WO2022208805A1

WO2022208805A1 - Deterioration detection device, deterioration detection system, deterioration detection method, weight measurement device, weight measurement method, and program

Info

Publication number: WO2022208805A1
Application number: PCT/JP2021/014021
Authority: WO
Inventors: 冬樹宮澤; 邦彦中島
Original assignee: 太陽誘電株式会社
Priority date: 2021-03-31
Filing date: 2021-03-31
Publication date: 2022-10-06
Also published as: JPWO2022208805A1

Abstract

The present invention continuously detects deterioration in a bridge. This deterioration detection device comprises: an acquisition unit for collecting time series data pertaining to a parameter that represents a travel-direction site on a subject portion of a bridge where a sensor is provided; an extraction unit for extracting, from the time series data, specific portion data from a time when a specific vehicle passes through a measured segment of the bridge on the basis of an assessment result indicating whether the specific vehicle has passed by, the assessment being made by a neural network that outputs the assessment result upon receiving input of the time series data; an amplitude calculation unit for calculating, on the basis of the specific portion data, an amplitude value pertaining to the amount of travel-direction expansion and contraction of the bridge at the time when the specific vehicle passes by; a deterioration assessment unit for assessing that the bridge has deteriorated when the amplitude value is greater than a preset criterion value; and a re-training command unit for determining whether to re-train the neural network on the basis of pre-established determination criteria, and outputting a re-training command that commands re-training of the neural network on the basis of the result of the determination as to whether to re-train the neural network.

Description

Deterioration detection device, deterioration detection system, deterioration detection method, weight measurement device, weight measurement method and program

The present invention relates to a deterioration detection device, a deterioration detection system, a deterioration detection method, a weight measurement device, a weight measurement method, and a program.

Conventionally, bridges are visually inspected about once every five years. In order to further improve the safety of bridges, it is desirable, for example, to constantly monitor bridges, predict the future state of bridges based on the monitoring results, and systematically manage bridges.

Patent Document 1 describes a technique for measuring the strain of a floor slab when a vehicle passes over a bridge using a strain gauge and detecting the characteristics of the passing vehicle based on the measured strain. In Patent Document 2, the center distance of the vehicle is calculated by calculating the center distance ratio of the vehicle from the waveform of the strain measured by the strain gauge, and comparing the calculated center distance ratio with the center distance ratio registered in the database. , vehicle speed and vehicle type. In Patent Document 3, a provisional axle weight value of the vehicle is calculated based on the strain of the longitudinal rib and the lateral rib when the vehicle passes, and the provisional axle weight value is corrected by the vehicle weight value calculated based on the strain of the lateral rib. technology is described. Patent Literature 4 describes a technique for detecting the strain of a floor slab when a vehicle passes over a bridge and calculating the weight of the vehicle based on the detected strain. Non-Patent Document 1 describes a technique for calculating the deflection characteristics of a bridge by installing an acceleration sensor under the rear wheel spring of a route bus.

JP 2006-084404 A JP 2009-237805 A JP 2014-228480 A JP 2017-106769 A

By the way, as the deterioration of bridges progresses, strain increases when vehicles pass over them. However, it is difficult to continuously measure how much the bridge is distorted by the influence of deterioration without relying on personnel.

Also, in order to monitor the bridge, it is conceivable to use a model that outputs output values according to multiple input values. However, detection of bridge deterioration is monitored over a long period of several years. For this reason, a deviation occurs between the characteristics of the bridge and the characteristics of the model during long-term operation, and deterioration of the bridge cannot be detected with high accuracy.

The present invention has been made in view of the above, and provides a deterioration detection device, a deterioration detection system, a deterioration detection method, a weight measurement device, a weight measurement method, and a program for accurately detecting deterioration of a bridge over a long period of time. intended to

In order to solve the above-described problems and achieve the object, a deterioration detection device according to the present invention provides a parameter representing displacement in the running direction of a target portion of the bridge where the sensor is installed, from a sensor installed on the bridge. and a neural network that inputs the time-series data and outputs a determination result indicating whether or not a specific vehicle has passed. an extracting unit for extracting specific portion data when a specific vehicle passes through the measured section of the bridge; and based on the specific portion data, calculating an amplitude value of the amount of expansion and contraction of the bridge in the running direction when the specific vehicle passes through. a deterioration determining unit that determines that the bridge has deteriorated when the amplitude value exceeds a preset reference value; and a re-learning of the neural network based on a predetermined criterion. and a relearning instruction unit for outputting a relearning instruction for instructing relearning of the neural network based on the determination result of whether or not the neural network should be relearned.

According to the present invention, deterioration of bridges can be accurately detected over a long period of time.

FIG. 1 is a diagram showing a deterioration detection system according to the first embodiment. FIG. 2 is a diagram showing the arrangement of sensors when the bridge is viewed from the side. FIG. 3 is a diagram showing the arrangement of sensors when the bridge is viewed from above. FIG. 4 is a diagram showing a portion of the sensor and main girder of the bridge. FIG. 5 is a diagram showing the displacement detection device together with the first member and the second member. FIG. 6 is a diagram showing the functional configuration of the deterioration detection device. FIG. 7 is a diagram showing an example of time-series data of the amount of expansion and contraction in the traveling direction when a route bus passes a bridge. FIG. 8 is a diagram showing temporal changes in amplitude values and error ranges. FIG. 9 is a diagram showing temporal changes in amplitude values and thresholds for re-learning the first neural network. FIG. 10 is a diagram illustrating a functional configuration of a deterioration detection device according to a modification; FIG. 11 is a flow chart showing the flow of processing of the deterioration detection device. FIG. 12 is a diagram showing a weight measurement system according to the second embodiment. FIG. 13 is a diagram showing the functional configuration of the weight measuring device. FIG. 14 is a diagram showing the relationship between the weight of the test vehicle and the amplitude value. FIG. 15 is a flow chart showing the flow of correction processing of relational information by the weight measuring device. FIG. 16 is a diagram showing the hardware configuration of the deterioration detection device and the weight measurement device.

Embodiments will be described below with reference to the drawings.

(First embodiment)
FIG. 1 is a diagram showing a deterioration detection system 10 according to the first embodiment. The deterioration detection system 10 outputs alarm information when the bridge deteriorates.

The deterioration detection system 10 includes a sensor 20, a transmission device 22, and a deterioration detection device 30.

The sensor 20 is provided at a predetermined target portion of the bridge. The sensor 20 detects a parameter representing displacement in the direction of travel at the target portion of the bridge where the sensor 20 is provided. In this embodiment, the sensor 20 measures the amount of expansion and contraction in the running direction of the target portion of the bridge. The amount of expansion and contraction is, for example, a change in the distance of several nanometers to several hundreds of nanometers between two points separated by a distance of several tens of centimeters.

It should be noted that the sensor 20 may detect a physical quantity other than the amount of expansion and contraction in the running direction, as long as it can detect a parameter representing displacement in the running direction. For example, sensor 20 may be a strain gauge that detects strain in the direction of travel in the target portion of the bridge. Further, for example, the sensor 20 may be a vibrometer that detects the magnitude of the natural frequency in the running direction in the target portion of the bridge or the magnitude of the vertical natural frequency in the target portion of the bridge.

The sensor 20 continuously detects a parameter representing displacement in the running direction of the target portion of the bridge at predetermined time intervals. For example, the sensor 20 detects parameters every few milliseconds. The sensor 20 continuously detects parameters at predetermined time intervals, for example, while the power is on. The sensor 20 may continuously detect parameters at predetermined time intervals all the time, for example, 24 hours a day.

The transmission device 22 transmits the parameters detected by the sensor 20 to the deterioration detection device 30 via the network. The network may be wired, wireless, or a mixture of wired and wireless. The network is, for example, a LAN (Local Area Network), a VPN (Virtual Private Network), or a WAN (Wide Area Network) in which LANs are connected via routers. The network may also include the Internet, telephone lines, or the like.

The deterioration detection device 30 receives time-series data of parameters representing the displacement in the running direction of the target portion of the bridge, transmitted from the transmission device 22 via the network. The deterioration detection device 30 determines whether or not the bridge has deteriorated based on the received time series data of the parameters. When the deterioration detection device 30 determines that the bridge has deteriorated, the deterioration detection device 30 outputs alarm information to, for example, an administrator or an information processing device held by the administrator.

The deterioration detection device 30 is a computer such as a server device that can be connected to a network. The deterioration detection device 30 may be one computer, or may be composed of a plurality of computers like a cloud system.

Note that the deterioration detection system 10 may be configured without the transmission device 22 . In this case, the deterioration detection device 30 directly obtains from the sensor 20 the parameter representing the displacement in the running direction of the target portion of the bridge. Further, in this case, the deterioration detection device 30 may be provided near the sensor 20, that is, near the bridge.

FIG. 2 is a diagram showing the arrangement of the sensors 20 when viewing the bridge from the side. FIG. 3 is a diagram showing the arrangement of sensors 20 when the bridge is viewed from above. The sensor 20 is attached, for example, on the end side of the bridge with respect to the center in the running direction. The sensor 20 is mounted, for example, on the lower surface of the bridge near the abutment. This allows workers to easily attach the sensor 20 to the bridge even after the bridge is completed. In addition, the sensor 20 may be attached at any position in the running direction on the bridge. For example, the sensor 20 may be installed in the center of the bridge in the direction of travel, although this may be difficult for the operator to install.

FIG. 4 is a diagram showing the sensor 20 and a portion of the main girder 54 of the bridge. The sensor 20 measures the amount of change in the distance between the first point 62 and the second point 64 on the lower surface 56 of the main girder 54 of the bridge as the amount of expansion and contraction in the direction of travel.

The first point 62 and the second point 64 are the same in the width direction of the bridge but different in the running direction. The distance between the first point 62 and the second point 64 is, for example, several tens of centimeters. In the example of FIG. 4, the distance between the first point 62 and the second point 64 is 35 centimeters. The sensor 20 measures the amount of change in distance in the direction of travel between the first point 62 and the second point 64 in units of, for example, several nanometers to several hundred nanometers.

The sensor 20 has a first member 66 , a second member 68 and a displacement detection device 70 .

The first member 66 is a cantilever beam having a support portion 66a and a beam portion 66b. One end of the support portion 66 a is a fixed end 66 c fixed to the first point 62 . The support portion 66 a extends vertically downward from the first point 62 with respect to the lower surface 56 of the main girder 54 by a predetermined distance. The beam portion 66b extends a predetermined distance toward the second point 64 in the traveling direction from the end portion of the support portion 66a opposite to the fixed end 66c. The end of the beam portion 66b that is not connected to the support portion 66a is a free end 66d that is not connected to any member. In this embodiment, the free end 66 d of the first member 66 is positioned near the approximate center of the line connecting the first point 62 and the second point 64 .

The second member 68 is a cantilever beam having a support portion 68a and a beam portion 68b. One end of the support portion 68 a is a fixed end 68 c fixed to the second point 64 . The support portion 68a extends vertically downward from the second point 64 to the lower surface 56 of the main girder 54 for a predetermined distance. The beam portion 68b extends a predetermined distance toward the first point 62 in the running direction from the end portion of the support portion 68a opposite to the fixed end 68c. The end of the beam portion 68b that is not connected to the support portion 68a is a free end 68d that is not connected to any member. In this embodiment, the free end 68 d of the second member 68 is positioned near the approximate center of the line connecting the first point 62 and the second point 64 .

Here, the free end 66d of the first member 66 and the free end 68d of the second member 68 are arranged at overlapping positions in the running direction without mechanical interference. As a result, the free end 66d of the first member 66 and the free end 68d of the second member 68 are arranged to face each other in the direction perpendicular to the lower surface 56 of the main girder 54 . When the distance between the first point 62 and the second point 64 changes, the relative positions of the free end 66d of the first member 66 and the free end 68d of the second member 68 shift in the running direction.

Note that the sensor 20 shown in FIG. 4 has a configuration in which both the first member 66 and the second member 68 are cantilever beams. However, the second member 68 may be cantilevered and the first member 66 may not be cantilevered. In this case, the second member 68 is arranged such that the free end 68d overlaps at least a portion of the first member 66 in the running direction without mechanical interference. Even with such a configuration, when the distance between the first point 62 and the second point 64 changes, the relative position between the first member 66 and the free end 68d of the second member 68 changes in the running direction. deviate.

The displacement detection device 70 is provided at a portion where the free end 66d of the first member 66 and the free end 68d of the second member 68 face each other. The displacement detection device 70 detects the displacement of the relative position between the free end 66d of the first member 66 and the free end 68d of the second member 68. As shown in FIG. Then, the displacement detection device 70 outputs the detected displacement as the amount of expansion and contraction between two points in the running direction of the bridge.

5 is a diagram showing the displacement detection device 70 together with the first member 66 and the second member 68. FIG. Displacement detection device 70 includes an optical element 72 and a detector 74 .

The optical element 72 is attached to either the free end 66 d of the first member 66 or the free end 68 d of the second member 68 . The detector 74 is attached to the other of the free end 66d of the first member 66 or the free end 68d of the second member 68 to which the optical element 72 is not attached.

The optical element 72 is an optical member whose reflected light amount or transmitted light amount changes according to the light irradiation position in the traveling direction. For example, the optical element 72 is a mirror whose surface is coated with a plurality of light absorbing materials spaced at predetermined intervals in the running direction. Also, the optical element 72 may be a diffraction grating in which a plurality of optical slits are formed at predetermined intervals in the running direction.

The detector 74 includes a half mirror 76 , a light emitter 78 , a light receiver 80 and a detection circuit 82 . The half mirror 76 reflects part of the irradiated light and transmits the other part.

The light emitting unit 78 irradiates the optical element 72 with light through the half mirror 76 . The light receiving section 80 receives the light reflected by the optical element 72 via the half mirror 76 . The detection circuit 82 outputs a signal representing the displacement of the relative position between the first member 66 and the second member 68 based on the change in the amount of light detected by the light receiving section 80 as the amount of expansion and contraction.

The position of the light irradiated to the optical element 72 shifts in the running direction according to the positional deviation of the relative positions of the first member 66 and the second member 68 in the running direction. Since the optical element 72 is formed with a plurality of light absorbing materials or a plurality of optical slits arranged in the running direction, the amount of light reflected by the optical element 72 increases or decreases according to the displacement of the light irradiation position in the running direction. Specifically, the amount of light reflected by the optical element 72 increases or decreases for one cycle when the position of the light irradiated to the optical element 72 shifts by the distance between a plurality of light absorbing materials or a plurality of optical slits. Therefore, for example, the detection circuit 82 can obtain the amount of change in the relative position between the first member 66 and the second member 68 by counting the increase or decrease in the signal output from the light receiving section 80 .

Further, the displacement detection device 70 includes two optical elements 72 shifted by 1/4 period from each other with respect to the pitch of the light absorbing material or the optical slit, and two light emitting units 78 and two light receiving units corresponding to the two optical elements 72. 80 may also be included. Thereby, the two light receiving sections 80 can output two periodic signals whose phases are shifted by 1/4 period with respect to the change in the relative positions of the first member 66 and the second member 68 . Therefore, for example, the detection circuit 82 detects the direction of change in the relative position of the first member 66 and the second member 68 based on the values of the two signals, and the direction of change in the relative positions of the first member 66 and the first member 66 at intervals shorter than the period of the stripes. The amount of change in relative position with the second member 68 can be detected.

Also, in the example of FIG. 5, the optical element 72 is configured to reflect light. Alternatively, the optical element 72 may be configured to transmit light. In this case, the optical element 72 changes the amount of transmitted light according to the irradiation position of the light with respect to the running direction. For example, the optical element 72 may be made of glass, plastic, or the like, on the surface of which a plurality of light absorbing materials are applied at predetermined intervals in the running direction. In such a case, the light receiving section 80 receives light that has passed through the optical element 72 .

Also, the detector 74 may be configured without the half mirror 76 . Here, the detector 74 is assumed to be provided on the first member 66 . It is also assumed that the optical element 72 is provided on the second member 68 . Let P be the position of the first member 66 facing the center of the optical element 72 in the width direction. In such a case, the light emitting portion 78 is arranged at a position shifted from P in the width direction by a predetermined distance in the first member 66 . Further, the light receiving section 80 is arranged at a position shifted by a predetermined distance from P on the side opposite to the light emitting section 78 in the width direction. The light emitting section 78 emits light in a direction toward the center of the optical element 72 in the width direction. The optical element 72 receives light from the light emitting section 78 at a predetermined angle and reflects the incident light toward the light receiving section 80 . The light receiving section 80 receives the light reflected by the optical element 72 . Such a detector 74 can have similar functionality as the configuration shown in FIG.

The displacement detection device 70 having such a configuration can be attached to the lower surface 56 of the main girder 54 in the bridge. For example, the displacement detection device 70 can be attached from the outside without embedding an expandable member in the bridge like a strain gauge. Thereby, the displacement detection device 70 can be attached to a completed bridge later. Moreover, the displacement detection device 70 can be installed without reducing the strength of the bridge. Also, the displacement detection device 70 can be easily maintained even after installation.

In addition, the displacement detection device 70 having such a configuration uses a cantilever beam to detect a change in the distance between two points with an optical sensor. As a result, the displacement detection device 70 can accurately detect very small expansion and contraction of the bridge using members with a simple configuration and low cost.

FIG. 6 is a diagram showing the functional configuration of the deterioration detection device 30. As shown in FIG. The deterioration detection device 30 includes an acquisition unit 112, a time series data storage unit 114, a cutout unit 116, an extraction unit 118, an amplitude calculation unit 120, a deterioration determination unit 122, an alarm output unit 124, and a collection unit. 132 , a relearning data storage unit 134 , a relearning instructing unit 136 , and a relearning unit 138 .

The acquisition unit 112 collects, from the sensors 20 provided on the bridge, time-series data of parameters representing the displacement in the traveling direction in the target portion of the bridge where the sensors 20 are provided. In this embodiment, the acquisition unit 112 acquires time-series data of parameters via a network. Further, in the present embodiment, the parameter is the amount of expansion and contraction in the running direction of the target portion. In the time-series data of parameters, detected times and parameters are associated with each other.

The time-series data storage unit 114 stores time-series data of parameters collected by the acquisition unit 112 .

The cutout unit 116 cuts out partial data by dividing the parameter time series data stored in the time series data storage unit 114 into units of a predetermined time length. Then, the extraction unit 116 sequentially supplies the extracted partial data to the extraction unit 118 one by one. The length of time of the partial data is at least longer than the time from when the specific vehicle passes through the measurement section of the bridge and the change in the amount of expansion and contraction in the running direction starts to when the change ends.

The partial data is data of a predetermined number of samples. More specifically, the partial data is data of the number of samples to be input to the first neural network used in the extraction unit 118 . In the present embodiment, the cutout unit 116 cuts out two pieces of partial data that are adjacent in the time direction while partially overlapping each other. That is, each partial data overlaps the latter half of the temporally previous partial data and the first half of the temporally immediately following partial data. For example, the first half data in each partial data may be the same as the second half data in the immediately preceding partial data. Also, the latter half data in each partial data may be the same as the first half data in the immediately following partial data. As a result, when the specific vehicle passes over the bridge, the extracting unit 116 stores any of the plurality of partial data as , can include everything up to the point where the change ends.

The extraction unit 118 extracts specific partial data, which is partial data when the specific vehicle passes over the bridge, from the time-series data of the parameters stored in the time-series data storage unit 114 based on the determination result by the first neural network. . The first neural network inputs target partial data and determines whether or not a specific vehicle passes through the target partial data in the measurement section. The first neural network acquires in advance time-series data obtained when a specific vehicle passes over a bridge, and learns using the acquired time-series data as teacher data. In this embodiment, the first neural network is a convolutional neural network (CNN). The extraction unit 118 uses a convolutional neural network to detect that the specific vehicle has passed the bridge even if data obtained when the specific vehicle has passed the bridge is included in any time part of the partial data. It can be detected with high accuracy. Then, the extraction unit 118 outputs the partial data input to the first neural network as the specific partial data in response to the determination result that the specific vehicle has passed the bridge by the first neural network. .

A specific vehicle is, for example, a vehicle that passes over a bridge on a regular basis, each time at approximately the same speed and approximately the same weight. For example, the specific vehicle is a route bus. A route bus runs according to a predetermined timetable every day. Therefore, the route bus is scheduled to pass the bridge at a predetermined time every day. Also, the specific vehicle may be, for example, a garbage truck or the like. The garbage truck has a predetermined travel route and time. Therefore, the garbage truck is scheduled to pass the bridge at a predetermined time on the garbage collection day.

Every time the extraction unit 118 extracts the specific portion data, the amplitude calculation unit 120 calculates the amplitude value of the expansion and contraction amount in the running direction of the bridge when the specific vehicle passes, based on the extracted specific portion data. For example, when the parameter represents the amount of expansion/contraction of the target portion, the amplitude calculator 120 calculates the difference between the maximum value and the minimum value in the extracted specific portion data as the amplitude value.

When the parameter is not the amount of expansion/contraction of the target portion, the amplitude calculator 120 may output a value correlated with the amplitude value of the amount of expansion/contraction of the target portion as the amplitude value. For example, when the parameter is the magnitude of the natural frequency, the amplitude calculator 120 may calculate the difference between the maximum value and the minimum value in the extracted specific partial data as the amplitude value.

The deterioration determination unit 122 compares the calculated amplitude value with a preset reference value each time the amplitude calculation unit 120 calculates the amplitude value. Then, the deterioration determining unit 122 determines that the bridge has deteriorated when the amplitude value becomes larger than the reference value.

The deterioration determination unit 122 may calculate the moving average value of the most recent predetermined sample number of amplitude values, and determine that the bridge has deteriorated when the moving average value is greater than the reference value. Further, the deterioration determination unit 122 executes predetermined filtering processing such as noise removal processing on the amplitude values of a plurality of samples. You may judge that it deteriorated. A plurality of different reference values may be set in advance in the deterioration determination unit 122 . Then, the deterioration determination unit 122 may determine that the bridge has deteriorated each time the amplitude value becomes larger than each reference value.

When it is determined that the bridge has deteriorated, the alarm output unit 124 outputs alarm information indicating that the bridge has deteriorated. Note that when a plurality of different reference values are preset in the deterioration determination unit 122, the alarm output unit 124 includes level information indicating the magnitude of the reference value each time the amplitude value becomes greater than each reference value. Alarm information may be obtained. Thereby, the alarm output unit 124 can notify the administrator or the like of the level of deterioration of the bridge.

The collection unit 132 collects the extracted specific partial data each time the extraction unit 118 extracts the specific partial data. The relearning data storage unit 134 stores the specific partial data collected by the collection unit 132 .

The relearning instruction unit 136 determines whether or not to relearn the first neural network based on predetermined determination criteria, and determines whether or not to relearn the first neural network based on the determination result of whether or not to relearn the first neural network. Outputs a relearning instruction to instruct the relearning of the network. For example, the relearning instruction unit 136 determines whether or not to relearn the first neural network based on a predetermined determination criterion each time the amplitude calculation unit 120 calculates an amplitude value.

When a relearning instruction is output from the relearning instruction unit 136, the relearning unit 138 relearns the first neural network using the specific partial data collected by the collecting unit 132 as teacher data. That is, the re-learning unit 138 re-learns the already learned first neural network. For example, the re-learning unit 138 may perform re-learning by error backpropagation or the like using network parameters such as weights and biases set immediately before re-learning as initial values. After changing the existing network parameters to random values or predetermined values, re-learning may be performed by error backpropagation or the like.

The re-learning unit 138 re-learns the first neural network using specific partial data collected after the last re-learning as teacher data. If re-learning has not been performed since the start of deterioration determination, the re-learning unit 138 may re-learn the first neural network using specific partial data collected after the start as teacher data. Thereby, the re-learning unit 138 can re-learn the first neural network so as to perform appropriate determination processing according to the latest state of the bridge. Note that the relearning unit 138 may relearn the first neural network using time-series data collected in advance as teacher data in addition to the specific partial data collected by the collecting unit 132 .

FIG. 7 is a diagram showing an example of time-series data of the amount of expansion and contraction in the running direction when a route bus passes a bridge. When a vehicle passes over a bridge, the bridge changes in the direction of travel, expanding, expanding to the maximum value, then contracting, contracting to the minimum value, and then returning to its original state. do.

When vehicles of the same type pass through a bridge with approximately the same weight and approximately the same speed, the waveforms of the amount of expansion and contraction of the target portion of the bridge will be approximately the same if the degree of deterioration of the bridge is the same. For example, when vehicles of the same type pass through a bridge with approximately the same weight and approximately the same speed, the amplitude value (difference between the maximum value and the minimum value) and the change time (change start The period from time to change end time) is almost the same if the degree of deterioration of the bridge is the same. Note that FIG. 7 shows an example in which the sensor 20 is provided at the end of the bridge on the vehicle entry side. The sensor 20 may be provided at the end of the bridge on the exit side of the vehicle. In this case, the bridge changes opposite to that in FIG. That is, in this case, the bridge changes in the direction of contraction with respect to the running direction, contracts to the minimum value, changes in the direction of extension, expands to the maximum value, and then returns to its original state.

FIG. 8 is a diagram showing the time change of the amplitude value and the error range of the amplitude value.

Bridges deteriorate over time. When the deterioration of the bridge progresses, the amplitude value of the expansion/contraction amount increases even when the same type of vehicle passes over the bridge with substantially the same weight and substantially the same speed. That is, when vehicles of the same type pass through the bridge at substantially the same weight and at substantially the same speed, the amplitude value of the amount of expansion and contraction in the target portion of the bridge increases over time.

Also, even if the same type of vehicle passes over the bridge with almost the same weight and almost the same speed, the surrounding environment and measurement conditions cause an error in the amplitude value of the amount of expansion and contraction. Moreover, when the specific vehicle is a route bus, the number of passengers and the passing speed also differ depending on the day. The amplitude value of the expansion/contraction amount calculated when the specific vehicle passes over the bridge includes errors due to weight and passing speed.

However, referring to the experimental results of Non-Patent Document 1, for example, as shown in FIG. The quantity is expected to be sufficiently large. For this reason, when a specific vehicle such as a route bus passes through a bridge, if the amplitude value of the expansion/contraction amount of the target portion becomes larger than the preset reference value, the reference value will be the error of the amplitude value at the start of measurement. If it is set sufficiently larger than the distribution range, it can be said that the bridge has deteriorated since the start of the measurement.

Therefore, the deterioration detection device 30 can determine that the bridge has deteriorated when the amplitude value of the amount of expansion and contraction in the running direction of the bridge when a specific vehicle passes is greater than the reference value.

FIG. 9 is a diagram showing changes in amplitude values over time and thresholds for re-learning the first neural network.

If the data input to the machine learning model changes over time, the neural network has the problem of deteriorating accuracy when operated for a long period of time. This is considered to be due to the divergence between the situation when the neural network was learned and the situation after a long period of time. Therefore, the neural network is re-learned when the prediction accuracy deteriorates. However, it is difficult to determine how much the prediction accuracy has deteriorated during operation.

When a bridge deteriorates, the amplitude value of the amount of expansion and contraction in the running direction of the bridge when a vehicle passes increases over time. Therefore, it is considered that the accuracy of the first neural network to which the amount of expansion/contraction in the running direction of the bridge is input deteriorates according to the change in the amplitude value of the amount of expansion/contraction. Therefore, the relearning instruction unit 136 outputs a relearning instruction when, for example, the amplitude value of the expansion/contraction amount in the running direction of the bridge becomes larger than a threshold value.

As a result, the deterioration detection device 30 can appropriately determine when to re-learn the first neural network without determining how much the prediction accuracy has deteriorated. Therefore, the deterioration detection device 30 can accurately detect the deterioration of the bridge over a long period of time.

For example, the relearning instruction unit 136 is set with one or more thresholds different from each other. One or more thresholds are values between the initial value, which is the amplitude value at the start of deterioration determination, and the reference value for determining that the bridge has deteriorated. In this case, the relearning instruction section 136 outputs a relearning instruction each time the amplitude value calculated by the amplitude calculation section 120 becomes larger than one or more threshold values. In addition, the relearning instruction unit 136 calculates a moving average value of the amplitude values of the most recent predetermined number of samples, and issues a relearning instruction each time the moving average value becomes greater than one or a plurality of threshold values. may be output.

In addition, when a plurality of different reference values are set in advance and the deterioration determination unit 122 determines that the bridge has deteriorated each time the amplitude value becomes larger than each reference value, the relearning instruction unit 136 sets a plurality of Each of the different reference values may match the threshold. In this case, the relearning instruction unit 136 may further set one or more thresholds between the reference values.

Further, for example, the relearning instruction unit 136 may determine to relearn the neural network and output a relearning instruction when the calculated amplitude value changes in a preset pattern. For example, the relearning instruction section 136 outputs a relearning instruction when the amplitude value calculated by the amplitude calculating section 120 changes beyond a preset range from the time of the last relearning. Note that if relearning has not been performed since the start of deterioration determination, the relearning instruction unit 136 causes the amplitude value calculated by the amplitude calculation unit 120 to change beyond a preset range from the start. If so, output a relearning instruction.

For example, the relearning instruction unit 136 may output a relearning instruction when the amplitude value calculated by the amplitude calculating unit 120 changes by a predetermined rate from the time of start or the time of the last relearning. In this case, the relearning instruction unit 136 may output a relearning instruction when the amplitude value changes by a predetermined rate or more in the direction of increasing, or may output a relearning instruction when the amplitude value changes by a predetermined rate or more in the direction of decreasing. , may output a re-learning instruction.

Note that the relearning instruction unit 136 may output a relearning instruction when a predetermined time has passed since the last relearning. If relearning has not been performed since the start of deterioration determination, the relearning instruction unit 136 outputs a relearning instruction when a predetermined time has elapsed since the start of the deterioration determination.

FIG. 10 is a diagram showing the functional configuration of the deterioration detection device 30 according to the modification.

The deterioration detection device 30 includes one or more of the passage time acquisition unit 142, the passage detection information acquisition unit 144, the scheduled time estimation unit 146, the date acquisition unit 148, the passage speed acquisition unit 152, and the weight acquisition unit 154. You may have more.

If the specific vehicle is scheduled to pass through the bridge at the same first time every day, the passage time acquisition unit 142 acquires information indicating the first time, which is the passage time of the specific vehicle, from an external device or the like. For example, when the specific vehicle is a route bus, the timetable information of the route bus or the like may be acquired, and the first time may be calculated based on the acquired timetable information. Then, when the passing time acquiring unit 142 acquires the first time, the extracting unit 118 extracts specific partial data in the parameter time-series data based on the acquired first time. For example, the extracting unit 118 cuts out a predetermined time range before and after the first time in the parameter time series data, and extracts specific partial data from the cut out range. As a result, the extracting unit 118 can extract the specific partial data with high precision with a small amount of processing, since the extracting unit 118 only needs to perform the extraction process on the partial data during a partial time period of the parameter time series data.

The passage detection information acquisition unit 144 acquires a detection signal detected by a passage detection device that detects that a specific vehicle is passing through a bridge. For example, the passage detection device acquires image data from a camera that captures an image of a vehicle passing over a bridge, analyzes the acquired image data, and determines whether or not a specific vehicle is passing through the bridge. When the passage detection device determines that the specific vehicle is passing through the bridge, it gives the deterioration detection device 30 a detection signal indicating that the specific vehicle is passing through the bridge. Further, for example, the passage detection device may be a receiving device that receives identification information for identifying the specific vehicle from a wireless communication device provided in the specific vehicle by a radio signal. In this case, the passage detection device is provided near the bridge, and upon receiving the identification information from the specific vehicle, gives the deterioration detection device 30 a detection signal indicating that the specific vehicle is passing through the bridge.

Then, when the passage detection information acquisition unit 144 receives detection information indicating that the specific vehicle is passing through the bridge from the passage detection device, it provides the extraction unit 118 with time information indicating the time at which the detection information was received. The extraction unit 118 extracts specific partial data in the parameter time-series data based on the received time information. For example, the extraction unit 118 cuts out a predetermined time range before and after the time indicated in the time information received from the parameter time-series data, and extracts specific partial data from the cut out range. As a result, the extracting unit 118 can extract the specific partial data with high precision with a small amount of processing, since the extracting unit 118 only needs to perform the extraction process on the partial data during a partial time period of the parameter time series data.

The scheduled time estimation unit 146 acquires passage information indicating that the specific vehicle has passed a predetermined first position. For example, if the specific vehicle is a fixed-route bus, the first location is the bus stop immediately before the bridge on the bus route or the specific bus stop. When the specific vehicle is a route bus, the scheduled time estimation unit 146 receives passage information indicating that the vehicle has passed the first position. The transit information is transmitted from, for example, a transmitter provided at a bus stop, a transmitter provided on a route bus, or a management device that manages route bus operation information.

Upon receiving the passage information, the scheduled time estimating unit 146 estimates the scheduled time for the specific vehicle to pass the bridge based on the passage information and the predicted travel time of the specific vehicle from the first position to the bridge. For example, the scheduled time estimating unit 146 calculates the scheduled time by adding the estimated running time to the time when the passage information is received.

Then, the scheduled time estimation unit 146 gives the estimated scheduled time to the extraction unit 118 . The extraction unit 118 extracts specific partial data in the parameter time-series data based on the received scheduled time. For example, the extracting unit 118 cuts out a predetermined time range before and after the time indicated by the scheduled time from the time-series data of the parameter, and extracts specific partial data from the cut out range. As a result, the extracting unit 118 can extract the specific partial data with high precision with a small amount of processing, since the extracting unit 118 only needs to perform the extraction process on the partial data during a partial time period of the parameter time series data.

The date acquisition unit 148 acquires date information indicating a preset date. The date acquisition unit 148 may acquire the day of the week as the date. In this case, the specific vehicle is a vehicle scheduled to pass the bridge at the same time every day. The date acquisition unit 148 gives the acquired date information to the amplitude calculation unit 120 .

When the date information is acquired, the amplitude calculation unit 120 calculates the amplitude value based on the specific partial data extracted from the time-series data on the date indicated in the acquired date information. For example, the amplitude calculation unit 120 calculates the amplitude value based on the specific partial data extracted from the time-series data on weekday dates, and calculates the amplitude value based on the specific partial data extracted from the time-series data on holidays. do not do. For example, if the specific vehicle is a route bus, the number of passengers on weekdays and holidays may differ greatly, resulting in a significant difference in weight. Further, when the specific vehicle is a route bus, the degree of traffic congestion differs between weekdays and holidays, and the passing speed over the bridge may differ greatly. Therefore, when the specific vehicle is a route bus, the amplitude value of the expansion/contraction amount may have greatly different waveform characteristics between weekdays and holidays.

Therefore, by calculating the amplitude value based on the specific partial data extracted from the time-series data on a preset date, the amplitude calculation unit 120 can detect that the same type of vehicle can move at substantially the same weight and at substantially the same speed. It is possible to output the amplitude value of the expansion/contraction amount when passing through. As a result, the deterioration detection device 30 can accurately determine whether or not the bridge has deteriorated.

The time period acquisition unit 150 acquires time period information indicating a preset time period within a day. The time zone acquisition unit 150 may acquire information such as, for example, from 5:00 am to 7:00 am as the time zone. The time period obtaining section 150 provides the obtained time period information to the amplitude calculating section 120 .

When the time period information is acquired, the amplitude calculation unit 120 calculates the amplitude value based on the specific partial data extracted from the time series data in the time period indicated by the acquired time period information. For example, the amplitude calculation unit 120 calculates the amplitude value based on the specific partial data extracted from the time series data in the designated time period, Amplitude value is not calculated based on For example, the degree of traffic congestion varies depending on the time of day, and the speed at which a particular vehicle passes over a bridge may vary greatly. Therefore, the amplitude value of the amount of expansion/contraction may vary greatly in waveform characteristics depending on the time period of the day.

Therefore, by calculating the amplitude value based on the specific partial data extracted from the time-series data in a preset time period, the amplitude calculation unit 120 can detect that the same type of vehicle has substantially the same weight and substantially the same speed. It is possible to output the amplitude value of the expansion/contraction amount when passing through. As a result, the deterioration detection device 30 can accurately determine whether or not the bridge has deteriorated.

The passing speed acquisition unit 152 acquires the speed of a specific vehicle when it passes through a bridge. For example, the passing speed acquisition unit 152 may acquire log data measured by a speedometer provided in a specific vehicle. Then, the passing speed acquisition unit 152 may acquire the speed at the time when the specific vehicle passed through the bridge from the log data. The passing speed acquisition unit 152 gives speed information indicating the acquired speed to the amplitude calculation unit 120 .

When the speed information is acquired, the amplitude calculation unit 120 calculates the amplitude value based on the specific partial data extracted from the time-series data in which the speed of the specific vehicle when passing over the bridge is within a preset range. As a result, the amplitude calculator 120 can output the amplitude value of the expansion/contraction amount when vehicles of the same type pass at substantially the same speed. Therefore, the deterioration detection device 30 can more accurately determine whether or not the bridge has deteriorated.

The weight acquisition unit 154 acquires the weight of the specific vehicle when it passes over the bridge. For example, garbage trucks are commonly equipped with scales that measure their own weight. The weight acquisition unit 154 may acquire log data measured by a weight scale provided in a specific vehicle such as a garbage truck. Then, the weight acquisition unit 154 may acquire the weight at the time when the specific vehicle passed the bridge from the log data. When the specific vehicle is a route bus, the weight acquisition unit 154 acquires the number of passengers at the bus stop immediately before the bridge on the bus route or at the specific bus stop, and estimates the weight based on the acquired number of passengers. good too. Weight acquisition section 154 then provides weight information indicating the acquired weight to amplitude calculation section 120 .

When the weight information is acquired, the amplitude calculation unit 120 calculates the amplitude value based on the specific partial data extracted from the time-series data in which the weight of the specific vehicle when passing over the bridge is within a preset range. As a result, the amplitude calculator 120 can output the amplitude value of the expansion/contraction amount when vehicles of the same type pass with substantially the same weight. Therefore, the deterioration detection device 30 can more accurately determine whether or not the bridge has deteriorated.

FIG. 11 is a flowchart showing the processing flow of the deterioration detection device 30. FIG. The deterioration detection device 30 executes processing according to the flow shown in FIG. 11, for example.

First, in S11, the deterioration detection device 30 determines whether or not a route bus, which is a specific vehicle, has passed the bus stop (or a specific bus stop) immediately before the bridge on the bus route. If the vehicle has not passed the last bus stop (No in S11), the deterioration detection device 30 waits in S11. When passing the last bus stop (Yes in S11), the deterioration detection device 30 advances the process to S12.

At S12, the deterioration detection device 30 estimates the scheduled time of passing through the bridge. Subsequently, in S13, the deterioration detection device 30 determines whether or not the time is a predetermined time before the scheduled time. If the time is not the predetermined time before the scheduled time (No in S13), the deterioration detection device 30 waits for the process in S13. If the time comes a predetermined time before the scheduled time (Yes in S13), the deterioration detection device 30 advances the process to S14.

In S14, the deterioration detection device 30 turns on the power of the sensor 20. For example, the deterioration detection device 30 gives an instruction signal to the sensor 20 through wireless communication or the like to turn on the power.

Subsequently, in S15, the deterioration detection device 30 receives and stores the time-series data of the parameters detected by the sensor 20. For example, the deterioration detection device 30 receives and stores time-series data of the amount of expansion and contraction of the target portion of the bridge.

Subsequently, in S16, the deterioration detection device 30 determines whether or not the time has come after a predetermined time from the scheduled time. If the time is not the predetermined time after the scheduled time (No in S16), the deterioration detection device 30 waits for the process in S16. If the predetermined time after the scheduled time comes (Yes in S16), the deterioration detection device 30 advances the process to S17.

In S17, the deterioration detection device 30 turns off the power of the sensor 20. For example, the deterioration detection device 30 gives an instruction signal to the sensor 20 through wireless communication or the like to turn off the power.

Subsequently, in S18, the deterioration detection device 30 extracts specific partial data, which is partial data when a route bus, which is a specific vehicle, passes through a bridge, from the stored time-series data of parameters. Subsequently, in S19, the deterioration detection device 30 calculates the amplitude value of the expansion/contraction amount in the traveling direction of the bridge when the route bus, which is the specific vehicle, passes through from the extracted specific portion data. For example, when the parameter represents the amount of expansion/contraction of the target portion, the deterioration detection device 30 calculates the difference between the maximum value and the minimum value in the extracted specific portion data as the amplitude value. Subsequently, in S20, the deterioration detection device 30 stores the calculated amplitude value.

Subsequently, in S21, the deterioration detection device 30 determines whether or not the calculated amplitude value is greater than the reference value. If the amplitude value is not greater than the reference value (No in S21), the deterioration detection device 30 returns the process to S11 and repeats the process from S11. If the amplitude value is greater than the reference value (Yes in S21), the deterioration detection device 30 advances the process to S22.

In S22, the deterioration detection device 30 outputs alarm information indicating that the bridge has deteriorated, for example, to an administrator or an information processing device held by the administrator. When S22 ends, the deterioration detection device 30 ends this flow. After S22, the deterioration detection device 30 may increase the reference value by a predetermined amount and return the process to S11.

As described above, the deterioration detection system 10 according to the first embodiment detects the amount of expansion and contraction of the bridge in the running direction when a specific vehicle such as a route bus passes through the bridge, based on the specific portion data. An amplitude value is calculated, and when the calculated amplitude value exceeds a preset reference value, it is determined that the bridge has deteriorated, and alarm information indicating that the bridge has deteriorated is output. Thereby, according to the deterioration detection system 10 according to the first embodiment, the deterioration of the bridge can be detected easily and continuously for a long period of time. Therefore, according to the degradation detection system 10 according to the first embodiment, the state of the bridge can be constantly monitored at low cost, and the bridge can be systematically managed.

Furthermore, the deterioration detection system 10 according to the first embodiment re-learns the first neural network used to detect a specific vehicle at an appropriate timing, so that deterioration of a bridge can be accurately detected over a long period of time. can.

(Second embodiment)
Next, a weight measuring system 210 according to the second embodiment will be described. In the description of the second embodiment, elements having substantially the same configuration as those of the first embodiment are denoted by the same reference numerals, and detailed description thereof will be omitted except for differences.

FIG. 12 is a diagram showing a weight measurement system 210 according to the second embodiment. The weight measurement system 210 accurately measures the weight of vehicles passing over the bridge while correcting the relevant information according to the deterioration of the bridge.

The weight measurement system 210 includes a sensor 20 , a transmission device 22 and a weight measurement device 230 .

The weight measuring device 230 receives time-series data of parameters representing the displacement in the running direction of the target portion of the bridge, transmitted from the transmitting device 22 via the network. The weight measuring device 230 calculates the vehicle amplitude value representing the amplitude value of the amount of expansion and contraction in the running direction of the bridge when the vehicle passes, based on the received time-series data of the parameters. Then, the weight measuring device 230 calculates the weight of the vehicle based on the relationship information representing the correspondence between the amplitude value and the weight and the calculated vehicle amplitude value. Furthermore, when the weight measuring device 230 determines that the bridge has deteriorated, it corrects the relational information representing the correspondence between the amplitude value and the weight.

The weight measuring device 230 is a computer such as a server device that can be connected to a network. The weight measuring device 230 may be one computer, or may be composed of a plurality of computers like a cloud system.

Note that the weight measurement system 210 may be configured without the transmission device 22 . In this case, the weight measuring device 230 directly obtains from the sensor 20 the parameter representing the displacement in the running direction of the target portion of the bridge. Also, in this case, the weight measuring device 230 may be provided in the vicinity of the sensor 20, that is, in the vicinity of the bridge.

FIG. 13 is a diagram showing the functional configuration of the weight measuring device 230. As shown in FIG. Weight measuring device 230 includes acquisition unit 112 , time-series data storage unit 114 , extraction unit 116 , vehicle extraction unit 242 , vehicle amplitude calculation unit 244 , relationship information storage unit 246 , and weight calculation unit 248 . , an extraction unit 118, an amplitude calculation unit 120, a correction unit 250, a vehicle collection unit 262, a vehicle relearning data storage unit 264, a collection unit 132, a relearning data storage unit 134, and a relearning instruction. A unit 136 , a relearning unit 138 , and a vehicle relearning unit 266 are provided.

The acquisition unit 112, the time-series data storage unit 114, and the extraction unit 116 have the same functions and configurations as in the first embodiment.

The vehicle extraction unit 242 extracts vehicle partial data, which is partial data when the vehicle passes over the bridge, from the time-series data of the parameters stored in the time-series data storage unit 114, based on the determination result by the vehicle determination neural network. Extract. Here, the vehicle is not limited to a specific vehicle such as a route bus, and may be any kind of vehicle.

The neural network for vehicle judgment inputs target partial data and judges whether or not the vehicle has passed the bridge based on the target partial data. The vehicle determination neural network acquires in advance time-series data obtained when a vehicle passes over a bridge, and learns using the acquired time-series data as teacher data. In this embodiment, the vehicle determination neural network is a convolutional neural network (CNN). By using a convolutional neural network, the vehicle extraction unit 242 detects that the vehicle has passed over the bridge even if waveform data obtained when the vehicle has passed over the bridge is included in any time portion of the partial data. It can be detected with high accuracy. When the vehicle determination neural network determines that the vehicle has passed the bridge, the vehicle extraction unit 242 extracts the partial data input to the vehicle determination neural network as vehicle partial data. Output. The neural network for vehicle determination may be a neural network having the same configuration as the first neural network. However, in this case, the parameters included in the vehicle determination neural network have different values from the parameters included in the first neural network as a result of learning.

Each time the vehicle extraction unit 242 extracts the vehicle part data, the vehicle amplitude calculation part 244 calculates a vehicle amplitude value representing the amplitude value of the amount of expansion and contraction of the bridge in the traveling direction when the vehicle passes, based on the extracted vehicle part data. calculate. For example, when the parameter represents the amount of expansion/contraction of the target portion, the vehicle amplitude calculator 244 calculates the difference between the maximum value and the minimum value in the extracted vehicle portion data as the vehicle amplitude value.

If the parameter is not the amount of expansion/contraction of the target portion, the vehicle amplitude calculator 244 may output a value correlated with the amplitude value of the amount of expansion/contraction of the target portion as the vehicle amplitude value. For example, when the parameter is the magnitude of the natural frequency, the vehicle amplitude calculator 244 may calculate the difference between the maximum value and the minimum value in the extracted vehicle partial data as the vehicle amplitude value.

The relationship information storage unit 246 stores relationship information representing the correspondence relationship between the amplitude value and the weight of the vehicle.

The amplitude value of the amount of expansion and contraction in the running direction of the bridge when the vehicle passes over the bridge is correlated with the weight of the vehicle. More specifically, the amplitude value of the amount of expansion and contraction of the bridge in the running direction when the vehicle passes over the bridge increases according to the weight of the vehicle. Therefore, the administrator or the like generates in advance relational information representing the correspondence between the amplitude value and the weight of the vehicle passing through the bridge, and causes the relational information storage unit 246 to store the relational information.

The administrator, for example, may generate relational information by running test vehicles of various weights on the bridge and measuring the correspondence between the weight of the vehicle and the amplitude value. Also, the administrator or the like may further use the simulation results to generate the relationship information. Also, the administrator or the like may use the relationship information or the like applied to other bridges to generate the relationship information of the bridge.

The relationship information storage unit 246 may store, as relationship information, a table that associates each of a plurality of amplitude values with the corresponding weight. Further, the relational information storage unit 246 may store, as the relational information, a function or an arithmetic expression for outputting a weight by inputting an amplitude value.

The weight calculator 248 calculates the weight of the vehicle each time the vehicle amplitude calculator 244 calculates the vehicle amplitude value. More specifically, the weight calculation unit 248 calculates the weight of the vehicle based on the calculated vehicle amplitude value and the relationship information indicating the correspondence relationship between the amplitude value and the weight stored in the relationship information storage unit 246. . The weight calculator 248 outputs the calculated weight to, for example, an information processing device or a server held by an administrator. In this case, the weight calculator 248 may also output the time when the vehicle passed the bridge. As a result, the administrator or the like can associate the vehicle that has passed through the bridge with the weight.

The extraction unit 118 has the same function and configuration as those of the first embodiment, and based on the determination result of the first neural network, the time-series data of the parameters stored in the time-series data storage unit 114 is used to determine whether the specific vehicle has passed the bridge. Specific partial data, which is partial data at the time of passage, is extracted.

The amplitude calculator 120 has the same function and configuration as those of the first embodiment. A vehicle specific amplitude value representing the amplitude value of the quantity is calculated.

The correction unit 250 compares the calculated specific vehicle amplitude value with a preset reference value each time the amplitude calculation unit 120 calculates the specific vehicle amplitude value. Then, when the specific vehicle amplitude value becomes larger than the reference value, the correction unit 250 corrects the correspondence stored in the relationship information storage unit 246 .

The correction unit 250 may calculate the moving average value of the specific vehicle amplitude values of the most recent predetermined number of samples, and correct the related information when the moving average value becomes greater than the reference value. Further, the correction unit 250 executes predetermined filtering processing such as noise removal processing on the specific vehicle amplitude values of a plurality of samples, and when the result of the filtering processing becomes larger than the reference value, the relational You may correct the information.

The vehicle collection unit 262 collects the extracted vehicle partial data each time the vehicle extraction unit 242 extracts the vehicle partial data. The vehicle relearning data storage unit 264 stores vehicle partial data collected by the vehicle collection unit 262 .

The collection unit 132 has the same function and configuration as in the first embodiment, and collects the extracted specific partial data every time the extraction unit 118 extracts the specific partial data. The relearning data storage unit 134 has the same function and configuration as those of the first embodiment, and stores the specific partial data collected by the collection unit 132 .

The re-learning instruction unit 136 determines whether or not to re-learn the first neural network and the vehicle determination neural network based on the same predetermined criteria as in the first embodiment. A re-learning instruction is output based on the determination result of whether or not to re-learn the judgment neural network. Re-learning instruction unit 136 supplies a relearning instruction to vehicle relearning unit 266 and relearning unit 138 .

When the re-learning instruction is output from the re-learning instruction unit 136, the vehicle re-learning unit 266 re-learns the vehicle determination neural network using the vehicle partial data collected by the vehicle collection unit 262 as teacher data. That is, the vehicle re-learning unit 266 re-learns the already learned neural network for vehicle determination. For example, the vehicle re-learning unit 266 may perform re-learning by error backpropagation or the like using network parameters such as weights and biases set immediately before re-learning as initial values. After changing the network parameter to a random value or a predetermined value, re-learning may be performed by error backpropagation method or the like.

The vehicle re-learning unit 266 re-learns the vehicle determination neural network using the vehicle part data collected after the last re-learning as teacher data. If re-learning has not been performed since the start of deterioration determination, the vehicle re-learning unit 266 may re-learn the vehicle determination neural network using the vehicle part data collected after the start as teacher data. good. As a result, the vehicle re-learning unit 266 can re-learn the vehicle determination neural network so as to perform appropriate determination processing in accordance with the latest state of the bridge. The vehicle re-learning unit 266 may re-learn the vehicle determination neural network using pre-collected time-series data as teacher data in addition to the vehicle part data collected by the vehicle collecting unit 262 .

The relearning unit 138 has the same configuration as that of the first embodiment, and when the relearning instruction is output from the relearning instruction unit 136, the specific partial data collected by the collecting unit 132 is used as teacher data, and the first neural Retrain the network.

The weight measuring device 230 includes the passage time acquisition unit 142, the passage detection information acquisition unit 144, the scheduled time estimation unit 146, the date acquisition unit 148, the time period acquisition unit 150, the passage speed acquisition unit 152, and the weight The configuration may further include one or more of the acquisition units 154 .

FIG. 14 is a diagram showing the relationship between the weight of the test vehicle and the amplitude value of the amount of expansion and contraction of the bridge. In FIG. 14, the graph with black circles represents the relationship when the deterioration of the bridge is small, and the graph with white squares represents the relationship when the deterioration of the bridge exceeds a predetermined reference value.

As the deterioration of the bridge increases, the amplitude value of the amount of expansion and contraction in the running direction of the bridge when a vehicle passes increases. Therefore, as shown in FIG. 14, the weight of the vehicle and the amount of expansion and contraction of the bridge in the running direction when the vehicle is passing are determined when the deterioration of the bridge exceeds, for example, the reference value and when the deterioration is less than the reference value. has a different relationship with the amplitude value of Therefore, for example, the administrator generates first relational information when the deterioration of the bridge is equal to or less than the reference value and second relational information when the deterioration of the bridge is greater than the reference value. 246.

Then, when the specific vehicle amplitude value is equal to or less than the reference value, the correction unit 250 outputs the first relationship information from the relationship information storage unit 246 to the weight calculation unit 248 . Further, when the specific vehicle amplitude value is greater than the reference value, correction unit 250 outputs the second relationship information from relationship information storage unit 246 to weight calculation unit 248 . Accordingly, the correction unit 250 can accurately measure the weight of the vehicle regardless of the degree of deterioration of the bridge.

Note that a plurality of different reference values may be preset in the correction unit 250 . Then, the correction unit 250 may adjust the correction amount of the related information according to the magnitude of the reference value each time the specific vehicle amplitude value becomes larger than each reference value. Thereby, the correcting unit 250 can appropriately correct the relationship information according to the level of deterioration of the bridge.

FIG. 15 is a flow chart showing the flow of related information correction processing by the weight measuring device 230 . The weight measuring device 230 executes processing according to the flow shown in FIG. 15, for example.

It should be noted that the processing from S11 to S21 in FIG. 15 is the same processing as the deterioration detection device 30 in FIG. However, in the weight measurement system 210, the power of the sensor 20 is always on in order to measure the weight of the vehicle passing over the bridge. Therefore, when the sensor 20 used for weight measurement and the sensor 20 used when the specific vehicle passes are the same, the weight measuring device 230 does not execute the processes of S13, S14, S16 and S17. you can However, in this case, in S18, the weight measuring device 230 cuts out a range from a predetermined time before the scheduled time to a predetermined time after the scheduled time from the parameter time series data, and extracts specific partial data from the cut out range. .

In S21, the weight measuring device 230 determines whether or not the calculated amplitude value is greater than the reference value. If the amplitude value is greater than the reference value (Yes in S21), the weight measuring device 230 advances the process to S31.

At S<b>31 , the weight measuring device 230 corrects the correspondence stored in the relationship information storage unit 246 . When S31 ends, the weight measuring device 230 ends this flow. After S31, the weight measuring device 230 may increase the reference value by a predetermined amount and return the process to S11.

The weight measurement system 210 according to the second embodiment as described above calculates the vehicle amplitude value representing the amplitude value of the amount of expansion and contraction of the bridge in the running direction when the vehicle passes, and the correspondence relationship between the amplitude value and the weight of the vehicle. and the calculated vehicle amplitude value, the weight of the vehicle is calculated. Therefore, the weight measurement system 210 can accurately measure the weight of vehicles passing over the bridge.

Furthermore, the weight measurement system 210 corrects the related information when the specific vehicle amplitude value representing the amplitude value of the amount of expansion and contraction of the bridge in the running direction when the specific vehicle passes exceeds the reference value. Therefore, the weight measurement system 210 can correct the relevant information according to the deterioration of the bridge.

As described above, according to the weight measurement system 210 according to the second embodiment, it is possible to accurately measure the weight of a vehicle passing over a bridge while correcting the relevant information as the bridge deteriorates.

(Hardware configuration of deterioration detection device 30 and weight measurement device 230)
FIG. 16 is a diagram showing the hardware configuration of the deterioration detection device 30 and the weight measurement device 230. As shown in FIG. The deterioration detection device 30 and the weight measurement device 230 are realized by, for example, a hardware configuration similar to that of a general computer. The deterioration detection device 30 and the weight measurement device 230 include a CPU (Central Processing Unit) 301, an operation device 302, a display device 303, a ROM (Read Only Memory) 304, a RAM (Random Access Memory) 305, and a storage device. 306 , a communication device 307 and a bus 309 . Each unit is connected by a bus 309 .

The CPU 301 uses a predetermined area of the RAM 305 as a working area to execute various processes in cooperation with various programs pre-stored in the ROM 304 or the storage device 306, and controls the operation of each part constituting the deterioration detection device 30 or the weight measurement device 230. Overall control. In addition, the CPU 301 operates the operating device 302, the display device 303, the communication device 307, etc. in cooperation with programs pre-stored in the ROM 304 or the storage device 306. FIG.

The operation device 302 is an input device such as a touch panel, a mouse, a keyboard, or the like, receives information input by a user as an instruction signal, and outputs the instruction signal to the CPU 301 .

The display device 303 is a display unit such as an LCD (Liquid Crystal Display). The display device 303 displays various information based on display signals from the CPU 301 .

The ROM 304 non-rewritably stores programs and various setting information used to control the deterioration detection device 30 or the weight measurement device 230 . The RAM 305 is a volatile storage medium such as SDRAM (Synchronous Dynamic Random Access Memory). A RAM 305 functions as a work area for the CPU 301 .

The storage device 306 is a rewritable recording device such as a semiconductor storage medium such as a flash memory, or a magnetically or optically recordable storage medium. Storage device 306 stores a program used to control deterioration detection device 30 or weight measurement device 230 .

The communication device 307 transmits and receives data to and from other devices. Also, the communication device 307 may transmit and receive data to and from a server or the like via a network.

The programs executed by the deterioration detection device 30 and the weight measurement device 230 are, for example, stored on a computer connected to a network such as the Internet, and provided by being downloaded via the network. Also, the programs executed by the deterioration detection device 30 and the weight measurement device 230 may be provided by being incorporated in advance in a portable storage medium or the like.

The programs executed by the deterioration detection device 30 include a collection module, an extraction module, an extraction module, an amplitude calculation module, a deterioration determination module, an alarm output module, a collection module, a re-learning instruction module, and a re-learning module. It has a module configuration including a learning module. The CPU 301 reads out such a program from a storage medium or the like and loads each module into the RAM 305 . By executing such a program, the CPU 301 executes the acquisition unit 112, the extraction unit 116, the extraction unit 118, the amplitude calculation unit 120, the deterioration determination unit 122, the alarm output unit 124, the collection unit 132, and the It functions as a learning instruction section 136 and a re-learning section 138 . Part or all of the acquisition unit 112, the extraction unit 118, the amplitude calculation unit 120, the deterioration determination unit 122, the alarm output unit 124, the collection unit 132, the relearning instruction unit 136, and the relearning unit 138 are configured by hardware. may be

The program executed by the weight measuring device 230 includes a collection module, an extraction module, a vehicle extraction module, a vehicle amplitude calculation module, a weight calculation module, an extraction module, an amplitude calculation module, a correction module, a vehicle It has a module configuration including a collection module, a collection module, a relearning instruction module, a vehicle relearning module, and a relearning module. The CPU 301 reads out such a program from a storage medium or the like and loads each module into the RAM 305 . By executing such a program, the CPU 301 executes the acquisition unit 112, the extraction unit 116, the vehicle extraction unit 242, the vehicle amplitude calculation unit 244, the weight calculation unit 248, the extraction unit 118, the amplitude calculation unit 120, the correction It functions as a unit 250 , a vehicle collection unit 262 , a collection unit 132 , a relearning instruction unit 136 , a vehicle relearning unit 266 and a relearning unit 138 . Part or all of acquisition unit 112, vehicle extraction unit 242, vehicle amplitude calculation unit 244, weight calculation unit 248, extraction unit 118, amplitude calculation unit 120, and correction unit 250 may be configured by hardware.

Although the embodiments of the present invention have been described above, these embodiments are presented as examples and are not intended to limit the scope of the invention. Embodiments may undergo various modifications.

10 deterioration detection system 20 sensor 22 transmission device 30 deterioration detection device 54 main girder 56 lower surface 62 first point 64 second point 66 first member 68 second member 70 displacement detection device 72 optical element 74 detector 76 half mirror 78 light emitting unit 80 Light receiving unit 82 Detection circuit 112 Acquisition unit 114 Time-series data storage unit 118 Extraction unit 120 Amplitude calculation unit 122 Deterioration determination unit 124 Alarm output unit 132 Collection unit 134 Re-learning data storage unit 136 Re-learning instruction unit 138 Re-learning unit 142 Passing time acquiring unit 144 Passing detection information acquiring unit 146 Scheduled time estimating unit 148 Date acquiring unit 150 Time period acquiring unit 152 Passing speed acquiring unit 154 Weight acquiring unit 210 Weight measuring system 230 Weight measuring device 242 Vehicle extracting unit 244 Vehicle amplitude calculating unit 246 Relationship information storage unit 248 Weight calculation unit 250 Correction unit 262 Vehicle collection unit 264 Vehicle relearning data storage unit 266 Vehicle relearning unit

Claims

an acquisition unit that collects, from sensors provided on a bridge, time-series data of parameters representing displacement in a running direction in a target portion of the bridge where the sensor is provided;
When the specific vehicle passes through the measured section of the bridge based on the determination result of a neural network that inputs the time-series data and outputs a determination result indicating whether the specific vehicle has passed or not, based on the time-series data an extraction unit for extracting specific partial data of
an amplitude calculation unit that calculates an amplitude value of the amount of expansion and contraction of the bridge in the running direction when the specific vehicle passes, based on the specific part data;
a deterioration determination unit that determines that the bridge has deteriorated when the amplitude value is greater than a preset reference value;
Determining whether or not to relearn the neural network based on predetermined determination criteria, and issuing a relearning instruction to instruct relearning of the neural network based on the determination result of whether or not to relearn the neural network. a relearning instruction unit to output;
A deterioration detection device comprising:
The deterioration detection device according to claim 1, further comprising an alarm output unit that outputs alarm information indicating that the bridge has deteriorated when it is determined that the bridge has deteriorated.
The relearning instruction unit is set with one or more threshold values between the initial value, which is the amplitude value at the start of the determination of the deterioration of the bridge, and the reference value,
The deterioration detection device according to claim 1 or 2, wherein the relearning instruction section outputs the relearning instruction each time the amplitude value becomes greater than each of the one or more threshold values.
The deterioration detection device according to claim 1 or 2, wherein the relearning instruction unit outputs the relearning instruction when the amplitude value changes in a preset pattern.
The deterioration detection device according to claim 1 or 2, wherein the re-learning instruction unit outputs the re-learning instruction when a predetermined time has elapsed from the start of deterioration determination or the previous re-learning.
a collection unit that collects the specific partial data;
a re-learning unit for re-learning the neural network using the collected specific partial data as teacher data when the re-learning instruction is output;
The deterioration detection device according to any one of claims 1 to 5, further comprising:
The deterioration detection device according to claim 6, wherein the re-learning unit re-learns the neural network using the specific partial data collected after the time of starting or immediately before re-learning as the teacher data.
The deterioration detection device according to any one of claims 1 to 7, wherein the neural network is a convolutional neural network.
a sensor provided at a target portion of a bridge for detecting a parameter representing displacement in the running direction at the target portion of the bridge;
an acquisition unit that collects time-series data of the parameter from the sensor;
When the specific vehicle passes through the measured section of the bridge based on the determination result of a neural network that inputs the time-series data and outputs a determination result indicating whether the specific vehicle has passed or not, based on the time-series data an extraction unit for extracting specific partial data of
an amplitude calculation unit that calculates an amplitude value of the amount of expansion and contraction of the bridge in the running direction when the specific vehicle passes, based on the specific part data;
a deterioration determination unit that determines that the bridge has deteriorated when the amplitude value is greater than a preset reference value;
Determining whether or not to relearn the neural network based on predetermined determination criteria, and issuing a relearning instruction to instruct relearning of the neural network based on the determination result of whether or not to relearn the neural network. a relearning instruction unit to output;
A degradation detection system comprising:
Collecting time-series data of parameters representing displacement in the running direction in a target portion of the bridge where the sensor is installed, from a sensor installed on the bridge;
When the specific vehicle passes through the measured section of the bridge based on the determination result of a neural network that inputs the time-series data and outputs a determination result indicating whether the specific vehicle has passed or not, based on the time-series data Extract specific partial data of
calculating an amplitude value of the amount of expansion and contraction of the bridge in the running direction when the specific vehicle passes, based on the specific portion data;
If the amplitude value is greater than a preset reference value, it is determined that the bridge has deteriorated,
Determining whether or not to relearn the neural network based on predetermined determination criteria, and issuing a relearning instruction to instruct relearning of the neural network based on the determination result of whether or not to relearn the neural network. Output deterioration detection method.
A program for causing an information processing device to function as a deterioration detection device,
the information processing device,
an acquisition unit that collects, from sensors provided on a bridge, time-series data of parameters representing displacement in a running direction in a target portion of the bridge where the sensor is provided;
When the specific vehicle passes through the measured section of the bridge based on the determination result of a neural network that inputs the time-series data and outputs a determination result indicating whether the specific vehicle has passed or not, based on the time-series data an extraction unit for extracting specific partial data of
an amplitude calculation unit that calculates an amplitude value of the amount of expansion and contraction of the bridge in the running direction when the specific vehicle passes, based on the specific part data;
a deterioration determination unit that determines that the bridge has deteriorated when the amplitude value exceeds a preset reference value, and determines whether or not to relearn the neural network based on a predetermined determination criterion, a relearning instruction unit that outputs a relearning instruction for instructing relearning of the neural network based on a determination result as to whether or not to relearn the neural network;
A program that works as
an acquisition unit that collects, from sensors provided on a bridge, time-series data of parameters representing displacement in a running direction in a target portion of the bridge where the sensor is provided;
Based on the determination result of a neural network for vehicle determination that inputs the time-series data and outputs a determination result indicating whether or not the vehicle has passed, the vehicle passes through the measured section of the bridge based on the time-series data. a vehicle extraction unit that extracts vehicle partial data that is partial data of time;
a vehicle amplitude calculation unit that calculates a vehicle amplitude value representing an amplitude value of the amount of expansion and contraction of the bridge in the traveling direction when the vehicle passes, based on the vehicle part data;
a weight calculation unit that calculates the weight of the vehicle based on relationship information representing a correspondence relationship between the amplitude value and the weight of the vehicle and the calculated vehicle amplitude value;
When the specific vehicle passes through the measured section of the bridge based on the determination result of a neural network that inputs the time-series data and outputs a determination result indicating whether the specific vehicle has passed or not, based on the time-series data an extraction unit for extracting specific partial data that is partial data of
an amplitude calculation unit that calculates, based on the specific portion data, a specific vehicle amplitude value representing an amplitude value of the amount of expansion and contraction of the bridge in the running direction when the specific vehicle passes;
a correcting unit that corrects the relevant information when the specific vehicle amplitude value exceeds a preset reference value;
Determining whether or not to relearn the vehicle determination neural network and the neural network based on predetermined determination criteria, and determining whether or not to relearn the vehicle determination neural network and the neural network a re-learning instruction unit that outputs a re-learning instruction instructing re-learning of the neural network for vehicle determination and the neural network based on
A weighing device comprising a
Collecting time-series data of parameters representing displacement in the running direction in a target portion of the bridge where the sensor is installed, from a sensor installed on the bridge;
Based on the determination result of a neural network for vehicle determination that inputs the time-series data and outputs a determination result indicating whether or not the vehicle has passed, the vehicle passes through the measured section of the bridge based on the time-series data. Extract the vehicle partial data, which is the hour partial data,
calculating a vehicle amplitude value representing an amplitude value of the amount of expansion and contraction of the bridge in the running direction when the vehicle passes, based on the vehicle part data;
calculating the weight of the vehicle based on the calculated vehicle amplitude value and relationship information representing the correspondence relationship between the amplitude value and the weight of the vehicle;
When the specific vehicle passes through the measured section of the bridge based on the determination result of a neural network that inputs the time-series data and outputs a determination result indicating whether the specific vehicle has passed or not, based on the time-series data an extraction unit for extracting specific partial data that is partial data of
calculating a specific vehicle amplitude value representing an amplitude value of the amount of expansion and contraction of the bridge in the running direction when the specific vehicle passes, based on the specific portion data;
correcting the relevant information when the specific vehicle amplitude value is greater than a preset reference value;
Determining whether or not to relearn the vehicle determination neural network and the neural network based on predetermined determination criteria, and determining whether or not to relearn the vehicle determination neural network and the neural network and outputting a re-learning instruction instructing re-learning of the neural network for vehicle determination and the neural network based on the weight measurement method.
A program for causing an information processing device to function as a weight measuring device,
the information processing device,
an acquisition unit that collects, from sensors provided on a bridge, time-series data of parameters representing displacement in a running direction in a target portion of the bridge where the sensor is provided;
Based on the determination result of a neural network for vehicle determination that inputs the time-series data and outputs a determination result indicating whether or not the vehicle has passed, the vehicle passes through the measured section of the bridge based on the time-series data. a vehicle extraction unit that extracts vehicle partial data that is partial data of time;
a vehicle amplitude calculation unit that calculates a vehicle amplitude value representing an amplitude value of the amount of expansion and contraction of the bridge in the traveling direction when the vehicle passes, based on the vehicle part data;
a weight calculation unit that calculates the weight of the vehicle based on relationship information representing a correspondence relationship between the amplitude value and the weight of the vehicle and the calculated vehicle amplitude value;
When the specific vehicle passes through the measured section of the bridge based on the determination result of a neural network that inputs the time-series data and outputs a determination result indicating whether the specific vehicle has passed or not, based on the time-series data an extraction unit for extracting specific partial data that is partial data of
an amplitude calculation unit that calculates, based on the specific portion data, a specific vehicle amplitude value representing an amplitude value of the amount of expansion and contraction of the bridge in the running direction when the specific vehicle passes;
a correcting unit that corrects the relevant information when the specific vehicle amplitude value exceeds a preset reference value;
Determining whether or not to relearn the vehicle determination neural network and the neural network based on predetermined determination criteria, and determining whether or not to relearn the vehicle determination neural network and the neural network a re-learning instruction unit that outputs a re-learning instruction instructing re-learning of the neural network for vehicle determination and the neural network based on
A program that works as