Performance evaluation system, contaminant removal device, information processing device

Abstract

The present invention predicts an appropriate replacement time for an adsorbent. Provided is a performance evaluation system 100 in which a contaminant removal device 40 that removes a contaminant from the air and an information processing device 60 can communicate over a network. The contaminant removal device comprises an adsorbent that adsorbs the contaminant, an imaging means 50 that captures images of the adsorbent, and a communication device 51 that transmits image data for the adsorbent that has been captured by the imaging means to the information processing device. A control unit inputs a feature quantity into correspondence information that associates the feature quantity and the performance of the adsorbent to infer the performance of the adsorbent and predicts a replacement time for the adsorbent on the basis of the relationship between the time elapsed from when use of the adsorbent began and the inferred performance of the adsorbent.

Description

Performance evaluation system, contaminant removal device, information processing device

[0001] The present invention relates to a performance evaluation system, a contaminant removal device, and an information processing device.

[0002] There are spaces where airborne pollutants (chemical substances, suspended particulate matter, airborne microorganisms, NOx, etc.) are controlled to a specific cleanliness level or lower. For example, in clean rooms, art museums, and museums, various pollutants that may be present in the air are removed before the air is supplied to the room. Pollutant removal devices that remove pollutants from the air in this way are known. Pollutant removal devices remove pollutants by adsorbing them onto an adsorbent that is built into the device. The performance of the adsorbent decreases as more pollutants are adsorbed.

[0003] Techniques for diagnosing the deterioration of the performance of contaminant removal devices are known (see, for example, Patent Document 1). Patent Document 1 discloses a technique for evaluating the performance of an ion exchange resin for inspection from inspection images of the ion exchange resin for inspection, using a trained model that has been machine-learned based on training images of the ion exchange resin captured for training and evaluation results of the appearance of the ion exchange resin.

[0004] International Publication No. 2022 / 080416

[0005] However, conventional technology has made it difficult to accurately predict when to replace the adsorbent. For example, if the timing of adsorbent replacement is delayed, there is a risk that air containing a large amount of pollutants will be supplied to the room. On the other hand, if the adsorbent is replaced too early, excessive replacement is likely to occur.

[0006] In view of the above issues, this disclosure provides a technology for predicting the appropriate replacement time for adsorbents.

[0007] A first aspect of the present disclosure is a performance evaluation system in which a pollutant removal device for removing pollutants from the air and an information processing device can communicate via a network, wherein the pollutant removal device comprises an adsorbent for adsorbing the pollutants, an imaging means for imaging the adsorbent, and a communication device for transmitting image data of the adsorbent imaged by the imaging means to the information processing device, the information processing device comprises a control unit, the control unit analyzes the image data to obtain feature quantities relating to the performance of the adsorbent, inputs the feature quantities into correspondence information that associates the feature quantities with the performance of the adsorbent to infer the performance of the adsorbent, and predicts the time to replace the adsorbent based on the relationship between the elapsed time since the start of use of the adsorbent and the inferred performance of the adsorbent.

[0008] According to a first aspect of this disclosure, it is possible to predict the appropriate time for replacing the adsorbent.

[0009] A second aspect of this disclosure is the performance evaluation system described in the first aspect, wherein the characteristic quantity relating to the performance of the adsorbent is a characteristic quantity that changes according to the amount of salt precipitated on the adsorbent.

[0010] A third aspect of this disclosure is a performance evaluation system according to the first or second aspect, wherein the feature quantity is the amount of change in the color of the adsorbent, and the control unit inputs the amount of change in color to the corresponding information and infers the performance of the adsorbent.

[0011] A fourth aspect of this disclosure is a performance evaluation system according to the first or second aspect, wherein the feature quantity is a change in the mass or volume of the adsorbent, and the control unit inputs the change in the mass or volume of the adsorbent into the corresponding information to infer the performance of the adsorbent.

[0012] A fifth aspect of this disclosure is a performance evaluation system according to the first or second aspect, wherein the feature quantity is the amount of change in particle size information of the adsorbent, and the control unit inputs the amount of change in particle size information into the corresponding information to infer the performance of the adsorbent.

[0013] A sixth aspect of this disclosure is a performance evaluation system according to any of the first to fifth aspects, wherein the control unit notifies the terminal device of the predicted replacement time.

[0014] A seventh aspect of this disclosure is a performance evaluation system according to any of the first to sixth aspects, wherein the correspondence information is a first model obtained by machine learning the correspondence between the feature quantities and the performance of the adsorbent.

[0015] An eighth aspect of this disclosure is a performance evaluation system as described in the seventh aspect, wherein the pollutant removal device has the adsorbent placed in a first location and a second location, the communication device transmits ambient environmental data of the pollutant removal device to the information processing device, and the control unit inputs the performance of the adsorbent in the first location and ambient environmental data of the pollutant removal device, inferred by the first model, into a second model that has learned the correspondence between the performance of the adsorbent in the first location and ambient environmental data and the performance of the adsorbent in the second location, and the ambient environmental data of the pollutant removal device, to infer the performance of the adsorbent in the second location.

[0016] A ninth aspect of the present disclosure is a performance evaluation system according to the eighth aspect, wherein the control unit weights the performance of the adsorbent at the first location inferred by the first model by the airflow rate or the degradation rate of the adsorbent, weights the performance of the adsorbent at the second location by the airflow rate or the degradation rate of the adsorbent, and calculates the overall performance of the adsorbent in the pollutant removal device by integrating the weighted performance of the adsorbent at the first location and the performance of the adsorbent at the second location.

[0017] A tenth aspect of this disclosure is a performance evaluation system according to any one of the first to ninth aspects, wherein the imaging means is sensitive to the wavelength of light used to image the salt precipitated on the adsorbent.

[0018] An eleventh aspect of the present disclosure is a pollutant removal device for removing pollutants from the air, comprising: an adsorbent for adsorbing the pollutants; an imaging means for imaging the adsorbent; and a control unit, wherein the control unit analyzes image data of the adsorbent captured by the imaging means to obtain feature quantities relating to the performance of the adsorbent; inputs the feature quantities into correspondence information that associates the feature quantities with the performance of the adsorbent to infer the performance of the adsorbent; and predicts the time to replace the adsorbent based on the relationship between the elapsed time since the start of use of the adsorbent and the inferred performance of the adsorbent.

[0019] According to the eleventh aspect of this disclosure, the appropriate time for replacing the adsorbent can be predicted.

[0020] A twelfth aspect of the present disclosure is a pollutant removal device that removes pollutants from the air, comprising: an adsorbent for adsorbing pollutants; an imaging means for imaging the adsorbent; and a communication device for transmitting image data of the adsorbent imaged by the imaging means to an information processing device, and an information processing device that can communicate via a network, comprising: a control unit, the control unit analyzes the image data to obtain feature quantities relating to the performance of the adsorbent; inputs the feature quantities into correspondence information that associates the feature quantities with the performance of the adsorbent to infer the performance of the adsorbent; and predicts the replacement time of the adsorbent based on the relationship between the elapsed time since the start of use of the adsorbent and the inferred performance of the adsorbent.

[0021] According to a twelfth aspect of this disclosure, the appropriate time for replacing the adsorbent can be predicted.

[0022] This figure shows an example of the system configuration of a performance evaluation system. This is an external view of an example of a pollutant removal device. This is a side view of an example of a pollutant removal device. This is a hardware configuration diagram of an example of an edge device. This is a hardware configuration diagram of an example of a server device. This is a functional block diagram of an example of a server device in a performance evaluation system, illustrating the functions of the server device in the performance evaluation system in blocks. This is a functional block diagram of a learning device according to one embodiment. This is a functional block diagram of an analysis unit according to one embodiment. This figure shows an example of the relationship between elapsed time, image data, and pollutant removal efficiency, which was determined experimentally. This is a scatter plot of an example showing the correspondence between the number of white pixels and the pollutant removal efficiency. This is a diagram schematically showing the correspondence between the change in the number of white pixels and the ammonia removal efficiency. This is a flowchart of an example explaining the process by which the feature detection unit quantifies the number of white pixels. This is an image diagram illustrating the increase in mass and volume due to the adsorption of pollutants. This figure shows an example of image data of an adsorbent. This figure shows an example of particle size information of an adsorbent. This is a flowchart of an example explaining the process of detecting particle size information. This is a diagram schematically showing the correspondence between the change in mass or volume and the removal efficiency. This figure shows an example of a performance evaluation model composed of a neural network. This figure shows an example configuration of a performance evaluation model using CNN. This is an example graph showing the relationship between elapsed time and removal efficiency. This is an example sequence diagram explaining the process flow in which the performance evaluation system predicts the replacement time of the adsorption filter. This figure shows an example arrangement of a temperature sensor, a humidity sensor, and an airflow sensor. This is an example functional block diagram explaining the functions of the performance evaluation system by dividing them into blocks. This is a functional block diagram of a learning device according to one embodiment. This is a functional block diagram of an analysis unit according to one embodiment. This figure shows an example of a second-stage model configured with a neural network. This figure shows a removal efficiency curve of an example adsorption filter against elapsed time. This is an example sequence diagram explaining the process flow in which the performance evaluation system predicts the replacement time of the adsorption filter.

[0023] Below, we will describe a performance evaluation system and the evaluation method performed by the performance evaluation system as an example of how to implement this disclosure.

[0024] <Performance Evaluation of General Adsorbents> Contaminant removal devices remove contaminants by adsorbing them onto an internal adsorbent. The performance of the adsorbent is evaluated by a service technician taking a sample of the adsorbent on-site and requesting the manufacturer to evaluate its performance. Performance refers to, for example, the efficiency of contaminant removal. This method involves costs for on-site sample collection and costs for performance evaluation by the manufacturer.

[0025] Furthermore, because there is a relatively long gap between the previous and current performance evaluations, it is difficult to continuously monitor the performance of the adsorbent. If the performance deteriorates between the previous and current evaluations, there is a risk that air containing pollutants that should be removed will be supplied to the room.

[0026] Furthermore, even if performance evaluation results indicate that the adsorbent has not deteriorated enough to warrant replacement, excessive replacement is likely to occur if further performance deterioration is expected before the next performance evaluation. Conversely, if it is expected that performance will not deteriorate before the next performance evaluation, there is a risk of delays in replacement.

[0027] [First Embodiment] In this disclosure, the performance evaluation system is equipped with a camera (imaging means) in the pollutant removal device, and the camera continuously transmits image data to a server device. The server device evaluates the performance of the adsorbent by analyzing the image data. The following is a description of the workflow or operation of the performance evaluation system. In the following description, ammonia is used as an example of a pollutant. Ammonia exists in the air as a gas. The adsorbent is sometimes referred to as a gas adsorbent.

[0028] (1) The adsorbent precipitates ammonium salts (hereinafter simply referred to as salts) upon adsorption of ammonia. When salt precipitates, the surface of the adsorbent turns white (changes color), and its mass increases. It is known that the performance of the adsorbent decreases as more ammonia is adsorbed.

[0029] (2) A camera for imaging the adsorbent is located inside the contaminant removal device. The camera repeatedly images the adsorbent at a preset frequency. The image data of the adsorbent is transmitted to a server device.

[0030] (3) The server device analyzes the image data and quantifies the color and particle size of the adsorbent as features. When ammonia is adsorbed onto the adsorbent, salt precipitates and the adsorbent turns white over time. The amount of white color correlates with the amount of salt precipitated. Similarly, when ammonia is adsorbed onto the adsorbent, the mass and volume of the adsorbent increase. An increase in mass or volume manifests as an increase in particle size information (diameter, surface area, etc.). Therefore, the server device can estimate the performance of the adsorbent based on its color, mass or volume, or particle size information.

[0031] (4) The server device uses a performance evaluation model (an example of the first model) that has learned the correspondence between features and performance to evaluate the performance of the adsorbent each time image data is transmitted. Since the performance evaluation is obtained over time, the server device can estimate the replacement time when the performance is expected to fall below a certain level.

[0032] (5) The server device predicts when the adsorbent needs to be replaced and automatically notifies the person in charge via email or other means.

[0033] Thus, the performance evaluation system of this disclosure has the following advantages: (i) The server device can evaluate the performance of the adsorbent remotely, allowing service personnel to predict when the adsorbent needs to be replaced without having to go to the site, thus reducing costs. (ii) The server device can continuously evaluate the performance of the adsorbent, making it easier to determine the appropriate replacement time. (iii) The server device can predict when the replacement time needs to be replaced, thus preventing excessive replacements or delays in replacement.

[0034] <About Terminology> Air pollutants refer to substances that are not present in pure air. Pollutants include chemical substances, suspended particulate matter, suspended microorganisms, NOx, etc. Specifically, pollutants include nitric oxide, sulfur oxides, nitrogen dioxide, sulfur compounds, methyl mercaptan, alkaline gases such as ammonia or amines, acidic gases such as hydrochloric acid, nitric acid, sulfuric acid or acetic acid, ozone, chlorine, organic solvents, or organic gases.

[0035] Adsorbents are substances that attract and store pollutants. The adsorbent may vary depending on the pollutant to be adsorbed. Examples of adsorbents include potassium permanganate, phosphoric acid, calcium hydroxide, potassium carbonate, manganese dioxide, activated alumina, activated carbon, zeolite, and activated carbon.

[0036] The features relating to the performance of the adsorbent are features obtained by analyzing image data and serve as indicators of the adsorbent's performance. In this disclosure, the features relating to the performance of the adsorbent are, as an example, features that correlate with the amount of salt precipitated. For example, the amount of change in color, the amount of change in particle size, or the amount of change in volume may be features relating to the performance of the adsorbent. Depending on the combination of the adsorbent and the pollutant, substances other than salt may precipitate, and features corresponding to the precipitated substance may be adopted.

[0037] <System Configuration of Performance Evaluation System> The system configuration of the performance evaluation system 100 will be explained with reference to Figure 1. Figure 1 is a diagram showing an example of the system configuration of the performance evaluation system 100.

[0038] The performance evaluation system 100 provides a variety of IoT-based services to users, from administrators to general users, by enabling communication between various devices 30 and a cloud-side server device 60 via a network N. The edge devices 10, devices 30, sensor switches 20, and user terminals 5 are mainly located on the customer's side, while the server device 60 is located in a cloud such as a data center or the internet.

[0039] The device 30 generally refers to all devices that consume power, such as a pollutant removal device 40, an air conditioner, security equipment, heat source equipment, a fire alarm, an AHU (Air Handling Unit), a power meter, lighting, etc. The device 30 may also be other devices. The sensor switches 20 include various sensors, lamps, relays, etc. The camera 50 may be included in the sensor switches 20, or the camera 50 may be built into the device 30. The device 30 and the sensor switches 20 are communicably connected to the edge device 10 via a dedicated cable or a network such as a LAN. The device 30 and the sensor switches 20 may be communicably connected to the edge device 10 by wireless communication.

[0040] The pollutant removal device 40 is a device that removes pollutants in the air. What constitutes a pollutant may vary depending on the purpose of use of the space. The installation location of the pollutant removal device 40 is, for example, a building, a commercial facility, an art museum, a museum, a food factory, a pharmaceutical factory, an electronics factory, a power generation facility, a nuclear disaster countermeasure facility, etc., but is not limited thereto.

[0041] The device 30 and the sensor switches 20 are controlled by the edge device 10. In other words, the edge device 10 performs the required operations on the device 30 and the sensor switches 20 so as to conform to the purposes of the device 30 and the sensor switches 2,0. The content of the control varies depending on the types of the device 30 and the sensor switches 20. For example, when the device 30 is an air conditioner, all controls related to the functions of the air conditioner, such as the cooling and heating modes, set temperature, air volume, humidity, air direction, etc., that can generally be set on the air conditioner may be included. Also, the control includes operation modes such as a pre-season inspection dedicated mode, microcomputer reset, operation stop, and function substitution. Further, when the device 30 is the pollutant removal device 40, the control for setting the imaging timing for the camera 50 may be included.

[0042] The device 30 collects operation data corresponding to the device 30 and mainly transmits it to the edge device 10 regularly. Regularly means, for example, once per minute, once per ten minutes, once per sixty minutes, etc., but it may be set by the user or the server device 60. Also, upon request from the edge device 10 or the user terminal 5, the device 30 can transmit the operation data to the edge device 10. The operation data varies depending on the device 30. For example, in the case of an air conditioner, it includes various data such as the high-pressure and low-pressure of the refrigerant, the refrigerant temperature, the rotation speed of the fan, and the CPU temperature of the microcomputer.

[0043] Also, when the device 30 detects an abnormality, it transmits an abnormality code to the edge device 10. The device 30 that has detected an abnormality stops operating. The edge device 10 transmits the abnormality code to the server device 60. The same processing by the edge device 10 can be applied to the sensor switches 20. The sensor switches 20 mainly transmit their own information and abnormality codes to the edge device 10 regularly. The camera 50 transmits image data to the edge device 10. The timing of transmission is regular, a set date and time, a time zone with low load, a difference from the last image data being a certain amount or more, etc.

[0044] The edge device 10 is a controller that controls the device 30 and the sensor switches 20. The edge device 10 has functions as a control device that controls the device 30 and the sensor switches, an information processing device that processes operation data, etc., and a communication means that communicates with the server device 60. The edge device 10, for example, transmits various information from the device 30 to the server device 60 and receives an instruction according to the information from the server device 60. Alternatively, even if the edge device 10 does not transmit any information to the server device 60, it can receive an instruction from the server device 60 (for example, when there is an instruction from the user terminal 5 to the server device 60). The edge device 10 converts the instruction into an appropriate instruction according to the models of the device 30 and the sensor switches 20 and transmits it to the device 30 and the sensor switches 20.

[0045] A server device 60 is one or more information processing devices. Although Figure 1 shows one server device 60, the server devices 60 may be divided into several units according to their functions. Alternatively, the functions of the server devices 60 may be consolidated into a single information processing device. Furthermore, multiple server devices 60 with the same functions may be provided, and these multiple server devices 60 may communicate with each other to process data, like a server cluster.

[0046] The server device 60 receives various information transmitted from the edge device 10 via the network N and generates necessary instructions. For example, the server device 60 evaluates the performance of the adsorbent image data using a performance evaluation model. Based on the evaluated performance, the server device 60 predicts the replacement time and transmits it to the user terminal 5. In addition, in response to an abnormal code from the edge device 10, the server device 60 instructs the edge device 10 to perform emergency operation regardless of the model of the equipment 30. Furthermore, the server device 60 can also transmit instructions to the edge device 10 to the equipment 30 according to the schedule and operations set by the user terminal 5.

[0047] Instead of the server device 60 evaluating the performance of the adsorbent, the contaminant removal device 40 may evaluate the performance of the adsorbent. The contaminant removal device 40 predicts the replacement time based on the evaluated performance and transmits it to the user terminal 5. Alternatively, the edge device 10 may have a performance evaluation model.

[0048] The server device 60 also has the functionality of a web server. The web server responds to requests from client software (web clients) such as a web browser operated by the user and provides the client with screen information written in HTML files, XML, CSS files, JavaScript (registered trademark), etc. An application that uses the web mechanism in this way is called a web application.

[0049] Furthermore, it is preferable that the server device 60 supports cloud computing. Cloud computing refers to a usage model in which network resources are utilized without the user being aware of specific hardware resources. Cloud computing provides users with data and software that they previously used on their own computers, as a service via the network. Users can access a variety of services from any device by providing a web browser that runs on a personal computer or mobile device, and an internet connection.

[0050] User terminal 5 is a terminal device that displays various screens provided by server device 60. User terminal 5 may be used by an administrator or by a general user. Administrators include customer-side administrators and management system administrators, but this disclosure does not distinguish between them. Furthermore, administrators are those who perform maintenance and management that are not performed by general users who use equipment 30 on a daily basis.

[0051] The user terminal 5 displays a wide variety of screens, but one example is a screen that identifies the contaminant removal device 40 and notifies the user of its replacement time. Other screens include a list of the devices 30 and sensor switches 20 connected to the customer's edge device 10, an internal company map showing the locations of the devices 30 and sensor switches 20, and operation screens for operating the devices 30 and sensors.

[0052] User terminal 5 can be, for example, a PC (Personal Computer), smartphone, tablet, PDA (Personal Digital Assistant), or wearable PC (sunglasses type, wristwatch type, etc.). However, it only needs to have communication capabilities and be able to run a web browser. Alternatively, instead of a web browser, a dedicated native application for the performance evaluation system may run on user terminal 5.

[0053] <Appearance of the Contaminant Removal Device> The contaminant removal device 40 will be described using Figures 2 and 3. Figure 2 shows an external view of an example of the contaminant removal device 40. Figure 3 shows a side view of an example of the contaminant removal device 40. The contaminant removal device 40 has a cubic appearance, and one or more drawers 25 to 27 can be attached and detached from the front. In the figure, there are three drawers 25 to 27, and the number of drawers is just one example. Adsorption filters 31A, 31B, and 31C, each filled with adsorbent 32, are installed in each of the drawers 25 to 27. Adsorption filter 31A is called the first-stage (an example of a first placement location) adsorption filter, adsorption filter 31B is called the second-stage (an example of a second placement location) adsorption filter, and adsorption filter 31C is called the third-stage adsorption filter. Since the adsorption filters 31A to 31C are placed in drawers 25 to 27, they are replaceable. Hereinafter, any of the adsorption filters 31A, 31B, and 31C will be referred to as adsorption filter 31.

[0054] In Figure 3, drawers 25 to 27 are arranged in a single vertical row, but the contaminant removal device 40 may have multiple rows of drawers.

[0055] In Figure 3, the arrows indicate the airflow. Air is supplied from the front of the pollutant removal device 40, passes through the adsorption filters 31A, 31B, and 31C, and flows out from the vent 33 on the rear. As the air passes through the adsorption filters 31, pollutants (e.g., ammonia) are adsorbed by the adsorbent 32. Although Figure 3 shows three adsorption filters 31, this number is just an example. It can be expected that the more adsorption filters 31 there are, the greater the amount of pollutants adsorbed per short time. Also, by changing the type of adsorbent 32 for each adsorption filter 31, different types of pollutants can be adsorbed. In this disclosure, it is assumed that the adsorbent 32 in adsorption filters 31A, 31B, and 31C is of the same type.

[0056] The airflow may flow in from the rear and out onto the surface, or it may flow in from one side and out from the other side. The airflow may also be a combination of these. Furthermore, the shape of the pollutant removal device 40 is just an example.

[0057] The contaminant removal device 40 has a camera 50 for imaging the adsorbent 32 and a communication device 51. For example, as shown in Figure 3, the camera 50 is installed on the ceiling of the contaminant removal device 40 with its optical axis pointing directly downwards. In addition, the camera 50 only needs to be installed with its optical axis oriented so that the adsorption filter 31 is within its field of view. At least one camera 50 is installed. A camera 50 may be installed for each adsorption filter 31A to 31C, or if it is known that the performance of each adsorption filter 31A to 31C deteriorates to a similar extent, a camera 50 may be installed for only one of the three adsorption filters 31A to 31C, namely adsorption filter 31A. In Figure 3, a camera 50 is installed for the first stage adsorption filter, but it may also be installed for the second or third stage.

[0058] One of the features of this disclosure is that it predicts when the adsorption filter 31 needs to be replaced, but it is preferable to replace the adsorption filters 31A to 31C at the same time. This is because when replacing them, the customer or service technician needs to work in the room where the contaminant removal device 40 is installed, and preparations such as removing contaminants from the body are necessary for working in this room.

[0059] If a camera 50 is installed on only one adsorption filter 31A, the replacement timing for the adsorption filters 31B and 31C, which do not have cameras 50 installed, is estimated by the model. The server device 60 evaluates the performance of the adsorption filters 31B and 31C by taking surrounding environment data (e.g., air temperature, humidity, airflow, etc.) along with image data as input to the model. The server device 60 calculates the overall performance of the adsorption filters 31A to 31C by integrating the performance of the adsorption filters 31A to 31C and predicts the replacement timing so that the adsorption filters 31A to 31C can be replaced at the same time.

[0060] Camera 50 is sensitive to wavelengths of light that can image the salt precipitated on the adsorbent 32. Camera 50 may have a light source that emits one or more of the following: visible light, near-infrared light, or ultraviolet light. Near-infrared light is light on the longer wavelength side of the wavelength range outside the visible light range and may capture discoloration that cannot be detected with visible light. Ultraviolet light is light on the shorter wavelength side of the wavelength range outside the visible light range and may capture discoloration that cannot be detected with visible light.

[0061] The camera 50 is either integrated with the communication device 51 or connected to the communication device 51. The communication device 51 transmits the image data of the adsorbent 32 captured by the camera 50 to the edge device 10 in real time. The edge device 10 transmits the image data to the server device 60 in real time. Alternatively, the communication device 51 transmits directly to the server device 60. The image data may be video.

[0062] <Hardware Configuration of Edge Device and Server Device> Next, the hardware configuration of the edge device 10 will be described with reference to Figure 4. Figure 4 is a diagram showing an example of the hardware configuration of the edge device 10. As shown in Figure 4, the edge device 10 has a processor 201, memory 202, auxiliary storage device 203, I / F (Interface) device 204, communication module 205, and drive device 206. Each piece of hardware in the edge device 10 is interconnected via a bus 207.

[0063] The processor 201 has various computing devices such as a CPU (Central Processing Unit). The processor 201 reads various programs into the memory 202 and executes them. The processor 201 corresponds to the control unit 110 that controls the entire edge device 10.

[0064] Memory 202 has main memory devices such as ROM (Read Only Memory) and RAM (Random Access Memory). The processor 201 and memory 202 form a so-called computer, and the processor 201 executes various programs read into memory 202.

[0065] The auxiliary storage device 203 stores various programs and various data used when those programs are executed by the processor 201.

[0066] The I / F device 204 is a connection device that connects the edge device 10 to an example of an external device, such as equipment 30 and sensor switches 20.

[0067] The communication module 205 is a communication device for communicating with the server device 60 via the network N.

[0068] The drive device 206 is a device for setting the recording medium 208. The recording medium 208 here includes media that record information optically, electrically, or magnetically, such as CD-ROMs, flexible disks, and magneto-optical disks. The recording medium 208 may also include semiconductor memory that records information electrically, such as ROMs and flash memory.

[0069] The various programs to be installed on the auxiliary storage device 203 are installed, for example, when the distributed recording medium 208 is set in the drive device 206 and the various programs recorded on the recording medium 208 are read by the drive device 206. Alternatively, the various programs to be installed on the auxiliary storage device 203 may be installed by downloading them from the network N via the communication module 205.

[0070] On the other hand, Figure 5 shows an example of the hardware configuration of the server device 60. Since the hardware configuration of the server device 60 is generally the same as that of the edge device 10, this explanation will focus on the differences from the hardware configuration of the edge device 10. Furthermore, the hardware configuration of the learning device 500, which will be described later, is the same as that of the server device 60.

[0071] The processor 221 reads various programs into the memory 222 and executes them. The processor 221 corresponds to the control unit 210 that controls the entire server device 60.

[0072] The I / F device 224 is a connection device that connects the server device 60 to an external device, such as a display device 230 and an operating device 240. The display device 230 displays the internal status of the server device 60. The operating device 240 is used by the administrator of the server device 60 to input various instructions to the server device 60.

[0073] The communication module 225 is a communication device for communicating with the edge device 10 and the user terminal 5 via the network N.

[0074] <About the Functions> Next, with reference to Figure 6, the functional configuration of each device in the performance evaluation system 100 will be explained in detail. Figure 6 is an example of a functional block diagram that explains the functions of the server device 60 in the performance evaluation system 100 by dividing them into blocks. The server device 60, the contaminant removal device 40, and the learning device 500 are connected to each other via the network N so as to be able to communicate. Note that the edge device 10 is omitted in Figure 6.

[0075] <<Contaminant Removal Device>> A camera 50 installed in the contaminant removal device 40 captures image data of the adsorbent 32 and transmits it to the communication device 51. The communication device 51 transmits the image data of the adsorbent 32 captured by the camera 50 to the server device 60. The communication device 51 may also instruct the camera 50 on the timing of image capture. This timing may be set by the customer operating the user terminal 5 to the server device 60, and the server device 60 instructs the communication device 51, or it may be set directly and manually on the communication device 51.

[0076] Although the camera 50 and communication device 51 are shown as separate components in the diagram, the camera 50 and communication device 51 may be configured as an integrated module. Furthermore, at least one of the communication device 51 and camera 50 may be located outside the contaminant removal device 40.

[0077] Furthermore, if the contaminant removal device 40, rather than the server device 60, is the one predicting the replacement time for the adsorption filter 31, the contaminant removal device 40 has the same analysis unit 62, prediction unit 63, and notification unit 64 as the server device 60.

[0078] <<Learning Device>> The learning device 500 is an information processing device that learns the correspondence between image data features and performance during the learning phase of machine learning. The learning device 500 is an arbitrary information processing device. The learning device 500 generates a performance evaluation model. The server device 60 may also have the functions of the learning device 500. Details of the functions of the learning device 500 will be described later.

[0079] <<Server Device>> The server device 60 includes a communication unit 61, an analysis unit 62, a prediction unit 63, and a notification unit 64. Each of these units in the server device 60 is a function or means realized by any of the components shown in Figure 5 operating according to instructions from the processor 221 in accordance with a program expanded from the auxiliary storage device 223 to the memory 222.

[0080] The communication unit 61 communicates with the edge device 10 (or may communicate directly with the communication device 51) via the network N. In this disclosure, the communication unit 61 receives image data of the suction filter 31A captured by the camera 50.

[0081] The analysis unit 62 analyzes the image data to evaluate the current performance. One example of performance is the efficiency of pollutant removal. Specifically, the analysis unit 62 analyzes the image data and outputs the efficiency of pollutant removal. Details of the functions of the analysis unit 62 will be described later.

[0082] The prediction unit 63 records the removal efficiency of pollutants in a time series and calculates an approximate curve from the change in removal efficiency with respect to the elapsed time since the start of use of the adsorbent. The prediction unit 63 extrapolates the approximate curve and predicts the time when the removal efficiency falls below a threshold as the replacement time.

[0083] The notification unit 64 notifies the person in charge of the replacement time. The person in charge may be notified of the replacement time each time it is predicted, or when the remaining time until the replacement time falls below a certain level.

[0084] <About the Functions of the Learning Device> Figure 7 is a functional block diagram of a learning device 500 according to one embodiment of the present disclosure. The learning device 500 includes a learning data acquisition unit 502, a learning data storage unit 503, a feature detection unit 504, and a learning unit 505. These functional units of the learning device 500 are functions or means realized by the processor 221 of the learning device 500 executing program instructions expanded in the memory 222. The learning device 500 is used in the learning phase of machine learning.

[0085] The learning data acquisition unit 502 acquires learning data 501. The learning data 501 consists of multiple sets of image data and pollutant removal efficiency (performance), with each set comprising image data and pollutant removal efficiency.

[0086] The learning data storage unit 503 stores the learning data 501 acquired by the learning data acquisition unit 502.

[0087] The feature detection unit 504 analyzes the image data and first quantifies the following: 1. Number of white pixels 2. Particle size information (area, perimeter, maximum diameter, minimum diameter, average particle size) The absolute values ​​of the number of white pixels and particle size information may vary even in the initial state immediately after the adsorption filter 31 is attached to the contaminant removal device 40. Therefore, in this disclosure, the amount of change in the number of white pixels and particle size information is used as features. Features may be referred to as explanatory variables, and removal efficiency as the target variable (sometimes called training data). The learning device 500 may learn the correspondence between the image data itself and the removal efficiency without extracting features from the image data.

[0088] The learning unit 505 learns from training data using various machine learning algorithms to generate a performance evaluation model 506. The performance evaluation model 506 is correspondence information that associates image data features with removal efficiency. In other words, the performance evaluation model 506 outputs removal efficiency for each input of image data features.

[0089] Machine learning is a technique for enabling computers to acquire human-like learning abilities. It involves a computer autonomously generating algorithms necessary for data identification and other judgments from pre-introduced training data, and then applying these algorithms to new data to make predictions. The learning method for machine learning can be supervised learning, unsupervised learning, semi-supervised learning, reinforcement learning, or deep learning, or a combination of these methods; the learning method for machine learning is not limited. The performance evaluation model 506 can be implemented using a regression model, which is a type of supervised learning. Regression models include multiple regression, neural networks, ridge regression, lasso regression, elastic network regression, etc., and are not limited to the methods described in this disclosure.

[0090] <About the Functions of the Analysis Unit> Next, with reference to Figure 8, the functions of the analysis unit 62 will be described. Figure 8 is a functional block diagram of the analysis unit 62 according to one embodiment of the present disclosure. The analysis unit 62 includes an image data acquisition unit 512, a feature detection unit 504, and an inference unit 513. The analysis unit 62 is used in the inference phase of machine learning.

[0091] The image data acquisition unit 512 acquires image data captured by the camera 50 from the communication device 51. The image data acquisition unit 512 may receive the image data from the communication device 51, or it may acquire image data stored in the auxiliary storage device 223. Since the performance of the adsorbent 32 does not deteriorate rapidly, performance evaluation may be performed as a batch process at a fixed time each day.

[0092] The feature detection unit 504 detects features from image data, similar to the learning device 500.

[0093] The inference unit 513 inputs the features detected from the image data into the performance evaluation model 506 and outputs the pollutant removal efficiency 514. The pollutant removal efficiency 514 is estimated to be the performance of the pollutant removal device 40.

[0094] <Relationship between change in the number of white pixels and removal efficiency> Next, we will explain the relationship between the change in the number of white pixels and the removal efficiency. When ammonia, for example, is adsorbed onto the adsorbent 32, salt precipitates over time and the adsorbent becomes discolored white. Since the performance decreases when ammonia is adsorbed, the performance of the pollutant removal device 40 can be estimated based on the color of the adsorbent 32.

[0095] Figure 9 shows an example of the relationship between the elapsed time since the start of adsorption use, image data, and the removal efficiency of contaminants, as experimentally determined. The elapsed time is the time elapsed since the adsorption filter 31 was attached to the contaminant removal device 40. The removal efficiency is a reliable value, such as a value measured by the manufacturer. The correspondence between elapsed time, the number of white pixels, and the removal efficiency is as follows: 15 minutes later: 28,531 pixels, 99% 30 minutes later: 30,949 pixels, 98% 60 minutes later: 30,957 pixels, 93% 120 minutes later: 32,395 pixels, 42% 180 minutes later: 32,469 pixels, 0% Thus, the performance of the adsorbent 32 is correlated with the discoloration state of the adsorbent 32. Note that the experimental results in Figure 9 are from an accelerated test. In reality, it takes six months to more than a year for the removal efficiency to reach 0%.

[0096] Figure 10 is a scatter plot showing the correspondence between the number of white pixels and the removal efficiency of contaminants. Figure 10 also shows an approximation curve that approximates the correspondence between the number of white pixels and the removal efficiency. According to Figure 10, the performance gradually decreases as the number of white pixels increases, and then the performance decreases sharply as the number of white pixels increases further. Note that such an approximation may differ depending on the combination of adsorbent and contaminant. If the learning unit 505 learns the correspondence between the number of white pixels and the removal efficiency using a regression model, it can infer the current removal efficiency from the number of white pixels. However, even if the elapsed time is 0 minutes, the number of white pixels will be a different value each time the adsorption filter 31 is attached. Therefore, in this disclosure, the change in the number of white pixels from the initial state is used as a feature.

[0097] In this disclosure, the change in the number of white pixels is used as a feature, but the change in each grayscale intensity may be used as a feature, rather than just white or black. Alternatively, a color histogram may be created and the change in each color may be used as a feature.

[0098] Figure 11 schematically shows the correspondence between the change in the number of white pixels and the ammonia removal efficiency. Note that Figure 11 is not an experimental result, but rather an estimated correspondence between the change in the number of white pixels and the removal efficiency (performance). The change in the number of white pixels is calculated as follows: Change in the number of white pixels = (Current number of white pixels - Number of white pixels in the initial state at 0 minutes elapsed) / Number of white pixels in the initial state at 0 minutes elapsed. The learning unit 505 generates a performance evaluation model 506 by learning the correspondence between the change in the number of white pixels and the removal efficiency of pollutants using a regression model. The performance evaluation model 506 outputs the removal efficiency of the adsorption filter 31 located in the pollutant removal device 40. Note that, as will be described later, the learning unit 505 may learn not only the change in the number of white pixels, but also the correspondence between the change in the number of white pixels and the change in mass (or volume) and the removal efficiency of pollutants.

[0099] Furthermore, the feature detection unit 504 may quantify the ratio of white pixels to the total number of pixels instead of the number of white pixels. If the correspondence between the change in the number of white pixels and the removal efficiency is learned, the learning device 500 may need to retrain if the type or performance of the camera 50 changes. In contrast, the ratio of white pixels to the total number of pixels does not change easily even if the type or performance of the camera 50 changes, thus reducing the need for retraining.

[0100] <Method for detecting the number of white pixels> Figure 12 is a flowchart illustrating the process by which the feature detection unit 504 quantifies the number of white pixels. The method for quantifying the number of white pixels can be the same in the learning phase and the inference phase.

[0101] First, the feature detection unit 504 converts the color image data captured by the camera 50 into grayscale (S1).

[0102] Next, the feature detection unit 504 binarizes the grayscale image data at a predetermined threshold (S2). Binarization methods include static binarization, variable thresholding, and Otsu's binarization, and an appropriate method is adopted.

[0103] The feature detection unit 504 counts the number of white pixels (S3). The number of white pixels can be counted as the number of pixels themselves, or multiple pixels can be treated as one block, and if a certain number of pixels or more within a block are white, the entire block can be counted as one white pixel.

[0104] <Relationship between change in mass and removal efficiency> Next, the relationship between the change in mass or volume and the removal efficiency will be explained. Since mass or volume cannot be directly detected from image data, in this disclosure, the feature detection unit 504 digitizes particle size information from the image data and converts the particle size information into mass or volume.

[0105] Figure 13 is an illustrative diagram illustrating the increase in mass and volume due to the adsorption of pollutants. As shown in Figure 13, when ammonia is adsorbed onto the adsorbent particles 125, a reaction occurs to produce salt 123 and water, increasing the mass and volume of the particles 125.

[0106] The mass or volume correlates with particle size information. Particle size information includes the area, perimeter, maximum diameter, minimum diameter, and average particle size of each individual particle of the adsorbent 32. Since the performance of the adsorbent 32 deteriorates when ammonia is adsorbed onto it, the performance of the adsorbent 32 correlates with the mass or volume of the adsorbent 32, or with the particle size information.

[0107] Figure 14A shows an example of image data of the adsorbent 32. Figure 14A is image data of the adsorbent 32, captured with a general-purpose camera 50 (visible light). Figure 14B is the particle size distribution of the adsorbent 32, a histogram of diameters with a BIN width of 10 mm. As will be described later, each individual particle of the adsorbent 32 is detected by image processing, so the area, perimeter, maximum diameter, minimum diameter, and average particle size can be determined.

[0108] Figure 15 is a flowchart illustrating the process for detecting granularity information.

[0109] First, the feature detection unit 504 performs image preprocessing on the image data (S11). For example, the feature detection unit 504 removes noise from the image data and adjusts the contrast and brightness to perform preprocessing in order to facilitate particle identification.

[0110] Next, the feature detection unit 504 uses segmentation techniques to separate the particles of the adsorbent 32 individually (S12). Known segmentation techniques include the watershed method and thresholding. The watershed method is an image processing algorithm that separates (segments) objects that come into contact. Thresholding is an algorithm that generates a binary image divided into pixels with intensity below a certain threshold and pixels with intensity above a certain threshold. Since these methods are existing or publicly known, their explanation will be omitted.

[0111] Next, the feature detection unit 504 calculates the area, perimeter, maximum diameter, minimum diameter, and average particle size of each individual particle separated by segmentation (S13). The area is calculated by the number of pixels occupying one particle, the perimeter by the number of pixels on the outer edge of one particle, and the diameter by the length of the straight line passing through the center of one particle, etc., through image processing.

[0112] Next, the feature detection unit 504 uses one or more of these particle size information to convert them into mass or volume (S14). In this disclosure, it is assumed that correlation coefficients p and q are obtained that associate particle size information with mass or volume. The feature detection unit 504 uses the correlation coefficients p and q to convert particle size information into mass or volume. Note that one or more of the following may be used as particle size information: area, perimeter, maximum diameter, minimum diameter, and average particle size. Mass = p × particle size information Volume = q × particle size information Note that the correlation coefficient may be obtained by machine learning. That is, a model that has learned the correspondence between particle size information and mass or volume takes particle size information as input and outputs mass or volume.

[0113] If the learning unit 505 learns the correspondence between mass or volume and removal efficiency using a regression model, it can infer the current removal efficiency from the mass or volume. However, the mass or volume of the adsorbent 32 packed in the adsorption filter 31 is unlikely to be the same value even in the initial state. Therefore, in this disclosure, the change in mass or volume from the initial state is used as a feature.

[0114] Figure 16 schematically shows the correspondence between the change in mass or volume and the removal efficiency. Note that Figure 16 is not an experimental result, but rather an estimated correspondence between the change in mass or volume and the removal efficiency (performance). The change in mass is calculated as follows: Change in mass = (Current mass - Mass in the initial state at 0 minutes elapsed) / The change in mass and volume in the initial state at 0 minutes elapsed is calculated as follows: Change in volume = (Current volume - Volume in the initial state at 0 minutes elapsed) / Volume in the initial state at 0 minutes elapsed The learning unit 505 generates a performance evaluation model 506 by learning the correspondence between the change in mass or volume and the removal efficiency of contaminants using a regression model. The learning unit 505 may also learn the correspondence between the change in the number of white pixels and the change in mass (or volume) and the removal efficiency of contaminants. In this way, only one performance evaluation model 506 is needed, which reduces the processing load.

[0115] <Example of a performance evaluation model> The performance evaluation model 506 obtained by regression analysis is as follows:

[0116] Let f(x) be the removal efficiency, Φ1(x) be the change in the number of white pixels, Φ2(x) be the change in mass or volume, and w0 be the bias. Φ(x) is the basis function. By determining the coefficients w0 to w2 using the least squares method, a performance evaluation model 506 can be generated.

[0117] Alternatively, a performance evaluation model 506 may be generated using a neural network, as shown in Figure 17. Figure 17 shows a performance evaluation model 506 configured with a neural network 170. The neural network 170 is a form of artificial intelligence (AI) model that trains a computer to process data in a way that mimics the workings of the human brain. An existing configuration will be used for the neural network 170.

[0118] The neural network 170 corresponds to the performance evaluation model 506. The neural network 170 has an input layer 171, a hidden layer 172, and an output layer 173. In the neural network 170 of Figure 17, the L layer is fully connected from node 175 of the input layer 171 to node 176 of the output layer 173. A neural network with deep layers is called a DNN (Deep Neural Network). The layer between the input layer 171 and the output layer 173 is called the hidden layer 172. The number of nodes in the hidden layer 172 is merely an example. The input to the input layer 171 is the change in the number of white pixels, the change in mass or volume, and a bias (set to 1). The output from the output layer 173 is the efficiency of pollutant removal.

[0119] Weights are assigned to the connections between nodes, and the output from a node multiplied by its weight is transmitted to the node in the next layer. The node in the next layer receives the outputs of all the nodes in the previous layer, so it sums up the outputs of all the nodes in the previous layer. The node in the next layer activates the summed output with an activation function and transmits it to the next node. This process is repeated until the values ​​are transmitted up to output layer 173.

[0120] In this embodiment, since we want to infer the removal efficiency, the neural network 170 is a regression model (a classification model is another option). For this reason, the output layer 173 is provided with one node 176 that outputs the removal efficiency.

[0121] It should be assumed that the neural network 170 was trained using existing methods such as backpropagation. In other words, the performance evaluation model 506 learns the correspondence between the change in the number of white pixels and the change in mass or volume, and the removal efficiency. However, only the change in the number of white pixels or the change in mass or volume may be used for training. Other information may also be input.

[0122] In the inference phase using the performance evaluation model 506, for example, the change in the number of white pixels and the change in mass or volume are input to the input layer 171. The output layer 173 calculates (infers) the removal efficiency. Note that since there is a correlation between mass or volume and particle size information, the change in particle size information may be input instead of mass or volume, or together with mass or volume.

[0123] In response to inputs of the change in the number of white pixels and the change in mass or volume detected from a single image data, the output layer 173 outputs a single removal efficiency. The performance evaluation model 506 repeatedly outputs the removal efficiency to obtain a removal efficiency associated with elapsed time. The prediction unit 63 calculates an approximation curve from the relationship between elapsed time and removal efficiency and can predict the lifetime until the extrapolated removal efficiency falls below a threshold.

[0124] <Performance Evaluation Model Using CNN> As shown in Figure 18, a CNN (Convolutional Neural Network) may be used in the performance evaluation model 506. Figure 18 shows an example configuration of the performance evaluation model 506 using a CNN 60. As an example, the CNN 60 has convolutional layers 72, 74, pooling layers 73, 75, and a fully connected layer 70. The input image 71 is image data captured by the camera 50. The input image 71 is processed in the following order: convolutional layer 72, pooling layer 73, convolutional layer 74, pooling layer 75, and fully connected layer 70.

[0125] The convolutional layers 72 and 74 perform a process that converts a grid-like numerical data called a kernel (or filter) into a single numerical value by calculating the sum of the element-wise products of the numerical data of a partial image (called a window) of the same size as the kernel. The convolutional layers 72 and 74 perform this conversion process by slightly shifting the window, thereby converting it into smaller grid-like numerical data (i.e., tensors). The grid-like numerical data is activated by an activation function and input to the pooling layers 73 and 75.

[0126] Pooling layers 73 and 75 are processes that create a single numerical value from activated numerical data. Examples include maximum value pooling, which selects the maximum value in a window, and average value pooling, which selects the average value in a window. Convolutional layers 72 and 74 extract features from the image data, and pooling layers 73 and 75 reduce the sensitivity of the object's position. The activation function is a function that transforms (activates) the input non-linearly (examples include ReLU, tanh, and sigmoid).

[0127] The output of the pooling layer 75 is input to the input layer 171 of the fully connected layer 70 (neural network). The information input to the input layer 171 propagates through the hidden layer 172, and the node 176 of the output layer 173 outputs the removal efficiency. The structure of the neural network is explained in Figure 17, but the number of layers and nodes should be optimized for a CNN.

[0128] Thus, even if image data is input directly (or preprocessed) into the performance evaluation model 506, the output layer 173 can infer the removal efficiency. However, it is expected that learning with extracted features will allow learning with less training data. Measuring performance at the stage of preparing training data is costly, so having less training data is one of its advantages.

[0129] <Predicting Replacement Time> Next, the method for predicting the replacement time of the adsorption filter 31 will be explained with reference to Figure 19, etc. Figure 19 is a graph showing the relationship between elapsed time and removal efficiency. Elapsed time is the time elapsed since the adsorption filter 31 was installed in the pollutant removal device 40. The vertical axis is the removal efficiency inferred by the inference unit 513 using the performance evaluation model 506. In Figure 19, the inference unit 513 inferred the removal efficiency at times t1, t2, and t3, respectively. The prediction unit 63 approximates the relationship between elapsed time and removal efficiency with a preset approximation curve 121. Since the approximation curve 121 intersects with the removal efficiency (for example, 50%) that serves as a guideline for replacing the adsorption filter 31, the time tx at which it intersects with the time axis is the time until replacement (also called the lifespan). Also, with t3 as the current time, tx - t3 is the remaining time until replacement.

[0130] <Operation or Processing> Figure 20 is a sequence diagram illustrating the processing flow of the performance evaluation system 100 in predicting when to replace the adsorption filter 31.

[0131] S31: The camera 50 images the adsorption filter 31 at a preset interval. The interval can be a frequency corresponding to the lifespan of the adsorbent 32, for example, once a day. The camera 50 may also image irregularly, or it may image upon instruction from the user terminal 5. The communication device 51 transmits the image data of the adsorption filter 31 to the server device 60.

[0132] S32: The communication unit 61 of the server device 60 receives image data. The feature detection unit 504 analyzes the image data and quantifies the number of white pixels and granularity information. The feature detection unit 504 further converts the granularity information into mass or volume. The feature detection unit 504 calculates the difference between the recorded initial number of white pixels (the number of white pixels in the initial state when the elapsed time is 0 minutes) and the current number of white pixels as the change in the number of white pixels. The feature detection unit 504 calculates the difference between the recorded initial mass or volume (the mass or volume in the initial state when the elapsed time is 0 minutes) and the current mass or volume as the change in mass or volume.

[0133] S33: The inference unit 513 inputs feature quantities into the performance evaluation model 506 to infer the efficiency of pollutant removal.

[0134] S34: The prediction unit 63 approximates the relationship between elapsed time and removal efficiency with a preset approximation curve 121 and predicts the replacement time (lifespan) at which it intersects with the removal efficiency (e.g., 50%) that serves as a guideline for replacing the adsorption filter 31.

[0135] S35: The notification unit 64 determines whether the remaining time until the end of life is below a threshold. If it is below the threshold, the notification unit 64 notifies the user terminal 5 that it is recommended to replace the adsorption filter. The notification unit 64 may send the notification by email or by SNS, etc. The recipient of the notification may be the customer's representative or a service technician.

[0136] <Main Effects> As described above, the performance evaluation system of this disclosure has the following advantages: (i) The server device 60 can evaluate the performance of the adsorbent 32 remotely, so service personnel can predict when to replace the adsorbent 32 without having to go to the site, thus reducing costs. (ii) The server device 60 can continuously evaluate the performance of the adsorbent 32, making it easier to determine the appropriate replacement time. (iii) Since the server device 60 can predict when to replace the adsorbent, it can suppress excessive replacements and delays in replacement.

[0137] [Second Embodiment] In this embodiment, the inference of the removal efficiency of the second and third stage adsorption filters 31B and 31C will be described. In the first embodiment, it was necessary to either place cameras 50 at all adsorption filters 31A to 31C, or to replace the adsorption filters 31B and 31C based on the performance of the first stage adsorption filter 31A. In this embodiment, the server device 60 uses the inferred removal efficiency for the first stage adsorption filter 31A and surrounding environment data to infer the removal efficiency of the second and third stage adsorption filters 31B and 31C.

[0138] <Outline> Assume that the pollutant removal device 40 is equipped with three adsorption filters 31A to 31C. The removal efficiency 531 of the first-stage adsorption filter 31A can be inferred in the same way as in the first embodiment. Since cameras 50 are not placed in the second-stage adsorption filter 31B and the third-stage adsorption filter 31C, in order to infer their performance, this disclosure uses the temperature, humidity, and airflow of the air passing through the adsorption filters 31B and 31C as ambient environmental data.

[0139] First, the removal efficiency of the first-stage adsorption filter 31A, the removal efficiency of the second-stage adsorption filter 31B, the removal efficiency of the third-stage adsorption filter 31C, and ambient environmental data (temperature, humidity, airflow) are obtained as training data. The removal efficiencies of the adsorption filters 31A to 31C are reliable values, such as those measured by the manufacturer. The learning device 500 generates the second-stage model and the third-stage model using the following as the target variable (training data) and explanatory variables. The second-stage model is a performance evaluation model that infers the removal efficiency of the second-stage adsorption filter 31B, and the third-stage model is a performance evaluation model that infers the removal efficiency of the third-stage adsorption filter 31C. Second-stage model (an example of the second model) Target variable: Removal efficiency of the second-stage adsorption filter Explanatory variables: Performance of the first-stage adsorption filter, temperature, humidity, airflow Third-stage model Target variable: Removal efficiency of the third-stage adsorption filter Explanatory variables: Performance of the first-stage adsorption filter, temperature, humidity, airflow During inference, the performance of the first-stage adsorption filter 31A and the surrounding environment data inferred by the performance evaluation model 506 are input into the second-stage and third-stage models, thereby inferring the performance of the second-stage adsorption filter 31B and the performance of the third-stage adsorption filter 31C.

[0140] Furthermore, since the surrounding environment data is controlled by the air conditioner, it is assumed that there will be little fluctuation throughout the year. However, more preferably, the surrounding environment data used as an explanatory variable should be the cumulative value since the adsorption filter 31 was installed, or time should be added to the explanatory variable.

[0141] <Example of arrangement of temperature sensor, humidity sensor, and airflow sensor> Figure 21 shows an example of the arrangement of the temperature sensor 21, humidity sensor 22, and airflow sensor 23. The temperature sensor 21, humidity sensor 22, and airflow sensor 23 only need to be in the airflow path, and the figure is just one example. A contaminant removal device 40 is installed at the end of the duct 35 from which the air flows out. The air flowing out of the duct 35 has contaminants removed by the contaminant removal device 40 and flows into the cleanroom after passing through the mesh 36. Therefore, the contaminant removal device 40 is installed in the containment space 37 in front of the cleanroom.

[0142] The temperature sensor 21, humidity sensor 22, and airflow sensor 23 are arranged, for example, in a mesh 36. The temperature sensor 21, humidity sensor 22, and airflow sensor 23 are connected to a communication device 51, which transmits temperature, humidity, and airflow to the server device 60.

[0143] <About the Functions> Figure 22 is an example of a functional block diagram that explains the functions of the performance evaluation system 100 of this embodiment by dividing them into blocks. Note that the explanation of Figure 22 mainly explains the differences from Figure 6. A temperature sensor 21, a humidity sensor 22, and an airflow sensor 23 are connected to the communication device 51. The communication device 51 transmits the temperature, humidity, and airflow along with the image data captured by the camera 50 to the server device 60.

[0144] The image acquisition cycle and the measurement cycles of the temperature sensor 21, humidity sensor 22, and airflow sensor 23 may be different. This is because temperature, humidity, and airflow do not fluctuate much throughout the year. If temperature, humidity, and airflow may fluctuate, it is preferable for the communication device 51 to transmit temperature, humidity, and airflow to the server device 60 at a shorter cycle than the image data. The server device 60, for example, integrates the received temperature, humidity, and airflow and uses the integrated value for inference.

[0145] The server device 60 now includes a removal efficiency integration unit 65. The removal efficiency integration unit 65 integrates the removal efficiency of the first-stage adsorption filter 31A, the removal efficiency of the second-stage adsorption filter 31B, and the removal efficiency of the third-stage adsorption filter 31C to calculate the overall removal efficiency (performance) of the pollutant removal device 40. Further details will be described later.

[0146] <About the Functions of the Learning Device> Figure 23 is a functional block diagram of a learning device 500 according to one embodiment of the present disclosure. The explanation of Figure 23 will mainly describe the differences from Figure 7. Of the functions of the learning device 500, the function of generating the performance evaluation model 506 may be the same as in Figure 7. Here, we will explain the functions of generating the second-stage model 507 and the third-stage model 508.

[0147] The training data 501 for generating the second-stage model 507 consists of the performance of the first-stage adsorption filter 31A, the performance of the second-stage adsorption filter 31B, and ambient environmental data (temperature, humidity, airflow). Multiple sets of these are available.

[0148] The training data 501 for generating the third-stage model 508 consists of the performance of the first-stage adsorption filter 31A, the performance of the third-stage adsorption filter 31C, and ambient environmental data (temperature, humidity, airflow). Multiple sets of these are available.

[0149] The learning unit 505 generates a second-stage model 507 by learning the correspondence between the performance and surrounding environment data of the first-stage adsorption filter 31A and the removal efficiency of the second-stage adsorption filter 31B. The second-stage model 507 is correspondence information that associates the performance and surrounding environment data of the first-stage adsorption filter 31A with the removal efficiency of the second-stage adsorption filter 31B.

[0150] The learning unit 505 generates a third-stage model 508 by learning the correspondence between the performance and surrounding environment data of the first-stage adsorption filter 31A and the removal efficiency of the third-stage adsorption filter 31C. The third-stage model 508 is correspondence information that associates the performance and surrounding environment data of the first-stage adsorption filter 31A with the removal efficiency of the third-stage adsorption filter 31C.

[0151] <About the Function of the Analysis Unit> Figure 24 is a functional block diagram of the analysis unit 62 according to one embodiment of the present disclosure. The explanation of Figure 24 will mainly describe the differences from Figure 8. The analysis unit 62 newly includes a surrounding environment data acquisition unit 522, a first inference unit 523, a second inference unit 524, and a third inference unit 525. The first inference unit 523 has the same function as the inference unit 513 in Figure 8, inputting the feature quantities detected from the image data 511 into the performance evaluation model 506 and outputting the removal efficiency of the first stage adsorption filter 31A.

[0152] The ambient environment data acquisition unit 522 acquires temperature, humidity, and airflow (ambient environment data 521) measured by the temperature sensor 21, humidity sensor 22, and airflow sensor 23 from the communication device 51.

[0153] The second inference unit 524 inputs the surrounding environment data 521 acquired by the surrounding environment data acquisition unit 522 and the removal efficiency 531 of the first stage adsorption filter 31A inferred by the first inference unit 523 to the second stage model 507, and outputs the removal efficiency 532 of the second stage adsorption filter 31B.

[0154] The third inference unit 525 inputs the surrounding environment data 521 acquired by the surrounding environment data acquisition unit 522 and the removal efficiency 531 of the first stage adsorption filter 31A inferred by the first inference unit 523 to the third stage model 508, and outputs the removal efficiency 533 of the third stage adsorption filter 31C.

[0155] <Examples of the second and third models> The second model 507 or third model 508 obtained by regression analysis may look like this: • Second model

[0156] ・Third model

[0157] Let g(x) be the removal efficiency 532 of the second-stage adsorption filter 31B, h(x) be the removal efficiency 533 of the third-stage adsorption filter 31C, Φ1(x) be the removal efficiency 531 of the first-stage adsorption filter 31A, Φ2(x) be the temperature, Φ3(x) be the humidity, Φ4(x) be the airflow rate, and w0 be the bias. Φ(x) is a basis function. The learning device 500 can generate a performance evaluation model 506 by determining the coefficients w0 to w4 using the least squares method or the like.

[0158] Figure 25 shows the second-stage model 507, which is composed of a neural network 182. Compared to Figure 17, the input layer 171 receives the removal efficiency 531 of the first-stage adsorption filter 31A, which was inferred by the first inference unit 523, as well as temperature, humidity, and airflow. The output layer 173 outputs the removal efficiency of the second-stage adsorption filter 31B.

[0159] In the inference phase, first, the first inference unit 523 uses the performance evaluation model 506 to infer the removal efficiency 531 of the first-stage adsorption filter 31A. Next, the second inference unit 524 inputs the removal efficiency 531 of the first-stage adsorption filter 31A, temperature, humidity, and airflow rate inferred by the first inference unit 523 into the second-stage model 507, and the output layer 173 can calculate (infer) the removal efficiency 532 of the second-stage adsorption filter 31B.

[0160] Each time image data and surrounding environment data are transmitted, the second inference unit 524 outputs the removal efficiency. Thus, a removal efficiency corresponding to the elapsed time is obtained. The prediction unit 63 calculates an approximation curve from the relationship between elapsed time and removal efficiency and can predict the replacement time (lifetime) when the removal efficiency falls below a threshold.

[0161] The configuration of the neural network 182 is the same for the third model, so the diagram is omitted.

[0162] <Overall Removal Efficiency of the Pollutant Removal System> Therefore, as shown in Figure 26, in this disclosure, removal efficiency curves 41 to 43 are obtained for each of the three adsorption filters 31A to 31C. Figure 26 shows an example of the removal efficiency curves of the adsorption filters 31A to 31C with respect to elapsed time. Figure 26 shows not only the removal efficiency curves 41 to 43 for each of the adsorption filters 31A to 31C, but also an integrated removal efficiency curve 44 which combines these curves.

[0163] A method for integrating multiple removal efficiencies is described. Equation (4) is the formula for calculating the integrated removal efficiency curve 44.

[0164] n is the total number of adsorption filters 31 (3 in this disclosure), w(i) is the weight of the i-th stage adsorption filter 31, and Li(t) is the removal efficiency curve of the i-th stage adsorption filter 31. At each elapsed time, the combined removal efficiency curve 44 is obtained by adding the weights of the removal efficiencies of the adsorption filters 31A to 31C.

[0165] The weights w(i) are calculated using equation (5). The sum of the weights w(i) is 1.

[0166] A. Contribution to air circulation: Score S １be set as. Score S １ is a value that evaluates how much each adsorption filter 31A to 31C contributes to the air circulation of the entire pollutant removal device 40 (flow rate, ease of air passage). The degree of contribution to air circulation varies depending on the positions (such as the central part and the outer peripheral part) of the adsorption filters 31A to 31C, but the contribution to air circulation can be reflected in the integrated removal efficiency curve 44 by the score S １ B. Past performance data is used as the score S ２ be set as. Score S ２ is a value obtained by analyzing the lifespan of each adsorption filter 31A to 31C from past data and evaluating the deterioration rate and performance variation. A larger weight is assigned to the adsorption filter 31 with a faster deterioration or a higher tendency of performance decline.

[0167] S １ +S ２ is determined so that = 1. Let the weighting coefficients of each factor be a1 and a2. Thus, the removal efficiency integration unit 65 can calculate the integrated removal efficiency curve 44. １ S ２

[0168] <Operation or process> FIG. 27 is a sequence diagram for explaining the flow of the process in which the performance evaluation system predicts the replacement timing of the adsorption filter

[0169] S51: The camera 50 images the adsorption filter 31 at a preset cycle. Also, the temperature sensor 21 detects the temperature, the humidity sensor 22 detects the humidity, and the air volume sensor 23 detects the air volume. The communication device 51 transmits the image data of the adsorption filter 31 and the peripheral environment data to the server device 60.

[0170] S52: The communication unit 61 of the server device 60 receives the image data and the peripheral environment data. The feature quantity detection unit 504 detects the feature quantity in the same manner as in step S32.

[0171] S53: The first inference unit 523 inputs the feature quantity into the performance evaluation model 506 and infers the removal efficiency 531 of the first-stage adsorption filter 31A.

[0172] S54: The second inference unit 524 inputs the removal efficiency 531 of the first-stage adsorption filter 31A and surrounding environment data into the second-stage model 507 and infers the removal efficiency 532 of the second-stage adsorption filter 31B.

[0173] S55: The third inference unit 525 inputs the removal efficiency 531 of the first-stage adsorption filter 31A and surrounding environment data into the third-stage model 508 and infers the removal efficiency 533 of the third-stage adsorption filter 31C.

[0174] S56: The removal efficiency integration unit 65 calculates the overall removal efficiency from the three removal efficiencies.

[0175] S57: The prediction unit 63 approximates the relationship between elapsed time and removal efficiency with a preset approximation curve and predicts the time (lifetime) when it intersects with the removal efficiency (e.g., 50%) which serves as a guideline for replacing the adsorption filter 31.

[0176] S58: The notification unit 64 determines whether the remaining time until the end of life is below a threshold. If it is below the threshold, the notification unit 64 notifies the user terminal 5 that it is recommended to replace the adsorption filter. The notification unit 64 may send the notification by email or by SNS, etc. The recipient of the notification may be the customer's representative or a service technician.

[0177] <Main Effects> As described above, the performance evaluation system 100 of this disclosure can infer the performance of adsorption filters 31B and 31C based on the performance of adsorption filter 31A and surrounding environment data. Furthermore, the performance evaluation system 100 can integrate the performance of adsorption filters 31A to 31C to calculate the overall performance of the pollutant removal device 40 (the entirety of adsorption filters 31A to 31C), allowing a service technician to replace adsorption filters 31A to 31C at once.

[0178] <Reasons for the effect> The first aspect of this disclosure "analyzes the image data to obtain feature quantities related to the performance of the adsorbent, inputs the feature quantities into correspondence information that associates the feature quantities with the performance of the adsorbent to infer the performance of the adsorbent, and predicts the replacement time of the adsorbent based on the relationship between the elapsed time since the start of use of the adsorbent and the inferred performance of the adsorbent," thus making it possible to predict the appropriate replacement time of the adsorbent.

[0179] A second aspect of this disclosure states that "the characteristic quantity relating to the performance of the adsorbent is a characteristic quantity that changes according to the amount of salt precipitated on the adsorbent," and therefore, the performance of the adsorbent can be inferred by using the amount of salt that correlates with the performance of the adsorbent as the characteristic quantity relating to the performance.

[0180] A third aspect of this disclosure states that "the feature quantity is the amount of change in the color of the adsorbent, and the control unit inputs the amount of change in color into the corresponding information to infer the performance of the adsorbent," so the amount of change in color due to salt precipitation can be used as a feature quantity to infer the performance of the adsorbent.

[0181] A fourth aspect of this disclosure is that "the change in mass or volume of the adsorbent, and the control unit inputs the change in mass or volume of the adsorbent into the corresponding information to infer the performance of the adsorbent," so the change in mass or volume due to salt precipitation can be used as a feature to infer the performance of the adsorbent.

[0182] A fifth aspect of this disclosure states that "the feature quantity is the change in particle size information of the adsorbent, and the control unit inputs the change in particle size information into the corresponding information to infer the performance of the adsorbent," so the change in particle size information that changes due to salt precipitation can be used as a feature quantity to infer the performance of the adsorbent.

[0183] - The sixth aspect of this disclosure "notifies the terminal device of the predicted replacement time," so that the user operating the terminal device can understand the replacement time.

[0184] The seventh aspect of this disclosure is that "the correspondence information is a first model that has been machine-learned to determine the correspondence between the feature quantities and the performance of the adsorbent," and therefore the correspondence information can be generated using a machine learning method implemented by a tool or the like.

[0185] - An eighth aspect of this disclosure is that "the control unit inputs the performance of the adsorbent at the first location and the surrounding environment data of the pollutant removal device, inferred by the first model, into a second model that has learned the correspondence between the performance of the adsorbent at the first location and the surrounding environment data of the pollutant removal device, and the performance of the adsorbent at the second location, and infers the performance of the adsorbent at the second location," so that the performance of the adsorbent at the second location can be inferred from the performance of the adsorbent at the first location and the surrounding environment data of the pollutant removal device, without having to provide a camera for each adsorbent.

[0186] The ninth aspect of this disclosure calculates the overall performance of the adsorbent in the contaminant removal device by integrating the performance of the adsorbent at the weighted first placement location with the performance of the adsorbent at the second placement location. Therefore, even if the inferred performance differs depending on the placement location, the overall performance of the adsorbent in the device can be calculated, and multiple adsorbents can be replaced together.

[0187] - A tenth aspect of this disclosure is that "the imaging means is sensitive to the wavelength of light used to image the salt precipitated on the adsorbent," making it easier to image the precipitated salt.

[0188] This application claims priority based on Japanese Patent Application No. 2024-168837, filed with the Japan Patent Office on September 27, 2024, and the entire contents of Japanese Patent Application No. 2024-168837 are incorporated herein by reference.

[0189] 40 Contaminant removal device 60 Server device 100 Performance evaluation system

Claims

1. A performance evaluation system in which a pollutant removal device for removing pollutants from the air and an information processing device can communicate via a network, wherein the pollutant removal device comprises an adsorbent for adsorbing the pollutants, an imaging means for imaging the adsorbent, and a communication device for transmitting image data of the adsorbent captured by the imaging means to the information processing device, the information processing device comprises a control unit, the control unit analyzes the image data to obtain feature quantities relating to the performance of the adsorbent, inputs the feature quantities into correspondence information that associates the feature quantities with the performance of the adsorbent to infer the performance of the adsorbent, and predicts the replacement time of the adsorbent based on the relationship between the elapsed time since the start of use of the adsorbent and the inferred performance of the adsorbent.

2. The performance evaluation system according to claim 1, wherein the characteristic quantity relating to the performance of the adsorbent is a characteristic quantity that changes according to the amount of salt precipitated on the adsorbent.

3. The performance evaluation system according to claim 1 or 2, wherein the feature quantity is the amount of color change of the adsorbent, and the control unit inputs the amount of color change to the corresponding information and infers the performance of the adsorbent.

4. The performance evaluation system according to claim 1 or 2, wherein the feature quantity is the change in mass or volume of the adsorbent, and the control unit inputs the change in mass or volume of the adsorbent into the corresponding information to infer the performance of the adsorbent.

5. The performance evaluation system according to claim 1 or 2, wherein the feature quantity is the amount of change in particle size information of the adsorbent, and the control unit inputs the amount of change in particle size information into the corresponding information to infer the performance of the adsorbent.

6. The performance evaluation system according to any one of claims 1 to 5, wherein the control unit notifies the terminal device of the predicted replacement time.

7. The performance evaluation system according to any one of claims 1 to 6, wherein the correspondence information is a first model obtained by machine learning the correspondence between the feature quantities and the performance of the adsorbent.

8. The performance evaluation system according to claim 7, wherein the pollutant removal device has the adsorbent placed in a first location and a second location, the communication device transmits ambient environmental data of the pollutant removal device to the information processing device, and the control unit inputs the performance of the adsorbent in the first location and ambient environmental data of the pollutant removal device, inferred by the first model, into a second model that has learned the correspondence between the performance of the adsorbent in the first location and ambient environmental data of the pollutant removal device and the performance of the adsorbent in the second location, and infers the performance of the adsorbent in the second location.

9. The performance evaluation system according to claim 8, wherein the control unit weights the performance of the adsorbent at the first location inferred by the first model by the airflow rate or the degradation rate of the adsorbent, weights the performance of the adsorbent at the second location by the airflow rate or the degradation rate of the adsorbent, and calculates the overall performance of the adsorbent in the pollutant removal device by integrating the weighted performance of the adsorbent at the first location and the performance of the adsorbent at the second location.

10. The performance evaluation system according to any one of claims 1 to 9, wherein the imaging means is sensitive to the wavelength of light used to image the salt precipitated on the adsorbent.

11. A pollutant removal device for removing pollutants from the air, comprising: an adsorbent for adsorbing the pollutants; an imaging means for imaging the adsorbent; and a control unit, wherein the control unit analyzes image data of the adsorbent captured by the imaging means to obtain feature quantities relating to the performance of the adsorbent; inputs the feature quantities into correspondence information that associates the feature quantities with the performance of the adsorbent to infer the performance of the adsorbent; and predicts the replacement time of the adsorbent based on the relationship between the elapsed time since the start of use of the adsorbent and the inferred performance of the adsorbent.

12. An information processing device that can communicate via a network with a pollutant removal device, which removes pollutants from the air, comprising: an adsorbent for adsorbing pollutants; an imaging means for imaging the adsorbent; and a communication device for transmitting image data of the adsorbent captured by the imaging means to an information processing device, the information processing device having a control unit, the control unit analyzing the image data to obtain feature quantities relating to the performance of the adsorbent; inputting the feature quantities into correspondence information that associates the feature quantities with the performance of the adsorbent to infer the performance of the adsorbent; and predicting the replacement time of the adsorbent based on the relationship between the elapsed time since the start of use of the adsorbent and the inferred performance of the adsorbent.

Images