JP7777653B2 - Structural Deterioration Diagnostic System - Google Patents

Description

This disclosure relates to a structural deterioration diagnosis system that can identify whether a structure has deteriorated or a sensor has malfunctioned when diagnosing the deterioration of a structure such as a bridge using acceleration information detected by a sensor.

For example, structures such as bridges over which vehicles pass will gradually deteriorate over time as vehicles pass by. It is important to detect the deterioration of structures such as bridges before they break.

Prior art for diagnosing bridge deterioration involves diagnosing sensor abnormalities (failures) from changes in tilt or natural frequency calculated based on multiple sensors (see, for example, Patent Document 1). Specifically, when the ratio of probability density distributions based on the detection results of multiple sensors deviates from an acceptable range, it becomes possible to diagnose a failure in the sensor itself. As a result, a structural deterioration diagnosis system can be realized that can confirm the soundness of the sensors and ensure the reliability of deterioration diagnosis.

Patent No. 6081867

However, the conventional techniques have the following problems.
In Patent Document 1, the detection results of a plurality of sensors are compared with each other to diagnose abnormalities (failures) of the sensors. That is, in order to identify whether deterioration of the structure or an abnormality in the sensor has occurred, it is necessary to compare the detection results of the plurality of sensors.

Furthermore, when comparing the probability density distribution of slopes between multiple sensors, it was sometimes impossible to determine whether the problem was a structural abnormality or a sensor failure, depending on the influence of the sensor position or temperature characteristics.

This disclosure has been made to solve the above-mentioned problems, and aims to provide a structural deterioration diagnosis system that can distinguish between structural abnormalities and sensor failure from the detection results of a single sensor.

The structure deterioration diagnosis system disclosed herein comprises a sensor that is installed in a structure to be diagnosed for deterioration and outputs acceleration information of the structure, and a sensor controller that determines whether deterioration of the structure has occurred based on the acceleration information output from the sensor.The sensor controller calculates live load displacement based on vibration components that do not contain DC components in the acceleration information , calculates time series data of the most frequent value at the maximum displacement amount from the time progression of the live load displacement as a feature that does not change depending on temperature , calculates an abnormality degree that serves as an indicator for determining deterioration of the structure based on the time series data of the most frequent value , and determines that deterioration of the structure has occurred if the abnormality degree exceeds a threshold value.

This disclosure provides a structural deterioration diagnosis system that can distinguish between structural abnormalities and sensor failure from the detection results of a single sensor.

1 is a configuration diagram of a structure deterioration diagnosis system according to a first embodiment of the present disclosure; 1 is an explanatory diagram showing a state in which a sensor is installed in a structure that is a diagnosis target of a structure deterioration diagnosis system according to a first embodiment of the present disclosure. 10A to 10C are diagrams illustrating a time transition of a slope, a relationship between the slope and temperature, and a slope abnormality degree when the slope is used as a feature amount in the first embodiment of the present disclosure. 10A to 10C are diagrams illustrating a time transition of the natural frequency, a relationship between the natural frequency and temperature, and a natural frequency anomaly degree when the natural frequency is used as a feature quantity in the first embodiment of the present disclosure. 10A and 10B are diagrams showing the time progression of live load displacement, the relationship between live load displacement and temperature, and the live load displacement abnormality degree when live load displacement is used as a feature quantity in embodiment 1 of the present disclosure. 10 is an explanatory diagram illustrating a case where a tilt, which is one of the feature quantities, is calculated using three-axis acceleration information output from a sensor according to the first embodiment of the present disclosure. FIG. FIG. 10 is an explanatory diagram showing a case where only the tilt anomaly degree is in an abnormal state, and the natural frequency anomaly degree and the live load displacement anomaly degree are not in an abnormal state, in the first embodiment of the present disclosure. FIG. 10 is a diagram comparing a histogram of each feature amount when the abnormality degrees for all feature amounts are determined to be normal and a histogram of each feature amount when only the slope abnormality degree is determined to be abnormal in the first embodiment of the present disclosure. FIG. 10 is an explanatory diagram for comparing a histogram of the degree of inclination abnormality with a histogram of natural frequencies using a statistical method in the first embodiment of the present disclosure. 1 is a flowchart showing a series of processes for identifying whether a structural abnormality or a sensor failure is present, which are executed by the structural deterioration diagnosis system according to the first embodiment of the present disclosure.

Hereinafter, preferred embodiments of the structure deterioration diagnosis system of the present disclosure will be described with reference to the drawings.
The present disclosure has a technical feature of having the ability to distinguish between a structural abnormality and a sensor failure by determining multiple feature values from the detection results of one sensor, and then determining the degree of abnormality corresponding to each of the multiple feature values, and performing deterioration diagnosis using a combination of the multiple degrees of abnormality.

Embodiment 1.
In the first embodiment, a specific configuration for distinguishing between a structural abnormality and a sensor failure when diagnosing deterioration of a structure based on the measurement results from one sensor will be described.

Figure 1 is a configuration diagram of a structure deterioration diagnosis system according to embodiment 1 of the present disclosure. The structure deterioration diagnosis system according to embodiment 1 is configured to include one sensor 10 and a sensor controller 20.

Figure 2 is an explanatory diagram showing a state in which a sensor 10 is installed in a structure that is the target of deterioration diagnosis by the structure deterioration diagnosis system according to embodiment 1 of the present disclosure. Figure 2 shows a bridge 30 as a specific example of a structure, with Figure 2(A) being a side view of the bridge 30 and Figure 2(B) being a rear view of the bridge 30.

Bridges 30 gradually deteriorate over time as vehicles 1 pass over them. Therefore, the structure deterioration diagnosis system according to the first embodiment diagnoses the constantly changing state of deterioration of bridges 30 by using a sensor controller 20 to analyze acceleration information output by sensors 10 installed in positions suitable for deterioration diagnosis.

Furthermore, the sensor controller 20 according to this embodiment 1 has the function of determining whether a sensor 10 is malfunctioning based on the acceleration information output by one sensor 10.

The sensor 10 is installed on the main girder 31, which is a component of the bridge 30. Here, the main girder 31 corresponds to the structural element of the structure to be diagnosed. As shown in Figure 2(B), the main girder 31 is configured as three main girders 31a, 31b, and 31c, as an example. The example in Figure 2(B) illustrates a case in which the sensor 10 is installed in the center of the main girder 31b (i.e., the center of the span corresponding to the distance between the left and right supports 2). This installation location corresponds to an example of a location suitable for deterioration diagnosis.

The following explanation will explain a specific method for diagnosing deterioration based on the detection results from a single sensor 10 and distinguishing between a structural abnormality and a sensor failure. However, the sensor 10 itself may be installed in a different location from that in the first embodiment, or multiple sensors 10 may be installed in multiple locations on the bridge 30.

When multiple sensors 10 are installed, the sensor controller 20 can identify whether a structural abnormality or sensor failure has been detected for each sensor 10 based on the individual detection results at the installation location of each sensor 10.

Sensor 10 detects acceleration information occurring on bridge 30 and outputs the acceleration information to sensor controller 20. One example of sensor 10 is a three-axis acceleration sensor that uses a thin-film quartz crystal oscillator, has excellent responsiveness, and can measure accelerations in the measurement range from DC to several tens of Hz.

In this way, by using a three-axis acceleration sensor as sensor 10, leveling is not necessary and sensor output can be obtained regardless of the direction of tilt or vibration. Therefore, if the direction of tilt can be determined, such as by leveling, a two-axis or one-axis acceleration sensor may be used.

The sensor 10 can also convert analog signals related to the three-axis acceleration of the structure at the installation location into digital signals at a predetermined sampling rate (e.g., a sampling rate of 50 Hz) and transmit them to the sensor controller 20 as acceleration information.

The sensor controller 20, which receives the acceleration information output from the sensor 10, is configured with a feature conversion unit 21, an abnormality degree calculation unit 22, and an abnormality/fault determination unit 23, as shown in Figure 1 above.

The feature conversion unit 21 converts the acceleration information received from the sensor 10 into the tilt, natural frequency, and displacement of the structure at the installation position of the sensor 10. Here, the tilt, natural frequency, and displacement each correspond to feature values that serve as index values for diagnosing deterioration of the structure.

In the present embodiment, "displacement" refers to live load displacement. Here, live load refers to a load whose magnitude is not constant and whose position of action changes. Factors that cause such live load displacement include the weight of a vehicle 1 passing over the bridge 30, the weight of the bridge 30 itself, and the inertial force acting on the bridge 30 due to an earthquake. In the following explanation, "displacement" and "live load displacement" are treated as synonyms.

The anomaly degree calculation unit 22 calculates the anomaly degree for each of the multiple feature quantities generated by the conversion process by the feature quantity conversion unit 21: tilt, natural frequency, and live load displacement. Below, the method for calculating the anomaly degree for each of tilt, natural frequency, and live load displacement will be explained in detail.

Figure 3 shows the time progression of the slope, the relationship between the slope and temperature, and the degree of slope abnormality when the slope is used as a feature in embodiment 1 of the present disclosure. Figure 3(A) shows the time progression of the slope, with the vertical axis representing the slope and the horizontal axis representing the date. It can be seen that the feature related to the slope changes depending on the temperature. Furthermore, Figure 3(A) shows that it is difficult to set a threshold for determining an abnormal state for the slope, which is a feature.

Figure 3(B) shows the relationship between slope and temperature, with the vertical axis representing slope and the horizontal axis representing temperature. Figure 3(B) shows the "predicted distribution," "normal data distribution," and "abnormal data distribution."

The "predictive distribution" is a distribution that can be created from the relationship between the slope and temperature when temperature is the input and the slope is the output, using time series data showing the transition state of the slope calculated during the "learning period" shown in Figure 3(A).

Specifically, the anomaly calculation unit 22 uses Gaussian process regression to model the relationship between the input temperature and the output slope from the time series data obtained during the learning period in the form of a response surface or regression line. Furthermore, the anomaly calculation unit 22 can create a "predicted distribution," for example, a 3σ range, from the modeled response surface or regression line.

In other words, the "predictive distribution" corresponds to a confidence interval determined by statistical processing using the covariance matrix of a Gaussian distribution estimated from the time series data obtained during the learning period.

On the other hand, the "normal data distribution" and "abnormal data distribution" are distributions plotted from time series data collected during the "assessment period" after the "learning period" shown in Figure 3(A), with the distribution of data that falls within the range of the predicted distribution corresponding to the normal data distribution, and the distribution of data that falls outside the range of the predicted distribution corresponding to the abnormal data distribution.

One of the important features of using Gaussian process regression is its nonlinearity, making it effective even when the input-output relationship cannot be fitted well using linear regression. Another important feature is that it uses Bayesian estimation, making it a nonparametric regression model.

Here, we will further explain, using mathematical formulas and Figure 3(C), a specific method for calculating the degree of anomaly, which serves as an index for determining whether or not an "abnormal data distribution" state like that shown in Figure 3(B) exists.

The procedure for calculating the degree of anomaly using Gaussian process regression when the temperature is the input and the slope is the output is as follows:
<Step 1> Consider N sets of training data, and then set the input training data as
Then, the output learning data is
Let the training data be
Let's say.

<Step 2> Let f(x) be the output of the response surface (or curve in the case of one dimension) at any coordinate x. Assuming that the model representing noise during observation follows a normal distribution with a mean value of 0 and a variance of σ2, the following equation (1) is obtained.

<Step 3> The distribution of the response surface value f(x) for input x is calculated from the data D.
If the above equation is obtained, the prediction distribution corresponding to the confidence interval is expressed as the following equation (2).

<Step 4> For a model related to the smoothness of the response surface, let f(x) and f(x') be the values of the response surface for arbitrary inputs x and x', respectively. At this time, f(x) and f(x') follow the probability distribution shown in the following equation (3).

Here, K is a kernel function, which intuitively expresses "how similar x and x' are." An RBF (radial basis function) kernel is often used as the kernel function.

Generally, N inputs shown in the following formula (4) are used.
If there is a prior distribution of fN, then fN follows the prior distribution of equation (5) below.

This is the basic assumption of a Gaussian process. However, if there are N inputs, K(x, x') becomes an NxN matrix, and if there are two inputs, it becomes a 2x2 matrix as shown in equation (3) above.

<Step 5> The degree of anomaly α(x') is the mean of the predictive distribution of GPR.
and dispersion
It can be calculated using the following formula (6).

In other words, the degree of anomaly corresponding to each feature can be calculated based on the predictive distribution (equivalent to the confidence interval) calculated during the learning period.

The degree of anomaly related to the slope calculated using the above procedure over the learning period and the evaluation period is the degree of slope anomaly shown in Figure 3(C). By setting an appropriate threshold and determining whether there is an anomaly that exceeds the threshold, it is possible to quantitatively determine whether the abnormal data distribution state shown in Figure 3(B) is occurring.

By performing this series of processes, the anomaly degree calculation unit 22 can convert the slope, which is a feature quantity, into an anomaly degree, which is an index value for determining with high accuracy whether an abnormal state exists. As a result, when using the slope as a feature quantity, it was difficult to set a threshold value for determining an abnormal state because the degree of temperature dependence of the slope differs for each sensor. However, for the slope anomaly degree calculated from the slope, a threshold value for determining an abnormal state can be easily set.

In other words, by having the function of making judgments using the degree of tilt abnormality, the sensor controller 20 is able to accurately identify abnormal conditions based on tilt while taking into account the effects of temperature fluctuations.

Next, we will explain the case where the natural frequency is used as a feature quantity. Figure 4 shows the time progression of the natural frequency, the relationship between the natural frequency and temperature, and the natural frequency anomaly degree when the natural frequency is used as a feature quantity in embodiment 1 of the present disclosure.

Figure 4(A) shows the time progression of the natural frequency, with the vertical axis representing the natural frequency and the horizontal axis representing the date. It can be seen that the characteristic quantities related to the natural frequency, like the slope, change depending on the temperature. Therefore, just like the slope, it is clear that it is difficult to set a threshold value for determining an abnormal state in the case of the natural frequency as well.

Therefore, in the same way as in the case of the slope described above in Figure 3, when the natural frequency is used as the feature, the relationship between the natural frequency and temperature shown in Figure 4(B) and the natural frequency anomaly degree shown in Figure 4(C) can be obtained.

Although a detailed explanation will be omitted, by performing a series of processes similar to those for the slope, the anomaly degree calculation unit 22 can convert the natural frequency, which is a feature quantity, into an anomaly degree, which is an index value for determining with high accuracy whether an abnormal state exists. As a result, when using the natural frequency as a feature quantity, it was difficult to set a threshold value for determining an abnormal state for the natural frequency. However, for the natural frequency anomaly degree calculated from the natural frequency, it is now easy to set a threshold value for determining an abnormal state.

In other words, by having the function of making judgments using the natural frequency abnormality degree, the sensor controller 20 is able to accurately identify abnormal conditions based on the natural frequency while taking into account the effects of temperature fluctuations.

Next, we will explain the case where live load displacement is used as a feature quantity. Figure 5 is a diagram showing the time progression of live load displacement, the relationship between live load displacement and temperature, and the live load displacement anomaly degree when live load displacement is used as a feature quantity in embodiment 1 of the present disclosure.

Figure 5 (A) shows the time progression of live load displacement, with the vertical axis representing live load displacement and the horizontal axis representing date. Unlike the inclination and natural frequency, the characteristic quantities related to live load displacement do not change with temperature. The reason that live load displacement does not change with temperature is that the most frequent value for maximum displacement is more influenced by factors such as the amount of travel, speed, and weight of the vehicle, while the influence of temperature is relatively small.

As shown in Figure 5(B), even when using live load displacement, a feature that is independent of temperature, the "prediction distribution" can be calculated as a distribution equivalent to the confidence interval using the covariance matrix of the Gaussian distribution estimated from the time series data obtained during the learning period.

As an example, when using live load displacement, which is a feature independent of temperature, the abnormality calculation unit 22 may use, for example, Hotelling's T2 method to calculate the abnormality α(x') using the following equation (7):

In the above equation (7), x' represents the data, μ with a ^ represents the mean value, and σ with a ^ represents the variance value.

When live load displacement is used as a feature, a threshold can be set to determine whether the live load displacement is in an abnormal state, as shown in Figure 5(A).

Furthermore, as shown in Figure 5(C), just as when the slope and natural frequency that change depending on temperature are used as feature quantities, when live load displacement is used as the feature quantity, abnormal conditions can be easily identified using a threshold value by calculating the live load displacement abnormality degree using the above equation (7) or the like.

The thresholds for the tilt anomaly, natural frequency anomaly, and live load displacement anomaly can be set in advance, or they can be set statistically based on the results of time-series data acquired during the learning period. For example, when the standard deviation is σ, the threshold can be 3σ or the value of the following formula (8):

In this way, by comparing the degree of abnormality calculated based on the time-series data of the feature with a threshold value for the degree of abnormality, it is possible to easily and quantitatively identify whether the feature indicates an abnormal state, regardless of whether the feature changes depending on the temperature.

Next, we will explain the method for distinguishing between a structural abnormality and a sensor failure, which is a technical feature of the structural deterioration diagnosis system according to the first embodiment. As described above, by using an index called the degree of abnormality, it is possible to easily distinguish whether a feature is in an abnormal state. However, such an abnormal state can occur both when a structural abnormality occurs and when a sensor fails.

The structural deterioration diagnosis system according to the first embodiment uses the results of abnormality assessments, which use the degree of tilt abnormality, the degree of natural frequency abnormality, and the degree of live load displacement abnormality calculated based on the three feature quantities of tilt, natural frequency, and live load displacement as deterioration diagnosis indices, to identify whether the problem is a structural abnormality or a sensor failure. This identification process will be explained in detail below.

In the structure deterioration diagnosis system according to the first embodiment, a series of processes from [Process 1] to [Process 7] described below are performed for the three abnormality levels to identify whether the abnormality is in the structure or a sensor failure.
[Process 1] Based on the acceleration information output from the sensor 10, the feature quantity conversion unit 21 calculates three feature quantities, namely, the tilt, natural frequency, and live load displacement of the structure.

[Process 2] The anomaly degree calculation unit 22 calculates the anomaly degree for each of the three feature quantities, i.e., inclination, natural frequency, and live load displacement, as anomaly degrees that serve as indicators for determining deterioration of the structure: inclination anomaly degree, natural frequency anomaly degree, and live load displacement anomaly degree.

[Process 3] The abnormality/fault determination unit 23 determines that an abnormal tilt has occurred if the degree of abnormal tilt exceeds the first threshold value.
[Process 4] The abnormality/failure determination unit 23 determines that an abnormality in the natural frequency has occurred if the degree of abnormality in the natural frequency exceeds the second threshold value.
[Process 5] The abnormality/fault determination unit 23 determines that a live load displacement abnormality has occurred if the live load displacement exceeds the third threshold value.

[Process 6] If the abnormality/failure determination unit 23 determines that an inclination abnormality has occurred and that neither a natural frequency abnormality nor a live load displacement abnormality has occurred, it determines that a sensor failure has occurred.
[Process 7] The abnormality/failure judgment unit 23 determines that deterioration of the structure has occurred if it determines that at least one of a natural frequency abnormality and a live load displacement abnormality has occurred, regardless of whether an inclination abnormality has occurred or not.

The reason why process 6 and process 7 make it possible to distinguish between a structural abnormality and a sensor failure will be explained using Figure 6. Figure 6 is an explanatory diagram of the case where tilt, which is one of the feature quantities, is calculated using triaxial acceleration information output from sensor 10 in embodiment 1 of the present disclosure.

ax, ay, and az are the acceleration values for each axis. However, to calculate the static tilt, excluding the effects of vibration, low-pass filtering is performed using a digital filter for each axis.

The tilt is therefore calculated from the direct current (DC) component of each axis, excluding the vibration component. This DC component can change due to deterioration of electronic components such as the power supply inside the sensor 10. On the other hand, the natural frequency and live load displacement are calculated from the vibration rather than the DC component, and are therefore less susceptible to changes in the DC component due to sensor failure, etc.

In principle, if there is an abnormality in the tilt, it is likely that there is a decrease in rigidity, such as damage to the girders or supports, and that a decrease in rigidity, like the tilt, also affects the live load displacement and natural frequency. Therefore, if the results show that only the tilt is abnormal, it is likely that there is deterioration or damage to the electronic components inside the sensor, in other words, a failure of the sensor 10.

On the other hand, it is unlikely that a sensor failure will result in an abnormality only in the natural frequency. The reason for this is that when the natural frequency is used as an indicator for determining the fault state, it deals with vibrations, as mentioned above, and is therefore less susceptible to the influence of DC components.

In addition, the natural frequency is greatly influenced by the moisture content of the entire bridge, the temperature of the entire bridge, the condition of the bearings, and other factors, and is therefore subject to considerable variation. Therefore, if an abnormality in the natural frequency is observed, it is assumed that a decrease in rigidity is the cause.

Similarly, it is unlikely that a sensor failure will result in an abnormality in only the live load displacement. The reason for this is that when live load displacement is used as an indicator for determining the fault state, the live load displacement is also calculated from the vibration component of the acceleration value (for example, around 1 Hz, which does not include a DC component), as mentioned above.

As a result, live load displacement is less susceptible to changes in the DC component due to deterioration of internal electronic components, such as deterioration around the power supply. Therefore, if abnormalities in live load displacement occur, it is assumed that the cause is a decrease in the rigidity of the structure.

Figure 7 is an explanatory diagram showing a case in which only the tilt anomaly degree is in an abnormal state, while the natural frequency anomaly degree and the live load displacement anomaly degree are not in an abnormal state, in embodiment 1 of the present disclosure. This case corresponds to process 6 described above, and the anomaly/fault determination unit 23 can determine that a fault has occurred in the sensor 10 because a tilt anomaly has occurred and neither the natural frequency anomaly nor the live load displacement anomaly has occurred.

When determining whether an inclination abnormality has occurred but neither a natural frequency abnormality nor a live load displacement abnormality has occurred, the abnormality/fault determination unit 23 can use a statistical determination method using a histogram or the like in addition to a determination method based on whether each abnormality level exceeds a threshold value. Therefore, a statistical determination method using a histogram will be explained using Figures 8 and 9.

Figure 8 compares (A) a histogram of each feature when the anomaly levels for all feature levels are determined to be normal, and (B) a histogram of each feature when only the slope anomaly level is determined to be abnormal, in embodiment 1 of the present disclosure. Comparing Figure 8(A) with Figure 8(B), the frequency distribution of the slope anomaly level is clearly different. On the other hand, there is not much difference between Figure 8(A) and Figure 8(B) in the distribution of the natural frequency anomaly level and the live load displacement anomaly level.

In this way, if only the anomaly degree of the slope has increased, it is possible to determine whether only the anomaly degree calculated from the slope has increased by using a statistical method to compare the anomaly degree of the slope with the anomaly degree calculated from other physical quantities. Specific statistical methods that can be used include the T-value from Welch's t-test or the density ratio.

9 is an explanatory diagram illustrating a case where a histogram of the degree of anomaly in the slope and a histogram of the natural frequency are compared using a statistical method in the first embodiment of the present disclosure. (A) to (D) in FIG. 9 show the following diagrams.
FIG. 9A: Histogram of each feature amount when the degree of abnormality for the two feature amounts of tilt and natural frequency is determined to be normal.
FIG. 9B: Histogram of each feature when, of the anomaly degrees for the two feature amounts of inclination and natural frequency, only the anomaly degree for inclination is determined to be in an abnormal state, and the anomaly degree for natural frequency is determined to be normal.
FIG. 9(C): Time series data of T-values from Welch's t-test calculated based on the histograms of FIG. 9(A) and FIG. 9(B).
FIG. 9(D): Time series data of density ratio calculated based on the histograms of FIG. 9(A) and FIG. 9(B).

The T-value of the Welch t-test can be calculated using the average value, variance, and number of samples for the slope anomaly and natural frequency anomaly, respectively, according to the following formula (9):

The density ratio can be calculated using the probability density of the tilt anomaly and the probability density of the natural frequency anomaly according to the following equation (10).

As is clear from Figures 9(C) and 9(D), the T value and density ratio are both larger in the transition state during the judgment period than in the transition state during the learning period. Therefore, the abnormality/fault judgment unit 23 can determine that only the degree of slope abnormality has become abnormal based on the results of comparing the transition state during the judgment period with a threshold value determined from the transition state during the learning period or a preset threshold value.

Note that in the example shown in Figure 9 above, a comparison was made using inclination and natural frequency as feature quantities, but similar statistical methods can also be applied when comparing using inclination and live load displacement as feature quantities, and detailed explanations will be omitted.

Next, a series of processes for identifying whether a structural abnormality or a sensor failure is present based on the detection results of a single sensor using the structural deterioration diagnosis system according to the first embodiment of the present disclosure will be described using a flowchart. Figure 10 is a flowchart showing a series of processes for identifying whether a structural abnormality or a sensor failure is present, performed by the structural deterioration diagnosis system according to the first embodiment of the present disclosure.

First, in step S1001, the feature conversion unit 21 acquires acceleration information output from the sensor 10 and sequentially acquires temperature information from the outside, which corresponds to the air temperature when the acceleration information was generated.

Next, in step S1002, the feature conversion unit 21 converts the acceleration information into physical quantities (tilt, natural frequency, and live load displacement) corresponding to the three feature quantities.

Next, in step S1003, the anomaly calculation unit 22 calculates the tilt anomaly degree, natural frequency anomaly degree, and live load displacement anomaly degree based on each converted physical quantity. Note that when calculating the tilt anomaly degree and natural frequency anomaly degree, the anomaly calculation unit 22 takes temperature information into account, thereby suppressing the influence of temperature-dependent changes in the judgment process using the tilt anomaly degree and natural frequency anomaly degree, and achieving stable judgment process.

Next, in step S1004, the abnormality/failure determination unit 23 determines whether the tilt abnormality degree has become abnormal. If the abnormality/failure determination unit 23 determines that the tilt abnormality degree has become abnormal, it proceeds to processing in step S1007, and if it determines that the tilt abnormality degree has not become abnormal, it proceeds to processing in step S1005.

If the process proceeds to step S1005, the abnormality/failure judgment unit 23 determines whether at least one of the live load displacement abnormality degree or the natural frequency abnormality degree has become abnormal. If the abnormality/failure judgment unit 23 determines that at least one of the abnormality degrees has become abnormal, the process proceeds to step S1006, where it determines that a structural abnormality has occurred.

Furthermore, if the abnormality/failure judgment unit 23 determines in step S1005 that both the live load displacement abnormality degree and the natural frequency abnormality degree are in a normal state, all abnormality degrees are in a normal state, so the process returns to step S1001 and repeats the processes from step S1001 onwards.

Furthermore, if the process proceeds to step S1007, the abnormality/failure judgment unit 23 determines whether both the live load displacement abnormality degree and the natural frequency abnormality degree are in a normal state. If the abnormality/failure judgment unit 23 determines that at least one of the abnormality degrees has become abnormal, the process proceeds to step S1006, where it determines that a structural abnormality has occurred.

On the other hand, if the abnormality/failure determination unit 23 determines in step S1007 that both the live load displacement abnormality degree and the natural frequency abnormality degree are normal, it proceeds to processing in step S1008 and determines that a sensor failure has occurred because only the tilt abnormality degree is in an abnormal state.

In addition, if the system is configured with multiple sensors 10 and determines that only the degree of tilt abnormality is abnormal based on the acceleration information from two or more sensors 10, the abnormality/failure determination unit 23 can also determine that a minor structural abnormality has occurred.

As described above, according to embodiment 1, multiple feature values are calculated from acceleration information detected by a single sensor, and the tilt anomaly degree, natural frequency anomaly degree, and live load displacement anomaly degree are determined corresponding to each feature value. The system is then able to identify whether there is a structural abnormality or a sensor failure based on the combination of these anomaly degrees. In this way, by performing an identification process using multiple anomaly degrees calculated based on acceleration information from a single sensor, it is possible to identify with high accuracy whether there is a structural abnormality or a sensor failure from the detection results of a single sensor.

In the above-described first embodiment, the case where three feature quantities - tilt, natural frequency, and live load displacement - are used is described, but this is not limited to this. By using two feature quantities, with tilt as the first feature quantity and a feature quantity other than tilt as the second feature quantity, it is possible to obtain a structural deterioration diagnosis system that can distinguish between a structural abnormality and a sensor failure from the detection results of a single sensor.

The natural frequency or live load displacement described in embodiment 1 can be used as the second feature, and any other feature can also be used as the second feature as long as it serves as an index value for diagnosing deterioration of the structure.

In this case, if only an inclination abnormality occurs, it can be determined that a sensor failure has occurred, and if it is determined that a natural frequency abnormality or live load displacement abnormality has occurred based on the second feature value, regardless of whether an inclination abnormality has occurred, it can be determined that deterioration of the structure has occurred.

10 Sensor, 20 Sensor controller, 21 Feature conversion unit, 22 Anomaly degree calculation unit, 23 Anomaly/fault determination unit.

Claims (3)

a sensor that is installed in a structure that is a deterioration diagnosis target and outputs acceleration information of the structure;
a sensor controller that determines whether or not deterioration of the structure has occurred based on the acceleration information output from the sensor,
The sensor controller
Calculating a live load displacement based on vibration components not including a DC component in the acceleration information;
calculating time series data of the most frequent value at the maximum displacement from the time transition of the live load displacement as a feature quantity that does not change depending on temperature ;
calculating an abnormality degree serving as an index for determining deterioration of the structure based on the time-series data of the mode ;
A structure deterioration diagnosis system that determines that deterioration of the structure has occurred when the degree of abnormality exceeds a threshold value. The sensor controller determines that deterioration of the structure has occurred based on vibration components that do not include a DC component in the acceleration information, and can therefore infer that deterioration of the structure has occurred due to a decrease in the rigidity of the structure in a state that is less susceptible to the influence of changes in the DC component.
The structure deterioration diagnosis system according to claim 1 . The sensor controller
creating a predictive distribution by performing statistical processing on the time series data of the mode during a predetermined learning period;
The degree of anomaly, which serves as an index for determining deterioration of the structure, is calculated based on the predictive distribution as an index for determining that a distribution of data outside the range of the predictive distribution is an abnormal data distribution among the time-series data of the mode collected during a determination period after the learning period.
The structure deterioration diagnosis system according to claim 1 or 2.