Near miss prediction system, near miss prediction method, and information processing device

Abstract

To predict an occurrence of a near-miss.SOLUTION: A near-miss prediction system includes an information processing device (1) for acquiring first information on a state of the body of a worker during work and second information including an occurrence time of a near-miss, and a server device (8) for constructing a prediction model for predicting an occurrence of a near-miss by machine learning using teacher data extracted on the basis of the first information and the second information, either one of the information processing device or the server device includes an extraction part (801) for extracting partial information within a prescribed period before the occurrence time of the near-miss in the first information, and the server device constructs a prediction model with the partial information extracted by the extraction part as the teacher data.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a near-miss prediction system, a near-miss prediction method, and an information processing device. [Background technology]

[0002] At work sites such as construction sites, in order to prevent industrial accidents, cases of near misses that could lead to industrial accidents are collected and workers are warned to be careful, etc. One method for collecting near miss cases is to detect near misses of subjects based on electrocardiogram data (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2021-019977 Summary of the Invention [Problem to be solved by the invention]

[0004] However, up until now, measures have been taken such as issuing warnings to workers based on cases of near misses that have occurred (detected), and no attempt has been made to predict the occurrence of near misses.

[0005] An object of the present invention is to provide a technique that can predict the occurrence of near misses. [Means for solving the problem]

[0006] A near-miss prediction system according to one embodiment includes an information processing device that acquires first information regarding the physical condition of a worker while working and second information including the time the near-miss occurred, and a server device that constructs a predictive model that predicts the occurrence of a near-miss through machine learning using training data extracted based on the first information and the second information, wherein either the information processing device or the server device includes an extraction unit that extracts partial information from the first information within a predetermined period prior to the time the near-miss occurred, and the server device constructs the predictive model using the partial information extracted by the extraction unit as training data. [Effects of the Invention]

[0007] According to the above-described aspect, it is possible to predict the occurrence of a near miss. [Brief explanation of the drawings]

[0008] [Figure 1] 1 is a block diagram illustrating a configuration example of a near-miss prediction system according to an embodiment. [Figure 2] 1 is a diagram illustrating an example of mounting various sensors used in a near-miss prediction system. FIG. [Figure 3] 10 is a flowchart illustrating an example of processing performed by an information processing device carried by a worker. [Figure 4] 10 is a flowchart illustrating an example of near-miss information collection processing performed by an information processing device carried by a worker. [Figure 5] 10 is a flowchart illustrating an example of a process for constructing a near-miss prediction model. [Figure 6] FIG. 10 is a diagram illustrating an example of a method for extracting teacher data. [Figure 7] 10 is a flowchart illustrating an example of a near-miss prediction process performed by an information processing device carried by a worker. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

[0010] 1 is a block diagram illustrating an example of the configuration of a near-miss prediction system according to an embodiment. The near-miss prediction system illustrated in FIG. 1 includes an information processing device 1, a posture detection sensor 2, a heart rate monitor 3, a sole pressure sensor 4, an eye camera 5, and a server device 8.

[0011] The information processing device 1 is carried by a worker while working, and may be, for example, a portable computer such as a tablet, a smartphone, or the like. The information processing device 1 is a collection device that collects near-miss information used to build a near-miss prediction model, and is also a prediction device that predicts the occurrence of a near-miss for a worker carrying the information processing device 1 using the built near-miss prediction model. The information processing device 1 is an example of a first computer in the near-miss prediction system of this embodiment. The information processing device 1 includes a control unit 100, a memory unit 110, an input unit 120, a display unit 130, an audio output unit 140, an information collection unit 150, and a communication unit 160.

[0012] The control unit 100 controls various operations of the information processing device 1. The control of operations by the control unit 100 is performed, for example, by having a processor such as a CPU (Central Processing Unit) execute an OS (Operating System) program and various application programs. The control unit 100 of the information processing device 1 according to this embodiment functions as an inference unit 101 and a notification processing unit 102. The inference unit 101 predicts the occurrence of a near miss using a trained model (near miss prediction model) described below. When the inference unit 101 predicts the occurrence of a near miss, the notification processing unit 102 notifies a worker or the like carrying the information processing device 1 that a near miss is predicted to occur and urges the worker to take a break.

[0013] The storage unit 110 stores various programs executed by the processor serving as the control unit 100, data referenced during program execution, data acquired during program execution, and the like. The data acquired during program execution includes a data log 111 and near-miss information 112. The data log 111 includes information related to the physical condition of a worker carrying the information processing device 1, detected (measured) using the posture detection sensor 2 or the like while the worker is performing work. The near-miss information 112 includes information indicating the time when a near-miss occurred to the worker. The storage unit 110 includes a random access memory (RAM) and a read-only memory (ROM). The storage unit 110 may include, for example, an auxiliary storage device such as a solid state drive (SSD) or a hard disk drive (HDD). The storage unit 110 may include, for example, a portable recording medium detachable from the information processing device 1, such as a card-type memory device (memory card) or a universal serial bus (USB) memory.

[0014] The input unit 120 accepts input of operation information related to the operation of the information processing device 1. The display unit 130 visualizes and displays various information related to the operation of the information processing device 1. The input unit 120 and the display unit 130 may be a touch panel display in which a position detector (digitizer) serving as the input unit 120 is superimposed on the display area of ​​a liquid crystal display serving as the display unit 130. The input unit 120 may include a keyboard, a mouse, etc. The display unit 130 may include a display device such as a liquid crystal display capable of displaying characters, images, etc., as well as a display device that presents information by the light emission state (e.g., off, on, and blinking) of an LED lamp or the like. Furthermore, the display unit 130 is not limited to a display device or display device built into the information processing device 1, but may also include an external device connected to the information processing device 1, such as an LED light worn by a worker during work or an HMD (Head Mounted Display). The audio output unit 140 outputs (emits sound) information related to the operation of the information processing device 1. The audio output unit 140 is not limited to a sound emitting device built into the information processing device 1, such as a speaker, but may also include an external device connected to the information processing device 1, such as earphones worn by a worker. The input unit 120 is used to input information indicating that a near miss has occurred. The display unit 130 and the audio output unit 140 are examples of output units that can be used to notify that a near miss is predicted to occur or to encourage a break.

[0015] The information collection unit 150 connects the information processing device 1 to an external device used to collect information about the physical condition of the worker that can be used to predict near misses. The information collection unit 150 of the information processing device 1 illustrated in FIG. 1 is connected to each of the posture detection sensor 2, heart rate monitor 3, sole pressure sensor 4, and eye camera 5. The information collection unit 150 and each of the posture detection sensor 2, heart rate monitor 3, sole pressure sensor 4, and eye camera 5 may be connected via wireless communication according to a known short-range wireless communication standard, or via a transmission cable such as a USB cable (i.e., wired). The posture detection sensor 2 is a sensor used to detect the posture of the worker, and for example, a set of multiple inertial measurement units (IMUs) 201, 202, 203, ... is used. The heart rate monitor 3 is a sensor that measures the worker's heart rate. The sole pressure sensor 4 is a sensor used to measure the sole pressure of the worker, and for example, a set of sole pressure sensor 401 for the right foot and sole pressure sensor 402 for the left foot is used. The eye camera 5 is a camera that captures images of the worker's eyes (eyeballs and surrounding areas) to obtain (measure) information related to the eyes, such as gaze and blinking.

[0016] The communication unit 160 connects the information processing device 1 to a communication network 9 such as the Internet or a LAN (Local Area Network), and communicates with a communication terminal such as a server device 8 via the communication network 9. The communication unit 160 transmits the data log 111 and near-miss information 112 acquired by the information processing device 1 to the server device 8. The communication unit 160 also receives a near-miss prediction model (trained model) from the server device 8.

[0017] The server device 8 constructs a trained model used to predict the occurrence of near misses based on the data log and near miss information acquired from the information processing device 1, and provides the trained model to the information processing device 1. The server device 8 is an example of a second computer in the near miss prediction system of this embodiment. The server device 8 includes a control unit 800, a storage unit 810, and a communication unit 860.

[0018] The control unit 800 controls various operations of the server device 8. The control unit 800 controls the operations by, for example, causing a processor such as a CPU to execute an OS program or various application programs. The control unit 800 of the server device 8 according to this embodiment functions as an extraction unit 801 and a learning unit 802. The extraction unit 801 extracts, from the data log, training data to be used for constructing a near-miss prediction model. Specifically, the extraction unit 801 extracts, from the data log, partial data from a predetermined period prior to the occurrence time of the near-miss identified by the near-miss information as training data. The learning unit 802 constructs a near-miss prediction model through supervised machine learning using the extracted training data. The learning unit 802 constructs a near-miss prediction model according to any of the known supervised machine learning learning methods.

[0019] The storage unit 810 stores various programs executed by the processor serving as the control unit 800, data referenced during program execution, data generated during program execution, etc. The data generated during program execution includes teacher data 811 and a trained model (near-miss prediction model) 812. The storage unit 810 includes RAM and ROM. The storage unit 110 may include an auxiliary storage device such as an HDD or SSD, for example.

[0020] The communication unit 860 connects the server device 8 to a communication network 9 and communicates with communication terminals such as the information processing device 1 via the communication network 9. The communication unit 860 receives data logs and near-miss information from the information processing device 1. The communication unit 860 also transmits a near-miss prediction model to the information processing device 1.

[0021] 1, the near-miss prediction system according to the present embodiment may include a plurality of information processing devices 1. The server device 8 can construct a near-miss prediction model based on the data logs and near-miss information received from each of the plurality of information processing devices 1.

[0022] As described above, the information processing device 1 according to this embodiment is a collection device that collects near-miss information to be used in building a near-miss prediction model, and is also a prediction device that uses the built near-miss prediction model to predict the occurrence of near-misses for workers carrying the information processing device 1. Information on the physical condition of the worker is used to collect near-miss information and predict near-misses. In this embodiment, a posture detection sensor 2, a heart rate monitor 3, a sole pressure sensor 4, and an eye camera 5 are used as sensors for acquiring information on the physical condition of the worker.

[0023] FIG. 2 is a diagram illustrating an example of how various sensors used in the near-miss prediction system are attached. A worker 10 carries the information processing device 1, for example, in a breast pocket (not shown) of his / her jacket while working. Using the information processing device 1, the worker 10 can, for example, check the situation of the work location, the work content, record work results, and record the occurrence of near-misses. In addition, the information processing device 1 acquires information regarding the physical condition of the worker 10, as described above.

[0024] The posture detection sensor 2 is a sensor used to detect the posture of the worker 10, and may be, for example, a set of the above-described multiple inertial measurement units (IMUs) 201, 202, 203, ... FIG. 2 shows an example in which five inertial measurement units 201 to 205 are worn on the worker 10. The first inertial measurement unit 201 is worn on the head of the worker 10 (for example, inside the helmet 11). The second inertial measurement unit 202 is worn on the chest of the worker 10 (for example, in a breast pocket of a jacket). The third inertial measurement unit 203 is worn on the waist of the worker 10 (for example, in the center of the back waist of the worker's trousers). The fourth inertial measurement unit 204 and the fifth inertial measurement unit 205 are worn on the right ankle and the left ankle of the worker 10, respectively. The five inertial measurement units 201-205 are connected to the information processing device 1, and output information from the inertial measurement units 201-205 is stored in the storage unit 110 of the information processing device 1 as part of a data log 111 in a state in which the measurement time can be identified. A method for detecting the posture of the worker 10 based on the output information from the inertial measurement units 201-205 is known, and therefore a detailed description of the posture detection method will be omitted in this specification. The posture of the worker 10 detected based on the output information from the inertial measurement units 201-205 can be used, for example, to determine the fatigue level of the worker 10. For example, if the detected posture of the worker 10 is hunched over or the leg movements (e.g., leg lift, stride length, etc.) become small, it is predicted that the fatigue level of the worker 10 is high and that a near-miss such as a stumble or fall is likely to occur. Note that the number of inertial measurement units worn by the worker 10 and the locations at which they are worn are not limited to the combinations exemplified in FIG. 2 and can be changed as appropriate.

[0025] The heart rate monitor 3 is a sensor that measures the heart rate and other parameters of the worker 10, and is worn, for example, on the left wrist of the worker 10. The heart rate monitor 3 is connected to the information processing device 1, and output information from the heart rate monitor 3 is stored in a storage unit 110 of the information processing device 1 as part of a data log 111 in a state in which the measurement time can be identified. The output information from the heart rate monitor 3 can be used, for example, to determine the level of tension of the worker 10 or to detect changes (deterioration) in physical condition at an early stage. For example, if the heart rate of the worker 10 increases, it is predicted that the worker 10's level of tension will increase due to fatigue or the like, or that his or her physical condition will deteriorate, making him or her more susceptible to near misses.

[0026] The sole pressure sensor 4 is a sensor used to detect the sole pressure of the worker 10, and may be, for example, a set of a sole pressure sensor 401 provided on the sole of the right shoe 12R and a sole pressure sensor 402 provided on the sole of the left shoe 12L. The output information of the sole pressure sensors 401 and 402 is stored in the storage unit 110 of the information processing device 1 as part of the data log 111 in a state in which the measurement time can be identified. The sole pressure sensors 401 and 402 are connected to the information processing device 1, and the output information of the sole pressure sensors 401 and 402 is stored in the storage unit 110 of the information processing device 1 as part of the data log 111 in a state in which the measurement time can be identified. The output information of the sole pressure sensors 401 and 402 can be used, for example, to determine the fatigue level of the worker 10. For example, if the pressure detected by the sole pressure sensors 401 and 402 becomes low, it is predicted that fatigue will make it difficult to lift the legs, and that near misses such as stumbling or falling will be more likely to occur. Also, for example, if the pressure detected by the sole pressure sensors 401 and 402 fluctuates irregularly, it is predicted that dizziness or the like will cause the body to become unsteady, resulting in a near miss.

[0027] The eye camera 5 is a camera that captures an image of the eyes 1001 of the worker 10, and is used to detect eye conditions of the worker 10, such as line of sight and blinking. The eye camera 5 is attached, for example, to the brim of the helmet 11 in an orientation that allows it to capture an image of the eyes 1001. The imaging range (angle of view) of the eye camera 5 may include parts of the face of the worker 10 that are further outside the eyes 1001. In that case, it is desirable to extract the part of the eyes 1001 by performing known image processing such as cropping on the image captured by the eye camera 5, so that it becomes difficult to identify the worker 10. The image of the eyes 1001 captured by the eye camera 5 is stored in the storage unit 110 of the information processing device 1 as part of the data log 111 in a state in which the measurement time can be identified. The image of the eyes 1001 can be used to analyze, for example, sudden changes in the gaze of the worker 10, changes over time, the number of blinks, whether the eyelids are open, etc. These analysis results can be used, for example, to determine whether a near miss has occurred for the worker 10, to determine the level of concentration, and to detect changes (deterioration) in physical condition at an early stage. For example, the gaze of the worker 10 will suddenly change if the worker 10 stumbles. Furthermore, if the change in the gaze of the worker 10 over time is significantly different from normal, it is predicted that a near miss such as stumbling is likely to occur due to a decrease in concentration. Furthermore, for example, if the number of blinks is significantly different from normal, it is predicted that a near miss is likely to occur due to drowsiness or a deterioration in physical condition.

[0028] The sensors used to acquire information about the physical condition of the worker 10 are not limited to the above combination and can be changed as appropriate. For example, in addition to the above combination of sensors, a temperature sensor that measures (detects) the facial surface temperature or body temperature of the worker 10 may be included.

[0029] Fig. 3 is a flowchart illustrating an example of processing performed by an information processing device carried by a worker. In the flowchart illustrated in Fig. 3, a pair of double lines extending horizontally indicates that the two processes between the pair of double lines are performed in parallel.

[0030] For example, when starting work for the day, the worker 10 wears at least one or more of the above-described sensors, such as the posture detection sensor 2, heart rate monitor 3, sole pressure sensor 4, and eye camera 5, and connects the worn sensors to the information processing device 1. Thereafter, when the worker 10 performs a predetermined operation on the information processing device 1, the information processing device 1 acquires and updates, for example, a near-miss prediction model in response to the predetermined operation by the worker 10 (step S1). In step S1, the information processing device 1 determines whether a prediction model has been acquired (whether the inference unit 101 is valid or not). If not, the information processing device 1 acquires the prediction model from the server device 8. If the prediction model has been acquired, the information processing device 1 communicates with the server device 8 and determines whether the prediction model applied to the inference unit 101 is identical to the trained model 812 of the server device 8. If they are not identical, the information processing device 1 acquires the trained model 812 of the server device 8 and updates the prediction model of the inference unit 101.

[0031] After step S1, the information processing device 1 starts recording a data log in response to a predetermined operation by the worker 10 (step S2). In step S2, the information processing device 1 starts a process of recording output information from sensors such as the posture detection sensor 2, heart rate monitor 3, sole pressure sensor 4, and eye camera 5 described above as a data log 111 in the storage unit 110. After recording of the data log has started, the worker 10 starts work.

[0032] After step S2, the information processing device 1 performs a near-miss information collection process (step S3) and a near-miss prediction process (step S4) in parallel, as illustrated in FIG. 3. Step S3 is a process of collecting near-miss information including the time of occurrence of a near-miss. A specific example of step S3 will be described later with reference to FIG. 4. Step S4 is a process of inputting the output information of each sensor recorded in the data log 111 into the inference unit 101 (a near-miss prediction model) and predicting the occurrence of a near-miss. A specific example of step S4 will be described later with reference to FIG. 7. The processes of steps S3 and S4 end, for example, when the worker 10 performs a predetermined end operation on the information processing device 1. The predetermined end operation is associated with, for example, an operation to end the recording of the data log. For example, when the worker 10 finishes his or her work for the day and performs an operation to end the recording of the data log, the information processing device 1 ends the recording of the data log (not shown in FIG. 3), and ends the processes of steps S3 and S4.

[0033] When the processes of steps S3 and S4 are completed, the information processing device 1 transmits data to the server device 8 (step S5). In step S5, the information processing device 1 transmits, for example, output information of each sensor recorded in the data log 111 to the server device 8. Furthermore, if the near-miss information 112 contains valid information, the information processing device 1 transmits the near-miss information together with the output information of each sensor to the server device 8. When the transmission of data to the server device 8 is completed, the information processing device 1 terminates the process related to predicting the occurrence of a near-miss, illustrated in FIG. 3.

[0034] In this way, the information processing device 1 according to this embodiment performs a near miss information collection process (step S3) that collects near miss information including the time of occurrence when a near miss occurs while the worker 10 is performing work, and a near miss prediction process (step S4) that predicts the occurrence of a near miss using the output information of each sensor.

[0035] As the near-miss information collection process in step S3, the information processing device 1 performs a process according to the flowchart illustrated in FIG. 4, for example.

[0036] 4 is a flowchart illustrating an example of a near-miss information collection process performed by an information processing device carried by a worker. In the flowchart of FIG. 4, hexagonal blocks represent blocks where determination is made.

[0037] In the near-miss information collection process, the information processing device 1 determines whether an input operation indicating the occurrence of a near-miss has been performed (step S301) and whether an operation to end the recording of the data log has been performed (step S302). When the operation to end the recording of the data log has been performed (step S302; YES), the information processing device 1 ends the near-miss information collection process.

[0038] If no input operation indicating the occurrence of a near miss has been performed (step S301; NO), and if no operation to end the recording of the data log has been performed (step S302; NO), the information processing device 1 repeats the judgments of steps S301 and S302.

[0039] If an input operation indicating the occurrence of a near miss is performed (step S301; YES), the information processing device 1 records near miss information including the time of the near miss occurrence based on the time the input operation was performed (step S303).If an operation to end the recording of the data log is not performed even after step S303 (step S302; NO), the information processing device 1 repeats the determinations of steps S301 and S302.

[0040] When the near-miss information collection process is completed, the information processing device 1 transmits data to the server device 8 as described above with reference to Fig. 3. If the near-miss information including the time of the near-miss occurrence is recorded in step S303, the information processing device 1 transmits the recorded near-miss information to the server device 8 together with a data log related to the physical condition of the worker 10.

[0041] The server device 8 receives data logs and near-miss information related to the physical condition of the workers 10 from the information processing device 1, and constructs a near-miss prediction model (trained model 812) through machine learning using the received data logs and near-miss information. The server device 8 receives data including data logs and near-miss information about the multiple workers 10 transmitted from each of the multiple information processing devices 1, and constructs a near-miss prediction model.

[0042] 5 is a flowchart illustrating an example of a process for constructing a near-miss prediction model. The server device 8 starts the process for constructing the prediction model illustrated in FIG. 5, for example, when an operation is performed by an operator (administrator) of the server device 8 or when a scheduled time arrives.

[0043] The server device 8 first determines whether or not new near-miss information is present in the data received from the information processing device 1 (step S11). If there is no new near-miss information (step S11; NO), the server device 8 ends the process related to the construction of the prediction model exemplified in FIG.

[0044] If there is new near-miss information (step S11; YES), the server device 8 extracts and stores teacher data to be used in building a prediction model from the data log associated with the new near-miss information, based on the information on the time the near-miss occurred included in the new near-miss information (step S12). In step S12, the server device 8 extracts and stores partial data from the data logs, which are output information from each sensor, within a predetermined period before the time the near-miss occurred, as teacher data. In step S12, the server device 8 extracts teacher data for each time the near-miss occurred included in the new near-miss information.

[0045] After step S12, server device 8 determines whether or not to perform machine learning (step S13). For example, if the number of new training data extracted in step S12 is small (below a threshold) and it is difficult to perform meaningful machine learning, server device 8 determines not to perform machine learning (step S13; NO) and terminates the process related to building the prediction model exemplified in Fig. 5. In this case, the training data extracted in step S12 is added to training data 811 in storage unit 810 of server device 8 and will be used for the next machine learning.

[0046] When machine learning is performed (step S13), the server device 8 constructs a near-miss prediction model by machine learning using the extracted teacher data (step S14). In step S14, the server device 8 constructs a near-miss prediction model according to a known supervised machine learning method. Furthermore, if a near-miss prediction model has already been constructed, in step S14 the server device 8 reconstructs (updates) the prediction model by re-learning using the teacher data extracted in step S12.

[0047] After step S14, the server device 8 stores the constructed prediction model as the trained model 812 (step S15), and ends the process relating to the construction of the prediction model illustrated in FIG.

[0048] In this way, the server device 8 of this embodiment constructs a near miss prediction model through machine learning using data from the data log regarding the physical condition of the worker 10 prior to the time of the near miss as training data, rather than data from the time of the near miss.

[0049] Fig. 6 is a diagram illustrating an example of a method for extracting teacher data. Fig. 6 shows an example of information associated with the occurrence of a near miss in a data log 111A related to the physical condition of a worker 10 detected (measured) by one sensor.

[0050] 6 is a data log showing, for example, the change over time in the heart rate of the worker 10 measured by the heart rate monitor 3, and D(0999) to D(1026) show the heart rate at each measurement time. For example, if the worker 10 experiences a near miss such as tripping or falling at time T1 while working, the worker 10 inputs the occurrence of the near miss into the information processing device 1 at time T2, which is later than time T1.

[0051] On the other hand, when a near miss occurs, there are often changes in the physical condition of the worker 10 that are a sign of the near miss before the time of the occurrence. If the occurrence of a near miss could be predicted based on such changes in the physical condition that are a sign of the occurrence of a near miss, the occurrence of the near miss (and the occurrence of an industrial accident) could be prevented. For this reason, in the near miss prediction system of this embodiment, partial data within a predetermined period before the time of the near miss occurrence is extracted from the data log 111A and used as training data for the prediction model. The time of the near miss occurrence recorded in the information processing device 1 is the time T2 when the worker 10 inputs the data into the information processing device 1, which is after the time T1 when the near miss actually occurred. Furthermore, the period during which changes that are a sign of the occurrence of a near miss become noticeable varies depending on the physical condition being measured. Therefore, the period T0 to T1 extracted from the data log 111A as teaching data is set based on, for example, the time required from when a near miss actually occurs until the worker 10 inputs it into the information processing device 1, and the physical condition represented by the data log 111A.

[0052] 5 and 6 is provided to the information processing device 1 and is used in the near-miss prediction process (step S4) performed by the information processing device 1. As the near-miss prediction process in step S4, the information processing device 1 performs processing, for example, in accordance with the flowchart illustrated in FIG.

[0053] 7 is a flowchart illustrating an example of near-miss prediction processing performed by an information processing device carried by a worker. In the flowchart of FIG. 7, hexagonal blocks represent blocks where determination is made.

[0054] In the near-miss prediction process, the information processing device 1 inputs data indicating the physical condition of the worker 10 collected via the information collecting unit 150 into a near-miss prediction model (inference unit 101) (step S401). In step S401, the information processing device 1 extracts partial data within a period corresponding to a predetermined period when teacher data is extracted in the server device 8 for each log of output information of each sensor recorded in the data log 111, and inputs the extracted partial data into the prediction model. The prediction model (inference unit 101) outputs a prediction result of the occurrence of a near-miss based on the input data. For example, the prediction model outputs either a value (e.g., "1") suggesting that there is a high possibility that a near-miss will occur in the near future, or a value (e.g., "0") suggesting that there is an extremely low possibility that a near-miss will occur at the present time, as the prediction result.

[0055] After step S401, the information processing device 1 determines whether or not the output of the prediction model suggests the occurrence of a near miss (step S402). If the output does not suggest the occurrence of a near miss, in other words, if the output suggests that the possibility of a near miss occurring is extremely low at the present time (step S402; NO), the information processing device 1 determines whether or not an operation to end the recording of the data log has been performed (step S403).

[0056] On the other hand, if the output of the prediction model suggests the occurrence of a near miss (step S402; YES), the information processing device 1 notifies the worker 10 or the like that there is a high possibility of a near miss occurring in the near future and encourages the worker 10 to take a break (step S404). In step S404, the information processing device 1 notifies the worker 10 of the high possibility of a near miss occurring and encourages the worker 10 to take a break, for example, by means of a display screen on the display unit 130 or a sound output (emitted as sound) from the sound output unit 140. The notification to the worker 10 in step S404 may be performed using an external device connected to the information processing device 1 that can function as the display unit 130 or the sound output unit 140, such as an LED light, HMD, or earphones worn by the worker 10. The notification to the worker 10 in step S404 may also be performed by vibration using a vibration function implemented in a smartphone or the like, or by a combination of vibration and another notification method. Furthermore, in step S404, the information processing device 1 may notify, for example, a worker other than the worker 10 carrying the information processing device 1, a supervisor, or the like, that there is a high possibility of a near miss occurring. After step S404, the information processing device 1 determines whether an operation to end recording of the data log has been performed (step S403).

[0057] If an operation to end the data log recording is not performed during the execution of the near-miss prediction process (step S403; NO), the information processing device 1 continues the near-miss prediction process. Then, if an operation to end the data log recording is performed (step S403; YES), the information processing device 1 ends the near-miss prediction process illustrated in FIG.

[0058] In this way, the information processing device 1 according to this embodiment predicts whether or not a near miss is likely to occur in the near future using a near miss prediction model (inference unit 101) constructed by machine learning using data related to the physical condition of the worker 10 measured (detected) before the near miss occurs. This makes it possible to prevent near misses from occurring.

[0059] Furthermore, since the occurrence of a near miss is predicted using the physical condition of the worker 10 detected (measured) by the posture detection sensor 2, heart rate monitor 3, sole pressure sensor 4, and eye camera 5, etc., it is possible to predict the occurrence of a near miss even when the worker 10 himself is not aware of a change in his physical condition, etc. Therefore, it is possible to prevent the occurrence of work-related accidents, etc. that are more serious than a near miss caused by, for example, a deterioration in the physical condition of the worker 10.

[0060] 1 to 7 are merely examples for facilitating understanding of the present invention. That is, the near-miss prediction system and near-miss prediction method according to the present invention are not limited to the above-described embodiments, and various modifications and changes are possible without departing from the scope of the claims.

[0061] For example, in the near-miss prediction system according to the present invention, the extraction unit 801 (see FIG. 1) that extracts teacher data from the data log may be provided in the information processing device 1 rather than in the server device 8. In this case, the data transmitted from the information processing device 1 to the server device 8 may be only the teacher data, or may be data including the teacher data, the data log, and near-miss information.

[0062] Furthermore, the teacher data extracted from the data log may be, for example, data based on actual measurements obtained while the worker 10 is working, or data indicating the difference between normal values ​​measured during normal times by the worker 10 and actual measurements obtained while working. Which data is to be used as teacher data can be determined for each type of data to be acquired.

[0063] In addition to the time when the near miss occurred, the near miss information may also include information such as the type of near miss that occurred and the location where the near miss occurred.

[0064] Furthermore, as a near miss prediction model, for example, multiple types of prediction models optimized for each work environment constructed from data on the physical conditions of multiple workers 10 working in similar work environments may be prepared, and a prediction model according to the work environment in which the worker will be working may be provided to the information processing device 1. The work environment may be, for example, the location where the work is performed (e.g., outdoors, indoors, high up, etc.), the work content, the season in which the work is performed, etc.

[0065] Furthermore, the information processing device 1 carried by the worker 10 during work is not limited to a general-purpose computer, but may be a device that combines dedicated hardware for executing the above-described processes or a device that includes dedicated hardware. For example, some of the functions of the information processing device 1 described above with reference to FIG. 1 may be realized by an FPGA (Field Programmable Gate Array) or an ASIC (Application Specific Integrated Circuit). Furthermore, the information processing device 1 may have one or more built-in sensors that can measure (detect) the physical condition of the worker 10.

[0066] Furthermore, the information processing device 1 in the near-miss prediction system according to the present invention may be configured, for example, by a first information processing device that performs the near-miss information collection processing as described above with reference to Fig. 4, and a second information processing device that performs the near-miss prediction processing as described above with reference to Fig. 7. Furthermore, the information processing device 1 and the server device 8 in the near-miss prediction system according to the present invention may be directly connected by wireless communication according to a known short-range wireless communication standard or by a transmission cable such as a USB cable, without going through the communication network 9 described above with reference to Fig. 1. [Explanation of symbols]

[0067] 1. Information processing equipment 8 Server equipment 9. Communication Networks 100, 800 control section 101 Reasoning section 102 Notification processing unit 110 Storage section 111 Data Log 112 Near miss information 120 Input section 130 Display section 140 Audio output section 150 Interface section 160 Communications Department 2. Attitude detection sensor 201, 202, 203, 204, 205 Inertial Measurement Unit 3. Heart rate monitor 4, 401, 402 Plantar pressure sensor 5. Eye camera 10 workers 1001 (Worker's) Eyes 11. Helmet 12R, 12L shoes

Claims

1. an information processing device that acquires first information regarding the physical condition of a worker during work and second information including the time of occurrence of a near miss; a server device that constructs a prediction model that predicts the occurrence of near misses by machine learning using teacher data extracted based on the first information and the second information; Including, One of the information processing device and the server device includes an extraction unit that extracts partial information within a predetermined period before the occurrence time of the near miss from the first information, The server device constructs the prediction model using the partial information extracted by the extraction unit as training data. A near-miss prediction system characterized by:

2. The information processing device includes an inference unit that inputs the first information into the prediction model to predict the occurrence of a near miss. The near-miss prediction system according to claim 1 .

3. The information processing device includes an output unit that outputs, when the inference unit predicts the occurrence of a near miss, information indicating that the occurrence of the near miss has been predicted. The near-miss prediction system according to claim 2 .

4. the first information includes information relating to a plurality of types of physical conditions; The predetermined period for extracting the partial information from the information on the physical condition in the extracting unit is set for each type of information.

4. The near-miss prediction system according to claim 1, wherein:

5. Acquiring, by a first computer, first information regarding the physical condition of the worker during work and second information including the time of occurrence of the near miss; Extracting partial information within a predetermined period before the occurrence time of the near miss from the first information by a second computer or the first computer; constructing, by the second computer, a prediction model for predicting the occurrence of near misses through machine learning using the partial information extracted based on the first information and the second information as training data; A near-miss prediction method characterized by carrying out the following.

6. a collection unit that collects information on the physical condition of the worker while working; an inference unit that inputs the information collected by the collection unit into a prediction model that predicts the occurrence of a near miss, the prediction model being constructed by machine learning using, as training data, partial information within a predetermined period prior to the occurrence of the near miss out of the first information extracted based on first information regarding the physical condition of the worker during work and second information including the time of the near miss; and predicts the occurrence of the near miss; an output unit that outputs information indicating that a near miss has been predicted when the inference unit predicts the occurrence of a near miss; 10. An information processing device comprising:

Images