Surveillance system

Abstract

Provided is a monitoring system that makes it possible to use sensors with a simple structure to accurately obtain information indicating at least one from among the type, scattering condition, source, and means of handling particles in air under monitoring.　This monitoring system for monitoring particles in air under monitoring is characterized by comprising a plurality of particle sensors that are arranged within a space and each have a polarization separation mechanism and detect particles within a medium collected at the location of the particle sensor by emitting light onto the medium and receiving light that has been scattered by particles in the medium and has passed through the polarization separation mechanism, and an analysis device that acquires, from each of the plurality of particle sensors, particle detection results including information indicating the degree of depolarization of the scattered light and information indicating the signal amplitude of the received light, and acquires and outputs information indicating at least one from among the type, source, scattering condition, and means of handling the particles in the air.

Description

[Technical field]

[0001] The present disclosure relates to a monitoring system for monitoring air quality in a work environment when work such as product manufacturing is carried out in a clean room or a production line building. [Background technology]

[0002] One method for detecting suspended minute substances (hereinafter referred to as particles) such as PM2.5, pollen, or dust is the light scattering method. The light scattering method detects the presence, size, type, amount, etc. of particles by irradiating light into a monitored space and detecting scattered light generated when the light hits particles. Furthermore, one example of a method that can perform highly accurate detection with a simple structure is one that performs the above detection by polarizing and separating the scattered light (for example, Patent Documents 1 to 3).

[0003] Until now, most of the sensors used for detecting particles in clean rooms or production line buildings have been large and expensive, classified as high-performance particle counters, and are not suitable for use in arranging many of them to grasp the air in the entire space. In contrast, the light scattering type sensor described above is small and relatively inexpensive, and can achieve high-performance detection with a relatively simple structure, so it is considered suitable for arranging many of them.

[0004] As an example of a method for monitoring the air quality of an entire space to be monitored without disposing multiple sensors, there is a dust particle generation position identifying device described in Patent Document 4. The dust particle generation position identifying device described in Patent Document 4 arranges sampling tubes in a lattice pattern within a space, sequentially and selectively connects each point of the lattice pattern to obtain particle density data in the air and identify the generation position. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent No. 6207722 [Patent Document 2] Japanese Patent No. 6270700 [Patent Document 3] Japanese Patent No. 6542385 [Patent Document 4] Japanese Patent No. 3470370 Summary of the Invention [Problem to be solved by the invention]

[0006] However, the method described in Patent Document 4 has the problem that it is difficult to grasp the accurate particle dispersion situation (e.g., concentration distribution at each time and change in concentration over time) because it cannot detect particles at all lattice points simultaneously.

[0007] In order to accurately detect invisible particles floating in space and obtain information on their source and countermeasures, it is important to accurately grasp the type of particle and its dispersion state. However, with the method described in Patent Document 4, it takes time to obtain detection results for the entire space, making it difficult to accurately grasp the particle dispersion state.

[0008] The present disclosure has been made to solve the problems described above, and aims to provide a monitoring system that can use a sensor with a simple structure to accurately obtain information indicating at least one of the type of particles occurring in a monitored space, the dispersion situation, the source of the particles, and how to deal with them. [Means for solving the problem]

[0009] One aspect of the monitoring system according to the present disclosure is a monitoring system for monitoring particles occurring in a space to be monitored, the monitoring system including a plurality of such systems arranged in the space, each of which has a polarization separation mechanism, and which irradiates light onto a medium to be collected at the locations where the systems are arranged, and scatters light caused by particles in the medium through the polarization separation mechanism. As receiving lightA plurality of particle sensors detect particles in a medium by receiving light, and a particle detection result including information indicating a depolarization degree of the scattered light and information indicating a signal amplitude of the received light is obtained from each of the plurality of particle sensors. a data acquisition unit that calculates the movement speed of the detected particles based on a sensor data set including the particle detection results acquired by the data acquisition unit and information indicating the detection time and the detection location of the particle detection results, and generates an inference data set by adding the calculated particle movement speed to the sensor data set; an inference unit that uses a predetermined trained model based on the inference data set to infer information indicating at least one of the type of particles occurring in the space, the source of the particles, the scattering situation, and a method of dealing with the particles as an inference result; Acquire and output information indicating at least one of the types of particles occurring in the space, their source, the state of dispersion, and the method of dealing with them. Result output section The present invention is characterized by comprising: Effect of the Invention

[0010] According to the present disclosure, it is possible to provide a monitoring system that can use a sensor with a simple structure to accurately obtain information indicating at least one of the type of particles occurring in a monitored space, the dispersion situation, the source of the particles, and how to deal with them. [Brief description of the drawings]

[0011] [Figure 1] 1 is a configuration diagram showing an example of a monitoring system according to a first embodiment. [Diagram 2] FIG. 1 is a configuration diagram illustrating an example of a particle sensor. [Diagram 3] FIG. 1 is a configuration diagram illustrating an example of a particle sensor. [Figure 4] 4 is a flowchart showing an example of the operation of the monitoring system according to the first embodiment. [Diagram 5] 1 is a block diagram showing an example of the configuration of an analysis device according to a first embodiment. [Figure 6] FIG. 2 is a block diagram showing an example of the configuration of a learning device. [Figure 7] FIG. 11 is an explanatory diagram showing an example of a result output by a result output unit. [Figure 8] FIG. 11 is a block diagram showing an example of the configuration of an analysis device according to a second embodiment. [Figure 9] FIG. 11 is a block diagram showing an example of the configuration of an analysis device according to a third embodiment. [Figure 10] FIG. 11 is a block diagram showing an example of the configuration of an analysis device according to a third embodiment. [Figure 11]FIG. 2 is a conceptual diagram showing an example of a data structure of data held in an analytical DB. [Figure 12] FIG. 2 is a configuration diagram showing an example of a hardware configuration of an analysis device and a learning device. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] Embodiment 1 A monitoring system according to a first embodiment of the present disclosure will be described below with reference to the drawings. Fig. 1 is a configuration diagram showing an example of a monitoring system 100 according to the first embodiment. The monitoring system 100 shown in Fig. 1 includes a plurality of particle sensors 2 arranged in a space 1 to be monitored, and an analysis device 3 communicatively connected to each of the particle sensors 2. Note that, although reference numeral 4 in the drawing represents a manufacturing device installed in the space 1, the manufacturing device 4 may or may not be present.

[0013] In the example shown in Fig. 1, a total of 24 particle sensors 2 are arranged in areas obtained by dividing a space 1 into a 6 x 4 grid. The arrangement and number of particle sensors 2 are not limited to this. For example, the number of particle sensors 2 may be less than or greater than the above. In this example, each particle sensor 2 communicates with an analysis device 3 wirelessly or via a wire, and each particle sensor 3 operates at the same time and transmits sensor data acquired at the same time to the analysis device 3.

[0014] 2 and 3 are configuration diagrams showing an example of the particle sensor 2. The particle sensor 2 shown in Fig. 2 and 3 has a polarization separation mechanism, and detects particles in a medium by irradiating light onto a medium collected at an installation location and receiving light scattered by particles in the medium via the polarization separation mechanism. Note that the particle sensor 2 is not limited to the configuration shown in Fig. 2 and 3, and may be, for example, one having a polarization separation mechanism and capable of determining the degree of depolarization, or one of the devices described in the above-mentioned Patent Documents 1 to 3.

[0015] The particle sensor 2 shown in Figure 2 includes a laser light source 20 that irradiates light, a lens 21 that focuses the light into a detection space 22 provided within the sensor 2, a fan 23 that takes particles into the detection space 22, a prism 24 that polarizes and separates the scattered light generated when the light hits a particle into s waves and p waves, a light receiving element 25 that receives each of the polarized light (s waves and p waves), an amplifier 26 that amplifies an electrical signal based on the received intensity of the polarized light output from the light receiving element 25, and a signal processing unit 27 that processes the amplified signal.

[0016] The particle sensor 2 shown in Fig. 3 is provided with a polarizing plate 28 as a polarization separation mechanism instead of the prism 24. The particle sensor 2 shown in Fig. 3 is provided with the polarizing plate 28 on each of two different optical paths capable of detecting scattered light, and thereby polarizes and separates the scattered light into s waves and p waves, and then receives the scattered light at the corresponding light receiving element 25.

[0017] In the particle sensor 2 shown in FIG. 2 or FIG. 3, the laser light emitted from the laser light source 20 is focused by the lens 21 toward the particles that are taken into the detection space 22 by the air flow generated by the fan 23. When the focused light hits the particle, scattered light is generated. When the scattered light passes through the prism 24 or the polarizing plate 28, it is separated into a p wave (first polarized light) and an s wave (second polarized light). Each of the separated polarized lights is received by the light receiving element 25, converted into an electric signal, and output to the signal processing unit 27 via the amplifier 26. The signal processing unit 27 outputs the depolarization degree of the scattered light and the signal amplitude (or peak intensity) of the scattered light that is a combination of the P signal and the S signal, based on the electric signal (P signal and S signal) based on the reception intensity of each polarized light. Here, the depolarization degree of the scattered light generated by the particle is an example of sensor data indicating the shape of the particle (more specifically, the aspect ratio), and the signal amplitude or peak intensity of the scattered light is an example of sensor data indicating the size of the particle. The signal processing unit 27 may have a function as a particle counter that detects the number of times particles are detected within a specified time based on the change over time of the signal amplitude or peak intensity.

[0018] In addition, the lens 21, the fan 23, and the amplifier 26 are not required for the particle sensor 2, and a light scattering type sensor with high accuracy and a simple structure may be provided with, for example, a light source, a polarization separation mechanism (such as the prism 24 or the polarizing plate 28 in the above example), a light receiving element 25, and a signal processing unit 27.

[0019] In this embodiment, the monitoring system 100 includes a plurality of light scattering type particle sensors 2 having such a polarization separation mechanism. Each particle sensor 2 transmits the above-mentioned sensor data to the analysis device 3 wirelessly or with priority at a predetermined timing, periodically, or upon request.

[0020] Each particle sensor 2 is provided with identification information in advance, and the analysis device 3 may store the identification information of each particle sensor 2 in association with information indicating the installation position. The analysis device 3 collects sensor data indicating the particle detection results from each particle sensor 2. The sensor data includes at least information correlated with the shape of the particle, such as the degree of depolarization, and information correlated with the presence or absence of detection and the size of the particle, such as signal amplitude or peak intensity. The sensor data may further include the detection time, the position of the sensor (or identification information capable of acquiring the same), and the number of particles detected at that detection time.

[0021] The analysis device 3 may collect sensor data, for example, via a database (not shown) that stores the sensor data output from each particle sensor 2. In this case, the analysis device 3 may access such a database to collect the sensor data output from each particle sensor 2. The analysis device 3 is assumed to be capable of collecting sensor data at the same time from at least two or more particle sensors 2. It is preferable to be able to collect sensor data at the same time from all particle sensors 3 arranged in the space 1. Here, the sensor data at the same time refers to sensor data detected or output at the same time. It is not necessary for the sensor data to be exactly the same time, and as long as the difference (for example, ± several seconds) is recognized as the same time within a range that matches the application of the monitoring system, it may be considered to be the same time. In other words, it is sufficient for the sensor data to be associated with a time recognized as the same time within a predetermined error range.

[0022] Fig. 4 is a flowchart showing an example of the operation of the monitoring system 100 according to the first embodiment. In the example shown in Fig. 4, the analysis device 3 first acquires sensor data from each particle sensor 2 arranged in the space 1 to be monitored (step a1). Next, the analysis device 3 performs a data analysis process (inference process to be described later) based on the sensor data acquired from each particle sensor 2 to acquire information indicating at least one of the types of particles occurring in the space 1, the scattering state, the source of generation, and a method of dealing with them (step a2). Finally, the analysis device 3 outputs the acquired information (step a3).

[0023] The manner in which information is output by the analysis device 3 is not particularly limited, but examples include outputting display information to an information display device such as a display, outputting control signals, audio data, etc. to output devices such as alarms and speakers, outputting control signals to various control devices (e.g., air conditioners, automatic door control devices, etc.) for implementing countermeasures, and outputting log information to an information storage device such as a database.

[0024] Next, an analysis process performed by the analysis device 3 will be described. Fig. 5 is a block diagram showing a configuration example of the analysis device 3 in the monitoring system 100 according to the embodiment 1. The analysis device 3 shown in Fig. 5 uses a trained model to obtain information indicating at least one of the types of particles occurring in the space 1, the state of dispersion, the source of the particles, and a method of dealing with them, from the collected sensor data.

[0025] Here, the information indicating the type of particle may be information indicating the cluster (classification) of the particle. Furthermore, the information indicating the particle scattering state may be information indicating the diffusion speed and diffusion direction of the particle, or information indicating the range affected by the scattering of the particle (hereinafter referred to as the affected range). Furthermore, the information indicating the source of the particle may be information indicating a position within the space 1. Furthermore, the information indicating how to deal with the particle may be instruction information instructing a specific process or action that has been determined in advance.

[0026] 5 includes a data acquisition unit 31, an inference unit 32, and a result output unit 33. The inference unit 32 is connected to a learning model storage unit 51 so as to be accessible thereto.

[0027] The data acquisition unit 31 acquires sensor data from each particle sensor 2 arranged in the monitored space 1, and outputs the acquired sensor data to the subsequent inference unit 32 in association with information indicating the position (i.e., the detection location) of the particle sensor 2 from which the sensor data was acquired and information indicating the detection time of the sensor data. Hereinafter, the information output from the data acquisition unit 31 may be referred to as a sensor data set. In this case, the data acquisition unit 31 may output the acquired sensor data processed or converted as a sensor data set. Herein, the processing or conversion includes not only conversion of individual data formats, but also statistical processing of two or more pieces of data, addition processing of adding previously acquired sensor data or other data (such as environmental data to be described later), and deletion processing of deleting some of a plurality of elements. In this example, the processing or conversion by the data acquisition unit 31 is limited to one-dimensional processing (such as linear calculation of the same type of data, or simple addition or deletion of elements) in order to prioritize data collection speed, and does not include complex calculations (such as correlation between different types of data). In addition, the data acquisition unit 31 can collect information other than sensor data that is required by the subsequent inference unit 32, such as environmental data indicating the state and position of equipment and workers, and the air flow (wind direction) within the space 1, and include it in the sensor data set.

[0028] When the inference unit 32 receives a sensor data set (here, information associating sensor data for at least two or more sensors with detection locations and detection times) including sensor data collected from each particle sensor 2 from the data acquisition unit 31, the inference unit 32 obtains, based on the input sensor data set, any predetermined information among information indicating the type, scattering state, source, and handling method of particles occurring in the space 1, using the trained model stored in the trained model storage unit 51. The inference unit 32 also outputs the obtained information to the result output unit 33.

[0029] The inference unit 32 infers, using the trained model, any predetermined information among information indicating the type, scattering state, source, and countermeasure of particles occurring in the space 1 based on the sensor data set input from the data acquisition unit 31. For example, the inference unit 32 may convert the input sensor data set into an input data set of the trained model, input the converted input data set to the trained model, and obtain the inference result output from the trained model as any predetermined information among information indicating the type, scattering state, source, and countermeasure of particles occurring in the space 1. Note that, if the format of the sensor data set already matches the format of the input data set, the inference unit 32 may input the input sensor data set as it is to the trained model as an input data set. The trained model may be one that has previously learned the relationship between various information included in the input data set and output data (any predetermined information among information indicating the type, scattering state, source, and countermeasure of particles occurring in the space 1) by a learning device 6 (to be described later) or the like. There is no particular limitation on the learning method (learning algorithm) used to generate the trained model. In addition, the inference unit 32 may use multiple trained models that can infer information indicating the type of particles occurring in space 1, the dispersion situation, the source of the particles, and how to deal with the particles, based on the input sensor data set, to obtain information indicating the type of particles occurring in space 1, the dispersion situation, the source of the particles, and how to deal with the particles, based on the input sensor data set.

[0030] The input data set used by the inference unit 32 for inference does not need to include sensor data from all particle sensors 2, and may be processed to include sensor data from only some of the particle sensors 2 arranged in the space 1. In that case, however, the inference unit 32 executes the inference process multiple times while changing the target sensor data so that all sensor data from the particle sensors 2 in which particles have been detected is used for inference at least once. Here, there is no particular limitation on which sensor data is targeted in one inference process. However, the sensor data may be determined by, for example, dividing an area in the space 1 in advance, etc., and targeting sensor data acquired from sensors located in each block for each block, or by preferentially selecting a sensor that has detected particles many times, and then treating sensors located close to that sensor as a group of sensors, and targeting sensor data from the group of sensors.

[0031] Furthermore, when the inference unit 32 determines that particles have been detected at substantially the same time in multiple different locations in the space 1 based on the input sensor data or data for inference, the inference unit 32 may group the detection locations and perform inference processing on the sensor data belonging to each detection location group after classification. Here, the inference unit 32 may make the above judgment, for example, based on whether particles have been detected at the same detection time by two or more particle sensors 2 installed at locations in the space 1 that are separated by a predetermined distance or more. By grouping the detection locations in this way and performing inference processing for each detection location group, when particles occur in multiple different locations, accurate information can be obtained for each detection location group without mixing up the information.

[0032] 4 shows an example in which the inference unit 32 is included in the analysis device 3, the inference unit 32 may be implemented as a device (such as an inference device not shown) separate from the analysis device 3. For example, the inference unit 32 may be connected to the analysis device 3 via a network. Furthermore, the inference unit 32 may exist on a cloud server.

[0033] The trained model may, for example, receive an input data set including particle detection results including the degree of depolarization and signal amplitude for each sensor, and the detection time and detection location of the detection results, and output information indicating the types of particles occurring in space 1 as an inference result for the input data set.

[0034] The trained model may be, for example, a model that receives an input data set including a particle detection result including the depolarization degree and signal amplitude for each sensor, and the detection time and detection location of the detection result, and outputs information indicating the scattering status of particles occurring in the space 1 as an inference result for the input data set. As another example, the trained model may be, for example, a model that receives an input data set including a particle detection result including the depolarization degree and signal amplitude for each sensor at a first detection time, a particle detection result including the depolarization degree and signal amplitude for a detection time immediately preceding the first detection time, and the detection time (or a representative time) and detection location of the detection result, and outputs information indicating the scattering status of particles occurring in the space 1 as an inference result for the input data set. By including the particle detection result at each detection location at the immediately preceding time in the input data set, it is possible to obtain information indicating the scattering status of particles with higher accuracy. As yet another example, the trained model may input an input data set including the number of detections per sensor within a predetermined time, the depolarization degree and signal amplitude for each detection or a representative detection, and the detection time (or representative time) and detection location of the detection result, and output information indicating the scattering status of particles occurring in the space 1 as an inference result for the input data set. By including the number of detections indicating the amount of particles at each detection location in the input data set, it is possible to obtain information indicating the scattering status of particles with higher accuracy. As yet another example, the trained model may input an input data set including the number of detections per sensor within a predetermined time, the depolarization degree and signal amplitude for each detection or a representative detection, the detection time (or representative time) and detection location of the detection result, and information indicating the wind direction included in the environmental data, and output information indicating the scattering status of particles occurring in the space 1 as an inference result for the input data set.

[0035] The trained model may be, for example, a model that receives an input data set including a particle detection result including the depolarization degree and signal amplitude for each sensor, and the detection time and detection location of the detection result, and outputs information indicating the source of the particle occurring in the space 1 as an inference result for the input data set. As another example, the trained model may be, for example, a model that receives an input data set including a particle detection result including the depolarization degree and signal amplitude for each sensor at a first detection time, a particle detection result including the depolarization degree and signal amplitude for a detection time immediately preceding the first detection time, and the detection time (or a representative time thereof) and detection location of each detection result, and outputs information indicating the source of the particle occurring in the space 1 as an inference result for the input data set. By including the particle detection result for the time immediately preceding the time series in the input data set, it is possible to obtain information indicating the source of the particle with higher accuracy. As yet another example, the trained model may input an input data set including, for example, the number of detections within a given time for each sensor, the detection depolarization rate and signal amplitude for each detection or a representative detection, and the detection time (or representative time) and detection location of the detection result, and output, as an inference result for the input data set, information indicating the source of particles occurring in space 1. By including the number of detections indicating the amount of particles at each detection location in the input data set, it is possible to obtain information indicating the source of particles with higher accuracy.

[0036] The trained model may be, for example, a model that receives an input data set including a particle detection result including the depolarization degree and signal amplitude for each sensor, and the detection time and detection location of the detection result, and outputs information indicating how to deal with particles occurring in the space 1 as an inference result for the input data set. As another example, the trained model may be, for example, a model that receives an input data set including a particle detection result including the depolarization degree and signal amplitude for each sensor at a first detection time, a particle detection result including the depolarization degree and signal amplitude for each sensor at a detection time immediately before the first detection time, and the detection time (or a representative time) and detection location of each detection result, and outputs information indicating how to deal with particles occurring in the space 1 as an inference result for the input data set. By including the particle detection result for the time immediately before in the time series in the input data set, it is possible to obtain information indicating how to deal with particles occurring with higher accuracy. As yet another example, the trained model may input an input data set including, for example, the number of detections within a given time for each sensor, the detection depolarization rate and signal amplitude for each detection or a representative detection, and the detection time (or representative time) and detection location of the detection result, and output, as an inference result for the input data set, information indicating how to deal with particles occurring in the space 1. By including the number of detections indicating the amount of particles at each detection location in the input data set, it is possible to obtain information indicating how to deal with particles occurring with higher accuracy.

[0037] The sensor data set input to the inference unit 32 is assumed to include all of the information contained in the input data set input to the learned model used by the inference unit 32 or information that can be converted into that information.

[0038] Such a trained model may be obtained, for example, by a learning device for a monitoring system 100 as shown in Fig. 6. The learning device 6 shown in Fig. 6 includes a data acquisition unit 61 and a model generation unit 62. The model generation unit 62 is connected to the trained model storage unit 51 so as to be accessible thereto.

[0039] As an example, the data acquisition unit 61 acquires, as learning data, the particle detection results including the depolarization degree and signal amplitude as sensor data collected from two or more particle sensors 2 arranged at different positions, a learning input data set including the detection time and detection location of the detection results, and information indicating the type of particles occurring in the space 1 as ground truth data.

[0040] Further, the model generating unit 62 learns information indicating the type of particles occurring in the space 1, which is taken as an inference result for an input data set including particle detection results including depolarization degree and signal amplitude collected from two or more particle sensors 2 arranged at arbitrary positions, and detection times and detection locations of the detection results, based on the learning data generated based on a combination of the learning input data set and the correct answer data collected by the data acquiring unit 61. That is, a trained model is generated that infers information indicating the type of particles occurring in the space 1, which is optimal from the learning input data set and the correct answer data, and is applied to the monitoring system 100. Here, the learning data is data in which the learning input data set and the correct answer data are associated with each other.

[0041] As another example, the data acquisition unit 61 acquires, as learning data, the particle detection results including the degree of depolarization and the signal amplitude as sensor data collected from two or more particle sensors 2 arranged at different positions, a learning input data set including the detection time and detection location of the detection results, and information indicating the particle dispersion situation occurring within the space 1 as ground truth data.

[0042] Further, the model generation unit 62 learns information indicating the particle dispersion state occurring in the space 1, which is inferred from an input data set including particle detection results including depolarization degree and signal amplitude collected from two or more particle sensors 2 arranged at arbitrary positions, and detection times and detection locations of the detection results, based on learning data generated based on a combination of the learning input data set and the correct answer data collected by the data acquisition unit 61. That is, it generates a trained model that infers information indicating the particle dispersion state occurring in the space 1, which is optimal from the learning input data set and the correct answer data, and is applied to the monitoring system 100.

[0043] As another example, the data acquisition unit 61 acquires, as learning data, the particle detection results including the degree of depolarization and the signal amplitude as sensor data collected from two or more particle sensors 2 arranged at different positions, a learning input data set including the detection time and detection location of the detection results, and information indicating the source of particles occurring within the space 1 as ground truth data.

[0044] Further, the model generation unit 62 learns information indicating the source of particles occurring in the space 1, which is inferred from an input data set including particle detection results including depolarization degree and signal amplitude collected from two or more particle sensors 2 arranged at arbitrary positions, and detection times and detection locations of the detection results, based on learning data generated based on a combination of the learning input data set and the correct answer data collected by the data acquisition unit 61. That is, a trained model is generated that infers information indicating the source of particles occurring in the space 1, which is optimal from the learning input data set and the correct answer data, and is applied to the monitoring system 100.

[0045] As another example, the data acquisition unit 61 acquires, as learning data, the particle detection results including the degree of depolarization and the signal amplitude as sensor data collected from two or more particle sensors 2 arranged at different positions, a learning input data set including the detection time and detection location of the detection results, and information indicating how to deal with particles occurring in the space 1 as ground truth data.

[0046] Further, the model generation unit 62 learns information indicating how to deal with particles occurring in the space 1, which is inferred for an input data set including particle detection results including depolarization degree and signal amplitude collected from two or more particle sensors 2 arranged at arbitrary positions and detection times and detection locations of the detection results, based on learning data generated based on a combination of the learning input data set and the correct answer data collected by the data acquisition unit 61. That is, a trained model is generated that infers information indicating how to deal with particles occurring in the space 1, which is optimal from the learning input data set and the correct answer data, and is applied to the monitoring system 100.

[0047] Note that the learning input data set acquired by the data acquisition unit 61 is not limited to the above example. The learning input data set may be the same data as the input data set input to the trained model.

[0048] The learning algorithm used by the model generation unit 62 may be a known algorithm such as supervised learning, reinforcement learning, or unsupervised learning. As an example, a case where a neural network is applied will be described. The model generation unit 62 learns output data (in this example, information indicating the type of particles generated in the space 1, the scattering state, the source of generation, or a method of dealing with them) that is an inference result for an arbitrary input data set, for example, by so-called supervised learning according to a neural network model. Here, supervised learning refers to a method of providing a learning device with a set of input and result (label) data, learning the features of the learning data, and inferring the result from the input.

[0049] A neural network is composed of an input layer consisting of multiple neurons, an intermediate layer (hidden layer) consisting of multiple neurons, and an output layer consisting of multiple neurons. The intermediate layer may be one layer or two or more layers.

[0050] As an example of a neural network, when a value is input to a plurality of input layers, the value is weighted and input to the following intermediate layer, and the process is repeated, and finally output from the output layer. At this time, the output result changes depending on the weight value assigned when moving through each layer. In this subject, the neural network learns output data that is an inference result for an arbitrary input data set by so-called supervised learning according to the learning data created based on a combination of the learning input data set and the correct answer data acquired by the data acquisition unit 61. In other words, the neural network learns by inputting the learning input data set into the input layer and adjusting the weight between neurons when moving between layers so that the result output from the output layer approaches the correct answer data.

[0051] The model generation unit 62 may generate a trained model by performing the above-mentioned learning.

[0052] The learning device 6 is used to learn any predetermined information among information indicating the type of particles occurring in the space 1, the dispersion state, the source of the particles, and a method of dealing with the particles, from the sensor data that the monitoring system 100 can collect, and may be included in the monitoring system 100 or may be implemented as a device separate from the monitoring system 100. The learning device 6 may be connected to the monitoring system 100 via a network. Furthermore, the learning device 6 may exist on a cloud server.

[0053] In the above example, a case where supervised learning is applied to the learning algorithm used by the model generation unit 62 has been described, but the present invention is not limited to this. As for the learning algorithm, reinforcement learning, unsupervised learning, or semi-supervised learning can also be applied in addition to supervised learning.

[0054] Next, we will explain the case where reinforcement learning is applied as a learning algorithm. For example, in reinforcement learning, an agent (acting subject) in a certain environment observes the current state (environmental parameters) and decides on the action to be taken. The environment changes dynamically depending on the actions of the agent (in this example, various control devices for the space 1 and manufacturing equipment 4, etc.), and the agent is given a reward according to the change in the environment. The agent repeats this process and learns the action policy that will obtain the most reward through a series of actions. Q-learning and TD-learning are known as representative methods of reinforcement learning. For example, in the case of Q-learning, the general update formula for the action value function Q(s, a) is expressed by Equation 1.

[0055]

number

[0056] In the formula 1, S t represents the state of the environment at time t, and a t represents the action at time t. Action a t Therefore, the state is s t+1 Changes to r t+1 represents the reward that can be obtained depending on the change in state, γ represents the discount rate, and α represents the learning coefficient. Note that γ is in the range of 0<γ≦1, and α is in the range of 0<α≦1. In this example, for example, the method of dealing with particles in space 1 is action a t The learning input data set showing the state in space 1 at that time is called state s t Let state s at time t be t Best Practices in a t Learn.

[0057] The update formula expressed by Equation 1 increases the action value Q if the action value Q of the action a with the highest Q value at time t+1 is greater than the action value Q of the action a executed at time t, and decreases the action value Q if the action value Q is greater than the action value Q of the action a executed at time t. In other words, the action value function Q(s, a) is updated so that the action value Q of the action a at time t approaches the best action value at time t+1. This allows the best action value in a certain environment to be propagated sequentially to the action value in the previous environment. A reward is calculated based on the input action and state in this way, and a function for determining the best action (how to deal with a particle) for an arbitrary state (input data set) may be updated according to the calculated reward and output to the trained model storage unit 51. For example, in the case of Q-learning, the action value function Q(s, a) expressed by Equation 1 is t ,a t ) is used as a function to obtain output data (information indicating how to deal with the problem) as an inference result for an arbitrary state (input data set).

[0058] The above learning is repeated, and the final updated action value function Q(s t ,a t ) may be stored in the trained model storage unit 51 as a trained model.

[0059] Next, a case where unsupervised learning is applied as a learning algorithm will be described. When unsupervised learning is used, for example, the model generating unit 62 may learn the type of particle based on learning data created based on a combination of the depolarization degree and the signal amplitude, which are mutually correlated sensor data acquired by the data acquiring unit 61. In other words, the model generating unit 62 may generate a trained model that infers the type of particle from the depolarization degree and the signal amplitude of the sensor data from the particle sensor 2 in the monitoring system 100. Here, the learning data is data in which the depolarization degree and the signal amplitude are associated with each other.

[0060] The unsupervised learning style refers to a method of learning features of learning data that does not include results (labels) by providing the learning device with the learning data. The model generation unit 62 may learn the types of particles by so-called unsupervised learning, for example, according to a clustering (grouping) method using the K-means method.

[0061] The K-means method is a non-hierarchical clustering algorithm, and is a method of classifying a given number of clusters into k by using the average of the clusters. More specifically, the K-means method is processed in the following manner. First, a cluster is randomly assigned to each data xi. Next, the center Vj of each cluster is calculated based on the assigned data. Next, the distance between each xi and Vj is obtained, and xi is reassigned to the cluster with the closest center. Then, if the cluster assignment of all xi does not change in the above processing, or if the amount of change falls below a certain threshold set in advance, it is determined that convergence has occurred and the processing is terminated. For example, the inference unit 32 may obtain information indicating the type of particle as an inference result by inferring to which cluster the depolarization degree and signal amplitude included in the sensor data for each particle sensor 2 belong, using the trained model generated in this way.

[0062] In each of the above examples, the model generation unit 62 may learn the output data to be the inference result according to the learning data created for the multiple monitoring systems 100. The model generation unit 62 may acquire the learning data from one or more monitoring systems 100 used in the same type of space, or may learn the output data to be the inference result by using the learning data collected from one or more monitoring systems 100 used in different types of spaces. It is also possible to add or remove the collection destination of the learning data midway. Furthermore, it is also possible to apply a learning device that has learned the output data to be the inference result for a certain monitoring system 100 to another monitoring system 100, and re-learn the output data to be the inference result for the other monitoring system 100 to update the learned model.

[0063] In addition, the learning algorithm used in the model generation unit 62 can be deep learning, which learns to extract the features themselves, or machine learning can be performed according to other known methods, such as a genetic program, a functional logic program, or a support vector machine.

[0064] The result output unit 33 outputs the information (inference result) input from the inference unit 32 to the outside. The result output unit 33 outputs the information (inference result) input from the inference unit 32 to a predetermined output destination according to the type of information. For example, if the input information is information indicating the type of particles or the scattering state, the result output unit 33 may process the information into information for display and output it to an information display device such as the display 71. Also, for example, if the input information is information indicating the source of the particles, the result output unit 33 may process the information into information for display and output it to an information display device such as the display 71, or may output a control signal to an alarm device 72 close to the source to sound an alarm. Also, for example, if the input information is information indicating a countermeasure, the result output unit 33 may process the information into information for display and output it to an information display device such as the display 71 to encourage the monitor to take action, or if automatic control is possible, output a control signal to a control device 73 that is previously associated with the content of the countermeasure. Possible ways to deal with this include, for example, increasing the flow rate of the exhaust port closest to the source, or pointing a spot suction machine at the source, or forcibly evacuating all the air in the target space when the particle concentration is very high. Note that, in Fig. 5, the display 71, alarm device 72, and control device 73 are shown as output destinations of the results, but these are merely examples and are not limiting.

[0065] The result output unit 33 may display the obtained inference result superimposed on a map in the space 1. For example, the result is visualized by displaying it like a bird's-eye view. For example, when information indicating the type of particle is obtained, the result output unit 33 may display characters, symbols, figures, etc. corresponding to the type of particle on the display 71 at a position corresponding to the detection location of the particle, superimposed on the map in the space 1. In addition, at this time, color coding according to the type may also be performed. Furthermore, when information indicating the source of particle generation is obtained, the result output unit 33 may display characters, symbols, figures, etc. indicating the source of particle generation at a position corresponding to the source of particle generation, superimposed on a map (such as a floor plan) or image in the space 1 on the display 71. Furthermore, when information indicating the scattering state of particle is obtained, the result output unit 33 may display information indicating the scattering state (affected range) of particle on the display 71, superimposed on a map or image in the space 1. In this case, the result output unit 33 may calculate the particle concentration in each block on the map or image from information such as the number of times particles are detected, which is included in each sensor data, and may color-code the affected area by concentration, for example, by making the display color darker as the particle concentration increases and by making the display color lighter as the concentration decreases. In other words, the result output unit 33 can vary the display mode by at least changing the characters, symbols, figures, or colors to be displayed according to the estimated particle type and particle concentration in the display destination area.

[0066] FIG. 7 is an explanatory diagram showing an example of the output of the result by the result output unit 33. The example shown in FIG. 7 is an example in which the display color of the affected range is divided according to the concentration of the particle when the affected range of the particle is displayed on the display 71 as the scattering state of the particle by superimposing it on a map or image in the space 1. Note that in FIG. 7, due to the drawing rules, the density of the shading is different instead of the display color. Also, in FIG. 7, information on the type of particle and the target method are displayed by characters along with information on the scattering state of the particle. In this way, the result output unit 33 may color the location where the particle is detected according to its concentration, and may display the color from dark to light from high to low according to the particle concentration. Furthermore, the result output unit 33 may allow the user to adjust the concentration threshold, the display update speed, the fineness of the grid, etc., so that the source of the particle and the state of the scattering of the particle can be understood. Note that there is no limit to the number of particle sources to be displayed, and it is preferable to display all the sources.

[0067] The result output unit 33 may also compare concentrations (number of detections per unit time) between particle sensors 2 to complement the data between the grids. In the example shown in Fig. 7, the result output unit 33 displays the type of detected particle and the action that the user should take as a countermeasure, below the bird's-eye view showing the particle concentration distribution in such space 1.

[0068] When displaying the concentration, the result output unit 33 or another instruction unit (not shown) may develop information indicating the timing of sampling when conducting quality results such as sampling inspection of products on a production line, etc. In other words, an instruction may be output to reduce the frequency of quality inspections when the particle concentration is low, and to increase the frequency of quality inspections when the particle concentration is high. It is also possible to improve production efficiency by changing the frequency of product quality inspections depending on the particle concentration.

[0069] In addition, when particles occur in multiple locations in the monitored space 1, the impact on product manufacturing, etc. can be minimized by prioritizing the countermeasures, for example by applying countermeasures to locations with high particle concentrations. In other words, product productivity can be improved by prioritizing countermeasures based on the obtained information.

[0070] The result output unit 33 can also identify the equipment or person included in the affected range by, for example, collating the scattering situation or the source outputted from the inference unit 32 with environmental data, and can also identify the equipment or person considered to be the source. The result output unit 33 can, for example, determine whether the source has moved based on the temporal distribution movement of the particles obtained by comparing the previous output result with the latest output result, and identify whether the source is a fixed equipment or a moving object. Another example of a method for identifying the source is a method in which the spatial distribution and temporal change of the concentration of the detected particles are obtained from the obtained result, and the scattering situation (diffusion speed, diffusion direction) indicated by the result and the object arrangement information in the space 1 are used to identify whether the source is, for example, a person (including other moving objects) or a manufacturing device by an analysis device. The result output unit 33 can also select an output method of a more optimal countermeasure from the information on the source obtained in this way. For example, if the source is considered to be a moving worker, it can be considered to inform the worker of the countermeasure or sound an alarm.

[0071] The result output unit 33 updates the display every time the sensor data acquired by the data acquisition unit 31 is updated and the inference result is updated, thereby visualizing the result in a manner that is closer to real time and easier to understand.

[0072] In this embodiment, the particle sensors 2 may be installed at regular intervals on the floor or on the equipment in the space 1. In most clean rooms, the cleanliness is maintained by flowing clean air from the ceiling to the floor. Therefore, by arranging the particle sensors 2 at regular intervals on the floor or on the manufacturing equipment 4, etc. as viewed from above, the generation of particles in the space 1 can be monitored with relatively little leakage. In addition, for example, the particle sensors 2 may be densely arranged in places in the space 1 where the possibility of particle generation is relatively high, such as places where friction occurs and their surroundings, and sparsely arranged in places where the possibility of particle generation is relatively low and their surroundings. In the latter case, that is, when the particle sensors 2 are arranged at intervals with a sparse arrangement, the data acquisition unit 31 may collect data from the sensors in the densely arranged area more frequently than the frequency of collecting data from the sensors in the sparsely arranged area. In this way, it is possible to strengthen monitoring of areas where particles are more likely to be generated. In addition, when the particle sensors 2 are arranged at equal intervals, the data acquisition unit 31 can change the frequency of collecting sensor data based on preset predicted information about the occurrence location or occurrence probability information, i.e., can change the frequency of collecting data from sensors installed in locations where occurrence is likely to be high to be higher than the frequency of collecting data from sensors installed in locations where occurrence is unlikely to be high.

[0073] As described above, according to the present embodiment, the inference unit 32 collects sensor data from at least two or more particle sensors 2 at the same time, and uses a trained model that has been generated in advance by machine learning or the like to adapt the collected sensor data to the monitoring system 100 to infer information indicating at least the type of particles occurring within the space 1, the dispersion situation, the source of the particles, or how to deal with them. Therefore, it is possible to obtain such information with higher accuracy using a particle sensor 2 with a simple structure, without having to place multiple expensive particle sensors.

[0074] For example, if particles are generated in a clean room or a production line building, the generation of particles and their effects can be stopped quickly if not only the detection location but also the type of particle, source of generation, scattering situation, and countermeasure method are known. The countermeasure method here can be manual or automatic. If the generation of particles and their effects can be stopped quickly, the decrease in productivity, quality, and yield of products manufactured in the clean room or production line building can be minimized. In addition, by recording the generation of particles over time, it is possible to use the information for future defect and defective product analysis.

[0075] Furthermore, according to this embodiment, the particle sensor 2 may be small and of simple structure, so that the particle sensor 2 can be flexibly adapted to various clean rooms or production line buildings by arranging the particle sensor 2 in a position suitable for the monitoring area.

[0076] In addition, according to this embodiment, invisible things such as the generation status of particles in the monitored space 1 can be visually captured, so that appropriate measures can be taken quickly and the impact of particle generation can be minimized. For example, when a worker takes an air shower and enters the monitored area, it can be confirmed whether the particles have been removed reliably. In addition, since it is possible to determine whether the source of particles is a moving object or not, for example, when the source of particles moves over time, it is possible to appropriately take measures, whether automatically or manually, such as sounding an alarm to urge the worker to take an air shower again when particles remain. In this way, the type of particles, the source of particles, the scattering status, or the method of dealing with the particles can be accurately presented in response to the detection information of invisible particles, so that the cleanliness of a clean room or the like can be maintained.

[0077] Embodiment 2 Next, a monitoring system 100 according to a second embodiment will be described with reference to the drawings. Fig. 8 is a block diagram showing a configuration example of an analysis device 3 in the monitoring system 100 according to the second embodiment. The analysis device 3 shown in Fig. 8 further includes an inference data generating unit 34 compared to the analysis device 3 shown in Fig. 5. Note that other than the analysis device 3, the components may be the same as those in the first embodiment.

[0078] The inference data generation unit 34 calculates the movement speed of the detected particles based on the sensor data set acquired by the data acquisition unit 31, and outputs a data set further including the calculated movement speed of the particles as an inference data set to the inference unit 32. The inference data generation unit 34 serves as a pre-processing unit for the inference unit 32 that prepares data required for inference by classifying information, performing arithmetic processing, etc.

[0079] The inference unit 32 infers, using a trained model based on the inference data set input from the inference data generation unit 34, any predetermined information among information indicating the type, scattering state, source, and countermeasure of particles occurring in the space 1. For example, the inference unit 32 may convert an input data set of the trained model from the input inference data set, input the converted input data set to the trained model, and obtain an inference result output from the trained model as any predetermined information among information indicating the type, scattering state, source, and countermeasure of particles occurring in the space 1. The inference unit 32 may obtain, based on the input inference data set, each of information indicating the type, scattering state, source, and countermeasure of particles occurring in the space 1 using a plurality of trained models capable of inferring each of information indicating the type, scattering state, source, and countermeasure of particles occurring in the space 1.

[0080] As an example, when an inference dataset including particle detection results including the degree of depolarization and signal amplitude for each sensor, the movement speed of the particle detected in the detection results, and the detection time and detection location of the detection results is input, the inference unit 32 may input the inference dataset as an input dataset to the trained model and infer any predetermined piece of information indicating the type of particle occurring in space 1, the dispersion situation, the source of occurrence, and the method of dealing with the particle.

[0081] As another example, when an inference dataset including the number of detections within a predetermined time for each sensor, the depolarization degree and signal amplitude for each detection or a representative detection, the movement speed of the particle detected in the detection result, and the detection time (or its representative time) and detection location of the detection result are input, the inference unit 32 may input the inference dataset as an input dataset to the trained model to infer any predetermined information among information indicating the type, scattering state, source, and handling method of particles occurring in the space 1. The trained model used here may be the input dataset to which the movement speed has been added in the above-mentioned trained model example.

[0082] As described above, the present embodiment differs from the first embodiment in that the particle movement speed is added to the input data set of the trained model. The movement speed is also added to the training input data set in the model generation unit 62 that generates the trained model.

[0083] The movement speed of the particle may be calculated, for example, by calculating the similarity between the sensor data of the sensor where the particle was detected and the past sensor data of the sensors arranged around the sensor, judging that the pair of sensor data with a similarity equal to or greater than a predetermined threshold is due to the movement of the same type of particle, and calculating the movement speed of the detected particle from the detection time and detection location of the pair of sensor data. Here, examples of the sensor data to be used for calculating the similarity include a pair of the depolarization degree and the signal amplitude, and a pair of the depolarization degree, the signal amplitude, and the number of detections.

[0084] As described above, according to the present embodiment, a data set including the movement speeds of detected particles obtained from the presence of a plurality of particle sensors 2 can be used to infer any predetermined information among information indicating the type, dispersion state, source, and countermeasures of particles occurring in the space 1. Therefore, such information can be obtained with higher accuracy by using a particle sensor with a simple structure, without the need to place a plurality of expensive particle sensors.

[0085] In particular, by adding information on the moving speed of particles to obtain inference results, for example, the particle scattering situation (or the range of its influence) can be obtained with high accuracy, so that, for example, where particles are occurring on a production line or the like can be grasped in real time. Then, based on such real-time information, it is also possible to obtain optimal countermeasures for each location. As described above, it is also possible to obtain information indicating countermeasures as inference results from the inference dataset.

[0086] Embodiment 3 Next, a monitoring system 100 according to a third embodiment will be described with reference to the drawings. Fig. 9 is a block diagram showing an example of the configuration of the analysis device 3 in the monitoring system 100 according to the third embodiment.

[0087] 9, the inference unit 32 includes a first inference unit 321 and a second inference unit 321. Note that the components other than the inference unit 32 may be similar to those in the first or second embodiment. That is, although not shown in the figure, the analysis device 3 may further include an inference data generation unit 34.

[0088] When the first inference unit 321 receives the sensor data set acquired by the data acquisition unit 31 or the inference data set generated by the inference data generation unit 34, the first inference unit 321 obtains information indicating the type of particles occurring in the space 1 based on the input data set, using the trained model stored in the trained model storage unit 51. The first inference unit 321 outputs the obtained inference result (information indicating the type of particles occurring in the space 1) to the second inference unit 322 together with the input sensor data set or inference data set.

[0089] The first inference unit 321 may obtain information indicating the type of particle by using a trained model that inputs an input data set including the detection result of a particle including the depolarization degree and signal amplitude for each sensor, and the detection time and detection location of the detection result, as described above, and outputs information indicating the type of particle occurring in the space 1 as an inference result for the input data set. Note that the trained model used is not limited to this. For example, the first inference unit 321 may input an input data set including the detection result of a particle including the depolarization degree and signal amplitude for each sensor, the moving speed of the particle detected in the detection result, and the detection time and detection location of the detection result, and outputs information indicating the type of particle occurring in the space 1 as an inference result for the input data set. Here, the sensor data set or the inference data set input to the first inference unit 321 includes all information included in the input data set input to the trained model used by the first inference unit 321 or information that can be converted into such information, and all information included in the input data set input to the trained model used by the second inference unit 321 described later or information that can be converted into such information. In addition, it is not necessary for all information contained in the input data set input to the learned model used by the second inference unit 321 or information convertible to such information to be included, in which case the information is output directly to the second inference unit 321.

[0090] When the second inference unit 322 receives the sensor data set acquired by the data acquisition unit 31 or the inference data set generated by the inference data generation unit 34 and the inference result of the first inference unit 321, which is information indicating the type of particles occurring in the space 1, the second inference unit 322 obtains information indicating the scattering state, source, or countermeasure of the particles occurring in the space 1 using the trained model stored in the trained model storage unit 51 based on the input data set. Here, the second inference unit 322 separates the input sensor data set or inference data set according to the type of particle identified by the information indicating the type of particle occurring in the space 1 obtained by the first inference unit 321, and obtains information indicating the scattering state, source, or countermeasure of the particles occurring in the space 1 using the trained model stored in the trained model storage unit 51 based on the classified data set for each type of particle. Note that the inference process may be the same as that of the inference unit 32 in the first or second embodiment, except that the inference process is performed by selecting the input sensor data set or inference data set according to the type of particle.

[0091] The second inference section 322 outputs the obtained inference results (information indicating the scattering state, source, or countermeasures for each type of particle) to the result output section 33.

[0092] In the first inference unit 321 and the second inference unit 322 as well, the input data set used for inference does not need to include sensor data from all particle sensors 2, and may be processed to include, for example, sensor data from only some of the particle sensors 2 arranged in the space 1. In that case, however, the first inference unit 321 and the second inference unit 322 execute the inference process multiple times while changing the target sensor data so that all of the sensor data from the particle sensors 2 in which particles have been detected is used for inference at least once.

[0093] Also, the first inference unit 321 and the second inference unit 322 may generate a detection location group when it is determined that particles are detected at substantially the same time at multiple different locations in the space 1 based on the input sensor data or data for inference, and may perform inference processing for each of the generated detection location groups on the sensor data belonging to the group. The second inference unit 322 may take over the grouping results (information on the detection location group and information on the data set belonging thereto) of the first inference unit 321. Here, the first inference unit 321 and the second inference unit 322 may, for example, make the above judgment based on whether particles are detected at the same detection time by two or more particle sensors 2 installed at locations in the space 1 that are separated by a predetermined distance or more. Also, the classification of the detection locations may be performed based on, for example, the positional relationship between the detection location of the particle and the sensor position of each sensor data, and a known clustering method may be used. By performing inference processing for each detection location group after grouping the detection locations in this way, when particles occur in multiple different locations, accurate information can be obtained for each detection location group without mixing up the information.

[0094] As described above, according to this embodiment, by configuring the inference unit 32 in a multi-stage (two-stage) configuration, more accurate and highly precise information on particles can be obtained.

[0095] 9 shows an example in which the inference unit 32 includes two inference units, but it is also possible to configure the inference unit to perform two inference processes in one data analysis process (a2). Specifically, the same effect can be obtained by performing the inference process by the first inference unit 321 in the first inference process and the inference process by the second inference unit 322 in the second inference process. In such a case, the inference unit that performs the first inference process is regarded as the first inference unit, and the inference unit that performs the second inference process is regarded as the second inference unit.

[0096] In addition, in the above-described example, the first inference unit 321 is configured to infer the type of particles occurring in the space 1, and the second inference unit 322 classifies and selects the input data set for each type of particle inferred by the first inference unit 321 to obtain the dispersion status, source of generation, or countermeasures for particles belonging to that type, but the combination of information inferred by each inference unit is not limited to this.

[0097] For example, the first inference unit 321 may be configured to infer the source of particles occurring in the space 1, and the second inference unit 322 may be configured to classify and select the input data set for each particle source estimated by the first inference unit 321, and obtain the dispersion status, source, or countermeasures for the particles believed to have been generated from the source. Note that the input data set may be classified for each source based on, for example, the positional relationship between the detection location of the sensor data and the source, and a known clustering method may be used.

[0098] Also, while FIG. 9 shows an example of an inference unit having two stages, the inference unit 32 may have a three or more stages. As an example, the first inference unit is configured to infer the type of particles occurring in the space 1. The second inference unit is configured to infer the source of particles belonging to each type of particle estimated by the first inference unit. The third inference unit may be configured to infer information indicating the scattering status of particles belonging to the type that are said to have occurred from the source or information indicating a method of dealing with the particles, for each type and source of particles estimated by the first and second inference units.

[0099] As another example, the first inference unit is configured to infer the source of particles occurring in the space 1. The second inference unit is configured to infer, for each source of particles estimated by the first inference unit, the scattering state (e.g., diffusion speed and scattering direction) of particles that are said to have been generated from the source. The third inference unit may be configured to infer, for each source of particles and scattering state estimated by the first and second inference units, information indicating a method of dealing with particles that are said to have been generated from the source in accordance with the scattering state.

[0100] In this way, by configuring the inference unit 32 in a multi-stage manner, more accurate and highly precise information on particles can be obtained.

[0101] Embodiment 4 Next, a monitoring system 100 according to an embodiment 4 will be described with reference to the drawings. Fig. 10 is a block diagram showing an example of the configuration of the analysis device 3 in the monitoring system 100 according to the embodiment 4.

[0102] The analysis device 3 shown in FIG. 10 includes a data analysis unit 35 instead of the inference unit 32. The data analysis unit 35 is connected to an analysis database 52 (hereinafter, referred to as an analysis DB 52) so as to be accessible. Note that the configuration may be the same as any of the first to third embodiments except for the inclusion of the data analysis unit 35 instead of the inference unit 32 and the inclusion of the analysis DB 52 instead of the trained model storage unit 51. That is, although not shown in the figure, the analysis device 3 may further include an inference data generation unit 34.

[0103] The analytical DB 52 stores information for identifying at least one of the types, sources, and countermeasures for particles detected by the sensor from the sensor data set (a data group generated from sensor data and environmental data acquired from two or more particle sensors 2) output from the data acquisition unit 31. For example, the analytical DB 52 may be a database that holds the size of a particle, the shape (aspect ratio) of the particle, and the source (device or person) in association with each other. Also, for example, the analytical DB 52 may be a database that holds the size of a particle, the shape (aspect ratio) of the particle, the number of detected particles, and countermeasures for the particle in association with each other.

[0104] Fig. 11 is an explanatory diagram conceptually showing an example of the data structure of data held in the analytical DB 52. In the example shown in Fig. 11, the data held in the analytical DB 52 has a three-axis data structure. That is, a combination of three types of data included in a sensor data set is considered as a collection of rectangular parallelepipeds in a three-dimensional space in which the value range of each data is appropriately divided, and at least one or all of the information indicating the type of particle and the method of dealing with it is linked to each rectangular parallelepiped (each area when the three-dimensional data space is divided into areas).

[0105] The data analysis unit 35, based on the values ​​of three predetermined types of data contained in the sensor data set input from the data acquisition unit 31, identifies which rectangular prism the sensor data set corresponds to, i.e., which area of ​​the three-dimensional data space it corresponds to, and acquires at least one or all of the information indicating the type of particle and the target method associated with the identified area.

[0106] In the example shown in FIG. 11, the degree of depolarization, which is correlated with the shape of the particle, the signal amplitude, which is correlated with the size of the particle, and the number of detections, which is correlated with the number of detected particles, are used as axis data in a three-dimensional data space to obtain information indicating the type of detected particle and how to deal with it for each sensor.

[0107] In addition to the above, the data analysis unit 35 may perform a source identification process to identify the location of the source as the location where the particle sensor 2 with the highest number of detections is installed, based on the information on the installation locations and the information on the number of detections of each particle sensor 2. In the source identification process, it is also possible to perform a process of identifying whether the source is a device or a person, and which device or person the source is, by comparing the location information of devices and people in the space 1 indicated by the environmental data with the location of the source.

[0108] In the above example, the data structure held by the analytical DB 52 is considered as a three-dimensional data space, and data to be the analysis result is associated with a combination of three types of data included in the acquired data set, but the data structure held by the analytical DB 52 may be such that the data to be acquired is associated with a combination of two types of data, or the data to be the analysis result is associated with a combination of four or more types of data. In other words, the number of dimensions of the data space for obtaining the data to be the analysis result, i.e., the number of combinations of the use data in the acquired data set, is not limited to three, and may be less than three or more than three.

[0109] The operation of this embodiment may be basically the same as that of embodiments 1 to 3. However, in step a2, the data analysis unit 35 performs data analysis processing using the database as described above based on the sensor data acquired from each particle sensor 2 to obtain information indicating at least one of the types of particles occurring in the space 1, the scattering state, the source of generation, and the method of dealing with them.

[0110] In addition, when the data analysis unit 35 determines that particles have been detected at multiple different locations in space 1 at approximately the same time based on the input sensor data or inference data, the detection locations may be divided into groups, and for each detection location group after classification, analysis processing may be performed on the sensor data belonging to that group. Furthermore, the data set used for analysis by the data analysis unit 35 does not need to include sensor data from all particle sensors 2, and may be processed to include, for example, sensor data from only some of the particle sensors 2 arranged in the space 1. In that case, however, the data analysis unit 35 executes the analysis process multiple times while changing the target sensor data so that all of the sensor data from the particle sensors 2 in which particles have been detected is used at least once for analysis (more specifically, as reference parameters of the DB).

[0111] As described above, according to this embodiment, with a simple configuration that does not use machine learning, it is possible to obtain information indicating at least one of the types of particles occurring in space 1, the scattering state, the source of occurrence, and a method of dealing with them. Other points are the same as those of the first to third embodiments.

[0112] Fig. 12 is a configuration diagram showing an example of the hardware configuration of the analysis device 3, learning device 6, and their peripherals according to this embodiment. As shown in Fig. 12, each function of the analysis device 3 and learning device 6 may be realized by a signal processing unit 9. Also, each function of the signal processing unit 9 may be realized by a processor 91 and a memory 92. Also, the signal processing unit 9 may be further connected to an external input / output I / F 110, which is an interface for external input / output of various information, an input device 120 such as a mouse or keyboard, and an output device 130 such as a display, via a bus or the like.

[0113] That is, the analysis device 3 and the learning device 6 at least include a processor 91 and a memory 92 for implementing the functions of the signal processing unit 9 described above.

[0114] The signal processing unit 9 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC, an FPGA, or a combination of these. The signal processing unit 9 may implement each function of the analysis device 3 or the learning device 6, or may implement all of the functions together.

[0115] The processor 91 may be, for example, a CPU (also called a central processing unit, processing device, arithmetic device, microprocessor, microcomputer, or DPS) that executes a program stored in a memory. In this case, each function of the signal processing unit 9 is realized by software, firmware, or a combination of software and firmware. The software or firmware is written as a program and stored in the memory 92. The processor 91 realizes each function, for example, by reading and executing a program stored in the memory 92. That is, the analysis device 3 and the learning device 6 or each means included therein include a memory 92 for storing a program that results in the execution of various steps for realizing the function of each means realized by the signal processing unit 9. In addition, it can be said that these programs cause a computer to realize the above-mentioned tracking method. Here, the memory 92 corresponds to, for example, a non-volatile or volatile semiconductor memory such as a RAM, a ROM, a flash memory, an EPROM, an EEPROM, a magnetic disk, a flexible disk, an optical disk, a compact disk, a mini disk, a DVD, etc.

[0116] It should be noted that some of the functions of the signal processing unit 9 may be realized by dedicated hardware (processing circuits, etc.) and some by software or firmware (more specifically, the processor 91 or the like that operates in accordance with the hardware or firmware). In this case, the hardware, software, or firmware can use information input via the external input / output I / F 110 and the input device 120, output information to the outside via the external input / output I / F 110 and the output device 130, etc.

[0117] In this way, analysis device 3 and learning device 6 can realize the above-mentioned functions by hardware, software, firmware, or a combination of these. [Explanation of symbols]

[0118] 1 Space (monitored space), 2 Particle sensor, 3 Analysis device, 4 Manufacturing device, 20 Laser light source, 21 Lens, 22 Detection space, 23 Fan, 24 Prism, 25 Light receiving element, 26 Amplifier, 27 Signal processing unit, 28 Polarizing plate, 100 Monitoring system, 31 Data acquisition unit, 32 Inference unit, 321 First inference unit, 322 Second inference unit, 33 Result output unit, 34 Inference data generation unit, 51 Learned model storage unit, 52 Analysis database (analysis DB), 6 Learning device, 61 Data acquisition unit, 62 Model generation unit, 71 Display, 72 Alarm device, 73 Control device, 9 Signal processing unit, 91 Processor, 92 Memory, 110 External input / output I / F, 120 Input device, 130 Output device

Claims

1. A monitoring system for monitoring particles generated in a space to be monitored, comprising: a plurality of particle sensors arranged in the space, each having a polarization separation mechanism, for detecting the particles in the medium by irradiating light onto a medium to be collected at a location where the sensors are arranged and receiving light scattered by the particles in the medium as received light via the polarization separation mechanism; a data acquisition unit that acquires, from each of the plurality of particle sensors, a particle detection result including information indicating a depolarization degree of the scattered light and information indicating a signal amplitude of the received light; an inference data generation unit that calculates a moving speed of the detected particle based on a sensor data set including the particle detection result acquired by the data acquisition unit and information indicating a detection time and a detection location of the particle detection result, and generates an inference data set by adding the calculated moving speed of the particle to the sensor data set; an inference unit that uses a predetermined trained model based on the inference dataset to infer, as an inference result, information indicating at least one of the type, source, scattering state, and countermeasure method of the particles occurring in the space; a result output unit that acquires and outputs information indicating at least one of the type, source, scattering state, and countermeasure method of the particles occurring in the space that is the inference result; Equipped A monitoring system comprising:

2. The inference unit is a first inference unit that infers information indicating at least one of a type, a source, and a scattering state of the particles generated in the space based on the inference data set; and a second inference unit that infers information other than the information acquired by the first inference unit, the information indicating at least one of the type, source, scattering state, and method of dealing with the particles generated in the space, based on the inference data set and the information acquired by the first inference unit. The monitoring system of claim 1 .

3. A monitoring system for monitoring particles generated in a space to be monitored, comprising: a plurality of particle sensors arranged in the space, each having a polarization separation mechanism, for detecting the particles in the medium by irradiating light onto a medium to be collected at a location where the sensors are arranged and receiving light scattered by the particles in the medium as received light via the polarization separation mechanism; a data acquisition unit that acquires, from each of the plurality of particle sensors, a particle detection result including information indicating a depolarization degree of the scattered light and information indicating a signal amplitude of the received light; an inference unit that uses a predetermined trained model based on a sensor data set including the particle detection results acquired by the data acquisition unit and information indicating the detection time and the detection location of the particle detection results to infer information indicating at least one of the type, source, dispersion state, and countermeasure method of the particles occurring in the space as an inference result; a result output unit that acquires and outputs information indicating at least one of the type, source, scattering state, and countermeasure method of the particles occurring in the space that is the inference result; Equipped with The inference unit is a first inference unit that infers information indicating at least one of a type, a source, and a scattering state of the particles generated in the space based on the sensor data set; and a second inference unit that infers, based on the sensor data set and the information acquired by the first inference unit, information other than the information acquired by the first inference unit, which indicates at least one of the type, source, scattering state, and method of dealing with the particles generated in the space. A monitoring system comprising:

4. the first inference unit infers the type of the particle detected by the particle sensor based on a data set including the particle detection results, the data set including information indicating the depolarization degree of the scattered light for each particle sensor and information indicating the signal amplitude of the received light; The second inference unit infers information indicating the scattering state, source, or method of dealing with the particles occurring within the space for each type of particle, based on a data set including the particle detection results including information indicating the depolarization degree of the scattered light for each particle sensor and information indicating the signal amplitude of the received light, information indicating the detection time and detection location of the particle detection results, and the moving speed of the particle indicated by the particle detection results. A monitoring system according to claim 2 or 3.

5. When it is determined that the particles are detected at a plurality of different locations in the space at substantially the same time, the inference unit performs the inference for each group of detection locations. A monitoring system according to any one of claims 1 to 4.

6. A monitoring system for monitoring particles generated in a space to be monitored, comprising: a plurality of particle sensors arranged in the space, each having a polarization separation mechanism, for detecting the particles in the medium by irradiating light onto a medium to be collected at a location where the sensors are arranged and receiving light scattered by the particles in the medium as received light via the polarization separation mechanism; a data acquisition unit that acquires, from each of the plurality of particle sensors, a particle detection result including information indicating a depolarization degree of the scattered light and information indicating a signal amplitude of the received light; a result output unit that acquires and outputs information indicating at least one of the type, source, scattering state, and countermeasure method of the particles generated in the space; and a data analysis unit that acquires, as an inference result, information indicating at least one of the type, source, scattering state, and handling method of the particles occurring in the space, using an analysis database that holds information indicating at least one of the type, source, scattering state, and handling method of the particles occurring in the space, which corresponds to a combination of value ranges of two or more types of data included in the sensor data set or the inference data set, based on the particle detection result acquired by the data acquisition unit and a sensor data set including information indicating a detection time and a detection location of the particle detection result, or an inference data set to which the movement speed of the particles detected is added to the sensor data set. A monitoring system comprising:

7. When information indicating a source of the particles is obtained as the inference result, the result output unit compares a previous output result with a latest output result to determine whether the source of the particles has moved and specifies whether the source of the particles is a fixed device or a moving object. A surveillance system according to any one of claims 1 to 6.

8. A monitoring system for monitoring particles generated in a space to be monitored, comprising: a plurality of particle sensors arranged in the space, each having a polarization separation mechanism, for detecting the particles in the medium by irradiating light onto a medium to be collected at a location where the sensors are arranged and receiving light scattered by the particles in the medium as received light via the polarization separation mechanism; a data acquisition unit that acquires, from each of the plurality of particle sensors, a particle detection result including information indicating a depolarization degree of the scattered light and information indicating a signal amplitude of the received light; a result output unit that acquires and outputs information indicating at least one of the type, source, scattering state, and countermeasure method of the particles generated in the space as an inference result, When information indicating a source of the particles is obtained as the inference result, the result output unit compares a previous output result with a latest output result to determine whether the source of the particles has moved and specifies whether the source of the particles is a fixed device or a moving object. A monitoring system comprising:

9. and an inference unit that infers information indicating at least one of the type, source, scattering state, and handling method of the particles occurring in the space, using a predetermined trained model based on a sensor data set including the particle detection result acquired by the data acquisition unit and information indicating the detection time and detection location of the particle detection result. The monitoring system according to claim 8.

10. The data acquisition unit changes a frequency of acquiring the particle detection results from each of the plurality of particle sensors depending on the likelihood of the particles being generated or the density of the arrangement of the plurality of particle sensors. A monitoring system according to any one of claims 1 to 9.

11. A monitoring system for monitoring particles generated in a space to be monitored, comprising: a plurality of particle sensors arranged in the space, each having a polarization separation mechanism, for detecting the particles in the medium by irradiating light onto a medium to be collected at a location where the sensors are arranged and receiving light scattered by the particles in the medium as received light via the polarization separation mechanism; a data acquisition unit that acquires, from each of the plurality of particle sensors, a particle detection result including information indicating a depolarization degree of the scattered light and information indicating a signal amplitude of the received light; a result output unit that acquires and outputs information indicating at least one of the type, source, scattering state, and countermeasure method of the particles generated in the space, The data acquisition unit changes a frequency of acquiring the particle detection results from each of the plurality of particle sensors depending on the likelihood of the particles being generated or the density of the arrangement of the plurality of particle sensors. A monitoring system comprising:

12. and an inference unit that infers information indicating at least one of the type, source, scattering state, and handling method of the particles occurring in the space, using a predetermined trained model based on a sensor data set including the particle detection result acquired by the data acquisition unit and information indicating the detection time and detection location of the particle detection result. The monitoring system of claim 11.

13. When it is determined that the particles are detected at a plurality of different locations in the space at substantially the same time, the inference unit performs the inference for each group of detection locations. A monitoring system according to claim 9 or 12.

14. The result output unit displays the inference result by superimposing it on a map or an image of the space. A monitoring system according to any one of claims 1 to 10.

15. The result output unit changes a display mode in superimposing the inference result on the map or image of the space depending on the type of the particle or the concentration of the particle detected in each display destination area.

15. The surveillance system of claim 14.

16. The plurality of particle sensors are disposed on the floor or on the equipment in the space at equal intervals or at intervals corresponding to the likelihood of the particles being generated. A monitoring system according to any one of claims 1 to 15.

17. The information indicating how to deal with the particles includes at least one of updating the timing of quality inspections, outputting an alarm, controlling a ventilation system, and providing information to workers in the space. A monitoring system according to any one of claims 1 to 16.

18. The plurality of particle sensors each include a polarizing plate as the polarization separation mechanism. A monitoring system according to any one of claims 1 to 17.

Images