JP2022029791A - Safety belt use state monitoring system - Google Patents

Abstract

To provide a safety belt use state monitoring system capable of determining a use state of a safety belt further accurately by using an acceleration sensor.SOLUTION: A safety belt use state monitoring system comprises: an acceleration sensor 10 fitted to a hook 5 or the like of a safety belt 4; a transmitter 11 and a receiver 12 for transmitting/receiving an information signal R of acceleration data detected on the acceleration sensor 10; and a hook use determination section 13 for, on the basis of the received acceleration data, determining whether the hook 5 is used. The hook use determination section 13 comprises: an arithmetic unit 25 for determining a feature amount from the acceleration data; a machine learning unit 26; and a determination unit 27 for determining whether the hook 5 is used, by estimation performed by the machine learning unit 26. The machine learning unit 26 includes: a learning unit 31 for generating a learning model on the basis of the feature amount determined on the arithmetic unit 25; and an estimation unit 32 for estimating whether the hook is used, on the basis of the learning model.SELECTED DRAWING: Figure 2

Description

This disclosure relates to a safety belt usage monitoring system.

Conventionally, workers who work at heights in construction work are obliged to use a safety belt to prevent a fall accident.

This type of safety belt comprises, for example, a safety belt (wearing body) worn on the worker's body, a rope connected to one end of the safety belt, and a hook connected to the other end of the rope. ing. Then, when a worker wears a safety belt and hooks a hook hook member (fixed support member) such as a main rope or a handrail provided at the work site to perform the work, such as when the worker steps off the foot. Even so, the worker's body is supported by the safety belt and the hook hooking member such as the main rope or the handrail on which the hook is hooked, and the worker's body can be avoided from falling.

On the other hand, the supervisor who manages and supervises the work (site supervisor (responsible) person, construction supervisor, etc.) needs to confirm whether the worker is using the safety belt, especially the hook properly.
However, there are many cases where it is difficult to visually confirm the situation of workers working at high places at a distance such as on the ground. For this reason, many methods for checking the usage status of safety belt hooks have been proposed and put into practical use.

For example, Patent Document 1 states that an acceleration sensor is attached to a hook of a safety belt, and when the acceleration sensor detects an acceleration of a predetermined value or more, the hook is locked (hooked) to a hook hooking member at a work site. A safety belt usage monitoring system (use of safety belt) that makes a judgment and uses a locking detection means to compare with the reference value according to the work pattern and judge whether or not the hook is properly locked. Status confirmation system) is disclosed.

In the safety belt usage monitoring system of Patent Document 1, when the worker locks the hook to the hook hook member, the acceleration sensor causes an acceleration of a predetermined gravitational acceleration or more (for example, 2G or more) downward in the plane direction of the hook. Detect the detected timing. Then, the timing at which the gravitational acceleration of a predetermined value or more is detected is stored downward in the plane direction, and it is determined whether or not the worker has locked the hook to the hook hook member.

In addition, a disengagement detection sensor is provided to detect that the hook has been disengaged from the locked portion of the mounted body, and the disengagement detection sensor detects when the hook is disengaged from the locked portion of the mounted body such as a safety belt. It is possible to clearly distinguish between the case where the hook is locked to the body and the case where the hook is locked to the hook hook member.

On the other hand, in the safety belt usage monitoring system of Patent Document 1, when the hook is locked to the hook hooking member at the work site, the hook is locked in order to detect whether or not the hook is locked "properly". The detection means is used to make the following judgments (a) to (e).

(A) Judgment by comparing the number of locks per predetermined time with the reference value For example, the number of locks per predetermined time such as 20 times / 5 minutes is stored as a reference value and exceeds this reference value. In some cases, it is judged to be "appropriate". In addition, a reference value is set for each work pattern, and the reference value of the worker corresponding to the work pattern is compared with the locking timing in which acceleration of a predetermined value or more such as 2G or more is detected. Determine if the stop state is appropriate.

(B) Judgment by comparing the grouped locking state with the reference value For example, a series of locking operations performed within a few seconds is judged as one locking, and the grouped number of locking times and the reference value are used. By comparing with a certain number of locking times, it is determined whether or not the locking is performed "appropriately".

(C) Judgment by comparing the movement distance and the number of locking times For example, when the movement in a certain direction is detected by the acceleration sensor, it is judged whether or not the locking operation is performed within the movement range of the predetermined distance by the reference value or more. Then, it is determined whether or not the locking is performed "appropriately".

(D) Judgment based on the time interval between the previous lock and the current lock For example, the timer is operated from the time of the previous lock to determine whether the next lock is performed within a predetermined time. By doing so, it is determined whether or not the locking is performed "properly".

(E) Judgment when multiple hooks are provided For example, when two hooks or ropes are provided and work is performed while alternately locking the hooks, the locking on one hook is followed by the next locking. Whether or not the locking is performed "properly" is determined based on whether or not the time until the hook overlaps with the time from the locking of the other hook to the next locking. At this time, an acceleration sensor is attached to each hook so that the locked state of each can be determined.

Then, in the safety belt usage status monitoring system of Patent Document 1, when it is detected that the safety belt is not locked "appropriately" by making the judgments (a) to (e) above, the worker In addition to notifying the alarm to, it is configured to store it in the storage unit as a history of the locked state of the worker, perform printing processing, and the like.

Japanese Unexamined Patent Publication No. 2019-5425

Here, in the safety belt usage monitoring system of Patent Document 1, a threshold value equal to or higher than a predetermined gravitational acceleration such as 2G is simply set, and at the timing when an acceleration exceeding this threshold value is detected, the main rope of the hook by the worker or the like. A locking operation (locking operation) on a hook hook member such as a handrail is detected and discriminated.

However, the thinking movements of workers who perform construction work and the like are naturally diverse, and it is very difficult to determine the hook locking operation with a certain threshold value. Therefore, the safety belt usage monitoring system of Patent Document 1 Then, the accuracy of specifying and detecting the locking operation of the hook is low, and it is difficult to put it into practical use.

Further, in the safety belt usage monitoring system of Patent Document 1, as described above, whether or not the hook is "properly" locked is determined by using the acceleration sensor and the locking detecting means. I try to judge by e).

However, it is difficult to accurately identify and discriminate the thinking movements of a wide variety of workers by the judgments such as (a) to (d) except for (e), that is, the other movements of the worker are hooked. It is easy to imagine that it is a locking operation, and it must be said that it is difficult to put it into practical use.

In addition, there are many construction sites where dozens, and in some cases hundreds, of workers work at heights.
On the other hand, in the safety belt usage monitoring system of Patent Document 1, since the discrimination accuracy of the hook locking operation is low, a large number of cases are detected that the locking operation is not "appropriate", and the worker is informed. There is a risk that frequent alarm notifications will occur, a huge amount of incorrect history will be recorded, and printing processing will occur. As a result, the worker will stop the work every time the alarm is falsely reported, and the supervisor may be overwhelmed by the information that the locking operation is not "appropriate", and the construction site may be overwhelmed. It can be confusing.

In view of the above circumstances, it is an object of the present disclosure to provide a safety belt usage status monitoring system that enables more accurate determination of the usage status of a safety belt using an acceleration sensor.

(1) One aspect of the safety belt usage status monitoring system of the present disclosure is a safety belt usage status monitoring system for monitoring the usage status of the safety belt in a worker who works at a high place, and is a rope of the safety belt. An acceleration sensor mounted on one end of the safety band or a hook of the safety band, a transmitter and a receiver for transmitting and receiving an information signal including at least the acceleration data detected by the acceleration sensor, and the received acceleration data. Based on this, the hook use determination unit for determining whether or not the hook is used is provided, and the hook use determination unit includes a calculation unit for obtaining a feature amount from the acceleration data, and the hook from the feature amount. The machine learning unit includes a machine learning unit that estimates whether or not the data is used, and the machine learning unit includes a preprocessing unit that extracts feature extraction data from the feature amount obtained by the calculation unit, and a preprocessing unit that processes the feature extraction data. It includes a learning unit that generates a learning model based on the processed data, and an estimation unit that estimates whether or not the hook is used based on the learning model.

According to the safety belt usage monitoring system of (1) above, a learning model can be created in which the acceleration data of the operation with the hook attached is accurately extracted, and evaluation and judgment are performed based on this learning model. It is possible to judge the hook usage status with much higher accuracy than the conventional safety belt usage monitoring system.

(2) One aspect of the safety belt usage status monitoring system of the present disclosure is the safety belt usage status monitoring system of (1) above, and the learning unit processes the feature amount using an SVM to perform a learning model. Is configured to generate.

According to the safety belt usage status monitoring system of (2) above, a learning algorithm of machine learning, SVM, is used to classify some data patterns and determine the usage status of hooks. This makes it possible to determine the usage status of the hook with even higher determination accuracy.

For example, when a learning model is generated using SVM, the accuracy of determining the hook usage status is as high as 70 to 80% or more even under adverse conditions, whereas the decision tree, LSTM (Long Term Short Memory) is used. ) Is used to generate a learning model, it has been confirmed that the judgment accuracy is about 60% to 70%, and by using SVM, the hook usage status can be judged with high judgment accuracy. Will be possible.

(3) One aspect of the safety band usage status monitoring system of the present disclosure is the safety band usage status monitoring system of (1) or (2) above, and the calculation unit is preset from the acceleration data in the time series. Multiple data sets extracted at regular time intervals are generated, and the position of the slide time preset from the start of acceleration data acquisition of the previous data set in time series is used as the acceleration data acquisition start time of the next data set. A plurality of data sets are superimposed on the time axis and normalized to data in a predetermined time unit, and the feature amount is obtained for each data set.

According to the safety belt usage monitoring system of (3) above, it is possible to suitably obtain the feature amount.

(4) One aspect of the safety belt usage status monitoring system of the present disclosure is the safety belt usage status monitoring system according to any one of (1) to (3) above, and the information signal is at least an identification of a worker. Further includes any one of information, worker position information, and work content information.

According to the safety belt usage monitoring system of (4) above, as an information signal, not only the information of the detection result of the acceleration sensor but also the identification information of the worker, the position information of the worker, and the work content information of the worker. By acquiring this information, it is possible to determine the hook usage status with even higher accuracy.

That is, for example, the usage status of the hook can be determined based on the determination criteria (learning data) for each process performed by the worker. As a result, machine learning can be performed for each process performed by the worker, and it becomes possible to further improve the accuracy of determining the usage status of the hook.

(5) One aspect of the safety belt usage status monitoring system of the present disclosure is the safety belt usage status monitoring system according to any one of (1) to (4) above, based on the estimation result of the machine learning unit. Further, a determination unit for finally determining whether or not the hook is used is provided, and an alarm unit for issuing an alarm when the determination unit determines that the hook is not used is "inappropriate".

According to the safety belt usage monitoring system of (5) above, when the hook is determined to be "inappropriate" due to improper use, the alarm unit issues an alarm such as sound, image display, or vibration to hook the hook. It is possible to promptly make workers and supervisors aware that the usage situation of the is unsuitable, and appropriately prevent the workers from falling.

(6) One aspect of the safety belt usage status monitoring system of the present disclosure is the safety belt usage status monitoring system according to any one of (1) to (5) above, to the site where work is performed at a high place or to the site. At least one of the routes is provided with a stationary transmitter that identifies the worker.

According to the safety belt usage monitoring system of (6) above, for example, a worker who carries a receiver (wearing an accessory on which the receiver is attached) that has received the activation radio wave is a hook. It can be seen that the worker who carries the receiver that does not receive the activation radio wave is in a place where it is not necessary to judge the usage status of the hook. As a result, it is possible to identify a worker who needs to determine the usage status of the hook among the workers who carry the receiver, and it is possible to reduce the processing load of the hook use determination unit.

(7) One aspect of the safety belt usage status monitoring system of the present disclosure is the safety belt usage status monitoring system according to any one of (1) to (6) above, in which the acceleration data during the operation of using the hook is obtained. Further, an input unit to be given to the machine learning unit as teacher data is provided.

According to the safety belt usage monitoring system of (7) above, for example, at the start of work, the worker is made to perform an appropriate safety belt and hook usage behavior, and acceleration data detected by a clear proper operation is obtained. Teacher data can be generated based on this and learning can be performed. This makes it possible to improve the accuracy of determining the usage status of the hook relatively easily and at an early stage.

(8) One aspect of the safety belt usage status monitoring system of the present disclosure is the safety belt usage status monitoring system of (7) above, and the input unit is an image captured by an image pickup device that captures the movement of a worker. Based on the above, the acceleration data at the time of an appropriate use operation of the hook is extracted, and the extracted acceleration data is given to the machine learning unit as the teacher data.

According to the safety belt usage monitoring system of (8) above, for example, the movement of the worker is imaged, and the acceleration data when the worker uses the appropriate safety belt and hook in the image is extracted. By generating teacher data based on this, it is possible to improve the accuracy of determining the hook usage status relatively easily and at an early stage.

According to the safety belt usage monitoring system of the present disclosure, it is possible to determine the usage status of the safety belt more accurately by using the acceleration sensor.

It is a figure which shows an example of the safety belt usage state monitoring system which concerns on one Embodiment. It is a block diagram which shows an example of the safety belt usage state monitoring system which concerns on one Embodiment. It is a block diagram which shows an example of the safety belt usage state monitoring system which concerns on one Embodiment. It is a block diagram which shows an example of the safety belt usage state monitoring system which concerns on one Embodiment. It is a figure which shows an example of the flow which performs machine learning in the safety belt usage state monitoring system which concerns on one Embodiment. It is a figure which shows an example of the management terminal in the safety belt usage state monitoring system which concerns on one Embodiment.

Hereinafter, the safety belt usage status monitoring system according to the embodiment will be described with reference to FIGS. 1 to 6.

Here, as shown in FIG. 1, the safety belt usage monitoring system 1 of the present embodiment is for monitoring the usage status of the safety belt 4 of the worker 2 who works at a high place in, for example, construction work. It is a thing.
Further, the "high place" in the present disclosure is a place where the worker 2 may fall (fall) at a distance where injury may occur, and is, for example, the highest of the two points having a height difference of 2 m or more. Refers to the point on the other side. For this reason, "high places" include one underground point.

(Installation)
First, the worker 2 who works at a high place wears the attachment 3. As shown in FIG. 1, the attachment 3 includes clothes (work clothes) 3a, a helmet 3b, a safety belt 4, a waist belt, and the like.

(Safety belt)
Of these, the safety belt 4 in the present embodiment includes a mounting portion 4a to be mounted on the body of the worker 2, at least one hook 5, at least one rope 6 connecting the mounting portion 4a and the hook 5, and a mounting portion. It is provided in 4a and includes at least one hook locked portion 4b for mounting and holding the hook 5 on the mounting portion 4a.

The mounting portion 4a of the present embodiment includes a pair of shoulder belts 4c and 4d wrapped around the shoulder of the worker 2 and having two tips hanging down, and a pair of shoulder belts 4c and 4d wrapped around the waist of the worker 2. A waist belt 4e that connects the waist belts 4e at the waist height position, a chest belt 4f that connects the pair of shoulder belts 4c and 4d at the chest height position, and a pair of shoulder belts 4c and 4d that are wrapped around the thigh of the worker 2. It is provided with a pair of thigh belts 4g and 4h connected to the tip portion.

The "safety belt 4" of the present disclosure does not necessarily have to be configured as described above. For example, the safety belt 4 may be configured to include a pair of shoulder belts 4c and 4d, a chest belt 4f, and a pair of thigh belts 4g and 4h, or may be configured to include only a waist belt 4e. May be good.

The hook 5 of the present embodiment is rotatably connected via a hook-shaped portion (hook main body portion) 5a curved in a substantially U shape and a hinge portion 5c on the base end side of the hook-shaped portion, and is hook-shaped. A rotating portion 5b that opens and closes the open portion of the portion 5a, a lock mechanism portion (not shown) that locks the rotating portion 5b with the open portion of the hook-shaped portion 5a closed, and a lock state released by the lock mechanism portion. It is configured with an unlocking part for doing so.

The hook 5 can be held by hooking the hook-shaped portion 5a on the hook-locked portion 4b of the safety band 4 and attaching it to the body of the worker 2 or by opening and closing the rotating portion 5b to the main rope 8a or the like. The hook-shaped portion 5a can be hooked on the hook hooking member (fixed support member) 8 such as a handrail, or can be removed from the hook hooking member 8.

It should be noted that the "hook 5" of the present disclosure is not necessarily the present embodiment as long as it can be detachably attached to the hook hook member 8 such as the main rope 8a and the handrail to prevent the worker 2 from crashing. It does not have to be configured as. For example, as another example of the "hook 5" of the present disclosure, a lolip used on an inclined surface or a vertical surface such as a slope work, that is, a hook body that rotates around the rotation axis of a hinge portion. A rotating portion 5b for opening and closing the open portion of the portion (5a) is provided, and a lolip or the like for attaching to a main rope hanging from above to prevent the worker 2 from falling can be mentioned.

The rope 6 is a lifeline, a lanyard, or the like, and is formed in a long shape. One end of the rope 6 is connected to the hook 5, and the other end of the rope 6 is connected to the mounting portion 4a of the safety belt 4.

The rope 6 may be indirectly connected to the mounting portion 4a or the hook 5 of the safety belt 4 via fasteners such as a D ring, a carabiner, and a rope winding device. The length of the rope 6 is preferably 0.5 m to 3.0 m, more preferably 1.0 m to 2.5 m, and even more preferably 1.5 m to 2.0 m.

The "rope 6" in the present disclosure includes all kinds of long objects provided in the conventionally well-known safety belt 4. Examples of the type of the "rope 6" in the present disclosure include a rope type such as a rope body, and a long object such as a winding type having the above-mentioned winding device and a bellows type.

(Safety belt usage monitoring system)
Next, the safety belt usage status monitoring system 1 of the present embodiment is a system for monitoring the usage status of the safety belt 4 of the worker 2 who works at a high place.

Specifically, the safety band usage monitoring system 1 of the present embodiment has one end of the rope 6 of the safety band 4 or the hook 5 of the safety band 4, as shown in FIGS. 1, 2, 3, and 4. The acceleration sensor 10 for detecting the acceleration (parameter for measuring the acceleration) of the hook 5 and the transmitter 11 for transmitting and receiving at least the information signal R of the acceleration data detected by the acceleration sensor 10 and The receiver 12, the hook use determination unit 13 for determining whether or not the hook 5 is used based on the information signal R received by the receiver 12, and the display unit 14 for displaying the determination result of the hook use determination unit 13. And an alarm unit 15 that issues an alarm when the determination result of the hook use determination unit 13 is "hook not used (inappropriate)".

The information signal R includes not only the acceleration information of the detection result of the acceleration sensor 10, but also necessary information such as discrimination information of the worker 2, position information of the worker 2, and work content information of the worker 2. It is desirable that it is.

The transmission means of the information signal R is preferably a beacon, but of course, other radio signal transmission means such as Wi-Fi may be used. Further, the position information of the worker 2 may be detected by using GPS (Global Positioning System).

The safety belt usage monitoring system 1 of the present embodiment works as a receiver 12 with at least the first receiver 12a that can be attached to the attachment 3 worn by the worker 2 within a predetermined distance range. It includes a second receiver 12b provided far from the member 2 and a plurality of receivers 12 including the second receiver 12b.

The safety belt usage monitoring system 1 of the present embodiment is attached to one end of the rope 6 of the safety belt 4 or the hook 5 of the safety belt 4 as the transmitter 11, and transmits the information signal R to the first receiver 12a. It is provided with a plurality of transmitters 11 such as transmitters 11a and 11b for transmitting and receiving the information signal R received by the receiver 12 at a place far away from the worker 2 within a predetermined distance range. ..

When the information signal R is transmitted / received between the plurality of transmitters 11 and the plurality of receivers 12, if the transmission / reception distance is large, a single or a plurality of repeaters 16 including the transmitter 11 and the receiver 12 may be used. It may be provided to configure the safety belt usage monitoring system 1. In this case, even when the beacon is used as the transmission means of the information signal R, the predetermined range can be significantly expanded.

The acceleration sensor 10 attached to one end of the rope 6 of the safety belt 4 or the hook 5 of the safety belt 4 is integrated with the transmitter 11 or the like, for example, an acceleration sensor / radio tag (acceleration radio tag) 17. You may be. In this case, a compact acceleration radio tag 17 that integrates the acceleration sensor 10 and the transmitter 11 or the like is attached to one end of the rope 6 of the safety belt 4 or at an arbitrary position of the hook 5 of the safety belt 4. It can be easily attached. That is, it can be easily attached to the existing safety belt 4.

In this embodiment, at least one receiver 12 (12b) is housed in a chest pocket (not shown) of the garment 3a worn by the worker 2. The receiver 12 (12b) may be stored in the pocket of the trousers of the clothes 3a, or may be mounted on, for example, a helmet 3b or a mounting portion 4a of the safety belt 4 other than the clothes 3a.

Further, the other receiver 12 (12c) is provided in a management / supervision room or in a chest pocket (not shown) of the clothes 3a worn by the supervisor. When the supervisor wears the receiver 12c, the receiver 12c may be stored in the pocket of the trousers of the clothes 3a, or may be worn on a helmet 3b or a mounting portion 4a of the safety belt 4 other than the clothes 3a. good.

Here, in the present embodiment, the receiver 12, the transmitter 11, the display unit 14, the alarm unit 15, the input / output unit 18, and the hook use determination unit 13 have a wireless communication function, for example, as shown in FIG. However, it is preferable that the mobile terminal 20 such as a smartphone owned by the worker 2 or the supervisor and the application software are applied.
Further, as shown in FIG. 3, the PC of the management terminal 21 provided in the management / supervision room or the like, application software, or the like may be applied and provided.

Further, as shown in FIG. 4, the hook use determination unit 13 may be provided in a separately provided server, main computer (host computer, cloud computer) 22, or the like.

Next, the hook use determination unit 13 of the present embodiment is based on, for example, i) the acceleration of the gravity axis, ii) the acceleration other than the gravity axis, and iii) the acceleration of the three-dimensional direction components of X, Y, and Z. Judgment of the usage status of (prediction of the usage status of the safety belt 4) is performed.

Specifically, first, as shown in FIG. 2 (FIGS. 3 and 4), the hook use determination unit 13 of the present embodiment includes a data acquisition unit 23, a data storage unit 24, a calculation unit 25, and machine learning. A unit 26 and a determination unit 27 are provided, and the machine learning unit 26 extracts features and estimates the usage status of the hook 5 of the worker 2 (and the usage status of the safety band 4 of the worker 2), and the determination unit 27 determines the usage status of the hook 5 of the worker 2 (and the usage status of the safety band 4 of the worker 2: omitted below) based on the feature extraction data and the estimation.

The data acquisition unit 23 continuously or periodically acquires acceleration data (acceleration information) received by the receiver 12 when the worker 2 is working at a high place. The data acquisition unit 23 of the present embodiment is configured to be able to acquire various data and information input from other input devices (input unit 29) such as an image pickup device 28 and a computer.

The data storage unit 24 is a non-volatile memory, and is composed of, for example, a battery-backed SRAM (Static Random Access Memory) or SSD (Solid State Drive), even if the power is turned off by the input / output unit 18. It is configured to retain its storage state. The data storage unit 24 of the present embodiment can store data acquired by the data acquisition unit 23, data processed by the calculation unit 25, learning data learned by the machine learning unit 26, and the like.

In the present embodiment, as shown in FIGS. 2 (3, 4) and 5, the calculation unit 25 is a time received by the receiver 12 and stored in the data storage unit 24 via the data acquisition unit 23. The historical acceleration data is divided by a predetermined time interval (window size) such as 20 seconds (Step1, Step2). At this time, the time center position of the next window is shifted from the time center position of the previous window (data set) by a slide size (movement size) such as 2 seconds with respect to the acceleration data over time (time history). Generate multiple windows, i.e. multiple datasets at regular time intervals, so that the windows before and after are shifted by slide size, such as 2 seconds. Then, the feature amount is calculated for the data in the windows of each data set divided at regular time intervals (Step 3).

That is, the calculation unit 25 generates a plurality of data sets extracted from the time-series acceleration data at predetermined fixed time intervals, and the slide time preset from the start of acquisition of the acceleration data of the previous data set in the time series. The position of is set as the acceleration data acquisition start time of the next data set, and a plurality of data sets are superimposed on the time axis and normalized to data in a predetermined time unit, and the feature amount is obtained for each data set.

Further, the calculation unit 25 of the present embodiment calculates the acceleration of the gravity axis and the acceleration in the horizontal direction with the gravity axis for each data of the plurality of windows (a plurality of data sets at a fixed time interval). At this time, the direction of the gravitational axis is obtained based on the average value of the acceleration data of X, Y, Z in three directions for a certain number of seconds (for example, 64 seconds) before a specific time t and the gravitational acceleration. That is, the direction of the gravitational axis is obtained based on the average value of the acceleration data of X, Y, and Z in each section and the gravitational acceleration, separated by a certain time interval.

Next, the calculation unit 25 obtains, for example, a rotation matrix from the direction of the gravity axis, calculates the acceleration in the direction of the gravity axis at each time, calculates the acceleration in the horizontal direction with the gravity axis, and obtains the feature amount.

In the safety belt usage status monitoring system 1 of the present embodiment, for example, the following data 1) to 28) are used as feature quantities.

1) x_grad std: x-axis acceleration differential standard deviation 2) y_grad std: y-axis acceleration differential standard deviation 3) z_grad std: z-axis acceleration differential standard deviation 4) fixed_xy max: horizontal axis direction Maximum value of acceleration 5) fixed_xy min: Minimum value of acceleration in the horizontal axis direction 6) fixed_xy std: Standard deviation of acceleration in the horizontal axis direction 7) fixed_z max: Maximum value of acceleration in the direction of gravity axis 8) fixed_z min: Gravity Minimum value of axial acceleration 9) fixed_z std: Standard deviation of acceleration in the direction of gravity 10) grad_distance max: Maximum value of differentiation of acceleration magnitude 11) grad_distance std: Standard deviation of differentiation of acceleration magnitude 12) grad_std_sum: sum of standard deviations of x, y, z-axis acceleration 13) grad_sum max: maximum value of sum of x, y, z-axis acceleration 14) grad_sum min: x, y, z-axis acceleration differential Minimum total value 15) grad_sum std: Standard deviation of total differential of x, y, z-axis acceleration 16) std_sum: Total standard deviation of x, y, z-axis acceleration 17) f_distance max: Value obtained by Fourier transforming grad_distance 18) f_distance min: Minimum value of grad_distance transformed into Fourier 19) f_distance std: Standard deviation of value obtained by transforming grad_distance into Fourier 20) f_sum max: Maximum value obtained by transforming grad_sum into Fourier 21) f_sum min: 22) f_sum std: Standard deviation of the value obtained by Fourier transforming grad_sum 23) f_xy max: Maximum value obtained by Fourier transforming the acceleration in the horizontal axis direction 24) f_xy min: In the horizontal axis direction Minimum value of acceleration converted by Fourier 25) f_xy std: Standard deviation of value obtained by transforming acceleration in the horizontal axis direction 26) f_z max: Maximum value of value obtained by transforming acceleration in the direction of gravity axis 27) f_z min: Minimum value of the value obtained by Fourier transforming the acceleration in the direction of the gravity axis 28) f_z std: The acceleration in the direction of the gravity axis is transformed by Fourier transform. Standard deviation of the values

It should be noted that these feature quantities are merely examples, and the feature quantities obtained by the safety belt usage status monitoring system 1 of the present disclosure do not necessarily have to be limited to the above feature quantities.

Next, the machine learning unit 26 has a learning unit 31 that generates a learning model based on the feature amount obtained by the calculation unit 25, and an estimation unit 32 that estimates whether or not the hook 5 is used based on the learning model. And have.

Further, the learning unit 31 of the present embodiment is configured to process features using an SVM (support vector machine) to generate (learn) a learning model (Step 3 and Step 4).

That is, the boundary plane (hyperplane, support vector) is calculated from the feature amount data by SVM, and which side of the boundary plane the newly given data (feature amount) is located is obtained, and binary classification (safety) is performed. Judgment of whether or not the belt is used).
Of course, the SVM in the present embodiment includes an SVM that maps a pattern to a finite or infinite dimensional feature space using a kernel function and performs linear separation on the feature space, a structured SVM, and the like.

The learning unit 31 performs supervised learning using the processing data generated by the preprocessing unit 30, and generates and learns a learning model in which the appropriateness (good / bad) of the usage status of the hook 5 is learned (Step 5). Further, this learning model is stored in the data storage unit 24 (or a learning model storage unit separately provided).

The estimation unit 32 estimates the usage status of the hook 5 using the learning model stored in the data storage unit 24 (learning model storage unit) based on the processing data input from the preprocessing unit 30 (Step 6). At this time, the estimation unit 32 estimates the learning model generated by the supervised learning by the learning unit 31 using the processing data input from the preprocessing unit 30 as input data.

Next, the determination unit 27 makes a determination based on the estimation result of the suitability of the usage status of the hook 5 by the estimation unit 32, and displays the determination result on the display unit 14 (see FIG. 6). If the hook 5 is not used properly, the alarm unit 15 outputs an alarm such as a sound, an image display, or a vibration. As a result, it is possible to promptly make the worker 2 and the supervisor recognize that the usage status of the hook 5 is unsuitable, and prevent the hook 5 from falling.
In addition, evaluation and judgment are performed using a learning model that continuously learns.

Here, the determination unit 27 may use the estimation result by the estimation unit 32 of the machine learning unit 26 as the determination result as it is.

On the other hand, the determination unit 27 calculates an evaluation index for evaluating the stored learning model, evaluates the learning model using this evaluation index, confirms and evaluates that the learning model is suitable, and then estimates. The estimation result by the unit 32 may be configured to be the determination result.
As an example of the evaluation method of this learning model, for example, ROC analysis (Receiver Operating Characteristic analysis) is performed on the training model, and the true positive rate (probability of correctly predicting the usage status of hook 5) and the false positive rate (wrong). (Probability of predicting positive) is obtained, and evaluation is performed based on this.

Further, for example, the learning model is evaluated for each preset learning cycle such as 100 times, and the learning model when the learning by the learning unit 31 is performed in the preset learning cycle (100 times) is set in advance. The evaluation index of the learning model was calculated for the learning model when it was performed twice (200 times) and the learning model when it was performed three times (300 times), and at each time point. It is preferable to store the learning model associated with the learning model in (1) and to perform evaluation and judgment using this associated learning model.

Then, when monitoring the usage status of the safety belt 4 of the worker 2 using the safety belt usage status monitoring system 1 of the present embodiment (in the safety belt usage status monitoring method of the present embodiment), the safety belt 4 is used. An acceleration radio tag 17 having an acceleration sensor 10 and a transmitter 11 is attached to the vehicle. The acceleration wireless tag 17 is a receiver 12 provided with a terminal device such as a smartphone (mobile terminal 20, management terminal 21, host computer, cloud computer 22, etc.) by the transmitter 11 for acceleration data detected by the acceleration sensor 10. Send to. The receiver 12 provided in the terminal devices 20, 21, 22 such as a smartphone receives the acceleration data detected by the acceleration sensor 10 of the acceleration wireless tag 17, and the application provided in the terminal devices 20, 21, 22 or the like is used. The program in the hook usage determination unit 13 predicts the state of the safety band 4 and determines the usage status of the hook 5.
Here, the power ON / OFF of the terminal devices 20, 21 and 22, the start / stop of the application software, and the like are performed by the input / output unit 18.

Then, the determination result determined by the determination unit 27 is displayed and output to the display unit 14, for example, or is used as a management PC 21 or another terminal such as a mobile terminal 20 or 21 through a host computer or a cloud computer 22 via the network 7. Send and output to the device.

Therefore, according to the safety belt usage monitoring system 1 of the present embodiment, it is possible to create a learning model in which the acceleration data of the operation in which the hook 5 is attached is accurately extracted, and evaluation and determination are performed based on this learning model. This makes it possible to determine the usage status of the hook 5 with much higher accuracy than the conventional safety belt usage status monitoring system.

Further, in the safety belt usage status monitoring system 1 of the present embodiment, a supervised learning algorithm of machine learning, SVM, is used to classify some data patterns and determine the usage status of the hook 5. This makes it possible to determine the usage status of the hook 5 with even higher determination accuracy.

Further, in the safety belt usage status monitoring system 1 of the present embodiment, by using the SVM, it is possible to determine the usage status of the hook with high determination accuracy.
For example, when a learning model is generated using SVM, the accuracy of determining the usage status of the hook 5 is as high as 70 to 80% or more even under adverse conditions, whereas the decision tree, LSTM (Long Term Short) When a learning model is generated using Memory), it has been confirmed that the determination accuracy is about 60% to 70%.

Then, by using such machine learning and SVM, even in a construction site where a large number of workers 2 such as hundreds or more perform work, the information signal of the acceleration sensor 10 of the large number of workers 2 is used. (Big data) R can be handled suitably, and the usage status of the hook 5 of each worker 2 can be appropriately evaluated, determined, and further notified. This eliminates the risk of confusion at the construction site, such as stopping work each time a false alarm is given, or the supervisor being overwhelmed by responding to information that is "inappropriate." ..

Here, in the case of using the supervised learning algorithm of machine learning, SVM as in the present embodiment, for example, at the start of work, the worker 2 is made to perform an appropriate action of using the safety band 4 and the hook 5. The teacher data may be generated based on the acceleration data detected by the clear proper operation, and the learning may be performed. For example, the input / output unit 18 may be operated to notify the hook 5 of the appropriate operation of hooking the hook 5 to the hook hooking member 8 and to give the teacher data before performing the work. In this case, it is possible to improve the accuracy of determining the usage status of the hook 5 relatively easily and at an early stage.
In addition, the image pickup device 28 images the work operation of the worker 2, confirms the proper use behavior of the safety belt 4 and the hook 5 of the worker 2 with an image, confirms the acceleration data, and supervises the teacher data based on the image. A member or the like may be configured to input in the input unit 29 and generate a learning model.

The machine learning algorithm does not necessarily have to be supervised learning, and algorithms such as unsupervised learning and semi-supervised learning may be used.

In the unsupervised learning algorithm, for example, SVM, SOM (self-organizing map), k-means clustering, neighborhood method mapping, singular value decomposition, etc. are used to obtain acceleration data corresponding to the proper safety zone 4 and hook 5 usage behavior. Identification / estimation, and conversely, identification / estimation of acceleration data (identification / estimation of outliers) that is not related to the proper usage behavior of the safety zone 4 and hook 5 are performed, and learning is performed using the acceleration data that are sequentially detected. It is possible to improve the accuracy of determining the usage status of the hook 5 as compared with the conventional case.

When applying the semi-supervised learning algorithm, it is possible to improve the determination accuracy as compared with the conventional case by using the supervised learning algorithm and the unsupervised learning algorithm in combination.

Further, in the safety belt usage status monitoring system 1 of the present embodiment, as the information signal R, not only the information of the detection result of the acceleration sensor 10, but also the discrimination information of the worker 2, the position information of the worker 2, and the worker 2 By acquiring the work content information and the like, it becomes possible to determine the usage status of the hook 5 with higher accuracy.

For example, by referring to the many processes (process names) shown in the Ministry of Land, Infrastructure, Transport and Tourism's construction work estimation standard and civil engineering work estimation standard, many work processes are individually reflected in machine learning, and the identified workers are identified. The usage status of the hook 5 may be determined according to the determination criteria (learning data) for each process carried out by 2. As a result, machine learning can be performed for each process performed by the worker 2, and it is possible to further improve the accuracy of determining the usage status of the hook 5.

Further, in the safety belt usage status monitoring system 1 of the present embodiment, the stationary transmitter 33 (transmitter) for discriminating the worker at at least one of the site where the work is performed at a high place or the route to the site. It is preferable that 11e and a receiver 12e) are provided.

In this case, for example, the worker 2 who carries the receiver 12 (wearing the wearing object to which the receiver 12 is attached) that has received the activation radio wave needs to determine the usage status of the hook 5. It can be seen that the worker 2 who carries the receiver 12 that does not receive the activation radio wave is in a place where there is no need to determine the usage status of the hook 5. As a result, among the workers 2 who carry the receiver 12, it is possible to identify the worker 2 who needs to determine the usage status of the hook 5, and it is possible to reduce the processing load of the hook use determination unit 13.

In other words, by providing the safety belt usage status monitoring system 1 of the present embodiment, it is possible to wirelessly know the movement of an object including the worker 2 detected by the acceleration sensor 10. Further, if the radio wave strength of the periodic beacon transmission can be measured together, it is possible to know the existence, proximity, and other conditions of the position of the object including the worker 2.

The embodiment of the safety belt usage monitoring system of the present disclosure has been described above, but the present invention is not limited to the above embodiment, and can be appropriately changed without departing from the spirit of the present embodiment.

For example, it may be possible to further improve the determination accuracy by changing the machine learning algorithm and the feature amount extraction method.

Further, the machine learning algorithm is not necessarily limited as in the present embodiment.

Further, in the present embodiment, the acceleration sensor / wireless tag 17 in which the acceleration sensor 10 and the transmitter 11 are integrated is attached to one end of the rope 6 of the safety belt 4 or to the hook 5 of the safety belt 4. Of course, the accelerometer 10 and the transmitter 11 and the receiver 12 may be provided separately.

Further, as shown in FIG. 1, the acceleration sensor 10 may be attached to the rotating portion 5b that opens and closes the open portion of the hook main body portion 5a (claw-shaped portion or the like) of the hook 5. ..
In this case, since the rotating portion 5b is configured to rotate forward and reverse around the rotation axis O1 of the hinge portion 5c to open and close the open portion of the hook main body portion 5a, the rotating portion 5b The acceleration sensor 10 mounted on the hook inevitably is centered on the rotation axis O1 of the hinge portion 5c when the worker 2 mounts the hook 5 on the hook hook member 8 and removes the hook 5. It will move forward and backward along the circumferential direction on the concentric circles. As a result, among the acceleration data detected by the acceleration sensor 10, the acceleration data indicating the hook 5 mounting operation and the hook 5 removing operation can be extracted more accurately. Therefore, it is possible to further improve the accuracy of determining the usage status of the hook 5. Of course, the same can be said for the lolip and the like included in the "hook 5" of the present disclosure.

1 Safety belt usage monitoring system 2 Workers 3 Attachments 4 Safety belts 5 Hooks 5a Hook-shaped part (hook body part)
5b Rotating part 5c Hinge part 6 Rope 8 Hook hooking member (fixed support member)
10 Acceleration sensor 11 Transmitter 12 Receiver 13 Hook use determination unit 14 Display unit 15 Alarm unit 17 Acceleration wireless tag 20 Mobile terminal 21 Management terminal 22 Host computer, cloud computer, etc. 23 Data acquisition unit 24 Data storage unit 25 Calculation unit 26 Machine learning unit 27 Judgment unit 28 Imaging device 29 Input unit 31 Learning unit 32 Estimating unit 33 Stationary transmitter O1 Rotation axis R Information signal (acceleration data, etc.)

Claims (8)

It is a safety belt usage monitoring system for monitoring the usage status of safety belts for workers who work at heights.
An accelerometer attached to one end of the safety belt rope or the hook of the safety belt,
A transmitter and a receiver for transmitting and receiving at least an information signal including acceleration data detected by the acceleration sensor, and
A hook use determination unit for determining whether or not the hook is used based on the received acceleration data is provided.
The hook use determination unit includes a calculation unit for obtaining a feature amount from the acceleration data, and a machine learning unit for estimating whether or not the hook is used from the feature amount.
The machine learning unit
A learning unit that generates a learning model based on the features obtained by the calculation unit, and a learning unit.
It comprises an estimation unit that estimates whether or not the hook is used based on the learning model.
Safety belt usage monitoring system. The learning unit is configured to process the feature amount using SVM to generate a learning model.
The safety belt usage monitoring system according to claim 1. The calculation unit generates a plurality of data sets extracted from the acceleration data in the time series at predetermined fixed time intervals, and has a slide time preset from the start of acquisition of the acceleration data of the previous data set in the time series. The position is set as the acceleration data acquisition start time of the next data set, and the plurality of data sets are superimposed on the time axis and normalized to data in a predetermined time unit, and the feature amount is obtained for each data set.
The safety belt usage monitoring system according to claim 1 or 2. The information signal further includes at least one of worker identification information, worker position information, and work content information.
The safety belt usage monitoring system according to any one of claims 1 to 3. Further, a determination unit for finally determining whether or not the hook is used based on the estimation result of the machine learning unit is provided.
The determination unit includes an alarm unit that issues an alarm when the determination unit determines that the hook is not used and is “inappropriate”.
The safety belt usage monitoring system according to any one of claims 1 to 4. An installed transmitter that identifies the worker is provided at the site where the aerial work is performed or at least one of the routes to the site.
The safety belt usage monitoring system according to any one of claims 1 to 5. Further, an input unit for giving the acceleration data at the time of using the hook to the machine learning unit as teacher data is provided.
The safety belt usage monitoring system according to any one of claims 1 to 6. The input unit extracts the acceleration data at the time of proper use operation of the hook based on the image captured by the image pickup device that captures the movement of the worker, and the extracted acceleration data is used as the teacher data in the machine. It is configured to give to the learning department,
The safety belt usage monitoring system according to claim 7.

Images