Odor detection system, odor detection method, and odor detection program

Abstract

To provide an odor detection system that suitably determines a state of an object using its odor.SOLUTION: An odor detection system 1A comprises: an odor sensor 12 reacting to an odor of an object; a speed calculation section 11 for calculating a travel speed of the object viewed from the odor sensor; and a determination section 24 for determining a state of the object, based on an output value of the odor sensor and a travel speed of the object. More specifically, a plurality of odor determination models 23A are provided that have been learnt by using an output value of the odor sensor for the plurality of travel speeds respectively, as an explanatory variable, and by using the state of an object as an objective variable. The determination section selects one odor determination model based on the travel speed calculated by the speed calculation section, and the selected odor determination model is used to determine the state of an object.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a technique for determining the state of an object using odor.

Background Art

[0002] As a technique for detecting odor, Patent Document 1 describes an odor detection device that specifies an odor component contained in a gas to be measured and its concentration based on output values of a plurality of odor sensors having different characteristics in response to odor.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] The odor detection device described in Patent Document 1 detects odor contained in the gas around the odor sensor. When it is desired to measure the odor of a moving object, the determination result changes at the speed at which the object to be measured passes near the sensor. That is, when such an odor detection device is used to determine the state of an object that is the source of the odor, there is a problem that it cannot be successfully determined if the moving speed of the object is different.

[0005] The present invention has been devised in view of the above circumstances, and an object thereof is to provide an odor detection system, an odor detection method, and an odor detection program capable of suitably determining the state of an object using odor.

Means for Solving the Problems

[0006] The present invention for solving the above problems has the following configuration.

[0007] (1) An odor detection system comprising an odor sensor that reacts to the odor of an object, a speed calculation unit that calculates the moving speed of the object as seen from the odor sensor, and a determination unit that determines the state of the object based on the output value of the odor sensor and the moving speed of the object.

[0008] (2) The odor detection system according to (1), wherein the determination unit determines whether the object is abnormal.

[0009] (3) The odor detection system according to (2), wherein the determination unit determines whether the object is damaged.

[0010] (4) The odor detection system according to (1), comprising a plurality of odor determination models learned with the output value of the odor sensor as an explanatory variable and the state of the object as a target variable for each of the plurality of moving speeds, wherein the determination unit selects one of the odor determination models based on the moving speed calculated by the speed calculation unit, and determines the state of the object using the selected odor determination model.

[0011] (5) The odor detection system according to (1), comprising an odor determination model learned with the moving speed and the output value of the odor sensor as explanatory variables and the state of the object as a target variable, wherein the determination unit determines the state of the object using the odor determination model from the moving speed calculated by the speed calculation unit and the output value of the odor sensor.

[0012] (6) The odor detection system according to (1), further comprising a distance calculation unit that calculates the distance between the object and the odor sensor, wherein the determination unit determines the state of the object based on the output value of the odor sensor, the moving speed of the object, and the distance.

[0013] (7) It includes a plurality of noise determination models that use the output value of the noise sensor as an explanatory variable and the state of the object as an objective variable for learning for each set of the plurality of the moving speeds and the distances. The determination unit selects one of the noise determination models based on the moving speed calculated by the speed calculation unit and the distance, and determines the state of the object using the selected noise determination model. The noise detection system according to (6).

[0014] (8) It includes a noise determination model that uses the moving speed, the distance, and the output value of the noise sensor as explanatory variables and the state of the object as an objective variable for learning. The determination unit determines the state of the object using the noise determination model from the moving speed calculated by the speed calculation unit, the distance calculated by the distance calculation unit, and the output value of the noise sensor. The noise detection system according to (6).

[0015] (9) It includes a plurality of the noise sensors with different characteristics that react to noise. The determination unit identifies the object based on the output values of the plurality of the noise sensors. The noise detection system according to (1).

[0016] (10) A noise determination method including: a step in which a noise sensor reacts to the noise of an object; a step in which a speed sensor detects the moving speed of the object; a step in which a speed calculation unit calculates the moving speed of the object as seen from the noise sensor; and a step in which a determination unit determines the state of the object based on the output value of the noise sensor and the moving speed of the object.

[0017] (11) A noise determination program for causing a computer to execute: a procedure for acquiring the output value of a noise sensor that has reacted to the noise of an object; a procedure for acquiring the moving speed of the object detected by a speed sensor; a procedure for a speed calculation unit to calculate the moving speed of the object as seen from the noise sensor; and a procedure for a determination unit to determine the state of the object based on the output value of the noise sensor and the moving speed of the object.

Advantages of the Invention

[0018] According to the present invention, since the state of the object detected by the odor sensor is determined in consideration of the moving speed of the object, the state of the object can be suitably determined using the odor.

Brief Description of the Drawings

[0019]

Figure 1

Figure 2A

Figure 2B

Figure 2C

Figure 3

Figure 4

Figure 5

Figure 6A

Figure 6B

Figure 7

Figure 8A

Figure 8B

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Best Mode for Carrying Out the Invention

[0020] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. The same reference numerals are assigned to the same components, and redundant explanations are omitted.

[0021] 《First Embodiment》 FIG. 1 is a configuration diagram of an odor detection system 1A according to a first embodiment for determining the state of an object 2 (see FIG. 2A) based on the odor of the object and the speed at which the object moves.

[0022] Examples of the object 2 include food and drink (such as fruits and vegetables), human body odor, etc. The state may be a level (such as a two-stage normal or abnormal), or a continuous value. Examples of abnormalities include that the container or packaging of food and drink is damaged (there is an odor of food and drink), fruits and vegetables are spoiled (there is a putrid smell), etc.

[0023] The odor detection system 1A includes an odor detection device 10A and a determination device 20A. The odor detection device 10A and the determination device 20A are provided with a short-range wireless communication unit 18, and can communicate with each other via Bluetooth (registered trademark). Note that the communication method of the odor detection device 10A and the determination device 20A is not limited to the above, and may be wireless communication other than Bluetooth (registered trademark), or wired network communication such as LAN.

[0024] The odor detection device 10A includes a control unit 11, a plurality of odor sensors 12 (12a, 12b, 12c, 12d), an ADC (Analog to Digital Converter) 13, a speed sensor 14, a display unit 15, an operation switch 16, a battery 17, and a communication unit 18.

[0025] The control unit 11 is composed of a CPU (Central Processing Unit) or the like, and comprehensively controls the processing operations of each part of the odor detection device 10A by executing an odor detection program (not shown). Specifically, the control unit 11 reads out various processing programs stored in a storage unit (not shown), and performs various processes in cooperation with the programs.

[0026] The odor sensors 12a to 12d are semiconductor gas sensors with different characteristics of reacting to odors, such as which odors they strongly react to. The odor sensors 12a to 12d convert the concentration of the gas to be detected (mainly the gas to react with) into an electrical quantity and output an electrical signal corresponding to the gas concentration. As the odor sensors 12a to 12d, for example, gas sensors for detecting VOC (Volatile Organic Compounds), gas sensors for detecting CO, gas sensors for detecting hydrogen, gas sensors for detecting hydrocarbons, gas sensors for detecting alcohol, gas sensors for detecting tobacco, etc. can be used. Each of the odor sensors 12a to 12d reacts to a plurality of odor components and does not react to only one odor component. Here, the odor component is a chemical substance constituting the odor. Note that the odor sensors 12a to 12d may be MEMS (Micro Electro Mechanical Systems) type sensors. In this specification, the case where the odor sensor 12 is composed of four sensors will be described, but it goes without saying that it is not limited to this.

[0027] The ADC 13 converts the analog signals output from the odor sensors 12a to 12d into digital signals and outputs the converted digital signals to the control unit 11.

[0028] The speed sensor 14 is a sensor that detects the speed of the movement of the object 2 to be state-determined, that is, a parameter related to the speed of the movement of the object 2 as seen from the speed sensor 14. As the speed sensor 14, for example, sensors using laser, infrared rays, electromagnetic waves, ultrasonic waves, etc. can be used. Also, when the object 2 moves by a conveyor belt or the like, instead of the speed sensor 14, a moving object speed acquisition unit that acquires the conveyance speed of the conveyor belt as the moving speed of the object 2 may be provided.

[0029] The control unit 11 acquires the detection result of the speed sensor 14 (parameter related to the speed of the object 2) and functions as a speed calculation unit that calculates the speed of the object 2 as seen from the odor sensors 12a to 12d based on the acquired detection result.

[0030] The display unit 15 is composed of a monitor or the like, and displays the determination result of the state of the object 2 of the determination device 20A described later based on the control of the control unit 11.

[0031] The operation switch 16 is composed of a power switch for turning the power on / off, a measurement switch for instructing the start of measurement, etc., and outputs an operation signal corresponding to the operation (pressing) of the operation switch 16 by the user to the control unit 11.

[0032] The battery 17 supplies power to each part of the odor detection device 10A. As the battery 17, a detachable dry battery, a rechargeable battery, etc. are used.

[0033] The communication unit 18 has an interface for performing data communication by wireless communication such as Bluetooth (registered trademark) with the terminal device 40 and the determination device 20A described later, and transmits and receives data with the determination device 20A according to the communication standard.

[0034] The determination device 20A is a server, a PC (Personal Computer), etc., and includes a control unit 21, a storage unit 22, a plurality of odor determination models 23A (23a, 23b, 23c), a determination unit 24, and a communication unit 25.

[0035] The control unit 21 is composed of a CPU or the like, and comprehensively controls the processing operations of each part of the determination device 20A by executing a determination program (not shown). Specifically, the control unit 21 reads out various processing programs stored in the storage unit 22 and performs various processes in cooperation with the program.

[0036] The storage unit 22 is composed of a non-volatile semiconductor memory or the like, and stores various processing programs, parameters necessary for the execution of the program, files, etc. The storage unit 22 stores an identifier 22a and an odor type discrimination table 22b.

[0037] The identifier 22a is the result of machine learning pre-generated by an external device during the development stage of the odor detection device 10A. During machine learning, each of a plurality of odor components (chemical substances) is prepared for each concentration. The external device acquires the output values (waveforms) of odor sensors of the same type as the odor sensors 12a to 12d (hereinafter referred to as learning odor sensors) for each concentration of each of the plurality of odor components. Then, by inputting the combination of the output values of each learning odor sensor when the odor component of the said concentration is targeted, and performing machine learning with the said odor component and its concentration as the output, the identifier 22a is generated. That is, machine learning is performed using, as teacher data, a set of data with the combination of the output values of a plurality of learning odor sensors as the input and the odor component and its concentration as the output.

[0038] The "odor sensors of the same type" are sensors having the same characteristics as the target odor sensors 12a to 12d (ones that can obtain the same output), for example, sensors with the same model numbers. As the machine learning model, for example, a neural network, particularly a learning vector quantization (LVQ) neural network is used. For example, the identifier 22a is constructed using, as feature quantities, the rising manner, peak value, etc. of the waveform obtained by plotting the gas concentration, which is the output value of the learning odor sensor, along the elapsed time from the start of measurement.

[0039] Examples of odor components include components that cause bad odors such as nonanal, diacetyl, isovaleric acid, and ammonia. Nonanal is a component that causes body odor associated with aging. Diacetyl is a component that causes middle-aged male body odor with a greasy smell. Isovaleric acid is a component that causes body odor due to sweat.

[0040] Here, when generating the identifier 22a in advance, learning odor sensors are used. However, it is also possible to perform machine learning using the odor sensors 12a to 12d themselves actually mounted on the odor detection device 10.

[0041] The odor type discrimination table 22b is a table that associates the type of odor and its intensity (level) for each concentration of each of a plurality of odor components. The odor type discrimination table 22b is information serving as a criterion for converting the concentration of each odor component into the type of odor (and its intensity). The type of odor is what humans classify and name for an odor having some characteristics. Examples of the type of odor include the odor of food and drink as the object 2, the odor of aging, the odor of middle-aged sebum, and the three major body odors of sweat odor, etc.

[0042] In addition, in the storage unit 22, for each concentration of each odor component, output values (waveforms) of the plurality of odor sensors 12a to 12d when the odor component at the concentration is the target are stored in advance. Also, in the storage unit 22, output values (waveforms) of the plurality of odor sensors 12a to 12d when a predetermined odor is the target are stored in advance. The data of the output values corresponding to each concentration of each odor component and the data of the output values corresponding to a predetermined odor, which are stored in the storage unit 22, are obtained in advance using a learning odor sensor. Note that they may be obtained in advance using the odor sensors 12a to 12d themselves actually mounted on the odor detection device 10.

[0043] In addition, in the odor type discrimination table 22b, the name of the object 2 and the type of odor are stored in association with each other.

[0044] The odor determination model 23A (23a to 23c) is a model learned with the output values (or feature amounts) of the odor sensor 12 as explanatory variables and the state of the object 2 as the objective variable for each of a plurality of speed ranges. The odor determination model 23A has a threshold value of the output value regarding the odor sensor 12 obtained by machine learning. Then, if the output values regarding the odor sensors 12a to 12c are less than the threshold value, it is determined that the object 2 is normal, and if the output values regarding the odor sensors 12a to 12c are equal to or greater than the threshold value, it is determined that the object 2 is abnormal. Hereinafter, three odor determination models will be described, but it goes without saying that it is not limited thereto.

[0045] As shown in FIG. 2A, the noise determination model 23a is a model learned by machine learning based on the output value (or feature amount) of the noise sensor 12 of the object 2 that is located at a predetermined distance as seen from the noise sensor 12 and moves at a slow speed in the speed range (0 to 120 mm / s).

[0046] As shown in FIG. 2B, the noise determination model 23b is a model learned by machine learning based on the output value (or feature amount) of the noise sensor 12 of the object 2 that is located at a predetermined distance as seen from the noise sensor 12 and moves at a standard speed in the speed range (120 to 180 mm / s).

[0047] As shown in FIG. 2C, the noise determination model 23c is a model learned by machine learning based on the output value (or feature amount) of the noise sensor 12 of the object 2 that is located at a predetermined distance as seen from the noise sensor 12 and moves at a high speed in the speed range (180 to mm / s).

[0048] In the noise determination models 23a to 23c, the threshold values are set individually. When the speed of the object 2 is slow, since the noise reaches the noise sensor 12 without being significantly attenuated, a high threshold value is set, and a robust determination can be made with a low false alarm rate (the ratio of misjudging normal as abnormal) and being less affected by disturbances. When the speed of the object 2 is fast, since the noise attenuates before reaching the noise sensor 12, a low threshold value can be set to reduce the miss rate (the ratio of misjudging abnormal as normal).

[0049] Returning to FIG. 1, the determination unit 24 is configured by a CPU or the like, acquires the output values of the noise sensors 12a to 12d via the communication units 18 and 25, and generates a feature amount of the noise based on the acquired output values of the noise sensors 12a to 12d. Examples of the feature amount include the rising manner, peak value, etc. of the waveform obtained by plotting the gas concentration, which is the output value of the noise sensors 12a to 12d, along the elapsed time from the start of measurement.

[0050] Further, the determination unit 24 calculates the odor components and their concentrations based on the output values (or feature amounts) of the odor sensors 12a to 12d and the identifier 22a. Further, the determination unit 24 refers to the odor type discrimination table 22b using the calculated odor components and their concentrations. The determination unit 24 thereby discriminates the odor type and identifies the object 2 based on the discrimination result (by reading out the name of the object 2 associated with the discriminated odor type).

[0051] The communication unit 25 has an interface for performing data communication by wireless communication such as Bluetooth (registered trademark) between the odor detection device 10A and a terminal device 40 described later, and transmits and receives data to and from the odor detection device 10A according to the communication standard.

[0052] FIG. 3 is another configuration diagram of the odor detection system according to the first embodiment The odor detection system of FIG. 3 is different in that the speed measurement device 30C performs the speed detection of the object 2 by the speed sensor 14 in the odor detection device 10A of the odor detection system of FIG. 1, and notifies the terminal device 40 of the determination result of the object 2.

[0053] The speed measurement device 30C includes a control unit 31, a speed sensor 32C, a battery 33, and a communication unit 34. The control unit 31 is configured by a CPU or the like, and comprehensively controls the processing operations of each part of the speed measurement device 30C. Specifically, the control unit 31 reads out various processing programs stored in a storage unit (not shown) and performs various processes in cooperation with the programs.

[0054] The speed sensor 32C is a sensor that detects a parameter related to the speed of the object 2 to be state-determined, that is, the speed of movement of the object 2 as seen from the speed sensor 32C. As the speed sensor 32C, for example, a sensor using laser, infrared ray, electromagnetic wave, ultrasonic wave, or the like can be used.

[0055] The control unit 31 acquires the detection result of the speed sensor 32C (parameter related to the speed of the object 2) and notifies the determination device 20A. Based on the detection result of the speed sensor 32C, the speed of the object 2 as seen from the noise sensors 12a to 12d is calculated.

[0056] The battery 33 supplies power to each part of the speed measurement device 30C. As the battery 33, a detachable dry battery, rechargeable battery, etc. are used. The communication unit 34 has an interface for performing data communication by wireless communication such as Bluetooth (registered trademark) with the determination device 20A, and transmits and receives data to and from the determination device 20A according to the communication standard.

[0057] Here, the terminal device 40 will be described. The terminal device 40 is a smartphone or the like that can be carried by the user, and includes a control unit 41, a display unit 42, an operation unit 43, a first communication unit 44, a second communication unit 45, a storage unit 46, a speaker 47, and a microphone 48.

[0058] The control unit 41 is composed of a CPU or the like, and comprehensively controls the processing operations of each part of the terminal device 40. Specifically, the control unit 41 reads out various processing programs stored in the storage unit 46, and performs various processes in cooperation with the programs.

[0059] The display unit 42 is composed of an LCD (Liquid Crystal Display) or the like, and displays various screens according to the instruction of the display signal input from the control unit 41. The operation unit 43 is composed of operation keys and a touch panel laminated on the display unit 42, and outputs an operation signal corresponding to the operation keys and an operation signal according to the position of the touch operation by the user's finger or the like to the control unit 41.

[0060] The first communication unit 44 wirelessly connects to a communication network including a mobile communication network via a base station or an access point, and communicates with an external device connected to the communication network. The second communication unit 45 has an interface for performing data communication with an external device such as the determination device 20A by wireless communication such as Bluetooth (registered trademark). The second communication unit 45 transmits and receives data to and from the external device according to the communication standard.

[0061] The storage unit 46 is composed of a non-volatile semiconductor memory or the like, and stores various processing programs, parameters and files necessary for the execution of the programs, etc. An application program for noise detection (hereinafter referred to as a noise detection application) for performing noise detection using the noise detection device 10C is installed in the storage unit 46.

[0062] The speaker 47 converts the electrical signal received from the external device via the first communication unit 44 into an audio signal and outputs the audio. The microphone 48 detects sound waves, converts them into electrical signals, and outputs them to the control unit 41 and the first communication unit 44.

[0063] By executing the noise detection application, the control unit 41 causes the display unit 42 to display the determination result of the state of the object 2 received from the determination device 20A via the second communication unit 45.

[0064] The determination unit 24 determines the state of the object 2 from the positional relationship between the noise detection device 10C, the speed measurement device 30C, and the object 2, the speed of the object 2 notified from the speed measurement device 30C, and the output value (or feature amount) of the noise sensor 12 notified from the noise detection device 10C. In the present embodiment, the determination unit 24 determines whether the object 2 is normal or abnormal. Although details will be described later, the determination unit 24 selects a noise determination model 23A corresponding to the speed of the object 2 and determines the state of the object 2 using the selected noise determination model 23A.

[0065] Next, an operation example of the noise detection system 1A in FIG. 1 will be described with reference to the flowchart of FIG. 4.

[0066] First, the odor sensor 12 (12a to 12d) reacts to the odor of the object 2 (step S1A). Subsequently, the determination unit 24 acquires the output values of the odor sensors 12a to 12d, and generates a feature amount of the odor based on the acquired output values (step S2A). On the other hand, the speed sensor 14 detects the speed of the object 2 (step S1B). Subsequently, the control unit 11 calculates the speed of the object 2 as seen from the odor sensors 12a to 12d based on the acquired detection result, and notifies the determination unit 24 (step S2B).

[0067] Subsequently, the determination unit 24 selects one of the odor determination models 23A (23a to 23c) based on the moving speed information of the object 2 as seen from the odor sensors 12a to 12d calculated by the control unit 11 of the odor detection device 10A (step S3B).

[0068] Steps S1A, S2A and steps S1B to S3B may be executed simultaneously, or either one may be executed first.

[0069] Subsequently, the determination unit 24 acquires the feature amount of the odor, predicts the state of the object 2 using the acquired feature amount and the selected odor determination model 23A (step S4), and performs state determination of the object (step S5). That is, the determination unit 24 causes the acquired feature amount to be input to the selected odor determination model 23A, so that the odor determination model 23A determines whether the object 2 is normal or abnormal, and obtains a determination result. The determination result is transmitted to the odor detection device 10A via the communication unit 25, and is displayed on the display unit 15 by the control unit 11.

[0070] According to the odor detection system according to the first embodiment, since the odor detection system 1A determines the state of the object 2 in consideration of the moving speed of the object 2, regardless of the magnitude of the speed of the object 2, the state of the object 2 can be suitably determined using the odor.

[0071] According to the odor detection system according to the first embodiment, since the odor determination model 23A determines whether the object 2 is normal or abnormal, the inspection of the object 2 can be suitably performed.

[0072] According to the odor detection system according to the first embodiment, the odor detection system 1A includes a plurality of odor determination models 23A that are learned with the output value of the odor sensor 12 as an explanatory variable and the state of the object 2 as an objective variable for each speed range of the plurality of objects 2. Then, the determination unit 24 selects one of the odor determination models 23A based on the calculated speed of the object 2, and determines the state of the object 2 using the selected odor determination model 23A. Thereby, when the speed of the object 2 is low, the odor detection system 1A can reduce the false alarm rate (the ratio of determining normal as abnormal) by setting a high threshold value. Further, when the speed of the object 2 is high, the odor detection system 1A can reduce the miss rate (the ratio of determining abnormal as normal) by setting a low threshold value.

[0073] 《Second Embodiment》 FIG. 5 is a configuration diagram of an odor detection system 1B according to a second embodiment that determines the state of an object 2 based on the odor of the object and the speed at which the object moves.

[0074] The odor detection system 1B according to the second embodiment is different in that it includes an odor determination model 23B instead of a plurality of odor determination models 23A (23a, 23b, 23c). Hereinafter, the odor determination model 23B will be described. Since the other configurations of the determination device 20B in FIG. 5 are the same as those of the determination device 20A in FIG. 1, the description thereof will be omitted.

[0075] The odor determination model 23B is a model that is learned using the output values of the odor sensors 12a to 12d and the speed of movement of the object 2 as explanatory variables and the state of the object 2 as the objective variable. The odor determination model 23B calculates an output value related to the state obtained by machine learning based on the speed of movement of the object 2 and the output value (or feature amount) of the odor sensor 12. Then, the odor determination model 23B can determine that the object is normal if the state is less than the threshold value, and determine that the object is abnormal if the state is greater than or equal to the threshold value.

[0076] As shown in FIG. 6A, the odor sensor 12 measures while changing the speed, and collects the output values (s1, s2, s3,...) of the odor sensor 12 for each speed (v1, v2, v3,...) (see FIG. 6B). By inputting the speed as an explanatory variable, the odor determination model 23B can handle the distance as a continuous value. At this time, it is assumed that the object 2 is located at a predetermined distance away from the odor sensor 12.

[0077] Since the optimal threshold value is automatically set by inputting the movement speed information, the robustness of the odor determination model 23B can be improved. The optimal threshold value is set based on the prediction accuracy of the model, business value, etc. The relationship between the movement speed information and the optimal threshold value can be calculated by regression analysis or the like.

[0078] Next, an operation example of the odor detection system 1B in FIG. 5 will be described with reference to the flowchart of FIG. 7.

[0079] First, the odor sensors 12a to 12d respond to the odor of the object 2 (step S11A). Subsequently, the determination unit 24 acquires the output values of the odor sensors 12a to 12d, and generates a feature amount of the odor based on the acquired output values (step S21).

[0080] On the one hand, the speed sensor 14 detects the speed of the object 2 (step S11B). Subsequently, based on the acquired detection result, the control unit 11 calculates the speed of the object 2 as seen from the noise sensors 12a to 12d and notifies the determination unit 24 (step S12B).

[0081] Steps S11A, S21, and steps S11B to S12B may be executed simultaneously, or either one may be executed first.

[0082] Subsequently, the determination unit 24 acquires the detection result of the speed sensor 14, more specifically, the speed of the object 2 calculated by the control unit 11, and performs (calculates) state prediction using the noise determination model 23B with the acquired speed and the feature amount as explanatory variables (step S22).

[0083] Subsequently, the determination unit 24 determines the state of the object 2 using the generated feature amount of the noise and the determination threshold by the noise determination model 23B (step S23). That is, the determination unit 24 inputs the acquired feature amount of the noise and the calculated speed of the object 2 into the noise determination model 23B to predict the state of the object 2. Then, the determination unit 24 causes the noise determination model 23B to determine whether the object 2 is normal or abnormal, and obtains a determination result. The determination result is transmitted to the noise detection device 10A via the communication unit 25 and is displayed on the display unit 15 by the control unit 11.

[0084] According to the noise detection system according to the first embodiment, since the optimal threshold is automatically set by inputting the moving speed information in the noise detection system 1B, the robustness can be improved.

[0085] 《Third Embodiment》 Next, with reference to FIGS. 8A and 8B, a noise detection system that determines whether the object 2 is normal or abnormal using an abnormality detection model that predicts the degree of abnormality instead of the noise determination model 23B will be described. The system configuration in this case may be the same as that of the noise detection system 1B in FIG. 5.

[0086] The anomaly detection model is an unsupervised learning that learns only from normal data, and sets an optimal threshold between abnormal data and normal data based on the prediction accuracy and business value of the model. Specifically, the anomaly detection model learns based on the feature amounts of the noise sensors, and is a model that determines whether it is normal or abnormal based on the output value (degree of anomaly) of the anomaly detection model and the movement speed information. As shown in FIG. 8B, it is also possible to obtain the relationship with the optimal threshold by regression analysis or the like from the movement speed information and the degree of anomaly at that time, and calculate the optimal threshold for the movement speed information. Instead of regression analysis, a machine learning model such as SVM may be used to obtain a boundary in the space between the degree of anomaly and the movement speed.

[0087] The operation example of the noise detection system according to the third embodiment will be described with reference to the flowchart of FIG. 8A.

[0088] First, the determination unit 24 acquires the output values of the noise sensors 12a to 12d, and generates the feature amounts of the noise based on the acquired output values (step S21).

[0089] Subsequently, the determination unit 24 predicts the degree of anomaly by an anomaly detection model using the feature amounts of the noise as explanatory variables.

[0090] Then, based on the movement speed information of the object 2 obtained based on the detection result of the speed sensor 14 and the degree of anomaly that is the prediction result of the anomaly detection model, an anomaly / normal threshold determination is performed (step S25).

[0091] <<Fourth Embodiment>> Next, with reference to FIGS. 9 to 11, a noise detection system that determines the state of the object 2 based on the feature amounts of the noise of the object, the position of the object, and the speed at which the object moves will be described. That is, the noise detection system according to the fourth embodiment determines the state of the object 2 by adding the positional relationship information between the object and the noise sensors calculated from the position of the object.

[0092] FIG. 9 is a configuration diagram of the noise detection system according to the fourth embodiment.

[0093] The odor detection system of the fourth embodiment adds a position sensor 19 for detecting the position of the object 2 to the odor detection system of the first embodiment in FIG. 1. Further, the odor detection system of the fourth embodiment includes a speed sensor 14, a position sensor 19, and a position-distance conversion unit 191, and a speed / distance measurement device 30D that detects the speed and position of the object 2 and further calculates the distance is configured separately from the odor detection device 10C in the same manner as the odor detection system in FIG. 3.

[0094] The position sensor 19 for detecting the position of the object 2 may measure the position with a sensor such as a camera or a laser, or may be directly input into the system by the user's hand. Alternatively, the object 2 may be imaged by a camera or the like, and the video information may be analyzed to extract the position and speed of the object 2. That is, the speed sensor 14 and the position sensor 19 may be means capable of calculating the position and speed of the object 2, not limited to the measuring device. Thereby, the position-distance conversion unit 191 can calculate the distance between the object 2 and the odor sensor 12 from the position of the object 2. The distance calculated here may be the distance between the odor sensor 12 and the object 2, or in the case of a production line, it may be the shortest distance when the object 2 passes near the odor sensor 12.

[0095] The determination device 20A includes a plurality of odor determination models 23A (23a to 23d), details of which will be described later. These plurality of odor determination models 23A (23a to 23d) are models learned with the output value (or feature amount) of the odor sensor 12 for each combination of the speed range and distance range of the object 2 as the explanatory variable and the state of the object 2 as the objective variable. The odor determination model 23A has a threshold value of the output value regarding the odor sensor 12 obtained by machine learning. If the output values regarding the odor sensors 12a to 12c are less than the threshold value, it is determined that the object 2 is normal. If the output values regarding the odor sensors 12a to 12c are equal to or greater than the threshold value, it is determined that the object 2 is abnormal.

[0096] Since the other configurations of the odor detection device 10C and the determination device 20A are the same as those in FIGS. 1 and 3, the description thereof is omitted here. Further, the speed / distance measurement device 30D may be configured by adding a position sensor 19 and a position / distance conversion unit 191 to the speed measurement device 30C in FIG. 3, and the other configurations of the speed / distance measurement device 30D may be the same as those of the speed measurement device 30C.

[0097] FIG. 10 shows the corresponding odor determination models for each combination of the speed range and the distance range of the object 2, and explains the method for generating the odor determination models.

[0098] Specifically, the odor determination model A (odor determination model 23a) is a model learned by machine learning based on the output value (or feature amount) of the odor sensor 12 for the object 2 moving at a slow speed (~120 mm / s) and in a close distance range (~0.5 cm) as seen from the odor sensor 12.

[0099] Similarly, the odor determination model B (odor determination model 23b), the odor determination model C (odor determination model 23c), and the odor determination model D (odor determination model 23d) are also models learned by machine learning based on the output value (or feature amount) of the odor sensor 12 for the object 2 at the speed in the speed range and the position in the distance range shown in FIG. 10.

[0100] Next, an operation example of the odor detection system according to the fourth embodiment of FIG. 9 will be described with reference to the flowchart of FIG. 11.

[0101] First, the determination unit 24 acquires the output values of the odor sensors 12 (12a to 12d), and generates an odor feature amount based on the acquired output values (step S2A).

[0102] On one hand, a control unit 31 (not shown) of the speed and distance measurement device 30D detects the speed of the object by the speed sensor 14, calculates the speed of the object 2 as seen from the noise sensor 12, and notifies the determination unit 24. Further, the control unit 31 detects the position of the object by the position sensor 19, calculates the distance of the object 2 as seen from the noise sensor 12 by the position distance conversion unit 191, and notifies the determination unit 24. The position sensor 19 and the position distance conversion unit 191 function as a distance calculation unit that calculates the distance between the object 2 and the noise sensor 12.

[0103] Subsequently, the determination unit 24 selects one of the noise determination models 23A (23a to 23d) based on the movement speed information and distance information of the object 2 as seen from the noise sensor 12 notified from the speed and distance measurement device 30D (step S3B).

[0104] Subsequently, the determination unit 24 acquires the feature amount of the noise, predicts the state of the object 2 using the acquired feature amount and the selected noise determination model 23A (step S4), and performs a state determination on the predicted state (step S5). That is, the determination unit 24 inputs the acquired feature amount into the selected noise determination model 23A, causes the noise determination model 23A to determine whether the object 2 is normal or abnormal, and obtains a determination result.

[0105] <<Fifth Embodiment>> Next, with reference to FIGS. 12 to 14, a noise detection system different from the fourth embodiment will be described, in which the state of the object 2 is determined based on the feature amount of the noise of the object, the position of the object, and the speed at which the object moves.

[0106] FIG. 12 is a configuration diagram of the noise detection system according to the fifth embodiment.

[0107] The noise detection system according to the fifth embodiment is different in that the determination device 20B includes a single noise determination model 23B instead of the plurality of noise determination models 23A (23a, 23b, 23c, 23d) as described in the second embodiment with reference to FIG. 5.

[0108] The odor determination model 23B of the fifth embodiment is a model that learns with the output values (or feature amounts) of the odor sensors 12a to 12d, the speed of movement of the object 2, and the position of the object 2 as explanatory variables and the state of the object 2 as the objective variable. The odor determination model 23B calculates an output value regarding the state obtained by machine learning based on the speed of movement of the object 2, the position of the object 2, and the output value (or feature amount) of the odor sensor 12. Then, if the state is less than the threshold value, the odor determination model 23B determines that the object is normal, and if the state is greater than or equal to the threshold value, the odor determination model 23B determines that the object is abnormal.

[0109] As shown in FIG. 13, the odor determination model 23B changes the speed (v1, v2, v3,...) and position (d1, d2, d3,...) of the object 2, and for each combination of speed and position, collects the output value (s1, s2, s3,...) of the odor sensor 12 and the state (a1, a2, a3,...) of the object 2.

[0110] Next, an operation example of the odor detection system according to the fifth embodiment of FIG. 12 will be described with reference to the flowchart of FIG. 14.

[0111] First, the determination unit 24 acquires the output values of the odor sensors 12a to 12d, and generates an odor feature amount based on the acquired output values (step S21).

[0112] On the other hand, a control unit 31 (not shown) of the speed / position measurement device 30F detects the speed of the object 2 by the speed sensor 14. Then, the control unit 31 calculates the speed of the object 2 as seen from the odor sensor 12 and notifies the determination unit 24. Further, the control unit 31 detects the position of the object by the position sensor 19, calculates the position of the object 2 as seen from the odor sensor 12, and notifies the determination unit 24.

[0113] Subsequently, the determination unit 24 acquires the speed and position of the object 2 as seen from the noise sensor 12 notified by the speed / position measurement device 30F. Then, the determination unit 24 performs state prediction (calculates the state) using the noise determination model 23B with the acquired speed and position of the object 2 and the feature amount of the noise as explanatory variables (step S22).

[0114] Subsequently, the determination unit 24 determines the state of the object 2 using the generated feature amount of the noise and the determination threshold by the noise determination model 23B (step S23). That is, the determination unit 24 predicts the state of the object 2 from the acquired feature amount of the noise and the calculated speed and position of the object 2 by the noise determination model 23B, determines whether the object 2 is normal or abnormal from the predicted state, and obtains a determination result.

[0115] According to the noise detection system of the fourth embodiment or the fifth embodiment, the noise determination model is learned with more explanatory variables, and the state is predicted by the noise determination model, so that the state of the object 2 can be determined with high accuracy.

[0116] Next, a specific application example of the noise detection system of the fourth embodiment or the noise detection system of the fifth embodiment will be described. FIG. 15 is an external view of a production line in which the object 2 is conveyed by a conveying device 50 to which the noise detection system of the embodiment is applied. The noise detection system determines whether the object 2 being conveyed is damaged by noise.

[0117] The conveying device 50 includes a plurality of rollers 51, and by rotating the rollers 51, the object 2 as a product placed on the rollers 51 is conveyed in the conveying direction x.

[0118] The odor detection device 10C is installed below the conveyor device 50. Although not shown in FIG. 15, the speed and position measurement device 30F in FIG. 9 or FIG. 12 is at the same coordinate as the odor detection device 10C in the conveyance direction x and is arranged at a position on the side of the conveyor device 50 at the same height as the object 2.

[0119] The camera 30E is arranged above the conveyor device 50, and in this embodiment, directly above the odor detection device 10C. The camera 30E only needs to be able to image the object 2 and detect its position and moving speed, and it may also be a sensor such as LiDAR, laser, or ultrasonic wave.

[0120] The camera 32E functions as a position sensor 19 and a speed sensor 14 for the object 2 to be state-determined. Specifically, the speed and position measurement device 30F extracts the object 2 from the captured image of the camera 32E and obtains the position of the object 2 as seen from the camera 32E. Then, based on the positional relationship between the camera 32E and the odor detection device 10C and the position of the object 2 as seen from the camera 32E, the position of the object 2 as seen from the odor detection device 10C is obtained and used as the position information of the object 2.

[0121] Also, the speed and distance measurement device 30F obtains the moving speed of the object 2 as seen from the camera 32E from the position change between frames of the captured image of the extracted object 2. Then, based on the positional relationship between the camera 32E and the odor detection device 10C and the moving speed of the object 2 as seen from the camera 32E, the moving speed of the object 2 as seen from the odor detection device 10C is obtained and used as the moving speed information of the object 2.

[0122] As described above, even in a production line where the placement position of the product as the object 2 changes and the conveyance speed of the conveyor device 50 changes, damage to the product can be detected with high accuracy.

[0123] Furthermore, the present invention is not limited to the above-described embodiments, and includes various modifications. The above embodiments have been described in detail for easy understanding of the present invention, and are not necessarily limited to those having all the configurations described. Also, a part of the configuration of one embodiment can be replaced with the configuration of another embodiment, and the configuration of another embodiment can also be added to the configuration of one embodiment.

Explanation of Reference Numerals

[0124] 1A, 1B, 1C Odor Detection System 2 Object 11 Control Unit (Speed Calculation Unit) 12, 12a, 12b, 12c, 12d Odor Sensors 14 Speed Sensor 23A, 23a, 23b, 23c Odor Judgment Models 23B Odor Judgment Model 24 Judgment Unit 30C Speed Measuring Device 30D Speed and Distance Measuring Device 30F Speed and Position Measuring Device 31 Control Unit (Speed Calculation Unit) 32C Speed Sensor

Claims

1. An odor sensor that reacts to the odor of an object, a speed calculation unit that calculates the moving speed of the object as seen from the odor sensor, a determination unit that determines the state of the object based on the output value of the odor sensor and the moving speed of the object, An odor detection system comprising the above.

2. The determination unit determines whether the object is abnormal, The odor detection system according to Claim 1.

3. The determination unit determines whether the object is damaged, The odor detection system according to Claim 2.

4. Comprising a plurality of odor determination models learned with the output value of the odor sensor as an explanatory variable and the state of the object as a target variable for each of the plurality of moving speeds, The determination unit selects any one of the odor determination models based on the moving speed calculated by the speed calculation unit, and determines the state of the object using the selected odor determination model, The odor detection system according to Claim 1.

5. Comprising an odor determination model learned with the moving speed and the output value of the odor sensor as explanatory variables and the state of the object as a target variable, The determination unit determines the state of the object using the odor determination model from the moving speed calculated by the speed calculation unit and the output value of the odor sensor, The odor detection system according to Claim 1.

6. Further comprising a distance calculation unit that calculates the distance between the object and the odor sensor, The determination unit determines the state of the object based on the output value of the odor sensor, the moving speed of the object, and the distance, The odor detection system according to Claim 1.

7. Comprising a plurality of odor determination models learned with the output value of the odor sensor as an explanatory variable and the state of the object as a target variable for each combination of the plurality of moving speeds and the distance, The determination unit selects any one of the odor determination models based on the moving speed and the distance calculated by the speed calculation unit, and determines the state of the object using the selected odor determination model, The odor detection system according to Claim 6.

8. Comprising an odor determination model learned with the moving speed, the distance, and the output value of the odor sensor as explanatory variables and the state of the object as a target variable, The determination unit determines the state of the object using the odor determination model based on the moving speed calculated by the speed calculation unit, the distance calculated by the distance calculation unit, and the output value of the odor sensor. The odor detection system according to claim 6.

9. Comprising a plurality of the odor sensors having different characteristics responsive to an odor. The determination unit identifies the object based on output values of the plurality of the odor sensors. The odor detection system according to claim 1.

10. An odor sensor reacting to an odor of an object; A speed sensor detecting a moving speed of the object; A speed calculation unit calculating a moving speed of the object as seen from the odor sensor; A determination unit determining a state of the object based on an output value of the odor sensor and the moving speed of the object; An odor detection method including the above.

11. Causing a computer to acquire an output value of an odor sensor that has reacted to an odor of an object; acquire a moving speed of the object detected by a speed sensor; a speed calculation unit calculate a moving speed of the object as seen from the odor sensor; a determination unit determine a state of the object based on an output value of the odor sensor and the moving speed of the object; An odor detection program for causing the above to be executed.

Images