CN102359056B - Detection method of bituminous pavement data - Google Patents

Abstract

The invention discloses a detection method of asphalt pavement data, which comprises the following steps: step 1: collecting detection data to form an original data sequence; step 2: checking and supplementing detection data to form a data sequence; step 3: smoothing processing; 4: Pre-segment the detection data; Step 5: Find outliers for the first time; Step 6: Merge segments for the first time; Step 7: Find outliers for the second time; Step 8: Merge segments for the second time; Step 9: Find local processing points; Step 10: Calculate the representative value of each detection index in each segment. The detection method can effectively detect the abnormal value and its location in the detection data, and can effectively distinguish the road segment of the distribution of the detection index in the entire road section, and measure the representative value of the detection index in each segment. Road maintenance or reconstruction and expansion programs provide a more scientific and economical basis.

Description

A detection method of asphalt pavement data

technical field

The invention relates to a data detection method, in particular to a detection method of asphalt pavement data.

Background technique

With the rapid growth of my country's highway traffic volume, the increasing load of vehicles, and the many adverse effects of my country's special and changeable terrain and climate on the durability of the road surface, my country's highway industry is facing more severe tests and more challenges. high demands. As we all know, under the repeated action of traffic load and natural factors, the pavement performance will gradually weaken, and the pavement structure will gradually be damaged, which will eventually fail to meet the use requirements; After the number of loads, depending on the damage degree of the pavement structure and the attenuation degree of the service performance, a reasonable, effective and economical maintenance or reconstruction and expansion plan should be formulated.

When formulating maintenance or reconstruction and expansion plans, it is necessary to use some test data to characterize the road condition level of existing roads. my country's current "Technical Specifications for Highway Asphalt Pavement Maintenance" (JTJ073.2-2001) points out that the current use quality evaluation of asphalt pavement includes: pavement damage, pavement driving quality, pavement strength and pavement anti-skid performance. Among them, the pavement damage status index PCI is calculated by the comprehensive pavement damage rate DR according to the formula; the road driving quality index RQI is calculated by the test result BI of the flatness test equipment according to the formula; the road strength SSI is calculated by the road section representative deflection according to the formula The anti-skid performance of the pavement uses the anti-skid coefficient as the index, and the anti-skid coefficient is expressed by the lateral force coefficient SFC or the pendulum value BPN of the pendulum instrument. In addition, in addition to the above maintenance detection indicators, the asphalt pavement detection indicators also include rut depth and other indicators.

为该路段检测数据的平均值；S为该路段检测数据的标准差；rv为该检测指标在该路段的代表值；a为正数，由统计学上的保证率决定。当路面性能与检测指标的大小成正比时，公式中用负号；当路面性能与检测指标的大小成反比时，公式中用正号。It is clear that, for a certain road section, in addition to the comprehensive damage rate DR of the road surface, the test result BI of the flatness test equipment of the road section, the road section represents deflection, the lateral force coefficient SFC, and the pendulum value BPN of the pendulum instrument are all Based on a certain amount of detection data, using statistical principles, using the formula Calculated. In this formula,

is the average value of the detection data of the road section; S is the standard deviation of the detection data of the road section; rv is the representative value of the detection index in the road section; a is a positive number, determined by the statistical guarantee rate. When the pavement performance is directly proportional to the size of the detection index, the negative sign is used in the formula; when the pavement performance is inversely proportional to the size of the detection index, the positive sign is used in the formula.

From the above content, it can be seen that the detection method of asphalt pavement data at this stage is relatively simple, and its shortcomings mainly include the following points:

1. Commonly used detection methods ignore the existence of outliers in the detection data, and cannot make full use of the information contained in the detection data.

2. It is obviously unscientific to use only one value as its representative value for the entire road section. First of all, this representative value cannot effectively reflect the distribution of the detection index in the entire road section; second, if this representative value is used for maintenance or reconstruction and expansion plan design, it will inevitably result in The maintenance or reconstruction and expansion plan cannot effectively improve the pavement performance of some unfavorable positions in the road section; similarly, the corresponding maintenance or reconstruction and expansion plan in some favorable positions in the road section will produce a certain amount of waste.

Contents of the invention

Technical problem: The technical problem to be solved by the present invention is to provide a detection method for asphalt pavement data, which can effectively detect abnormal values and their locations in the detection data, and can effectively distinguish the distribution of detection indicators in the entire road section. Segmentation, measure the representative value of the detection index in each segment, so as to provide a more scientific and economical basis for road maintenance or reconstruction and expansion plans.

Technical scheme: in order to solve the above technical problems, the technical scheme adopted in the present invention is:

A detection method for asphalt pavement data, comprising the following steps:

Step 1: Collect test data and form the original data sequence: collect test data for the asphalt pavement test indicators, and each test data corresponds to a stake number information; sort the stake number information from small to large to form the original data sequence of the test data;

Step 2: Check and retest the detection data to form a data sequence: check the detection data collected in step 1 and perform supplementary measurement on the wrong data to form a data sequence d;

Step 3: smoothing processing: smoothing the detection data obtained in step 2 to obtain the smoothed detection data;

Step 4: Pre-segment the detection data: take the range as the control index, pre-segment the smoothed detection data obtained in step 3, the range in each segment is less than or equal to the maximum allowable range, And record the stake number information of each segment point in the pre-segmentation to form a pre-segmentation;

Step 5: Find outliers for the first time: According to the pre-segmentation obtained in step 4, segment the test data obtained in step 2, search for outliers in each segment, and mark them in the test data;

Step 6: Merge segmentation for the first time: according to the pre-segmentation obtained in step 4, segment the detection data obtained in step 2, and exclude the outliers obtained in step 5, based on the two-sample Kolmogorov- Smolov Kolmogorov-Smirnov test, merge the pre-segmentation obtained in step 4, and record the stake number information of each segment point after the merger;

Step 7: Find outliers for the second time: Obtain the chain number information of the segmentation points according to step 6, segment the detection data obtained in step 2, search for abnormal values for each segment, and mark them in the detection data ;

Step 8: Merge segmentation for the second time: according to the stake number information of the segmentation point obtained in step 6, segment the detection data obtained in step 2, and exclude the outliers obtained in step 7, based on the rank sum test, the The segments obtained in step 7 are merged, and record the stake number information of each segment point after the merger;

Step 9: Search for local processing points: according to the stake number information obtained in step 8, segment the detection data obtained in step 2, and search for local processing points;

Step 10: Calculate the representative value of each detection indicator in each segment: according to the stake number information obtained in step 8, segment the detection data obtained in step 2, and exclude the local processing points found in step 9, and calculate each The representative value of each detection indicator in each segment.

Beneficial effect: Compared with the prior art, the beneficial effect of the present invention is that it provides a more scientific and economical basis for road maintenance or reconstruction and expansion schemes. First, the technical solution of the present invention can obtain the segmentation of road detection data, which can effectively distinguish the distribution of road detection data in the road section. Specifically, the detection data is merged and divided twice through steps 6 and 8. Segment, which combines data with similar distribution types and similar sizes into one segment. Secondly, the representative value of a certain detection index in each segment can be obtained, which provides decision-making basis for the design of subsequent maintenance or reconstruction and expansion schemes. Specifically, step 10 calculates the representative value of each detection index in each segment , is based on smoothing, pre-segmentation, outlier search twice, merging segments twice and local processing point search on the collected original data sequence, so that the representative value of the calculation has a certain guarantee rate , which can effectively represent the value of the detection index in a certain segment. Third, the position of the abnormal point of the index in each section of the road can be obtained, that is, the position of the point that needs local treatment, which can effectively improve the economy of the maintenance or reconstruction and expansion plan.

Description of drawings

Fig. 1 is a flowchart of the detection method of the present invention.

Fig. 2 is a flowchart of step 4 in the present invention.

Fig. 3 is a flowchart of step 6 in the present invention.

Fig. 4 is a flowchart of step 8 in the present invention.

Detailed ways

The technical solutions of the present invention will be described in detail below in conjunction with the accompanying drawings.

As shown in Figure 1, a kind of detection method of asphalt pavement data of the present invention comprises the following steps:

Step 1: Collect test data to form the original data sequence: collect test data for the asphalt pavement test indicators, and each test data corresponds to a stake number information, and sort the stake number information from small to large to form the original data sequence of the test data. These detection data are directly collected data without any processing.

In this step 1, the detection indicators include road surface damage, driving quality, road surface strength, road surface skid resistance and rut depth. According to the requirements of "Technical Specifications for Highway Asphalt Pavement Maintenance" (JTJ073.2-2001), the detection data of the aforementioned detection indicators were collected. It is required to collect test data evenly at a certain interval, and the size of the sampling interval should be selected according to the requirements of "Technical Specifications for Highway Asphalt Pavement Maintenance" (JTJ073.2-2001). Each detection index forms a set of original data sequences respectively. In order to ensure the representativeness of the test data samples, it is required that each set of test data to be processed, that is, the number of data in each set of original data sequences is greater than 100. Sorting the collected detection data according to the pile size, from small to large, to form the original data sequence.

Step 2: Check the detection data collected in step 1 and perform supplementary measurement on the wrong data to form the data sequence d.

In this step 2, the inspection of the detection data includes error value inspection and uniformity inspection:

a. Error value check: Select a data sample of a detection indicator and calculate it by referring to the following formula:

$|x_i-\stackrel{\‾}{x}|\&Greater\; Equal;6S$ Formula 1)

In formula (1):

x _i represents the i-th data in the data sample;

表示数据样本的均值；

represents the mean of the data sample;

S represents the standard deviation of the data sample.

If x _i satisfies the formula (1), the supplementary measurement is performed on this point. If the supplementary test results still do not meet the above conditions, the value will be regarded as a normal value for subsequent data processing.

b. Uniformity check: check the uniformity of the data that has been checked for error values and has been supplemented by the following formula:

Uniformity=|Difference between two adjacent stakes-sampling interval|/sampling interval Formula (2)

The uniformity between two adjacent points should satisfy formula (3):

Uniformity＜=1 Formula (3)

If the above conditions are not met, supplementary testing shall be carried out at the corresponding position until the above conditions are met.

The detection data collected in each detection indicator is inspected and supplemented to form a data sequence d, which contains n data, that is, d consists of d ₁ , d ₂ , d ₃ , d ₄ , d ₅ ...d _i-1 , d _i , d _i+1 ...d _n-1 , d _n .

Step 3: smoothing processing: performing smoothing processing on the detection data obtained in step 2 to obtain smoothed detection data.

In Step 3, data smoothing refers to eliminating the influence of the peak and valley values of a set of data on the overall distribution, so as to make the data curve as smooth as possible.

The specific smoothing process is carried out according to formula (4):

ds ₁ =d ₁ ;

ds ₂ =(d ₁ +d ₂ +d ₃ )/3;

ds ₃ =(d ₁ +d ₂ +d ₃ +d ₄ +d ₅ )/5;

ds _i =(d _i-(span-1)/2 +...+d _i-1 +d _i +d _i+1 +...+d _i+(span-1)/2 )/span; (4)

ds _n-2 = (d _n-4 + d _n-3 + d _n-2 + d _n-1 + d _n )/5;

ds _n-1 = (d _n-2 + d _n-1 + d _n )/3;

ds _n =d _n ;

In formula (4):

In this formula: the data sequence ds composed of ds ₁ , ds ₂ , ds ₃ ... ds _i ... ds _n-2 , ds _n-1 , ds _n represents the data sequence after smoothing, and the total amount is n, where ds ₁ represents the first smoothed data in the data sequence, ds ₂ represents the second smoothed data in the data sequence, ds ₃ represents the third smoothed data in the data sequence, . .., ds _i represents the i-th smoothed data in the data sequence, ds _n-2 represents the n-2 smoothed data in the data sequence, ds _n-1 represents the n-1th data in the data sequence The smoothed data, ds _n represents the nth smoothed data in the data sequence. From d ₁ , d ₂ , d ₃ , d ₄ , d ₅ ...d _i(span-1)/2 ...d _i-1 , d _i , d _i+1 ...d _{i+(span- 1)/2} ... d _n-4 , d _n-3 , d _n-2 , d _n-1 , d _n consists of data sequence d, which is the data sequence obtained in step 2, the total amount is n, and the data The total amount of sequence ds is consistent. span is a smooth range, the value is less than or equal to n, and the value range is an odd number from 5 to 20. If the value of span is too large, the segmentation length will be too large, which will lose the meaning of segmentation. If the value of span is too small, the segmentation will be too fine, which is not conducive to the scheme design.

Step 4: Pre-segment the detection data: take the range as the control index, pre-segment the smoothed detection data obtained in step 3, the range in each segment is less than or equal to the maximum allowable range, And record the stake number information of each segmentation point in the pre-segmentation to form a pre-segmentation.

In this step 4, performing preliminary segmentation on the smoothed data sequence ds formed in step 3 is called pre-segmentation of detection data. This pre-segmentation method mainly uses the range (referring to the difference between the maximum value and the minimum value in a set of data) as the control index, that is, to pre-segment ds under the given maximum allowable range condition, and in the pre-segmentation After segmentation, the range of each segment is not greater than the maximum allowable range. The flow chart of the pre-segmentation principle is shown in Figure 2:

First, set i=1, create a temporary empty array TA, set the total amount of data in the data sequence ds in step 3 as n, and perform the following operations:

(a) Initialize the temporary array TA, that is, clear all elements in the array TA;

(b) insert the i-th data ds _i in the data sequence ds into the temporary array TA;

(c) judging whether the range of the array TA is less than or equal to the maximum allowable range rmax;

(d) If the range of the array TA is less than or equal to the maximum allowable range, add 1 to the value of i; if i is less than or equal to n at this time, return to (b); if i is greater than n at this time, the segment ends , organize the returned chainage information;

(e) If the extreme difference of the array TA is greater than the maximum allowable extreme difference, then record the chain number k _i-1 corresponding to the i-1th data; if at this time i is less than n, return to (a); if at this time i If it is equal to n, the segmentation ends, and the returned chainage information is sorted out.

The maximum allowable range rmax mentioned in step 4 determines the number of segments. The larger the maximum allowable range rmax, the smaller the number of segments obtained by pre-segmentation, and the smaller the maximum allowable range rmax, the more segments obtained by pre-segmentation. The value of the maximum allowable range rmax should be determined by the subsequent design scheme. In other words, if a certain detection index can cause significant changes in the design scheme of the two road sections after the difference between the two road sections reaches a certain critical value, then the critical The value can be used as the maximum allowable range rmax of the pre-segmentation. This method recommends that the maximum allowable range rmax of the deflection index be 10-30 (0.01mm), the maximum allowable range rmax of the rutting index be 0.5mm-2mm, and the maximum allowable range rmax value selection of other indicators refer to the above.

According to the returned chainage information, the obtained chainage can be used as a segmentation point to segment the data sequence d obtained in step 2. After the data sequence d is pre-segmented, a finer segment can be obtained, but because this segment is often too fine, it is not conducive to the scheme design. Especially in some local positions, the segment length is often very small. This means that these smaller segments only need to be processed locally, and there is no need to list them as a separate segment. Therefore, in order to make the segmentation meet the needs of practical applications, it is necessary to combine the pre-segmentation results to obtain new segments that are convenient for scheme design.

Step 5: Find outliers for the first time: According to the pre-segmentation obtained in step 4, segment the test data obtained in step 2, search for outliers in each segment, and mark them in the test data.

Before merging the pre-segmentation in step 4, it is necessary to find the abnormal data in each pre-segmentation in step 4 to prevent it from interfering with the merging operation. Refer to the "Highway Subgrade Pavement Field Test Regulations" (JTG E60-2008), if the size of the detection index is inversely proportional to the road condition level, all the detection data in the section that is greater than or equal to the average value of the section plus twice the standard deviation are regarded as abnormal values , which is shown in formula (5), and recorded; if the size of the detection index is proportional to the road condition level, then all the detection data in the segment that is less than or equal to the mean value of the segment minus twice the standard deviation are taken as outliers, that is, the formula (6) and record it.

The data sequence d obtained in step 1 is divided into segments obtained in step 4, and the following judgments are made:

$x_{\mathrm{ij}}\&Greater\; Equal;{\stackrel{\‾}{x}}_i+2S_i$ Formula (5)

$x_{\mathrm{ij}}\≤{\stackrel{\‾}{x}}_i-2S_i$ Formula (6)

In formula (5) and formula (6):

x _ij represents the j-th data in the i-th segment in the data sequence d;

表示数据序列d中第i分段中的数据的均值；

Indicates the mean value of the data in the i-th segment in the data sequence d;

S _i represents the standard deviation of the data in the i-th segment in the data sequence d.

According to formula (5) and formula (6), each data in each segment of the data sequence d is judged, if x _ij satisfies formula (5) or formula (6), it is regarded as the first abnormal value Find outliers, and the rest of the data except the outliers are the normal values obtained in the first outlier lookup, and mark them in the data sequence d respectively.

Step 6: Merge segmentation for the first time: according to the pre-segmentation obtained in step 4, segment the detection data obtained in step 2, and exclude the outliers obtained in step 5, based on the two-sample Kolmogorov- Smirnov Kolmogorov-Smirnov test, merge the pre-segments obtained in step 4, and record the chainage information of each segment point after the merger.

In step 6, the two-sample Kolmogorov-Smirnov Kolmogorov-Smirnov test can determine whether the two groups of samples belong to the same distribution (not necessarily a normal distribution). Using this control method, the data sequence can be Neighboring segments with the same distribution are initially merged.

The flow chart of the merging principle in step 6 is shown in Figure 3, and the merging principle is as follows:

First, let i=1, set up a temporary empty array TA, m groups of data sequences formed by the normal values in the data sequence d obtained in step 5, m is the number of pre-segmentation, set the m groups of data sequences as X ₁ , X ₂ ....X _m-1 , X _m ; m sets of data sequences composed of all data in the data sequence d can also be obtained, m is the number of pre-segments, and these m sets of data sequences are respectively Z ₁ , Z ₂ .... Z _m-1 , Z _m , where Z _i should be composed of Xi _and the outlier in the i-th segment obtained from step 5. Do the following:

(a) Initialize the temporary array TA, that is, clear all elements in the array TA;

(b) insert the i-th data sequence _Xi into the temporary array TA;

(c) Perform a two-sample Kolmogorov-Smirnov Kolmogorov-Smirnov test with a significance level of p on the array TA and the i+1th data sequence X i ₊₁ , which is H in Figure 3 =ks2test(TA, X _i+1 , p), if the two samples pass the test, then H=0, it is considered that the hypothesis is accepted, indicating that Xi ₊₁ and TA belong to the same distribution; if the two samples fail the test, then H= 1. It is considered that the hypothesis is rejected, indicating that Xi ₊₁ and TA do not belong to the same distribution;

(d) if H=0, then add 1 to the value of i; if this moment i is less than m, then return to b; if this moment i is equal to m, then segmentation finishes, arranges the stake information that returns;

(e) if H=1, then record in the i-th data sequence Z _i , the stake number Ki corresponding to the last data; Make i equal to i+1, if i is less than m at this moment, then return to (a); Otherwise, the segmentation ends, and the returned chainage information is sorted out.

The significance level p mentioned in step 6 is a probability value that determines how harsh the two-sample Kolmogorov-Smirnov test is. Generally speaking, the larger p is, the harsher the test is, that is, the easier it is for the hypothesis to be rejected. For common engineering applications, p is recommended to be 0.05.

According to the returned stake number information, the obtained stake number can be used as a segmentation point to segment the data sequence d obtained in step 2 again, and the segment obtained at this time is the segment after the first merge.

Since only the combination method based on the two-sample Kolmogorov-Smorov-Smirnov test is used, although two adjacent segments are distinguished due to different distributions, there is no significant difference in their means or medians Case. However, in practical engineering applications, only when the representative values of two adjacent segments are significantly different, the design scheme will change. In other words, if there is no difference between the representative values of adjacent segments, they can be divided into the same segment. Therefore, it is necessary to use the rank sum test, that is, the median test method, to merge adjacent paragraphs whose mean or median values are not significantly different from each other.

Step 7: Find outliers for the second time: Obtain the chain number information of the segmentation points according to step 6, segment the detection data obtained in step 2, search for abnormal values for each segment, and mark them in the detection data .

In this step 7, the process of finding outliers is: if the size of the detection index is inversely proportional to the road condition level, then all the detection data in the section that are greater than or equal to the mean value of the section plus twice the standard deviation are taken as outliers and recorded ; If the size of the detection index is directly proportional to the road condition level, then all the detection data less than or equal to the average value of the segment minus twice the standard deviation in the segment will be recorded as abnormal values.

Before merging the segments obtained in step 6 again, in order to prevent abnormal data from interfering with the merging operation, it is necessary to search for outliers on the data sequence d obtained in step 2. The difference between step 7 and step 5 is that the road segment in step 5 is obtained in step 4, while the road segment in step 7 is obtained in step 6.

According to step 6 to obtain the stake number information, segment the data sequence d obtained in step 2, and perform the second search for abnormal values: if the size of the detection index is inversely proportional to the road condition level, then add all the values in the segment that are greater than or equal to the mean value of the segment to The detection data of the last two standard deviations is taken as an abnormal value, which is shown in formula (7), and is recorded; if the size of the detection index is proportional to the road condition level, then subtract twice the mean value of the segment that is less than or equal to the segment The detection data of the standard deviation is regarded as an abnormal value, which is shown in formula (8), and recorded.

$x_{\mathrm{ij}}\&Greater\; Equal;{\stackrel{\‾}{x}}_i+2S_i$ Formula (7)

$x_{\mathrm{ij}}\≤{\stackrel{\‾}{x}}_i-2S_i$ Formula (8)

In formula (7) and formula (8):

x _ij represents the j-th data in the i-th segment in the data sequence d;

表示数据序列d中第i分段中的数据的均值；

Indicates the mean value of the data in the i-th segment in the data sequence d;

S _i represents the standard deviation of the data in the i-th segment in the data sequence d.

According to the above method, each data in each segment of the data sequence d is judged. If x _ij satisfies formula (7) or formula (8), it will be regarded as the outlier obtained by the second outlier search, except The rest of the data except the outliers are the normal values obtained from the second outlier search, and they are respectively marked in the data sequence d.

The reason why step 7 is necessary is that the outlier value obtained in the first outlier search is only for the segment obtained in step 4, and the outlier value obtained in the second outlier search is for the segment obtained in step 6. This also means that for the same data sequence d, as long as the segments obtained in step 4 and step 6 are different, the abnormal value obtained after the abnormal value search operation may change.

Step 8: Merge segmentation for the second time: according to the stake number information of the segmentation point obtained in step 6, segment the detection data obtained in step 2, and exclude the outliers obtained in step 7, based on the rank sum test, the The segments obtained in step 7 are merged, and the stake number information of each segment point after the merge is recorded.

In step 8, the rank sum test can effectively distinguish two groups of samples with significant differences in the values of each data. Compared with the two-sample Kolmogorov-Smirnov Kolmogorov-Smirnov test, it pays more attention to Whether the difference between the two groups of samples is significant. According to this property, the segments obtained in step 6 can be further merged.

The flowchart of the merging principle in step 8 is shown in Figure 4, and the merging principle is as follows:

First, set i=1 to create a temporary empty array TA. Step 7 obtains r groups of data sequences composed of normal values in data sequence d, where r is the number of segments obtained in step 6, and these r groups of data sequences are respectively Y ₁ , Y ₂ .... _{Y r-1} , Y _r ; It is also possible to obtain r groups of data sequences composed of all the data in data sequence d, where r is the number of pre-segments, and these r groups of data sequences are respectively W ₁ , W ₂ ....W _r-1 , W _r , where W _i should be composed of Y _i and the outliers in the i-th segment obtained from step 7. Do the following:

(a) Initialize the temporary array TA, that is, clear all elements in the array TA;

(b) Insert the i-th data sequence Y _i into the temporary array TA;

(c) Perform a rank sum test with a significance level of u on the array TA and the i+1th data sequence Y _i+1 , that is, H=ranksum(TA, Y _i+1 , u) in Figure 4, if If the two samples pass the test, then H=0, the hypothesis is considered accepted, indicating that the difference between Y _i+1 and TA is not significant; if the two samples fail the test, then H=1, the hypothesis is considered rejected, indicating that Y _i Significant difference between ₊₁ and TA;

(d) if H=0, then add 1 to the value of i; if this moment i is less than r, then return to (b); if this moment i is equal to r, then segment ends, arranges the stake number information that returns;

(e) If H=1, then record the stake k _i corresponding to the last data in the i-th data sequence W _i ; make i equal to i+1, if i is smaller than r at this time, return to (a) ; Otherwise, the section ends, and the returned chainage information is sorted out.

The significance level u mentioned in step 8 is a probability value which determines the harshness of the rank sum test. Generally speaking, the larger u is, the harsher the test is, that is, the easier the hypothesis is to be rejected. For common engineering applications, u is recommended to be 0.05.

According to the chainage information returned after the operation in step 8, the obtained chainage information can be used as the segmentation point, and the data sequence d obtained in step 2 is segmented again, and the segmentation obtained at this time is the Merged segments.

Step 9: Search for local processing points: According to the stake number information obtained in step 8, segment the detection data obtained in step 2, and search for local processing points.

The segment obtained in step 8 is the final segment of the detection data sequence d obtained in step 2. Local processing points should also be looked up in the final segment. In this step 9, the process of finding the local processing point is: if the size of the detection index is inversely proportional to the road condition level, then all the detection data in the segment that are greater than or equal to the mean value of the segment plus twice the standard deviation are used as the local processing point, That is, the mean value plus twice the standard deviation is used as the critical value for judging the local processing point, as shown in formula (9), and it is recorded; The detection data whose mean value minus twice the standard deviation is used as the local processing point, that is, the mean value minus twice the standard deviation is used as the critical value for judging the local processing point, as shown in formula (10), and recorded.

$x_{\mathrm{ij}}\&Greater\; Equal;{\stackrel{\‾}{x}}_i+2S_i$ Formula (9)

$x_{\mathrm{ij}}\≤{\stackrel{\‾}{x}}_i-2S_i$ Formula (10)

In formula (9) and formula (10):

x _ij represents the j-th data in the i-th segment in the data sequence d;

表示数据序列d中第i分段中的数据的均值；

Indicates the mean value of the data in the i-th segment in the data sequence d;

S _i represents the standard deviation of the data in the i-th segment in the data sequence d.

If x _ij satisfies Equation (9) or Equation (10), it is regarded as a point requiring local processing, and it is marked in the data sequence d.

Step 10: Calculate the representative value of each detection indicator in each segment: according to the stake number information obtained in step 8, segment the detection data obtained in step 2, and exclude the local processing points found in step 9, and calculate each The representative value of each detection indicator in each segment.

From step 9, the normal value in the data sequence d, that is, the data sequence after excluding the local processing points, can be obtained. The t group of data sequences is composed of t is the number of segments obtained in step 9. Let the t groups of data sequences be respectively N ₁ , N ₂ .... N _t-1 , N _t , and t sets of data sequences composed of all the data in the data sequence d can also be obtained, t is the number of pre-segments, and these t sets of data sequences are respectively M ₁ , M ₂ .... M _t-1 , M _t , where M _i should consist of N _i and the local processing points in the i-th segment obtained from step 9.

In this step 10, the process of calculating the representative value of each detection index in each segment is: if the size of the detection index is inversely proportional to the road condition level, then the mean value of the detection data excluding local processing points in this segment Add one standard deviation as the representative value, as shown in formula (11); if the size of the detection index is proportional to the road condition level, then subtract one standard deviation from the mean value of the detection data in the segment excluding local processing points The difference is taken as a representative value, as shown in formula (12).

${\mathrm{RV}}_i={\stackrel{\‾}{x}}_i+S_i$ Formula (11)

${\mathrm{RV}}_i={\stackrel{\‾}{x}}_i-S_i$ Formula (12)

In formula (11) and formula (12):

表示第i段中排除了局部处理点的检测数据的均值；

Indicates the mean value of the detection data excluding local processing points in the i-th segment;

S _i represents the standard deviation of the detection data excluding local processing points in the i-th paragraph;

RV _i represents the representative value of an indicator in the i segment.

Using normal values to calculate representative values can avoid the interference of abnormal data on the overall distribution of data, because after excluding abnormal values, the data in a certain segment is generally smooth and there will be no sudden changes. The representative value is measured by the method of adding and subtracting one standard deviation of the mean value, which is obtained by comprehensively considering the design and construction feasibility.

Claims (7)

1. a detection method of asphalt pavement data, is characterized in that, this detection method comprises the following steps: Step 1: Collect test data and form the original data sequence: collect test data for the asphalt pavement test indicators, and each test data corresponds to a stake number information; sort the stake number information from small to large to form the original data sequence of the test data; Step 2: Check and retest the detection data to form a data sequence: check the detection data collected in step 1 and perform supplementary measurement on the wrong data to form a data sequence d; Step 3: smoothing processing: smoothing the detection data obtained in step 2 to obtain the smoothed detection data; Step 4: Pre-segment the detection data: take the range as the control index, pre-segment the smoothed detection data obtained in step 3, the range in each segment is less than or equal to the maximum allowable range, And record the stake number information of each segment point in the pre-segmentation to form a pre-segmentation; Step 5: Find outliers for the first time: According to the pre-segmentation obtained in step 4, segment the test data obtained in step 2, search for outliers in each segment, and mark them in the test data; Step 6: Merge segmentation for the first time: according to the pre-segmentation obtained in step 4, segment the detection data obtained in step 2, and exclude the outliers obtained in step 5, based on the two-sample Kolmogorov- Smolov Kolmogorov-Smirnov test, merge the pre-segmentation obtained in step 4, and record the stake number information of each segment point after the merger; Step 7: Find outliers for the second time: Obtain the chain number information of the segmentation points according to step 6, segment the detection data obtained in step 2, search for abnormal values for each segment, and mark them in the detection data ; Step 8: Merge segmentation for the second time: according to the stake number information of the segmentation point obtained in step 6, segment the detection data obtained in step 2, and exclude the outliers obtained in step 7, based on the rank sum test, the The segments obtained in step 7 are merged, and record the stake number information of each segment point after the merger; Step 9: Search for local processing points: according to the stake number information obtained in step 8, segment the detection data obtained in step 2, and search for local processing points; Step 10: Calculate the representative value of each detection indicator in each segment: according to the stake number information obtained in step 8, segment the detection data obtained in step 2, and exclude the local processing points found in step 9, and calculate each The representative value of each detection index in each segment. 2. according to the detection method of asphalt pavement data according to claim 1, it is characterized in that, in described step 1, detection index comprises road surface damage condition, running quality, road surface strength, road surface skid resistance performance and rutting depth, each The detection indicators form a set of original data sequences respectively, and the number of data in each set of original data sequences collected is greater than 100. 3. according to the detection method of asphalt pavement data according to claim 1, it is characterized in that, in said step 2, the inspection to detection data comprises error value inspection and uniformity inspection, wherein: The error value check described is according to the formula

In the formula, x _i represents the i-th detection data in the sample,

Indicates the mean value of the sample, and S represents the standard deviation of the sample. If the formula is established, a supplementary measurement is performed on the i-th point to obtain a supplementary measurement value, and the supplementary measurement value is used as a normal value for subsequent processing; The uniformity check includes: first, according to the formula: uniformity=|difference between two adjacent stakes-sampling interval|/sampling interval, measure the uniformity, and then judge whether the uniformity meets the requirement of less than or equal to 1; if If the uniformity is greater than 1, a supplementary measurement shall be carried out at the corresponding stake position until the requirement that the uniformity is less than or equal to 1 is met. 4. according to the detection method of asphalt pavement data according to claim 1, it is characterized in that, in described step 3, smoothing process is to carry out the following formula of detection data: ds ₁ = d ₁ ; ds ₂ =( d ₁ +d ₂ +d ₃ )/3; ds ₃ =( d ₁ +d ₂ +d ₃ +d ₄ +d ₅ )/5; ...... ds _i =(d _i-(span-1)/2 +...+d _i-1 +d _i +d _i+1 +...+d _i+(span-1)/2 )/span; ...... ds _n-2 =( d _n-4 +d _n-3 +d _n-2 +d _n-1 +d _n )/5; ds _n-1 =(d _n-2 + d _n-1 + d _n )/3; ds _n = d _n ; In this formula: the data sequence composed of ds ₁ , ds ₂ , ds ₃ ... ds _i ... ds _n-2 , ds _n-1 , ds _n represents the data sequence after smoothing, and the total amount is n , where ds ₁ represents the first smoothed data in the data sequence, ds ₂ represents the second smoothed data in the data sequence, ds ₃ represents the third smoothed data in the data sequence, .. . , ds _n represents the nth smoothed data in the data sequence; by d ₁ , d ₂ , d ₃ , d ₄ , d ₅ ...d _i-(span-1)/2 ... d _{i -1} , d _i , d _i+1 ...d _i+(span-1)/2 ... d _n-4 , d _n-3 , d _n-2 , d _n-1 , d _n Sequence is the data sequence obtained in step 2, the total amount is n; span is the smooth range, the value is less than or equal to n, and the value range is an odd number from 5 to 20. 5. according to the detection method of the asphalt pavement data described in claim 1, it is characterized in that, in described step 5 and step 7, the process of finding abnormal value is: if the size of detection index is inversely proportional to road condition level, then will All the detection data in this segment that are greater than or equal to the mean value of the segment plus twice the standard deviation are recorded as abnormal values; The detection data with twice the standard deviation were recorded as outliers. 6. according to the detection method of the asphalt pavement data described in claim 1, it is characterized in that, in described step 9, the process of searching local processing point is: if the size of detection index is inversely proportional to road condition level, then this section All the detection data greater than the mean value of the segment plus twice the standard deviation are taken as local processing points and recorded; if the size of the detection index is proportional to the road condition level, all the detection data in the segment smaller than the mean value of the segment minus twice the standard Poor detection data are treated as local processing points and recorded. 7. according to the detection method of the asphalt pavement data described in claim 1, it is characterized in that, in described step 10, the process of measuring and calculating the representative value of each kind of detection index in each segment is: if the size of detection index and If the road condition level is inversely proportional, then the mean value of the detection data excluding local processing points in the segment plus one standard deviation is used as the representative value; if the detection index is proportional to the road condition level, then the segment is excluded The mean value of the detection data of the local processing point minus one standard deviation is used as the representative value.

Images