas to the sensor unit begins to when the discharge of the target gas from the sensor unit ends.

[0163] A detection device according to aspect 6 of the present disclosure is any of aspects 1 to 5 above, and includes a data acquisition unit that acquires the second time series data from the detection device, wherein the identification unit, each time the data acquisition unit acquires the second time series data, inputs the second time series data into the identification model and outputs the difference information, and the abnormality determination unit determines whether the detection device is operating normally based on a compilation result of compiling the magnitudes of the differences indicated by the multiple pieces of difference information output from the identification unit.

[0164] A detection device according to a seventh aspect of the present disclosure is any one of the first to sixth aspects, wherein the identification model is a neural network that has been machine-learned to output first output data that reconstructs the first time series data that has been input, and the identification unit may output the difference information based on a comparison result obtained by comparing second output data that is output from the identification model when the second time series data is input to the identification model with the second time series data that has been input to the identification model.

[0165] A detection device according to an eighth aspect of the present disclosure is any one of the first to seventh aspects, wherein the target gas may include gas originating from a user of the detection device.

[0166] A detection device according to aspect 9 of the present disclosure is any one of aspects 1 to 8 above, wherein the sensor unit may include a plurality of sensor elements capable of outputting a detection value corresponding to the concentration of the target component.

[0167] A detection device according to aspect 10 of the present disclosure is any one of aspects 1 to 9, wherein the abnormality determination unit may output a plurality of candidates for the state of the detection device as the determination result.

[0168] The estimation system according to aspect 11 of the present disclosure, in any one of aspects 1 to 10 above, further includes a supply mechanism that supplies the target gas or the standard gas to the sensor unit and discharges it from the sensor unit, and the supply mechanism may supply the target gas or the standard gas to the sensor unit, and then supply an exhaust gas to the sensor unit to discharge the target gas or the standard gas from the sensor unit.

[0169] An estimation system according to aspect 12 of the present disclosure comprises a detection device having a sensor unit, and an estimation device communicatively connected to the detection device, wherein the sensor unit is capable of outputting time series data of detection values ​​corresponding to the concentration of a target component contained in a target gas, an identification unit that inputs second time series data output by the detection device at a second time point after the first time point when the detection device detects the target component contained in the target gas into an identification model that has been machine-learned using first time series data output by the sensor unit when the detection device detects the target component contained in a standard gas at a first time point when the detection device is in a first state, and outputs discrepancy information regarding the difference between the first time series data and the second time series data, and an abnormality determination unit that determines a second state, which is the state of the detection device at the second time point, based on the magnitude of the difference indicated by the discrepancy information.

[0170] The estimation system according to aspect 13 of the present disclosure may further include, in aspect 12 above, an electronic device connected to the estimation device, and the electronic device may include a display screen that displays the judgment result output from the abnormality judgment unit.

[0171] A detection method according to aspect 14 of the present disclosure includes an identification step of inputting second time series data output by a detection device at a second time point after the first time point when the detection device detects the target component contained in the target gas, into an identification model that has been machine-learned using first time series data output by the detection device, which has a sensor unit capable of outputting time series data of detection values ​​corresponding to the concentration of the target component contained in the target gas, when the sensor unit detects the target component contained in the standard gas at a first time point when the detection device is in a first state, and outputting difference information regarding the difference between the first time series data and the second time series data; and a determination step of determining a second state, which is the state of the detection device at the second time point, based on the magnitude of the difference indicated by the difference information.

[0172] A control program according to aspect 15 of the present disclosure is a control program for causing a computer to function as the detection device of aspect 1, and causes the computer to function as the identification unit and the abnormality determination unit.

[0173] A recording medium according to aspect 16 of the present disclosure is a computer-readable recording medium on which the control program of aspect 15 above is recorded.

[0174] An estimation system according to a seventeenth aspect of the present disclosure includes a sensor unit including a plurality of sensor elements capable of outputting an output signal corresponding to the concentrations of a first gas component and a second gas component different from the first gas component contained in a target gas; a supply mechanism capable of supplying the target gas to the sensor unit; an estimation unit that estimates the concentrations of the first gas component and the second gas component contained in the target gas based on the output signals; and a determination unit that outputs a determination result of the estimation accuracy by the estimation unit based on the output signals, wherein the plurality of sensor elements include sensor elements that have different detection sensitivities to at least one of the first gas component and the second gas component. The determination unit uses an estimation model that has been machine-learned to learn feature information including: (1) information created based on a normal signal, which is the output signal output by the sensor unit when all of the sensor elements included in the plurality of sensor elements are operating normally; (2) information created based on an abnormal signal, which is the output signal output by the sensor unit including at least one sensor element that is not operating normally; or (3) information created to imitate the abnormal signal based on the normal signal; and outputs information regarding the estimation accuracy of the estimation result based on the output signal at the time of interest from the feature information in the output signal output by the sensor unit at the time of interest.

[0175] In the estimation system according to aspect 18 of the present disclosure, in the above-mentioned aspect 17, the feature information may be at least one of a peak value of the output signal and a ratio of the peak values ​​between the plurality of sensor elements.

[0176] An estimation system according to aspect 19 of the present disclosure may be such that, in aspect 17 or 18 above, the supply mechanism supplies the target gas to the sensor unit, and then supplies exhaust gas to the sensor unit for exhausting the target gas from the sensor unit, and the determination unit uses the output signal output from the sensor unit at the timing when the supply mechanism switches the gas supplied to the sensor unit from the target gas to the exhaust gas.

[0177] In the estimation system according to aspect 20 of the present disclosure, in any of aspects 17 to 19 above, the determination unit may output information indicating whether the state of the multiple sensor elements included in the sensor unit is normal or not, as information regarding the estimation accuracy of the estimation unit.

[0178] In the estimation system according to Aspect 21 of the present disclosure, in any one of Aspects 17 to 20, the first gas component and the second gas component may be hydrogen sulfide or methyl mercaptan.

[0179] In the estimation system according to Aspect 22 of the present disclosure, in any one of Aspects 17 to 20, the plurality of sensor elements may have different detection methods.

[0180] In the estimation system according to aspect 23 of the present disclosure, in any one of aspects 17 to 22, the plurality of sensor elements may be electrochemical sensors, semiconductor sensors, or combustion sensors.

[0181] The estimation system according to aspect 24 of the present disclosure comprises a sensor unit having a sensor element capable of outputting an output signal corresponding to the concentration of a target component contained in a target gas, a supply mechanism capable of supplying the target gas to the sensor unit, an estimation unit that estimates an estimated value of the concentration of the target component contained in the target gas based on the output signal, and a judgment unit that determines the accuracy of the estimation of the concentration of the target component by the estimation unit based on the estimated value, and if the estimated value is a negative value, the judgment unit determines that the accuracy of the estimation of the concentration of the target component by the estimation unit has decreased.

[0182] In the estimation system according to Aspect 25 of the present disclosure, in Aspect 24 above, the determination unit may determine that the sensor unit is abnormal if the estimated value is a negative value.

[0183] In the estimation system according to aspect 26 of the present disclosure, in aspect 24 or 25 above, the determination unit may determine that the sensor element is abnormal if the estimated value is a negative value and the absolute value of the estimated value is greater than or equal to a first threshold value.

[0184] In the estimation system according to aspect 27 of the present disclosure, in the above aspect 26, the judgment unit may judge that the sensor element is abnormal if the output signal is a negative value and the pattern in which the absolute value of the estimated value is less than or equal to a first threshold value occurs a predetermined number of times or more.

[0185] In the estimation system according to aspect 28 of the present disclosure, in any of aspects 24 to 27 above, the judgment unit may judge that the sensor element is abnormal if a pattern in which the output signal is a negative value and the absolute value of the estimated value is equal to or less than a second threshold value that is smaller than the first threshold value occurs a predetermined number of times or more.

[0186] In the estimation system according to Aspect 29 of the present disclosure, in any one of Aspects 24 to 28, the first gas component may be hydrogen, methyl mercaptan, or hydrogen sulfide.

[0187] In the estimation system according to aspect 30 of the present disclosure, in any one of aspects 24 to 29 above, the sensor element may be an electrochemical sensor, a semiconductor sensor, or a combustion sensor.

[0188] An estimation system according to aspect 31 of the present disclosure includes a detection device including a sensor unit including a plurality of sensor elements capable of outputting an output signal corresponding to the concentrations of target components contained in a target gas, the target components including a first gas component and a second gas component different from the first gas component; a supply mechanism capable of supplying the target gas to the sensor unit; an estimation unit that estimates the concentrations of the first gas component and the second gas component contained in the target gas based on the output signal; an identification unit that inputs second time series data output by the detection device at a second time point after the first time point into an identification model that has been machine-learned using first time series data output when the sensor unit detects the target component contained in a standard gas at a first time point when the detection device is in a first state, and outputs discrepancy information regarding a difference between the first time series data and the second time series data; a first determination unit that determines a second state, which is the state of the detection device at the second time point, based on the magnitude of the difference indicated by the discrepancy information; a second determination unit that outputs a determination result of the estimation accuracy by the estimation unit based on the output signal; the second determination unit uses an estimation model that has been machine-learned to acquire feature information from the output signal output by the sensor unit at a time point of interest, the feature information including: (1) information created based on a normal signal that is the output signal output by the sensor unit when all of the sensor elements included in the plurality of sensor elements are operating normally; (2) information created based on an abnormal signal that is the output signal output by the sensor unit when at least one of the sensor elements is not operating normally; or (3) information created to imitate the abnormal signal based on the normal signal,The third determination unit outputs information regarding the estimation accuracy of the estimation result based on the output signal at the time point of interest, and if the estimation value is a negative value, the third determination unit determines that the estimation accuracy of the concentration of the target component by the estimation unit has decreased, and the fourth determination unit determines whether the accuracy of the concentration estimation in the detection device is normal or not based on the determination results of the first determination unit, the second determination unit, and the third determination unit.

[0189] REFERENCE SIGNS LIST 1 detection device 5 estimation device 9 electronic device 30 sensor unit 100 estimation system 102 data acquisition unit 103 identification unit 104 abnormality determination unit 201 identification model S2 identification step S3 determination step

Claims

1. A detection device comprising: an identification unit that inputs second time series data output by the detection device at a second time point after the first time point when the detection device detects the target component contained in the target gas, into an identification model that has been machine-learned using first time series data output by the sensor unit that is capable of outputting time series data of detection values ​​corresponding to the concentration of a target component contained in a target gas when the sensor unit detects the target component contained in the target gas at a first time point when the detection device is in a first state, and outputs discrepancy information regarding the difference between the first time series data and the second time series data; and an abnormality determination unit that determines a second state, which is the state of the detection device at the second time point, based on the magnitude of the difference indicated by the discrepancy information.

2. The detection device according to claim 1, wherein the first state is a state in which the detection device is operating normally.

3. The detection device according to claim 1, further comprising a supply mechanism that supplies the target gas or the standard gas to the sensor unit and discharges it from the sensor unit, and the abnormality determination unit determines whether or not an abnormality has occurred in the sensor unit or the supply mechanism based on the difference information.

4. A detection device as described in claim 1, further comprising a supply mechanism that supplies the target gas or the standard gas to the sensor unit and discharges it from the sensor unit, and the sensor unit outputs the first time series data corresponding to the detection results during the period from when the supply of the standard gas to the sensor unit begins to when the discharge of the standard gas from the sensor unit ends.

5. A detection device as described in claim 1, further comprising a supply mechanism that supplies the target gas or the standard gas to the sensor unit and discharges it from the sensor unit, and the sensor unit outputs the second time series data corresponding to the detection results during the period from when the supply of the target gas to the sensor unit begins to when the discharge of the target gas from the sensor unit ends.

6. A detection device according to claim 1, further comprising a data acquisition unit that acquires the second time series data from the detection device, wherein the identification unit inputs the second time series data into the identification model and outputs the difference information each time the data acquisition unit acquires the second time series data, and the abnormality determination unit determines whether the detection device is operating normally based on a compilation result of aggregating the magnitude of the difference indicated by the multiple pieces of difference information output from the identification unit.

7. The detection device according to claim 1, wherein the identification model is a neural network that has been machine-trained to output first output data that is obtained by restoring the first time series data that has been input, and the identification unit outputs the difference information based on a comparison result obtained by comparing second output data that is output from the identification model when the second time series data is input to the identification model with the second time series data that has been input to the identification model.

8. The detection device according to claim 1, wherein the target gas includes a gas originating from a user of the detection device.

9. The detection device according to claim 1, wherein the sensor section comprises a plurality of sensor elements capable of outputting a detection value corresponding to the concentration of the target component.

10. The detection device according to claim 1, wherein the abnormality determination unit outputs a plurality of candidates for the state of the detection device as a determination result.

11. A detection device as described in claim 1, further comprising a supply mechanism that supplies the target gas or the standard gas to the sensor unit and discharges it from the sensor unit, wherein the supply mechanism supplies the target gas or the standard gas to the sensor unit, and then supplies discharge gas to the sensor unit for discharging the target gas or the standard gas from the sensor unit.

12. An estimation system comprising: a detection device having a sensor unit; and an estimation device communicatively connected to the detection device, wherein the sensor unit is capable of outputting time series data of detection values ​​corresponding to the concentration of a target component contained in a target gas; an identification unit that inputs second time series data output by the detection device at a second time point after the first time point into an identification model that has been machine-learned using first time series data output when the sensor unit detects the target component contained in a standard gas at a first time point when the detection device is in a first state, and outputs discrepancy information regarding the difference between the first time series data and the second time series data; and an abnormality determination unit that determines a second state, which is the state of the detection device at the second time point, based on the magnitude of the difference indicated by the discrepancy information.

13. The estimation system according to claim 12, further comprising an electronic device connected to the estimation device, the electronic device having a display screen for displaying the determination result output from the abnormality determination unit.

14. A detection method comprising: an identification step of inputting second time series data output by a detection device at a second time point after the first time point when the detection device detects the target component contained in the target gas, into an identification model that has been machine-learned using first time series data output by the detection device, which has a sensor unit capable of outputting time series data of detection values ​​corresponding to the concentration of the target component contained in the target gas, when the sensor unit detects the target component contained in the target gas at a first time point when the detection device is in a first state, and outputting difference information regarding the difference between the first time series data and the second time series data; and a determination step of determining a second state, which is the state of the detection device at the second time point, based on the magnitude of the difference indicated by the difference information.

15. A control program for causing a computer to function as the detection device according to claim 1, the control program causing the computer to function as the identification unit and the abnormality determination unit.

16. A computer-readable recording medium on which the control program according to claim 15 is recorded.

17. A gas analyzer comprising: a sensor unit including a plurality of sensor elements capable of outputting an output signal corresponding to the concentration of a first gas component and a second gas component different from the first gas component contained in a target gas; a supply mechanism capable of supplying the target gas to the sensor unit; an estimation unit that estimates the concentration of the first gas component and the concentration of the second gas component contained in the target gas based on the output signal; and a determination unit that outputs a determination result of the estimation accuracy by the estimation unit based on the output signal, wherein the plurality of sensor elements include sensor elements that differ in at least one of the detection sensitivity to the first gas component and the detection sensitivity to the second gas component, and the determination unit: An estimation system that outputs information regarding the estimation accuracy of an estimation result based on the output signal at a time of interest from the feature information of the output signal output by the sensor unit at the time of interest using an estimation model that has been machine-learned to learn feature information including: (1) information created based on a normal signal, which is the output signal output by the sensor unit when all of the sensor elements included in the plurality of sensor elements are operating normally; (2) information created based on an abnormal signal, which is the output signal output by the sensor unit including at least one sensor element that is not operating normally; or (3) information created to imitate the abnormal signal based on the normal signal.

18. The estimation system according to claim 17, wherein the characteristic information is at least one of a peak value of the output signal and a ratio of the peak values ​​between the plurality of sensor elements.

19. The estimation system described in claim 17, wherein the supply mechanism supplies the target gas to the sensor unit, and then supplies exhaust gas to the sensor unit to exhaust the target gas from the sensor unit, and the determination unit uses the output signal output from the sensor unit at the timing when the supply mechanism switches the gas supplied to the sensor unit from the target gas to the exhaust gas.

20. The estimation system according to claim 17, wherein the determination unit outputs information indicating whether the state of the plurality of sensor elements included in the sensor unit is normal or not as information regarding the estimation accuracy of the estimation unit.

21. The estimation system of claim 17, wherein the first gas component and the second gas component are hydrogen sulfide or methyl mercaptan.

22. The estimation system according to claim 17, wherein the plurality of sensor elements have different detection methods.

23. The estimation system according to claim 17, wherein the plurality of sensor elements are electrochemical sensors, semiconductor sensors, or combustion sensors.

24. An estimation system comprising: a sensor unit having a sensor element capable of outputting an output signal corresponding to the concentration of a target component contained in a target gas; a supply mechanism capable of supplying the target gas to the sensor unit; an estimation unit that estimates an estimated value of the concentration of the target component contained in the target gas based on the output signal; and a judgment unit that judges the accuracy of the estimation of the concentration of the target component by the estimation unit based on the estimated value, wherein the judgment unit judges that the accuracy of the estimation of the concentration of the target component by the estimation unit has decreased if the estimated value is a negative value.

25. The estimation system according to claim 24, wherein the determination unit determines that the sensor element is abnormal when the estimated value is a negative value.

26. The estimation system according to claim 24, wherein the determination unit determines that the sensor element is abnormal if the estimated value is a negative value and the absolute value of the estimated value is equal to or greater than a first threshold value.

27. The estimation system according to claim 26, wherein the determination unit determines that the sensor element is abnormal if a pattern in which the output signal is a negative value and the absolute value of the estimated value is equal to or less than a first threshold occurs a predetermined number of times or more.

28. The estimation system according to claim 27, wherein the determination unit determines that the sensor element is abnormal if a pattern in which the output signal is a negative value and the absolute value of the estimated value is equal to or less than a second threshold value that is smaller than the first threshold value occurs a predetermined number of times or more.

29. The estimation system of claim 23, wherein the first gas component is hydrogen, methyl mercaptan, or hydrogen sulfide.

30. The estimation system according to claim 23, wherein the sensor element is an electrochemical sensor, a semiconductor sensor, or a combustion sensor.

31. A detection device comprising: a sensor unit including a plurality of sensor elements capable of outputting an output signal corresponding to the concentration of target components contained in a target gas, the target components including a first gas component and a second gas component different from the first gas component; a supply mechanism capable of supplying the target gas to the sensor unit; an estimation unit that estimates the concentration of the first gas component and the concentration of the second gas component contained in the target gas based on the output signal; an identification unit that inputs second time series data output by the detection device at a second time point after the first time point into an identification model that has been machine-learned using first time series data output when the sensor unit detects the target component contained in a standard gas at a first time point when the detection device is in a first state, and outputs discrepancy information regarding the difference between the first time series data and the second time series data; a first determination unit that determines a second state, which is the state of the detection device at the second time point, based on the magnitude of the difference indicated by the discrepancy information; and a second determination unit that outputs a determination result of the estimation accuracy made by the estimation unit based on the output signal. a third determination unit that determines the accuracy of the estimation of the concentration of the target component by the estimation unit based on estimated values ​​indicating the concentrations of the first gas component and the second gas component; a fourth determination unit that determines whether the accuracy of the concentration estimation in the detection device is normal; and an estimation device that is communicably connected to the detection device, wherein the sensor unit is capable of outputting time-series data of detection values ​​corresponding to the concentration of the target component contained in the target gas, and the plurality of sensor elements include sensor elements that differ in at least one of detection sensitivity to the first gas component and detection sensitivity to the second gas component, and the second determination unitan estimation system that outputs information related to the estimation accuracy of an estimation result based on the output signal at a time point of interest from the feature information of the output signal output by the sensor unit at the time point of interest using an estimation model that has been machine-learned to learn feature information including: (1) information created based on a normal signal, which is the output signal output by the sensor unit when all of the sensor elements included in the plurality of sensor elements are operating normally; (2) information created based on an abnormal signal, which is the output signal output by the sensor unit including at least one sensor element that is not operating normally; or (3) information created to imitate the abnormal signal based on the normal signal; wherein the third determination unit determines that the estimation accuracy of the concentration of the target component by the estimation unit has decreased if the estimated value is a negative value; and the fourth determination unit determines whether the accuracy of the concentration estimation in the detection device is normal based on the determination results of the first determination unit, the second determination unit, and the third determination unit.

Images