Asphalt pavement disease correlation analysis method based on data mining

Abstract

The invention discloses an asphalt pavement disease correlation analysis method based on data mining. The method comprises the following steps: firstly, acquiring detection data of 15 diseases including cracking, block crack, longitudinal / transverse crack (light / heavy), pit slot, track, strip repair and the like, deleting a single disease or disease-free road section, and constructing a Boolean matrix; through an improved Apriori algorithm, frequent 1-item sets are scanned and screened for the first time, and all frequent item sets are generated through row-column compression optimization (row compression deletes non-frequent item set supersets based on anti-monotonicity and column compression eliminates transactions incapable of forming k-item sets) and secondary scanning. And calculating the support degree, the confidence coefficient and the lifting degree, screening a strong association rule meeting a threshold value, identifying key diseases and high-frequency induced diseases, and formulating combined maintenance measures. According to the method, the efficiency is improved through row and column compression and binary operation, the disease symbiosis mode is accurately mined, a data-driven scientific basis is provided for preventive maintenance, the service life of the pavement is prolonged, and the whole-cycle cost is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of highway maintenance, and in particular to a method for analyzing the correlation between asphalt pavement defects based on data mining. The method mines the symbiotic patterns between defects through an improved association rule algorithm, thereby providing a scientific basis for preventive maintenance. Background Art

[0002] In the field of highway maintenance, scientific maintenance plans are crucial for extending the service life of asphalt pavements and reducing lifecycle costs. However, traditional maintenance decisions focus solely on single defects, ignoring the interactions between them. This results in a lack of systematic and forward-looking maintenance measures. For example, if early cracks are not promptly addressed, they can lead to more serious defects such as rutting and potholes, increasing subsequent maintenance costs.

[0003] Existing association rule algorithms, such as Apriori and FP-Growth, are used to mine data associations, but they suffer from significant shortcomings: the Apriori algorithm requires multiple database scans, generating a large number of candidate item sets, resulting in low computational efficiency; while the FP-Growth algorithm reduces the number of scans, its data compression process is complex and its effectiveness in processing high-dimensional, sparse data is limited. Furthermore, existing research has largely focused on single disease prediction, lacking in-depth analysis of inter-disease symbiosis patterns and evolutionary mechanisms, making it difficult to meet the needs of preventive maintenance for multi-disease association analysis.

[0004] Therefore, there is an urgent need for an efficient data mining method that can accurately identify the correlation between asphalt pavement diseases, reveal the inducing mechanism of hidden diseases, and provide precise support for the formulation of maintenance strategies. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for analyzing the correlation of asphalt pavement defects based on data mining. In order to address the problems of neglecting the interactive relationships between defects and the low efficiency of association rule algorithms in traditional asphalt pavement maintenance decisions, the present invention realizes the efficient identification of defect symbiosis patterns through data preprocessing, row and column compression optimization and strong association rule mining, providing accurate technical support for preventive maintenance.

[0006] The present invention is achieved through the following technical solutions:

[0007] A method for analyzing the correlation between asphalt pavement defects based on data mining includes the following steps:

[0008] Step S1, data preprocessing: obtaining asphalt pavement disease detection data, deleting road section data with no disease or only a single disease, and forming an initial transaction data set D containing two or more diseases;

[0009] Step S2, constructing a Boolean matrix: sorting the initial transaction data set D by the column of the damage type and the row of the mileage pile number, constructing an m×n Boolean matrix M, where m is the number of transactions, n is the number of damage types, and the presence of a damage is marked as 1, otherwise it is marked as 0;

[0010] Step S3: Improved Apriori algorithm to mine frequent itemsets:

[0011] Scan the Boolean matrix M for the first time, calculate the support of all single item sets, and filter out the frequent 1-item sets L1 whose support is greater than or equal to the minimum support threshold;

[0012] Generate candidate item sets based on the frequent 1-item set L1 and optimize through row and column compression:

[0013] Row compression: If an item set is not frequent, its superset is directly deleted;

[0014] Column compression: If the number of transactions is less than k, delete transactions that cannot form a frequent k-item set;

[0015] Scan the compressed matrix for the second time, calculate the support of the candidate item set, generate the frequent k-item set Lk, repeat until no k+1-item set can be generated, and obtain all frequent item sets;

[0016] Step S4, generating strong association rules: calculating the confidence and lift for all the frequent itemsets, and screening strong association rules that simultaneously meet the confidence ≥ minimum confidence threshold and lift ≥ 1.25;

[0017] Step S5: Formulate a preventive maintenance strategy: Based on the strong association rule results, identify key diseases and high-frequency induced diseases, and formulate a combination of maintenance measures.

[0018] As a further improvement of the technical solution of the present invention, the types of defects include cracks, deformations, and other types, specifically covering 15 types of defects and their light and heavy degrees, including cracks, block cracks, longitudinal cracks (light / severe), transverse cracks (light / severe), potholes, looseness, rutting, oil spills, strip repairs, and block repairs, among which the light and heavy degrees are marked as L and H, respectively.

[0019] As a further improvement to the technical solution of the present invention, in the row-column compression optimization, row compression is based on the anti-monotonicity of frequent itemsets: if an itemset is infrequent, all its supersets are infrequent itemsets; column compression is based on the correlation between the number of transactions and the dimension of frequent itemsets. When the number of transactions is less than k, the corresponding transactions are deleted.

[0020] As a further improvement to the technical solution of the present invention, during the construction of the Boolean matrix M, the disease types are sorted in columns according to a preset order, and the preset order includes crack type first, deformation type second, and other types last. Each type of disease is arranged in order of severity and the first letter of the pinyin of the name.

[0021] As a further improvement to the technical solution of the present invention, the support calculation formula is:

[0022] The confidence calculation formula is: The lift calculation formula is:

[0023] As a further improvement of the technical solution of the present invention, the minimum support threshold is set in the range of 0.05 to 0.2, the minimum confidence threshold is set to 0.5 to 0.8, and the lift threshold is set to 1.25 to 2.0. The specific parameters are determined by scatter plot analysis based on the distribution of disease data.

[0024] As a further improvement to the technical solution of the present invention, after the strong association rules are generated, the rules are visualized, and a visualization map of the association rules is formed by color mapping enhancement and point size mapping support, which intuitively displays the association strength between diseases.

[0025] As a further improvement to the technical solution of the present invention, when formulating the preventive maintenance strategy, combined maintenance measures are taken for the strong association rules with a confidence level ≥ 70%. The combined maintenance measures include combining crack grouting with surface sealing, and combining milling and resurfacing with anti-cracking layer laying.

[0026] As a further improvement to the technical solution of the present invention, the improved Apriori algorithm uses binary bit operations instead of set operations when generating candidate item sets, and represents the existence of item sets through bit masks, thereby improving operation efficiency.

[0027] As a further improvement to the technical solution of the present invention, in the data preprocessing step, the damage data is cleaned to remove noise data with abnormal mileage pile numbers and inconsistent damage degree markings, thereby ensuring the integrity and accuracy of the initial transaction data set D.

[0028] In summary, the beneficial effects of the present invention are as follows:

[0029] Efficiency: Through row and column compression and Boolean matrix optimization, the multiple database scans of the traditional Apriori algorithm are reduced to two, significantly improving computational efficiency. This is particularly suitable for high-dimensional and sparse pavement disease data.

[0030] Accuracy: It can uncover implicit correlations between different disease types and degrees, such as the confidence level of "mild transverse cracks → severe longitudinal cracks" reaching 62.7%, providing an accurate disease evolution path for maintenance decisions.

[0031] Practicality: Integrating preventative maintenance needs, a multi-disease coordinated treatment plan is developed based on strong association rules, avoiding the limitations of a single measure and reducing lifecycle maintenance costs. For example, based on the association rule of "symbiosis of transverse cracks and strip repairs," the surface seal can be strengthened simultaneously with crack treatment to prevent further damage to the repaired area. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments with reference to the following drawings:

[0033] Figure 1 is a comparison chart of the running time of an embodiment of the present invention;

[0034] Figure 2 This is a statistical diagram of the distribution of disease types and degrees according to an embodiment of the present invention;

[0035] Figure 3 is a scatter plot of association rule distribution according to an embodiment of the present invention;

[0036] Figure 4 These are the first 20 groups of strong association rule results of the embodiment of the present invention. DETAILED DESCRIPTION

[0037] In order to make the purpose, features, and advantages of this application more obvious and easy to understand, the technical solutions in the embodiments of this application are clearly and completely described. Obviously, the embodiments described below are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0038] A method for analyzing the correlation between asphalt pavement defects based on data mining includes the following steps:

[0039] Step S1, data preprocessing: obtaining asphalt pavement disease detection data, deleting road section data with no disease or only a single disease, and forming an initial transaction data set D containing two or more diseases;

[0040] Step S2, constructing a Boolean matrix: sorting the initial transaction data set D by the column of the damage type and the row of the mileage pile number, constructing an m×n Boolean matrix M, where m is the number of transactions, n is the number of damage types, and the presence of a damage is marked as 1, otherwise it is marked as 0;

[0041] Step S3: Improved Apriori algorithm to mine frequent itemsets:

[0042] Scan the Boolean matrix M for the first time, calculate the support of all single item sets, and filter out the frequent 1-item sets L1 whose support is greater than or equal to the minimum support threshold;

[0043] Generate candidate item sets based on the frequent 1-item set L1 and optimize through row and column compression:

[0044] Row compression: If an item set is not frequent, its superset is directly deleted;

[0045] Column compression: If the number of transactions is less than k, delete transactions that cannot form a frequent k-item set;

[0046] Scan the compressed matrix for the second time, calculate the support of the candidate item set, generate the frequent k-item set Lk, repeat until no k+1-item set can be generated, and obtain all frequent item sets;

[0047] Step S4, generating strong association rules: calculating the confidence and lift for all the frequent itemsets, and screening strong association rules that simultaneously meet the confidence ≥ minimum confidence threshold and lift ≥ 1.25;

[0048] Step S5: Formulate a preventive maintenance strategy: Based on the strong association rule results, identify key diseases and high-frequency induced diseases, and formulate a combination of maintenance measures.

[0049] Specifically, in this embodiment, the types of defects include cracks, deformations, and other types, specifically covering 15 types of defects and their light and heavy degrees, including cracks, block cracks, longitudinal cracks (light / heavy), transverse cracks (light / heavy), potholes, looseness, rutting, oil spills, strip repairs, and block repairs, among which the light and heavy degrees are marked as L and H, respectively.

[0050] Specifically, in the row-column compression optimization of this embodiment, row compression is based on the anti-monotonicity of frequent itemsets: if an itemset is infrequent, all its supersets are infrequent itemsets; column compression is based on the correlation between the number of transactions and the dimension of frequent itemsets. When the number of transactions is less than k, the corresponding transactions are deleted.

[0051] Specifically, in the present embodiment, during the construction of the Boolean matrix M, the types of defects are sorted in columns according to a preset order, which includes crack types first, deformation types second, and other types last. Each type of defect is arranged in order of severity and the first letter of the pinyin of the name.

[0052] Specifically, in this embodiment, the support calculation formula is:

[0053] The confidence calculation formula is: The lift calculation formula is:

[0054] Specifically, in this embodiment, the minimum support threshold is set in the range of 0.05 to 0.2, the minimum confidence threshold is set to 0.5 to 0.8, and the lift threshold is set to 1.25 to 2.0. The specific parameters are determined by scatter plot analysis based on the disease data distribution.

[0055] Specifically, in this embodiment, after the strong association rules are generated, the rules are visualized, and a visualization map of the association rules is formed by mapping the enhancement degree with color and mapping the support degree with point size, which intuitively displays the association strength between diseases.

[0056] Specifically, in this embodiment, when formulating the preventive maintenance strategy, combined maintenance measures are taken for the strong association rules with a confidence level ≥ 70%. The combined maintenance measures include combining crack grouting with surface sealing, and combining milling and resurfacing with anti-cracking layer laying.

[0057] Specifically, in the solution of this embodiment, the improved Apriori algorithm uses binary bit operations instead of set operations when generating candidate item sets, and expresses the existence of item sets through bit masks, thereby improving operation efficiency.

[0058] Specifically, in the present embodiment, in the data preprocessing step, the damage data is cleaned to remove noise data with abnormal mileage pile numbers and inconsistent damage degree labels, thereby ensuring the integrity and accuracy of the initial transaction data set D.

[0059] In summary, the present invention provides a method for analyzing the correlation between asphalt pavement defects based on data mining. The core steps are:

[0060] 1. Data Preprocessing

[0061] Data collection: Obtain asphalt pavement disease detection data, covering 15 types of diseases, including cracks (cracking, block cracking, longitudinal / transverse cracks and light / severe cracks), deformations (potholes, looseness, rutting), and other types (oil overflow, strip / block repairs), with light and severe levels marked as L and H, respectively.

[0062] Data cleaning: Remove noise data with abnormal mileage pile numbers and inconsistent damage degree markings, and delete road section data with no damage or only a single damage, to form an initial transaction data set D containing two or more damages.

[0063] 2. Boolean matrix construction

[0064] Structured processing: Sort the dataset D by the damage type column (cracks first, deformations second, and other types last; similar damages are sorted by severity and initial pinyin letters), sort the mileage pile number range data by row, and construct an m×n Boolean matrix M (m is the number of transactions, n is the number of damage types).

[0065] Binary mapping: If the i-th disease exists in the j-th transaction, the matrix element M[i,j] = 1, otherwise it is 0, forming a binary representation of the disease existence.

[0066] 3. Improved Apriori algorithm for mining frequent itemsets

[0067] First scan and single item set screening:

[0068] The Boolean matrix M is scanned for the first time, the support of all single-item sets is calculated (Support(X) = number of transactions containing X / total number of transactions), and frequent 1-item sets L1 with support ≥ minimum support threshold (0.05-0.2) are screened out.

[0069] Row and column compression optimization:

[0070] Row compression: Based on the anti-monotonicity of frequent itemsets, if an itemset is infrequent, all its supersets are directly deleted to reduce the number of candidate itemsets.

[0071] Column compression: When the number of transactions is less than k, delete transactions that cannot form frequent k-itemsets to reduce the matrix dimension.

[0072] Iteratively generate frequent itemsets:

[0073] The compressed matrix is ​​scanned for the second time, candidate item sets are generated based on L1, support is calculated and frequent k-item sets Lk are screened, and this process is repeated until no k+1-item sets can be generated, and all frequent item sets are obtained.

[0074] 4. Generation of strong association rules

[0075] Three indicators calculation:

[0076] Confidence (Confidence(X→Y) = Number of transactions containing both X and Y / Number of transactions containing X) measures conditional probability;

[0077] Lift (X→Y) = confidence / support (Y)) measures the reliability of the association;

[0078] Rule screening: Set the minimum confidence threshold (0.5-0.8) and the lift threshold (≥1.25) to screen strong association rules from the frequent item set to reveal the disease symbiosis relationship and induction probability.

[0079] 5. Maintenance strategy formulation

[0080] Key disease identification: Based on the cause and effect of strong association rules, key diseases that are induced frequently (such as severe longitudinal cracks CLC(H)) and high-frequency diseases that are easily induced (such as mild transverse cracks CTC(L)) are located.

[0081] Formulate a combination of measures: Based on the rules with a confidence level of ≥70%, formulate a multi-disease coordinated treatment plan, for example:

[0082] For the “CTC(L)→CLC(H)” rule, crack grouting and anti-crack layer laying are carried out simultaneously;

[0083] For the “CTC(L), CTC(H)→DSP” rule, a surface sealing and strip repairing process is adopted.

[0084] Technological innovations

[0085] Row and column compression optimization: Through anti-monotonic pruning and dimension screening, the multiple scans of the traditional Apriori algorithm are reduced to two, significantly reducing computational complexity.

[0086] Multi-dimensional disease modeling: Incorporating 15 types of diseases and their severity levels, a high-dimensional disease association model is constructed, covering multiple types of interactive relationships such as cracks, deformation, and repairs.

[0087] Binary operation acceleration: Use bit masks to represent item set existence, replace set operations with binary bit operations, and improve the efficiency of candidate item set generation.

[0088] Application Value

[0089] This method achieves quantitative analysis of the symbiotic pattern of asphalt pavement diseases through standardized data processing procedures, efficient algorithm optimization, and precise rule screening, providing a traceable and verifiable scientific basis for maintenance decisions and promoting the transformation of preventive maintenance from experience-driven to data-driven.

[0090] Example:

[0091] A method for analyzing the correlation between asphalt pavement defects based on data mining, the research method includes:

[0092] Association rules

[0093] Association rules are used to represent the degree of association between item sets in a database. Association rule algorithms use a large amount of data in the database to find the relevant attributes of item sets. The description of the association rule mining algorithm is as follows:

[0094] I={I1,I2,…I m} is a set of data items, T={T1,T2,…T n} is a set of transactions, where T is a subset of the entire data set, i.e.

[0095] First, frequently occurring items are selected from a set of candidate items, generating frequent itemsets based on the high-frequency occurrences. Then, within the largest frequent itemset, association rules are searched for with a confidence level no less than a threshold. Users can obtain desired strong association rules by configuring the three association rule algorithm parameters: support, confidence, and lift.

[0096] Support refers to the ratio of the number of occurrences of a particular item set to the total number of occurrences. Higher support indicates a greater probability of occurrence for that combination, as shown in Equation (2-1). This metric serves as the first threshold for association rules. By setting a minimum support threshold, we can eliminate meaningless rules with low "appearance rates." Consequently, we retain the frequent item set Z, as shown in Equation (2-2).

[0097]

[0098] Support(Z)≥min_support (2-2)

[0099] Confidence is the second threshold for establishing strong association rules. It is expressed as the probability of Y appearing under the condition that the premise X of the association rule appears, that is, the probability of the item set of X and Y appearing at the same time. It is called the confidence level of the association rule X, as shown in formula (2-3).

[0100]

[0101] Lift is the ratio of the probability of including Y in the item set under condition X to the probability of including Y in the item set without condition X. This metric can be considered a complementary metric to confidence, also used to measure the reliability of the association rule, as shown in Equation (2-4).

[0102]

[0103] In practical applications, associations with high support and confidence levels can serve as useful association rules, which are defined by the minimum support threshold (min_support) and the minimum confidence threshold (min_confidence). Only data item sets that meet the min_support threshold appear in association rules; these data item sets are called frequent itemsets. Min_confidence represents the minimum reliability of an association rule, and only rules that meet both min_support and min_confidence are considered strong association rules. The task of association rule mining is actually frequent pattern mining, with the Apriori algorithm and FP-Growth algorithm being classic association rule algorithms.

[0104] Association rule algorithms

[0105] Apriori algorithm

[0106] The Apriori algorithm is one of the most influential algorithms for mining frequent itemsets in association rules. It uses an iterative, breadth-first, layer-by-layer search approach. First, it finds the set of frequent 1-item sets, denoted as L1. L1 is then used to find the set of frequent 2-item sets, and L2 is then used to find L3. This iteration continues until no frequent k+1 itemsets can be found, at which point the algorithm terminates. However, in large datasets, the Apriori algorithm may generate a large number of candidate sets, and finding each Lk requires rescanning the database. As the database becomes larger, its runtime increases accordingly. Finally, all possible itemsets are displayed through permutations and combinations, and the support of each item set is calculated.

[0107] FP-growth algorithm

[0108] The FP-Growth algorithm is a well-known method for addressing the efficiency issues of the Apriori algorithm. Both are classic association rule algorithms. The FP-growth algorithm does not generate candidate patterns when generating frequent itemsets. Instead, it compresses the provided frequent itemset data into a frequent pattern tree (FP-Tree), while still preserving the itemset association information. First, a database scan is performed to obtain the counts of all items. Frequent single-item sets are retained based on a minimum support threshold and arranged in descending order to form an item header table. Next, a root node is created, followed by a second database scan, with each corresponding item inserted into the tree. The count defaults to 1; if the node already exists, the count is incremented by 1. Finally, frequent itemsets are mined sequentially from the item header table, starting from the bottom up. The elements used to find frequent itemsets serve as conditional pattern bases, and frequent itemsets with the same base are then retrieved based on the conditional pattern bases.

[0109] The advantage of the FP-Growth algorithm over the Apriori algorithm is that even with large amounts of data, only two scans are required, significantly improving operational efficiency. However, the Apriori algorithm is more scalable and can be used in parallel computing.

[0110] Improved Apriori algorithm

[0111] The improved Apriori algorithm can be achieved by reducing the number of candidate itemsets generated or the number of transactions that need to be compared when obtaining the support of frequent itemsets. In the process of generating frequent itemsets, the Apriori algorithm needs to scan the dataset multiple times. By removing irrelevant transaction items, the number of scans can be reduced to a certain extent, thereby improving operation efficiency.

[0112] Optimized algorithm steps:

[0113] Traverse the database once, removing data records for sections with only one defect or no defects. The total number of recorded transactions is m, and the dataset is D. When Dx(x=1,2…,m)=1, Dx is deleted and the traversal is repeated to obtain a new dataset D′.

[0114] After constructing the new data set D′, it is compressed into a matrix Mat by rows and columns, where the disease types are arranged in columns T = {T1, T2, ...T n} sorting, the disease data of the mileage pile number range is sorted by row I={I1,I2,...I m} are sorted, and the matrix Mat is shown in formula (2-5):

[0115]

[0116] If the i-th item set is in the j-th transaction, then the matrix d ij =1, otherwise it is 0, forming a Boolean matrix.

[0117] The support of each itemset is calculated using a Boolean matrix to obtain an itemset index table. The obtained itemset support is then compared with a set minimum support threshold to determine the frequent itemsets. The properties of frequent itemsets indicate that if an itemset is infrequent, its superset is also not a frequent itemset and can be deleted directly. This is called row compression.

[0118] Each transaction in the Boolean matrix corresponds to a column vector. If the number of transactions is less than k, no frequent item set Lk will be formed and the transaction will be deleted during the search, which is column compression.

[0119] Scan the compressed Boolean matrix again to calculate the support and create an index table. Repeat the above steps until k+1 frequent item sets are found. All the frequent item sets obtained are presented in the form of an index table.

[0120] Finally, the strong association rule parameters are set, and the confidence and lift are calculated for all the frequent item sets obtained in the form of an index table to generate strong association rules.

[0121] The optimized algorithm only needs to scan the database twice to replace data records with coded sets, effectively avoiding the repetitive steps of multiple database scans. Row and column compression is performed using Boolean matrices, and candidate item sets are replaced with index tables during the scanning process, avoiding the generation of a large number of candidate item sets. During the execution of the Apriori algorithm, binary operations are used instead of operations between sets, improving the algorithm's efficiency.

[0122] Comparison of algorithms before and after improvement

[0123] Models for the Apriori algorithm, the FP-Growth algorithm, and an improved Apriori algorithm were created. The program was written using Python 3.11 on the Jupyter Notebook data analysis platform to simulate and analyze various algorithm parameters, including support, confidence, and lift. Based on the experimental results, a highly applicable algorithm was selected and its parameters were set appropriately to mine association rules for preventive maintenance of asphalt pavement damage data. The simulation hardware environment consisted of a 12th Gen Intel(R) Core(TM) i5-12450H CPU @ 2.00GHz and 16.0GB of RAM.

[0124] Input sample data and modeling parameters, output association rules between pavement diseases, and compare the running time of Apriori algorithm, FP-Growth algorithm and improved Apriori algorithm under different parameter settings, such as Figure 1 As shown:

[0125] Figure 1 The changes of the Apriori algorithm and FP-Growth algorithm before and after improvement with the change of minimum support are compared. Figure 1 As can be seen, the runtime of the improved algorithm decreases significantly as support increases. When the minimum support threshold is small, the runtime of the improved Apriori algorithm is shorter than both the pre-optimized Apriori algorithm and the FP-Growth algorithm. Furthermore, the larger the minimum support, the stronger the association rule and the shorter the runtime.

[0126] In summary, the improved Apriori algorithm significantly outperforms the unmodified Apriori algorithm and the FP-Growth algorithm on the same dataset, validating its effectiveness. Therefore, this paper conducts modeling and analysis based on the improved Apriori algorithm, further exploring the correlations between highway asphalt pavement defects, exploring the interactions between defects, and the mechanisms that induce latent defects.

[0127] Asphalt pavement disease data and models

[0128] To minimize the impact of varying lane damage severity, this study selected asphalt pavement main lanes from five different road sections for research, randomly sampling 359 units of pavement damage data. The main damage types used to evaluate the asphalt pavement damage index are: cracks (cracking, block cracking, longitudinal cracking, and transverse cracking), deformations (potholes, looseness, and rutting), and other types (oiling, strip repairs, and block repairs). This study, using association rule algorithms, examines the relationships between different damage types and the varying severity of each, providing a basis for developing scientific preventive maintenance strategies.

[0129] Data processing

[0130] A total of 15 major types of defects and their severity were extracted from the actual defect detection data of the main lanes of asphalt pavements on five different road sections, as shown in Table 1. The transaction set of defect types and severity in the dataset is the item set of the actual data recorded within the mileage pile number range.

[0131] Table 1 Asphalt pavement damage index disease type and degree

[0132]

[0133] Note: L stands for mild and H stands for severe.

[0134] 359 sample units were randomly selected from all the disease detection databases to form an association rule mining database, as shown in Table 2. The distribution statistics of these 15 items were performed, and it was found that the top three disease types in terms of the number of diseases in these sample units were: strip repair, transverse cracks (serious), and longitudinal cracks (serious). The disease distribution details are shown in Table 2. Figure 2 shown.

[0135] Table 2 Asphalt pavement disease database

[0136]

[0137] Data Model

[0138] Statistics based on these sample units revealed that several road sections had no defects or only a single defect. Therefore, data preprocessing was performed on the transaction dataset, removing sections with no defects or only a single defect. The analysis then continued on sections with two or more defects. The pavement defect database was converted into an association rule database, as shown in Table 3.

[0139] Table 3 Asphalt pavement disease association rule model dataset

[0140]

[0141] In order to obtain reasonable support and confidence, a total of 458 groups of association rules were statistically investigated. These grouped item sets are intuitively presented based on scatter plots, such as Figure 3 The results show that the majority of lift values ​​are above 1.25, while support values ​​are mostly distributed between 0.05 and 0.2. Therefore, we set the minimum support threshold of the model to 0.1 for association rule analysis. Then, we set the minimum confidence value to 0.5 and the lift value to 1.25 as input parameters to obtain frequent itemsets.

[0142] Results and discussion

[0143] Asphalt pavement disease results

[0144] Based on the optimized Apriori algorithm model, the top 20 strong association rules for asphalt pavement disease types were extracted. The association rules between different individual diseases were sorted in descending order by the number of occurrences. Association rules with high confidence levels are considered to have a significant role in inducing factors. For example, the confidence level for "CTC(L), CTC(H)" -> "CLC(H), DSP" is 72.5%, indicating that 72.5% of cells with stripe repairs and severe longitudinal cracks also have mild and severe transverse cracks. Therefore, the confidence level that the coexistence of mild and severe transverse cracks induces stripe repairs and severe longitudinal cracks is 72.5%. The table also shows that among all strong association rules, "CTC(L)" -> "CLC(H)" occurs the most frequently in the dataset, 52 times. This indicates that 52 of the cells with mild transverse cracks in the dataset also have moderate to severe longitudinal cracks. Furthermore, the confidence level that the presence of mild transverse cracks induces severe longitudinal cracks is 62.7%. The most frequent occurrence of strong association rules for defects indicates that this defect combination is the most likely to occur in the actual pavement defects investigated and is also the factor that has the greatest impact on pavement smoothness. In terms of pavement comfort, smoothness is a key indicator of comfort. Identifying the key defects affecting smoothness and how to maintain it are key priorities for current maintenance efforts. From the perspective of maintenance professionals, early maintenance efforts should not only rationally select a combination of maintenance measures based on actual pavement defect results rather than relying on a single measure, but also fully consider the potential for existing defects to induce other defects under long-term, complex conditions.

[0145] Table 4 Results of the top 20 strong association rules

[0146]

[0147] Visualize these 20 sets of strong association rules, such as Figure 4 As shown in the figure. Lighter colors indicate lower lift for the association rule, meaning lower confidence; smaller dot sizes indicate lower support for the association rule. The top three strong association rules with the highest confidence are: "CLC(L)->CTC(L)," "CTC(L), DSP->CLC(L)," and "CTC(L)->CLC(L)." Although "CLC(L)->CTC(L)" and "CTC(L)->CLC(L)" both have similarly high confidence, the figure shows that "CTC(L)->CLC(L)" has higher support than "CLC(L)->CTC(L)." Therefore, both defects have an interactive relationship, and the likelihood of mild transverse cracks causing mild longitudinal cracks is higher than the likelihood of mild longitudinal cracks causing mild transverse cracks.

[0148] Comparing the antecedents of strong association rules reveals that severe longitudinal cracks (CLC(H)) have the highest probability of coexisting with other defects, making them one of the primary factors influencing their development into other defects. Analysis of strong association rules reveals six sets of rules that contribute to the occurrence of mild transverse cracks (CTC(L)), indicating that mild transverse cracks are most likely to be induced by other defects. By identifying the defects most likely to affect other defects and the defects most likely to be induced by them through association analysis, maintenance professionals should prioritize addressing these defect types and severity levels when formulating preventive maintenance plans, thereby improving the sustainability and accuracy of maintenance plans.

[0149] Inducing mechanism of hidden diseases of asphalt pavement

[0150] The results of the strongly correlated defect combinations in the previous section indicate that mild transverse cracks, mild longitudinal cracks, and strip-shaped repairs occur in a large number of units, with a high probability of transforming into potholes, rutting, a network of cracks, and severe unidirectional cracks. Asphalt mixtures gradually age under complex environments and loads, surface water seeps into the asphalt mixture through cracks. The compressive stress of vehicle loads and the vacuum suction generated by high-speed driving create a repeated pumping action, exacerbating loosening and threshing. Tire rollup and water scour weaken the bond between asphalt and aggregate, leading to the development of potholes in the asphalt mixture. At this point, surface water infiltrates not only from reflective cracks from below but also through potholes, causing water damage, weakening local structural strength, and continuously compacting wheel tracks, resulting in rutting. The interaction of these various defects ultimately leads to structural problems that compromise road safety, such as subsidence or a network of cracks.

[0151] Each pavement disease and the different degrees of the same disease are continuously evolved from early diseases. The induction mechanism of this latent disease always exists with the use of the pavement. In general, the evolution characteristics of asphalt pavement diseases are from light to heavy, from shallow to deep, from surface to inside, from slow to fast, and from non-structural damage on the surface to structural damage inside the pavement

[18] . From the evolution results of asphalt pavement diseases, it can be seen that cracking is the main form of disease that appears in the early stage of the pavement, and the disease combination formed by the accompanying diseases is very important for us to formulate scientific maintenance plans, especially early preventive maintenance measures. If scientific maintenance is not carried out during the early investigation and maintenance process, the degree and area of ​​the disease will increase exponentially, and the pavement structural performance will be damaged, which will increase the difficulty of maintenance and repair, and waste a lot of money and time costs. Scientific pavement maintenance management throughout the entire service life of the pavement is one of the key technologies for developing long-life pavements, slowing down the process of one disease inducing another or a more serious disease. Therefore, understanding the initial and development stages of asphalt pavement diseases and comprehensively utilizing maintenance measures are the focus of preventive maintenance research and the key points to prevent the transformation of various diseases.

[0152] in conclusion

[0153] By conducting an association rule algorithm test on the main lane defects of asphalt pavement on five different highways, the following conclusions were obtained:

[0154] The Apriori algorithm is a traditional association rule algorithm. Another traditional algorithm, the FP-Growth algorithm, has an advantage over the Apriori algorithm in that it only requires two scans, even for large data volumes, significantly improving operational efficiency. To address the Apriori algorithm's inefficiency, an optimization algorithm was proposed that reduces the number of scans required by the original Apriori algorithm.

[0155] The computational efficiency of the Apriori algorithm and the FP-Growth algorithm before and after optimization was analyzed and compared. By setting different minimum support thresholds and comparing the changes in computational time, it was found that as the minimum support increases, the algorithm runtime ranks as follows: optimized Apriori > FP-Growth > Apriori. Furthermore, the greater the support, the stronger the association rule and the shorter the runtime.

[0156] By using the asphalt pavement disease database and selecting appropriate support, confidence and lift using an improved algorithm, an asphalt pavement disease association rule model was established. The strong association rules obtained reflect the main types and degrees of pavement diseases, as well as the conversion probability and evolution relationship between each disease and different degrees of disease.

[0157] Combining the strong association rule model of asphalt pavement diseases with the scientific formulation of preventive maintenance plans, the induction mechanism of hidden pavement diseases is explained, and a reasonable combination of maintenance measures is formulated to reduce the possibility of one disease transforming into another.

[0158] In summary, the present invention has the following beneficial effects:

[0159] Efficiency: Through row and column compression and Boolean matrix optimization, the multiple database scans of the traditional Apriori algorithm are reduced to two, significantly improving computational efficiency. This is particularly suitable for high-dimensional and sparse pavement disease data.

[0160] Accuracy: It can uncover implicit correlations between different disease types and degrees, such as the confidence level of "mild transverse cracks → severe longitudinal cracks" reaching 62.7%, providing an accurate disease evolution path for maintenance decisions.

[0161] Practicality: Integrating preventative maintenance needs, a multi-disease coordinated treatment plan is developed based on strong association rules, avoiding the limitations of a single measure and reducing lifecycle maintenance costs. For example, based on the association rule of "symbiosis of transverse cracks and strip repairs," the surface seal can be strengthened simultaneously with crack treatment to prevent further damage to the repaired area.

[0162] The technical solutions provided by the embodiments of the present invention are introduced in detail above. Specific examples are used herein to illustrate the principles and implementation methods of the embodiments of the present invention. The description of the above embodiments is only applicable to help understand the principles of the embodiments of the present invention. At the same time, for those skilled in the art, according to the embodiments of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the contents of this specification should not be understood as limiting the present invention.

Claims

1. A method for analyzing the correlation between asphalt pavement defects based on data mining, characterized in that: The following steps are involved: Step S1, data preprocessing: obtaining asphalt pavement disease detection data, deleting road section data with no disease or only a single disease, and forming an initial transaction data set D containing two or more diseases; Step S2, constructing a Boolean matrix: sorting the initial transaction data set D by the column of the damage type and the row of the mileage pile number, constructing an m×n Boolean matrix M, where m is the number of transactions, n is the number of damage types, and the presence of a damage is marked as 1, otherwise it is marked as 0; Step S3: Improved Apriori algorithm to mine frequent itemsets: Scan the Boolean matrix M for the first time, calculate the support of all single item sets, and filter out the frequent 1-item sets L1 whose support is greater than or equal to the minimum support threshold; Generate candidate item sets based on the frequent 1-item set L1 and optimize through row and column compression: Row compression: If an item set is not frequent, its superset is directly deleted; Column compression: If the number of transactions is less than k, delete transactions that cannot form a frequent k-item set; Scan the compressed matrix for the second time, calculate the support of the candidate item set, generate the frequent k-item set Lk, repeat until no k+1-item set can be generated, and obtain all frequent item sets; Step S4, generating strong association rules: calculating the confidence and lift for all the frequent itemsets, and screening strong association rules that simultaneously meet the confidence ≥ minimum confidence threshold and lift ≥ 1.25; Step S5: Formulate a preventive maintenance strategy: Based on the strong association rule results, identify key diseases and high-frequency induced diseases, and formulate a combination of maintenance measures.

2. The method for analyzing the correlation between asphalt pavement defects based on data mining according to claim 1, characterized in that: The types of defects include cracks, deformations, and other types, specifically covering 15 types of defects including cracking, block cracking, longitudinal cracks, transverse cracks, potholes, looseness, rutting, oil spills, strip repairs, and block repairs, as well as their light and heavy degrees, where the light and heavy degrees are marked as L and H respectively.

3. The method for analyzing the correlation between asphalt pavement defects based on data mining according to claim 1, characterized in that: In the row and column compression optimization, row compression is based on the anti-monotonicity of frequent itemsets: if an itemset is infrequent, all its supersets are infrequent itemsets; column compression is based on the correlation between the number of transactions and the dimension of frequent itemsets. When the number of transactions is less than k, the corresponding transactions are deleted.

4. The method for analyzing the correlation between asphalt pavement defects based on data mining according to claim 1, characterized in that: During the construction of the Boolean matrix M, the types of defects are sorted in columns according to a preset order, which includes cracks first, deformations second, and other types last. Each type of defect is arranged in order of severity and the first letter of the pinyin of the defect name.

5. The method for analyzing the correlation between asphalt pavement defects based on data mining according to claim 1, characterized in that: The support calculation formula is: The confidence calculation formula is: The lift calculation formula is:

6. The method for analyzing the correlation between asphalt pavement defects based on data mining according to claim 1, characterized in that: The minimum support threshold is set in the range of 0.05 to 0.2, the minimum confidence threshold is set in the range of 0.5 to 0.8, and the lift threshold is set in the range of 1.25 to 2.

0.

7. The method for analyzing the correlation between asphalt pavement defects based on data mining according to claim 1, characterized in that: After the strong association rules are generated, the rules are visualized, and a visualization map of the association rules is formed by mapping the enhancement degree with color and mapping the support degree with point size.

8. The method for analyzing the correlation between asphalt pavement defects based on data mining according to claim 1, characterized in that: When formulating the preventive maintenance strategy, combined maintenance measures are taken for the strong association rules with a confidence level of ≥70%. The combined maintenance measures include combining crack grouting with surface sealing, and combining milling and resurfacing with anti-cracking layer laying.

9. The method for analyzing the correlation between asphalt pavement defects based on data mining according to claim 1, characterized in that: The improved Apriori algorithm uses binary bit operations instead of set operations when generating candidate item sets, and expresses the existence of item sets through bit masks.

10. The method for analyzing the correlation between asphalt pavement defects based on data mining according to claim 1, characterized in that: In the data preprocessing step, the damage data is cleaned to remove noise data with abnormal mileage pile numbers and inconsistent damage degree markings.

Images