CN109813269A - On-line calibration data sequence matching method for structural monitoring sensors - Google Patents

Abstract

The present invention relates to pick up calibration fields, disclose a kind of bridge deformation monitoring sensor on-line calibration data sequence matching process, which comprises step 1, obtain the measurement value sequence of the sensor and reference sensor to be calibrated under same incentive action；Step 2, the measurement value sequence of the sensor to be calibrated and the reference sensor is segmented；Step 3, the measured value of the sensor to be calibrated and the reference sensor is matched in each piecewise interval；Step 4, the corresponding relationship of the sensor to be calibrated and the reference sensor between the measurement value sequence under the same incentive action is obtained.The out of step conditions such as the advanced, lag that the measurement value sequence of sensor to be calibrated and reference sensor occurs in technical solution provided by the invention have good adaptability, are remarkably improved the efficiency and accuracy of the on-line calibration of bridge deformation monitoring sensor.

Description

On-line calibration data sequence matching method for structural monitoring sensors

technical field

The invention relates to the field of sensor calibration, in particular to an online calibration data sequence matching method for bridge deformation monitoring sensors.

Background technique

Bridge deformation monitoring is a key link in bridge safety performance evaluation. In order to ensure bridge safety, sensors used for bridge deformation monitoring need to have high monitoring accuracy. In order to ensure the accuracy of the bridge deformation monitoring sensor, the bridge deformation monitoring sensor needs to be calibrated.

Taking the laser displacement meter in the bridge deformation monitoring sensor as an example, the laser displacement meter is designed according to the linear propagation characteristics of the laser. The relative deformation variable has strong practicability and high accuracy, so it is widely used in the actual bridge monitoring process. However, in the process of use, due to the influence of the sensor's own performance and complex environmental conditions, there will be a decrease in sensitivity and accuracy, which must be calibrated regularly.

The traditional calibration method uses higher-precision professional measuring equipment (such as laser interferometer, etc.) as a reference instrument. Within the allowable error range, if the indication value of the laser displacement meter to be calibrated is consistent with the reference value, it is considered that the laser displacement The metering performance of the meter meets the requirements for use. Due to the special requirements of the bridge deformation monitoring system running all day, the laser displacement meters in use are usually not allowed to be dismantled into the laboratory for calibration, so the above methods are often not universal.

In engineering practice, the combined loading effect of driving vehicles is generally used as the excitation source to perform online calibration of bridge health monitoring sensors. The comparison and calibration of the sensor (hereinafter referred to as RS) under the same excitation source condition. Due to the differences between SBC and RS in sampling frequency, response characteristics, etc., the matching between the two value sequences is poor, which is an important problem affecting on-line calibration.

The time series matching problem has a wide range of applications in positioning systems, environmental monitoring, Internet of Things, data mining, informatics and human psychology, and has gradually become a research hotspot in various fields in recent years. Among them, dynamic time warping (hereinafter referred to as DTW) is an important method for analyzing time series. It provides a distance measure that is insensitive to local compression and extension for the sequence, and is widely used in the field of data analysis and mining. However, the DTW method has the following problems: (1) The computational complexity is high. For time series with lengths of n and m respectively, the time complexity of O(nm) is required to accurately calculate the DTW distance; (2) The triangular inequality of distance is not satisfied. The degree of pruning and filtering is limited when querying similar time series, which may cause missed queries when using index query.

Aiming at the problem of unstable difference between the sensor value sequence and the reference value sequence in the online calibration of bridge sensors, the matching effect under the traditional Euclidean distance metric is poor, and the dynamic time warping using DTW takes a long time. If the distance lower bound function is introduced to To speed up the comparison process based on DTW, new requirements are put forward for the determination of the lower bound function of the distance.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a method for matching the online calibration data sequence of bridge deformation monitoring sensors with strong adaptability and high accuracy in order to overcome the above-mentioned defects existing in the prior art at least to a certain extent.

In order to achieve the above object, the technical scheme provided by the present invention is:

An online calibration data sequence matching method of a bridge deformation monitoring sensor, the online calibration data sequence matching method comprising:

Step 1, obtain the measurement value sequence of the sensor to be calibrated and the reference sensor under the same excitation;

Step 2, segment the measurement value sequence of the sensor to be calibrated and the reference sensor;

Step 3, matching the measured values of the sensor to be calibrated and the reference sensor in each segment interval;

Step 4: Obtain the correspondence between the measured value sequences of the sensor to be calibrated and the reference sensor under the same excitation action.

Preferably, the measurement value sequence of the sensor to be calibrated and the measurement value sequence of the reference sensor are both waveform sequences; the step 2 includes:

Step B1, extracting the target peak point in the measurement value sequence of the sensor to be calibrated and the target peak point in the measurement value sequence of the reference sensor;

Step B2: Segment the measurement value sequences of the sensor to be calibrated and the reference sensor by taking the target peak points in the measurement value sequences of the sensor to be calibrated and the reference sensor as nodes respectively.

Preferably, the step B1 includes:

Step S1, extracting the initial peak point in the measurement value sequence of the sensor to be calibrated and the initial peak point in the measurement value sequence of the reference sensor and forming the initial peak point sequence respectively;

In step S2, the initial peak point sequence of the sensor to be calibrated and the initial peak point sequence of the reference sensor are smoothed by the following formulas (1)-(3):

P={p ₁ ,p ₂ …p _n } (1)

j=(span-1)/2 (3)

Among them, P is the initial peak point sequence of the sensor to be calibrated or the reference sensor; n is the number of peak points in the initial peak point sequence; span is the smoothing scale; P _i ' is the ith in the smoothed initial peak point sequence P' a peak point;

Step S3, save P _i ' in turn to form a smoothed initial peak point sequence P';

Step S4, searching for the initial peak point P _i in the original initial peak point sequence P corresponding to each peak point in the smoothed initial peak point sequence P', and using this initial peak point P _i as the target peak point .

Preferably, both the measurement value sequence of the sensor to be calibrated and the measurement value sequence of the reference sensor are divided into K+1 segments; wherein the measurement value sequence of the sensor to be calibrated and the measurement value sequence of the reference sensor are divided into K+1 segments; The segment sequences formed after segmentation are {t ₀ ,…,t ₁ },{t ₁ ,…,t ₂ }…{t _k-1 ,…,t _k },{t _k ,…,t _{k +1} } and {t' ₀ ,…,t' ₁ },{t' ₁ ,…,t' ₂ }…{t' _k-1 ,…,t' _k },{t' _k ,…,t ' _k+1 }; where,

t ₀ is the starting measurement point of the measurement value sequence of the sensor to be calibrated;

t _k+1 is the termination measurement point of the measurement value sequence of the sensor to be calibrated;

t _n is the target peak point in the measurement value sequence of the sensor to be calibrated, wherein k≥n≥1;

t' ₀ is the starting measurement point of the measurement value sequence of the reference sensor;

t' _k+1 is the termination measurement point of the measurement value sequence of the reference sensor;

t' _n is the target peak point in the measurement value sequence of the reference sensor, where k≧n≧1.

Preferably, the step 3 includes: sequentially selecting the measurement values p _i to be matched in the measurement value sequence of the sensor to be calibrated, and locating the measurement value p to be matched according to the position information of the measurement values p _i to be matched The segment interval where _i is located {t _r-1 ,...,t _r };

Find the position of the measurement value q _j that matches the measurement value p _i to be matched in the measurement value sequence of the reference sensor by the following formula (4),

Wherein, j is the abscissa of the measurement value q _j that matches the measurement value p _i to be matched in the measurement value sequence of the reference sensor;

tr _-1 | _x is the abscissa of the measurement point tr _-1 in the segmented sequence of the sensor to be calibrated;

_tr | _x is the abscissa of the measurement point _tr in the segmented sequence of the sensor to be calibrated;

t' _r-1 | _x is the abscissa of the measurement point t' _r-1 in the segmented sequence of the reference sensor;

t' _r | _x is the abscissa of the measurement point t' _r in the segmented sequence of the reference sensor.

Preferably, the step 3 further includes: outputting and saving pi and _{q j matching} the _pi .

Preferably, before step 1, the method further includes: building a calibration platform for the bridge deformation monitoring sensor to simulate the random change of the load on the bridge when the actual vehicle passes through the bridge.

Preferably, the calibration platform includes a simply supported girder model bridge for simulating a bridge, an excitation source capable of acting on the simply supported girder model bridge to simulate actual vehicles passing on the bridge, and an excitation source installed on the simply supported girder model A sensor to be calibrated and a reference sensor for monitoring the deformation of the simply supported beam model bridge at the bottom of the bridge.

Preferably, the excitation source is a linear module.

Preferably, both the sensor to be calibrated and the reference sensor are laser displacement meters.

Compared with the prior art, the technical solution provided by the present invention has the following beneficial effects:

In the method for matching the online calibration data sequence of the bridge deformation monitoring sensor provided by the present invention, the measurement value sequences of the sensor to be calibrated and the reference sensor are firstly segmented, and then the sensor to be calibrated and the measured value sequence of the reference sensor are segmented in each segment interval. This method is not sensitive to the detection frequency of the sensor to be calibrated and the reference sensor, and has good adaptability to the situation that the measured value sequences of the sensor to be calibrated and the reference sensor often lead and lag each other.

Compared with the method in the prior art, the on-line calibration data sequence matching method provided by the present invention has obvious reduction in relative error and improved accuracy, and the matching accuracy can be maintained above 98% in different scenarios .

Other features and advantages of the present invention will be described in detail in the detailed description that follows.

Description of drawings

The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification, and together with the following specific embodiments, are used to explain the present invention, but do not constitute a limitation to the present invention. In the attached image:

FIG. 1 is a sequence diagram of original measurement values of SBC and RS provided by an embodiment of the present invention;

Fig. 2 is the figure after the original measurement value sequence of SBC and RS in Fig. 1 is corrected by the algorithm of traditional matching;

3 is a flowchart of a method for matching an online calibration data sequence of a bridge deformation monitoring sensor according to an embodiment of the present invention;

4 is a node diagram before preprocessing provided by an embodiment of the present invention;

FIG. 5 is a preprocessed node diagram provided by an embodiment of the present invention;

6 provides a node diagram of the extracted SBC and RS sequences according to an embodiment of the present invention;

Fig. 7 is the SBC and RS sequence node diagrams extracted after the occurrence of a larger offset provided by an embodiment of the present invention;

Fig. 8 is the realization flow chart of the matching algorithm provided by the embodiment of the present invention;

FIG. 9 is a partial interval sampling point diagram after segmentation provided by an embodiment of the present invention;

10 is a node-based segmentation matching diagram provided by an embodiment of the present invention;

11 is a reverse matching diagram of an SBC and an RS provided by an embodiment of the present invention;

12 is a DFLI and LDM sampling sequence and a node location diagram provided by an embodiment of the present invention;

13 is a local interval sampling point diagram after DFLI and LDM sampling sequences are segmented according to an embodiment of the present invention;

FIG. 14 is a segmentation matching diagram of DFLI and LDM sampling sequences provided by an embodiment of the present invention;

15 is a reverse matching diagram of DFLI and LDM sampling sequences provided by an embodiment of the present invention;

16 is a front view of a calibration platform provided by an embodiment of the present invention;

17 is a side view of a calibration platform provided by an embodiment of the present invention;

FIG. 18 is a matching result diagram of a reference sensor and a sensor to be calibrated under low-frequency excitation provided by an embodiment of the present invention;

19 is a matching result diagram of a reference sensor and a sensor to be calibrated under intermediate frequency excitation provided by an embodiment of the present invention;

FIG. 20 is a matching result diagram of a reference sensor and a sensor to be calibrated under high-frequency excitation provided by an embodiment of the present invention;

FIG. 21 is a node-based segmentation matching result diagram of a reference sensor and a sensor to be calibrated provided by an embodiment of the present invention;

FIG. 22 is a matching result diagram based on Euclidean metric of a reference sensor and a sensor to be calibrated provided by an embodiment of the present invention;

FIG. 23 is a relative error diagram under different matching modes provided by an embodiment of the present invention.

Among them, 5- excitation source; 6- sensor to be calibrated and reference sensor; 7- sensor mounting plate; 8- simply supported beam model bridge; 9- laser displacement meter receiving end; 10- laser displacement meter transmitting end.

Detailed ways

The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present invention, but not to limit the present invention.

In the present invention, unless otherwise stated, the use of directional words such as "up, down, left, right" generally refers to up, down, left, right, "inside, outside" as shown in the accompanying drawings Refers to the inside and outside of the contour relative to the part body.

As shown in Figure 1 and Figure 2, the traditional matching algorithm often uses the Euclidean metric to measure the degree of acquaintance between curves. The principle is intuitive and the calculation is simple. When the sequence undergoes common transformations (such as Fourier transform), its coefficients can keep the Euclidean The distance is invariant, so it is widely used in the time series similarity problem. However, the matching based on the Euclidean distance metric requires that the two sequences to be matched have the same length, which is more sensitive to the mutation point of the time series. In view of the problem in the embodiment of the present invention, due to the random excitation from the excitation source, the data collected by the SBC and the RS will inevitably undergo sudden changes. At the same time, due to the difference in sampling frequency, the SBC and RS data sequences will also exist. Dislocation, according to Figure 1, it can be found that the data collected by SBC and RS are not completely synchronized. With the increase of sampling points, the deviation of data will gradually increase, which is caused by the slight sampling frequency difference (SFD) between SBC and RS. Yes, in the same time, the number of SBC sampling points is more than the number of RS sampling points. At the same time, under the same excitation conditions, the response peaks of the two also have obvious differences, which are personality differences accompanying the sensor manufacturing process, and are usually kept within a certain threshold range.

The traditional matching method is generally to perform preprocessing such as sampling point calibration and inertia compensation on the SBC and RS value sequences, and then perform data matching on this basis. The principle is as follows:

Data collection is performed on SBC and RS under the action of the same excitation source per unit time, assuming that the SBC sampling sequence is:

P={p ₁ ,p ₂ …p _n } (1)

The RS sampling sequence is:

Q={q ₁ ,q ₂ ...q _m } (2)

Among them, n is the number of SBC sampling points, m is the number of RS sampling points, and the sampling number of SBC is more than the sampling number of RS, that is, n>m.

The sampling points that are synchronized with the RS are retained by the method of comparison sampling, and the redundant sampling points in the SBC are removed, which is obtained by formula (1) and (2), and the sequence of points to be removed is:

P'={p _[n·i/d] }(i=1,2,...d) (3)

Among them, d=n-m, in unit time, its value is the sampling frequency difference SFD, if the sampling sequence set of SBC before calibration is taken as the complete set U, that is:

U=P (4)

Then the calibrated SBC sampling sequence can be expressed as the absolute complement of P' on U, namely:

For SBC inertia compensation value (ICV) is generally determined by the following formula:

Among them, k is the number of sampling segments, and nj is the number of sampling points in the jth segment. The calibrated SBC and RS data sequences are shown in Figure 2.

It can be found that the overall matching effect of the SBC and RS value sequences after sampling correction and inertia compensation is improved compared with that before processing, but when the data is matched point by point, due to the discrete type of the sampled data, as long as there is one that cannot be eliminated. The sampling error will cause the matching sequence to be misaligned, and the Euclidean distance will increase significantly. At the same time, due to the interference of uncertain factors, if a sensor has an instantaneous sampling delay, the above calibration will also cause matching misalignment. It can be seen that the Euclidean distance is simply used as a metric Standards are very limited.

In view of the above problems, the embodiment of the present invention provides a new online calibration data sequence matching method for bridge deformation monitoring sensors. Matching is carried out in sub-intervals. The specific process is shown in Figure 3, including:

Step 1, obtain the measurement value sequence of the sensor to be calibrated and the reference sensor under the same excitation;

Step 2, segment the measurement value sequence of the sensor to be calibrated and the reference sensor;

Step 3, matching the measured values of the sensor to be calibrated and the reference sensor in each segment interval;

Step 4: Obtain the correspondence between the measured value sequences of the sensor to be calibrated and the reference sensor under the same excitation action.

As shown in FIG. 1 and FIG. 2 , the measurement value sequence of the sensor to be calibrated and the measurement value sequence of the reference sensor are both waveform sequences, generally irregular waveforms;

In step 2, in order to segment the measurement value sequence of the sensor to be calibrated and the reference sensor, it is first necessary to locate the segment node. The inventor of the present application found that the difference between the two sets of data peaks is large, which is the result of different sensors for the same excitation It is caused by the response of different intensities. At the moment of the response, the sampled data of the two sensors match. If it is the node, the data here will be matched first, and then the data between the two sets of peak points will be processed more conveniently. At the same time, there is no Similar to DTW one-to-many matching and other issues.

Due to the retention of the excitation source signal, the traditional extreme point extraction algorithm often obtains two or more peak points with equal values. At the same time, during the signal retention process, the sensor will fluctuate in a small range due to noise interference, as shown in Figure 4. As shown, the sampling sequence needs to be processed accordingly to find a relatively stable peak point as the target peak point, as follows:

Taking the SBC sampling sequence as an example, formula (1):

P={p ₁ ,p ₂ …p _n }

Smooth it out:

Where: j=(span-1)/2

In the formula, span is the smoothing scale, and p _i ' is the value point after smoothing; under the condition of no repetition, the extreme points in the sequence P' are stored in sequence, and the original sequence sampling data p _i corresponding to this point is found. , with _pi as the real node, as shown in Figure 5, the pseudo-code of the implementation process is as follows:

if p' _i =max(p' _it ～p' _i+t )or

min(p' _it ～p' _i+t )

i++

while w _j -1≠p _i

j++

The above operations are performed on the SBC and RS sequences in turn, and the node extraction results obtained are shown in Figure 6 and Figure 7 . In order to adapt to the difference of sampling frequency of different sensors, the smooth scale in the algorithm is associated with the actual frequency during extraction, which makes the algorithm more general.

The sequence is segmented by using nodes, and the sequence segment between two consecutive nodes is used as the segment interval. For SBC, let the total number of collected nodes be k, and the node sequence is:

{t ₀ ,t ₁ ,t ₂ …t _k ,t _k+1 } (8)

Among them, in order not to lose the data integrity, the sequence start point p1 is set as the sequence initial node t0, and the sequence end point pn is set as the last sequence node tk+1. The initial sequence P is divided into k+1 segments:

{t ₀ ,…,t ₁ },{t ₁ ,…,t ₂ }…{t _k-1 ,…,t _k },{t _k ,…,t _k+1 }

For the RS sequence, select the first k nodes corresponding to the SBC to form the node sequence:

{t' ₀ ,t' ₁ ,t' ₂ …t' _k ,t' _k+1 } (9)

And divide the RS sequence into k+1 segments:

{t' ₀ ,…,t' ₁ },{t' ₁ ,…,t' ₂ }…{t' _k-1 ,…,t' _k },{t' _k ,…,t' _k+1 } Let the input be any point pi on the SBC sampling sequence, the target output is the point q _j on the RS that matches pi, and _j indicates that it is the _jth sampling point of the RS sampling sequence.

According to the position information of _pi , the segment interval to which the point pi to be _matched belongs is located against the node sequence, and it is assumed that it is located in {t _r-1 ,...,t _r }, that is, it is located in the rth segment of the original sampling sequence, then j It is determined by the following formula:

Among them, t _r-1 | _x , t _r | _x , t' _r-1 | _x , t' _r | _x are SBC sequence nodes t _r-1 , t _r and RS sequence nodes t' _r-1 , t respectively The abscissa of ' _r , the algorithm flow is shown in Figure 8. It should be noted that the abscissa of the sequence node refers to the position coordinate of the sequence node in the entire measurement value sequence, and the position coordinate is generally the ordinal number of the node.

As shown in Figure 9, it is the distribution diagram of SBC and RC measurement points in a segmented interval, in which the node information has been marked in the figure. j value, use the j value index to find the target point corresponding to the point to be matched, as shown in Figure 10.

Compared with Fig. 9, because the sampling frequency of RS is lower than that of SBC, the sampling data of RS is ahead of SBC. However, since the node extraction is performed on the two sequences before matching, the lead-lag relationship between the magnitude sequences will no longer affect the matching of sequences. It is enough to perform point-by-point matching of the points to be matched starting from one end node.

Examining Figure 10, it can be found that when the SBC sequence value is matched, different sampling points may correspond to the same sampling point in the RS sequence. The 14th reference point of the sequence matches, and the 15th to-be-matched point of the SBC sequence also matches the 14th reference point of the RS sequence. This is due to the difference in the number of sampling points in the corresponding subsections of any SBC and RS, and the fundamental reason is that the sampling frequency between SBC and RS is inconsistent. Considering whether there is a one-to-many match in this method, which leads to the missing index and other problems, the original SBC and RS sampling sequences are reversely matched, and the RS sampling sequence is used as the sequence to be matched, and the SBC sequence is used as the reference sequence. Matching is performed, as shown in Figure 11. From the results, in this segment interval, the 13th point to be matched starting from the left node of the RS sequence matches the 13th reference point of the SBC sequence, and the thirteenth reference point of the RS sequence matches. The 14 points to be matched are matched with the 15th reference point of the SBC sequence, and the 14th reference point of the SBC sequence has no corresponding point to be matched. For the RS sequence to be matched, each point in the node interval can find the corresponding matching point in the reference sequence. Therefore, the many-to-one matching of the sequence to be matched will not cause problems such as missed checking and repeated matching.

In fact, due to the node location and extraction of the two sets of sampling sequences, the above method can adapt to the matching of sampling sequences of different frequencies. In the embodiment of the present invention, the measurement sequence collected by the dual-frequency laser interferometer (DFLI) is used for the comparison test with the above-mentioned laser displacement meter (LDM). It can be intuitively found in DFLI that the sampling data of DFLI is ahead of LDM by nearly half the excitation period. With the continuous accumulation of sampling data, the Euclidean distance between the node of DFLI and the previous node of LDM will be minimized, and the traditional method will inevitably cause matching misalignment. In this embodiment of the present invention, the method of node segmentation is used to first extract the target peak points of the two sets of sequences, and then segment the two sets of sequences according to the target peak points, and select a segment ( FIG. 13 ) to perform intra-area matching. It can be seen that the node-based segmentation matching method has good applicability to the matching of different frequency sensor magnitude sequences. The test results are shown in Figure 14, where the DFLI and LDM sampling points connected by the labels are the matching sampling points, the DFLI sampling sequence is used as the sampling sequence to be matched, the LDM sampling sequence is used as the reference sequence, and there are 15 sampling points in the DFLI segment interval. All points can find matching sampling points in the LDM node interval. Similarly, through reverse matching (Figure 15), the LDM sampling sequence is used as the sampling sequence to be matched, the DFLI sampling sequence is used as the reference sequence, and the LDM node interval is 21 Each sampling point can also find a matching sampling point in the DFLI node interval.

In order to implement the online calibration data sequence matching method provided by the present invention, a calibration platform for the bridge deformation monitoring sensor needs to be built first to simulate the random change of the load on the bridge when the actual vehicle passes through the bridge.

As shown in Figures 16 and 17, the calibration platform includes a simply supported girder model bridge 8 for simulating a bridge, an excitation source 5 that can act on the simply supported girder model bridge 8 to simulate actual vehicles passing on the bridge, and The sensor to be calibrated and the reference sensor 6 installed at the bottom of the simply supported beam model bridge 8 to monitor the deformation of the simply supported beam model bridge 8, the sensor 6 can be installed on the simply supported beam model bridge 8 through the sensor mounting plate 7 At the bottom, the monitoring data of the sensor 6 is collected by the data collection unit.

In a specific embodiment, the SBC and the RS can both use the STP-DM-ST laser displacement meter, the Y-axis measurement range: -24mm to 24mm, and the X-axis measurement range: -15mm to 15mm; as shown in FIG. 16 , Each laser displacement meter includes a laser displacement meter receiving end 9 and a laser displacement meter transmitting end 10. The receiving end 9 is installed at the bridge monitoring point, and the transmitting end 10 is installed at the reference elevation point. The monitoring of bridge deformation is realized through the interaction of the two. . The data acquisition unit can be a FAROPTFAR-J-1/01 online monitoring data acquisition instrument. The low-frequency random load excitation source is generated by the LySeiKi ZP140-300H linear module. The control software can convert the input random number sequence into motion control parameters and send it to the USB9030 motion control card to control the linear module to generate corresponding motion.

By adjusting the excitation frequency given by the program-controlled linear module, the data sequences of two sets of laser displacement meters under three different conditions are obtained, and the three sets of sequences are respectively matched in magnitude. Randomly select 12 (12, 27, 45, 58, 68, 84, 101, 122, 143, 166, 179, 196) sampling points in the matching sequence for matching, and observe whether the corresponding matching can be accurately found in the RS reference sequence point, the results are shown in Figure 18-Figure 20.

Quantitative analysis is performed on the three groups of sequences respectively. The traditional analysis method generally uses the DTW distance to judge the matching effect. The smaller the DTW distance, the better the matching effect. However, this method is more suitable for short-wave matching (for example, the audio sequence of the pronunciation of a single letter in speech recognition). For the experimental object of the embodiment of the present invention, on the one hand, the total amount of the two groups of sampling sequences is relatively large. The DTW distance is high. On the other hand, although the DTW algorithm has good applicability to the time scale, it is sensitive to the magnitude scale, so it is affected by the inherent inertia, which will reduce the calculated DTW distance reference value. Considering the above-mentioned limitations of the sensor magnitude matching applied to the traditional DTW metric standard in the embodiment of the present invention, the embodiment of the present invention introduces the concept of the mean value difference (hereinafter referred to as MVD), and the algorithm matching effect is made by calculating the mean value difference. judge.

For any point pi in the sequence to be matched, let its corresponding point in the reference sequence be q _j , the distance between the point _pi to be matched and the reference point _{q j is} known from the Euclidean distance formula:

Since the matching is performed between different frequency value sequences, the difference on the horizontal axis is the sampling frequency difference, and its value should not be added to the calculation, so take In order to remove the influence of the inherent inertia of the sensor on the evaluation accuracy, inertia compensation is performed on the difference between the two sequence values in the calculation, and formulas (6) (11) are as follows:

The matching results of the three sets of magnitude sequences under different excitation states are analyzed, and the obtained results are shown in Table 1:

Table 1 Analysis of matching results under different frequency excitation states

It can be seen from the above Table 1 that the relative error of SBC and RS sequence matching after node segmentation can be controlled within 2%.

In order to verify the superiority of the node-based segmentation matching algorithm (FPS) over the traditional Euclidean metric matching (EM), a comparative experiment is carried out in the embodiment of the present invention, and the experimental results are shown in FIGS. 21-23 .

Comparing Figures 21 and 22, it can be seen that the node-based segmentation matching has good adaptability to the sensor sampling frequency drift, and the corresponding relationship between the point to be matched and the reference point is basically correct, while the matching based on the Euclidean metric is affected by the sampling frequency and inherent Due to the influence of inertia, the matching accuracy is low and the relative error is large, and as the sampling points gradually increase, the error will gradually increase, while the relative error based on node segmentation matching is low and relatively stable, as shown in Figure 23. In fact, if the excitation source generates periodic excitation, the matching error will reach a peak when the two sequence values lead and lag each other by half a cycle. By comparing and matching three sets of value sequences under different excitation states, the matching accuracy of different methods is quantitatively analyzed. The experimental results are shown in Table 2.

Table 2 Matching accuracy under different matching methods

Comparing the data in Table 2, it can be found that the matching accuracy of the node-based segmentation matching algorithm is improved compared with the traditional Euclidean matching algorithm, and it is less affected by factors such as frequency and sensor inertia.

The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the specific details of the above-mentioned embodiments. Within the scope of the technical concept of the present invention, various simple modifications can be made to the technical solutions of the present invention, These simple modifications all belong to the protection scope of the present invention.

In addition, it should be noted that each specific technical feature described in the above-mentioned specific implementation manner may be combined in any suitable manner under the circumstance that there is no contradiction. In order to avoid unnecessary repetition, the present invention will not describe various possible combinations.

In addition, the various embodiments of the present invention can also be combined arbitrarily, as long as they do not violate the spirit of the present invention, they should also be regarded as the contents disclosed in the present invention.

Claims (10)

1. An online calibration data sequence matching method for a bridge deformation monitoring sensor is characterized by comprising the following steps:

step 1, acquiring measurement value sequences of a sensor to be calibrated and a reference sensor under the same excitation action;

step 2, segmenting the measured value sequences of the sensor to be calibrated and the reference sensor;

step 3, matching the measured values of the sensor to be calibrated and the reference sensor in each subsection interval;

and 4, acquiring the corresponding relation between the measured value sequences of the sensor to be calibrated and the reference sensor under the same excitation action.

2. The on-line calibration data sequence matching method of the bridge deformation monitoring sensor according to claim 1, wherein the measured value sequence of the sensor to be calibrated and the measured value sequence of the reference sensor are both waveform sequences; the step 2 comprises the following steps:

step B1, extracting a target peak point in the measured value sequence of the sensor to be calibrated and a target peak point in the measured value sequence of the reference sensor;

and step B2, segmenting the measured value sequences of the sensor to be calibrated and the reference sensor by taking the target peak point in the measured value sequences of the sensor to be calibrated and the reference sensor as a node respectively.

3. The on-line calibration data sequence matching method for bridge deformation monitoring sensors according to claim 2, wherein the step B1 comprises:

step S1, extracting an initial peak point in the measured value sequence of the sensor to be calibrated and an initial peak point in the measured value sequence of the reference sensor and respectively forming an initial peak point sequence;

step S2, smoothing the initial peak point sequence of the sensor to be calibrated and the initial peak point sequence of the reference sensor by the following equations (1) - (3):

P＝{p₁，p₂…p_n} (1)

j＝(span-1)/2 (3)

wherein, P is an initial peak point sequence of the sensor to be calibrated or the reference sensor; n is the initial peak point sequenceThe number of peak points in; span is a smooth scale; p_i'is the ith peak point in the smoothed initial peak point sequence P';

step S3, storing P in sequence_i'to form a smoothed initial sequence of peak points P';

step S4, finding the initial peak point P in the original initial peak point sequence P corresponding to each peak point in the smoothed initial peak point sequence P_iAnd using the initial peak point P_iAs the target peak point.

4. The method for matching the online calibration data sequence of the bridge deformation monitoring sensor according to claim 1, wherein the measurement value sequence of the sensor to be calibrated and the measurement value sequence of the reference sensor are divided into K +1 segments; wherein, the measured value sequence of the sensor to be calibrated and the measured value sequence of the reference sensor are segmented to form segmented sequences respectively { t }₀,…,t₁},{t₁,…,t₂}…{t_k-1,…,t_k},{t_k,…,t_k+1And { t'₀,…,t'₁},{t'₁,…,t'₂}…{t'_k-1,…,t'_k},{t'_k,…,t'_k+1}; wherein,

t₀a starting measurement point of the measurement value sequence of the sensor to be calibrated;

t_k+1a measurement point which is the end of the measurement value sequence of the sensor to be calibrated;

t_nthe target peak point in the measured value sequence of the sensor to be calibrated is shown, wherein k is more than or equal to n and more than or equal to 1;

t’₀a starting measurement point of a sequence of measurement values for the reference sensor;

t’_k+1an end measurement point of the sequence of measurement values for the reference sensor;

t’_nthe peak value is a target peak value point in the measured value sequence of the reference sensor, wherein k is more than or equal to n and more than or equal to 1.

5. The on-line calibration data sequence matching method for bridge deformation monitoring sensors according to claim 4, wherein the step 3 comprises: sequentially selecting the measured value p to be matched in the measured value sequence of the sensor to be calibrated_iAnd according to said measured value p to be matched_iPosition information of the position of the measured value p to be matched_iSegment interval t_r-1,…,t_r}；

The measured value p to be matched is searched in the measured value sequence of the reference sensor by the following formula (4)_iMatched measured values q_jIn the position of (a) in the first,

wherein j is a measured value p to be matched in a measured value sequence of the reference sensor_iMatched measured values q_jThe abscissa of (a);

t_r-1|_xfor measuring points t in a segmented sequence of sensors to be calibrated_r-1The abscissa of (a);

t_r|_xfor measuring points t in a segmented sequence of sensors to be calibrated_rThe abscissa of (a);

t'_r-1|_xis a measurement point t 'in a segmented sequence of reference sensors'_r-1The abscissa of (a);

t'_r|_xis a measurement point t 'in a segmented sequence of reference sensors'_rThe abscissa of (a).

6. The on-line calibration data sequence matching method for bridge deformation monitoring sensors according to claim 5, wherein the step 3 further comprises: output and save p_iAnd with said p_iMatched q_j。

7. The on-line calibration data sequence matching method for bridge deformation monitoring sensors according to claim 1, wherein the step 1 is preceded by: and (3) setting up a calibration platform of the bridge deformation monitoring sensor to simulate the random change of the load borne by the bridge when an actual vehicle passes through the bridge.

8. The method for matching the on-line calibration data sequence of the bridge deformation monitoring sensor according to claim 7, wherein the calibration platform comprises a simply supported beam model bridge for simulating a bridge, an excitation source capable of acting on the simply supported beam model bridge to simulate an actual vehicle passing through the bridge, and a sensor to be calibrated and a reference sensor which are installed at the bottom of the simply supported beam model bridge to monitor the deformation of the simply supported beam model bridge.

9. The on-line calibration data sequence matching method for the bridge deformation monitoring sensor according to claim 8, wherein the excitation source is a linear module.

10. The on-line calibration data sequence matching method for the bridge deformation monitoring sensor according to any one of claims 1 to 9, wherein the sensor to be calibrated and the reference sensor are both laser displacement meters.