A Highway Engineering On-site Management Method Based on Process Inspection

Abstract

This invention provides a highway engineering site management method based on process inspection reporting. It acquires highway construction site data, standards, and technical specifications through multiple channels; utilizes big data analytics to analyze the data, constructs a quality monitoring model to monitor the construction process, compares the data with standard data, predicts whether the quality is up to standard, and generates a quality monitoring report; executes corresponding strategies based on the quality prediction results: if the predicted quality is up to standard, relevant data is entered into the process inspection reporting form; if the predicted quality is down to standard, a risk assessment is conducted based on the quality monitoring report, and the risk assessment results are scored, triggering early warnings based on the scores; it analyzes correlations to pinpoint the influencing factors of non-compliance, and implements rectification and control measures based on the identified factors; this method enables quality monitoring of the project during the process inspection reporting process, improving project quality and facilitating later data review.

Description

Technical Field

[0001] This invention relates to the field of building information management technology, and more particularly to a method for on-site management of highway engineering based on process inspection. Background Technology

[0002] Highway engineering site management based on process inspection refers to the standardized inspection and acceptance of construction processes through digital means to ensure that the project quality meets the specifications. However, current technologies are lacking in the management of highway engineering data, especially in the process inspection procedures.

[0003] For example, inspection applications, quality inspection forms, and images of concealed works are still mainly in paper or Excel format. This requires on-site staff to fill out the forms, take photos, print, sign, scan, and archive them, taking an average of 2-3 days per instance. The process relies heavily on manual labor, resulting in low efficiency. Furthermore, the reliance on manual inspections for quality checks leads to fragmented results, increasing the risk of missed inspections and misjudgments, and making it difficult to create a comprehensive quality archive throughout the entire project lifecycle. The long cycle from project inspection to acceptance makes it impossible to detect quality anomalies in real time. In addition, the inconsistent data formats uploaded by different personnel prevent the correlation of key data, making it difficult to provide timely warnings of quality defects, pinpoint responsibility, and trace data.

[0004] Therefore, the technical problem to be solved by this invention is: how to monitor the quality of the project during the process of inspection and acceptance, improve the quality of the project, and facilitate data backtracking in the later stage. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a highway engineering on-site management method based on process inspection. By establishing a quality monitoring model, it is possible to detect potential quality problems in advance, implement timely rectification, improve the project's pass rate, and ensure a unified data format, making the data traceability process simple and fast.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] This invention provides a method for on-site management of highway engineering based on process inspection, comprising the following steps:

[0008] S1: Obtain highway construction site data, standards and technical specifications through multiple channels, and preprocess the collected data. The construction site data includes construction data, process flow data, equipment data and personnel data.

[0009] S2: Utilize big data analytics to analyze the collected data, construct a quality monitoring model to monitor the construction process, compare it with standard data, predict whether the quality is up to standard, and generate a quality monitoring report.

[0010] S3: Execute the corresponding strategy based on the quality prediction results. If the predicted quality is qualified, fill in the relevant data in the process inspection application form. The relevant data includes physical parameters and process parameters during construction. If the predicted quality is unqualified, conduct a risk assessment in conjunction with the quality monitoring report, score the risk assessment results, and trigger an early warning based on the score.

[0011] S4: Based on correlation analysis, identify the influencing factors of non-compliance and carry out rectification and prevention based on the specific factors identified.

[0012] Furthermore, a cloud-based data collaboration platform is constructed to receive collected data via the network, store the data using relational and non-relational databases, establish relationships between data, and perform cross-regional synchronization.

[0013] S2 includes the following steps:

[0014] S21: Based on the association rule mining algorithm, analyze the potential associations between highway construction site data;

[0015] S22: Use clustering algorithms to classify quality data during construction and group processes with similar quality characteristics into one category;

[0016] S23: Using highway construction site data and environmental data as input, and engineering quality acceptance results as output labels, construct a machine learning model and train it using historical data;

[0017] S24: Construct a comprehensive quality monitoring indicator system based on the highway engineering quality acceptance specifications and design requirements;

[0018] S25: Combine big data analytics with a quality monitoring indicator system to build a quality monitoring model, and incorporate the association rules mined in S21 and the cluster features of clustering in S22;

[0019] S26: Input the current highway construction site data and environmental data into the quality monitoring model to predict the current project quality acceptance results, and automatically generate a quality inspection report.

[0020] Furthermore, step S22 includes the following steps:

[0021] S221: Collect process quality data and extract quality features, including numerical quality features and indirect quality features, and use correlation analysis to retain core features that are highly correlated with quality.

[0022] S222: Combining the differences in quality characteristics of highway engineering processes, the elbow method is used to determine the number of clusters, and clustering is performed based on the K-Means algorithm;

[0023] S223: Perform statistical analysis on the quality data of each cluster, extract common characteristics, and analyze common influencing factors;

[0024] S224: Iterate the clustering process continuously and update the cluster features.

[0025] Furthermore, step S223 includes the following steps:

[0026] S2231: Align the quality data of all processes within the same cluster in terms of time and space, and perform a second screening of the data within the cluster to remove outliers;

[0027] S2232: Calculate descriptive statistics for each quality index within the cluster, and extract common quality characteristics based on the statistical results;

[0028] S2233: By combining the influencing factor data corresponding to the processes within the cluster, we can identify the common causes that lead to common quality characteristics;

[0029] S2234: Verify the correlation between the identified common features and influencing factors;

[0030] S2235: Extract cluster feature labels, sort them by influence intensity, and generate a list of common influences.

[0031] The quality monitoring indicator system constructed by S24 is used as the judgment standard for the model. Specifically, it includes indicator thresholds and indicator weights. The model needs to compare the machine learning prediction results with the thresholds of the indicator system, and calculate the comprehensive quality score by combining the indicator weights. The calculation formula is as follows:

[0032] Z = ∑α × G;

[0033] Where Z represents the overall quality score, α represents the weight, and G represents the indicator compliance rate.

[0034] Furthermore, when the quality monitoring model predicts that the construction quality is substandard, a risk assessment is conducted based on the quality inspection report and from three dimensions: severity, scope of impact, and probability of occurrence. Each dimension is quantified into a score, and the weight of each dimension is determined using the analytic hierarchy process (AHP). The comprehensive risk score is calculated using the following formula:

[0035] M = Attention(S)×S+Attention(R)×R+Attention(P)×P;

[0036] Where Attention(S), Attention(R), and Attention(P) represent the weights of each item, M represents the risk score, S represents the severity, R represents the scope of influence, and P represents the probability of occurrence.

[0037] Furthermore, an attention mechanism is introduced to dynamically calculate the weights of severity, scope of impact, and probability of occurrence, including the following steps:

[0038] Collect historical risk assessment data that does not meet quality standards. Each data point contains quantitative values ​​for three dimensions: severity, scope of impact, and probability of occurrence, as well as corresponding risk consequence labels. The features are then preprocessed.

[0039] We construct a lightweight model that includes a feature embedding layer, an attention computation layer, and a weight output layer, and map features into high-dimensional feature vectors to enhance the model's ability to capture the correlation between features.

[0040] Global features strongly correlated with risk consequences are used as query vectors to calculate attention scores. The scores are then normalized using the Softmax function to obtain attention weights.

[0041] Attention weights are combined with original features to predict risk consequences through a fully connected layer and output weight values.

[0042] S4 includes the following steps:

[0043] S41: Identify the dimensions of potential influencing factors and refine specific indicators, linking the specific manifestations of quality non-compliance with the influencing factor data to form an analysis dataset;

[0044] S42: Using the quantitative indicators of quality non-compliance as the dependent variable and the indicators of each factor as independent variables, construct a multiple linear regression model, and judge the intensity of influence by the magnitude of the absolute value of the regression coefficients;

[0045] S43: Calculate and rank the correlations of each factor based on SHAP technology;

[0046] S44: Based on the results of correlation analysis and combined with on-site investigation, accurately locate the root cause of the problem;

[0047] S45: Based on the specific factors of the positioning, formulate an executable and traceable rectification plan and clarify the rectification elements;

[0048] S46: The construction unit shall implement the rectification plan and upload the progress in real time.

[0049] Furthermore, the method also includes:

[0050] S5. Based on BIM technology, a visualization platform is built to display the overall construction status and key indicators in a visual way in real time, and to provide data interaction and sharing functions between different user roles.

[0051] The beneficial effects of this invention are as follows:

[0052] This invention provides a highway engineering site management method based on process inspection. By acquiring highway construction site data, standards, and technical specifications from multiple channels, it enables comprehensive data analysis and processing. The collected data is standardized, facilitating subsequent data comparison and improving the accuracy of data analysis. Big data analytics is used to analyze the collected data, constructing a quality monitoring model to monitor the construction process. By comparing with standard data, the method predicts whether the quality is up to standard. Quality monitoring during construction allows for timely detection and correction of quality problems, improving overall project quality. Corresponding strategies are implemented based on the quality prediction results. If the predicted quality is up to standard, a process inspection application form is completed; if the predicted quality is down to standard, a risk assessment is conducted based on the quality monitoring report, and the risk assessment results are scored, triggering early warnings. By assessing the risks of predicted quality problems and classifying risk levels based on the results, different countermeasures are taken for different risk levels, ensuring that risks are addressed early and improving project safety. Correlation analysis is used to pinpoint the influencing factors of non-compliance. Rectification and prevention are implemented based on the identified specific factors. Targeted rectification of the identified factors yields excellent results and saves resources. This invention achieves quality control of highway engineering during the construction and inspection process through the above-mentioned method. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of a highway engineering on-site management method based on process inspection according to the present invention. Detailed Implementation

[0054] It should be noted that this invention primarily addresses the management of on-site data in highway engineering projects, dividing the process into several steps, including construction layout, final pile hole drilling, reinforcement cage processing and installation, pre-pouring acceptance, concrete pouring, and pile completion acceptance. Among these, the concrete pouring step is most prone to quality problems, which are often subtle and difficult to detect. This is due to the characteristics of the process, its complexity, and various on-site influencing factors. For example, deviations in the on-site execution of the concrete mix design (such as failure to adjust the moisture content of sand and gravel in real time, resulting in an excessively high water-cement ratio); or failure to cure the concrete promptly after initial setting (especially in high-temperature environments where water loss is too rapid) can lead to insufficient strength, failing to meet the design strength grade, and the pile foundation being unable to withstand the superstructure load, posing a structural safety hazard. Problems in this step directly affect the quality of the entire project, highlighting its importance. However, existing technologies rely solely on manual inspection, making it difficult to detect inherent quality problems and carrying the risk of missed inspections and misjudgments. Therefore, how to achieve quality inspection of concrete pouring during construction is the main problem this invention aims to solve.

[0055] Please see Figure 1As shown, this invention relates to a method for on-site management of highway engineering based on process inspection, comprising the following steps:

[0056] S1: Obtain highway construction site data, standards and technical specifications through multiple channels, and preprocess the collected data. The construction site data includes construction data, process flow data, equipment data and personnel data.

[0057] In order to fully understand the situation at the construction site, it is necessary to collect data from multiple aspects for analysis. Specifically, this invention collects data from three aspects: highway construction site data, standards, and technical specifications. The collected relevant standards are used to compare subsequent data to make the comparison results more accurate. For example, the standards and construction technical specifications include the "Highway Engineering Quality Inspection and Evaluation Standard", "Highway Engineering Technical Standard", "Highway Subgrade Construction Technical Specification", "Highway Pavement Base Construction Technical Details", "Highway Bridge and Culvert Construction Technical Specification" and "Highway Tunnel Construction Technical Specification".

[0058] The highway construction site data includes construction data, process flow data, equipment data, and personnel data:

[0059] Construction data includes environmental data from the construction site, allowing for the deployment of various types of sensors to acquire key parameters. For example, pressure and temperature sensors can be installed in the concrete pouring area. Pressure sensors can accurately monitor pressure changes during pouring, reflecting the density of the concrete; temperature sensors can detect temperature fluctuations during pouring and curing, preventing quality defects such as cracks caused by abnormal temperatures.

[0060] For process flow data, specialized management software, such as a smart construction site app, can be developed to cover the entire process from construction preparation to final acceptance. Based on highway engineering construction specifications and the actual needs of the project, the software sets the sequence, logical relationships, and inspection milestones for each process. After completing each process, the construction unit submits an online inspection application through the software, detailing the start and end times, construction team, and other information. Upon receiving the application, the supervision unit records the review time, acceptance results, and rectification opinions in the software, ensuring a complete traceability of the process flow data.

[0061] For equipment data, the machinery and equipment at the construction site are connected to the management system via IoT technology. Equipment operating data is directly collected from the equipment's control system, such as the paving speed and thickness of the paver, the number of passes and compaction speed of the roller, and the raw material ratio and mixing time of the mixing plant. This data intuitively reflects the equipment's operating status and construction process parameters, providing crucial information for judging the quality of each process. Simultaneously, equipment fault alarm information is collected to facilitate timely maintenance and ensure construction continuity.

[0062] For personnel data, construction and supervision personnel are equipped with multi-functional mobile smart terminals. A customized app is used to input personnel information, including basic details such as name, job type, qualification certificate number, and contact information, as well as dynamic information such as attendance records and training status. During construction, the app's location function is used to obtain real-time personnel location information, facilitating personnel scheduling and management.

[0063] Furthermore, a cloud-based data collaboration platform is constructed to receive collected data via the network, store the data using relational and non-relational databases, establish relationships between data, and perform cross-regional synchronization.

[0064] Leveraging the powerful computing capabilities and storage resources of cloud computing, the platform processes and stores the collected data. It can also scale elastically according to business needs. Utilizing cloud computing's distributed computing capabilities, it enables rapid analysis and statistical processing of large amounts of data. Since highway engineering involves multiple parties, including owners, designers, construction companies, supervisors, and material suppliers, a unified data collaboration platform is built to ensure that all participants can receive project information synchronously. This platform provides unified data storage and establishes connections between data points, facilitating data retrieval and traceability.

[0065] Before data storage, the collected data undergoes preprocessing, including data cleaning, data format conversion, and data standardization. Data cleaning identifies and removes noise, outliers, and duplicate data. For example, abnormal pressure data from pressure sensors caused by momentary interference is identified and removed by comparing it with historical data and data from adjacent sensors. Duplicate data transmitted due to network fluctuations is also accurately identified and deleted. Missing data is supplemented using interpolation or based on statistical patterns of relevant data. Since data from different data sources has varying formats, format standardization is necessary. Binary data collected by sensors and specific format data exported from equipment control systems are converted to standard formats that the system can process uniformly, such as CSV and JSON, facilitating subsequent data storage and analysis. Different types of data are standardized to ensure comparability. Sensor data with different ranges are normalized to the [0,1] interval to eliminate the influence of dimensions, laying the foundation for subsequent data analysis and model building.

[0066] During data acquisition, data verification rules are set to verify the input data in real time. During data transmission, SSL / TLS encryption protocols are used to encrypt the transmitted data to prevent it from being stolen or tampered with, thus ensuring data security. Sensitive data stored is encrypted using the AES encryption algorithm, and only authorized users with the correct key can decrypt and read the data.

[0067] S2: Utilize big data analytics to analyze the collected data, construct a quality monitoring model to monitor the construction process, compare it with standard data, predict whether the quality is up to standard, and generate a quality monitoring report.

[0068] S2 includes the following steps:

[0069] S21: Based on the association rule mining algorithm, analyze the potential associations between highway construction site data;

[0070] Association rule mining algorithms are used to analyze potential relationships between construction data, process flow data, equipment data, and personnel data. For example, the relationship between concrete pouring temperature, raw material mix proportions, and final concrete strength can be mined to identify combinations of factors that significantly affect strength, providing a basis for optimizing construction techniques. Specifically, the Apriori algorithm can be used to analyze historical data and combine it with actual project needs, filtering frequent itemsets based on a set support threshold and retaining association rules with a confidence threshold.

[0071] S22: Use clustering algorithms to classify quality data during construction and group processes with similar quality characteristics into one category;

[0072] S23: Using highway construction site data and environmental data as input, and engineering quality acceptance results as output labels, construct a machine learning model and train it using historical data;

[0073] Machine learning models, such as Support Vector Machines (SVM) and Random Forests (RF), are built and trained on a large amount of historical construction data. Construction parameters, environmental factors, personnel and equipment information are used as input features, and the project quality acceptance results are used as output labels. The model is trained to learn the inherent patterns of quality formation, and the model parameters are continuously optimized through methods such as cross-validation to improve the accuracy and generalization ability of the model, so that it can accurately predict the quality status under different construction conditions.

[0074] S24: Construct a comprehensive quality monitoring indicator system based on the highway engineering quality acceptance specifications and design requirements;

[0075] The indicators cover key quality parameters such as subgrade compaction, pavement smoothness, concrete strength, and rebar spacing. A reasonable threshold range is set for each indicator as the basis for judging whether the quality is up to standard. Each indicator is quantified into a score, and the quantified score is compared with the preset threshold. If the score exceeds the preset threshold, the construction quality is judged as substandard.

[0076] S25: Combine big data analytics with a quality monitoring indicator system to build a quality monitoring model, and incorporate the association rules mined in S21 and the cluster features of clustering in S22;

[0077] Specifically, the quality monitoring model uses highway construction site data and environmental data as core inputs, including construction data, environmental data, and related data. The related data includes factor-quality correlations obtained through S21 association rule mining, and common characteristics of similar processes obtained through S22 cluster analysis. The model outputs the project quality acceptance results, specifically including qualitative results (whether the process is qualified or unqualified) and quantitative results (the deviation between the predicted values ​​of each quality indicator and the standard thresholds in the quality monitoring indicator system).

[0078] The machine learning model trained in S23 is used as the core computational module. This model has learned the mapping relationship between input parameters and quality results through historical data. The quality monitoring indicator system constructed in S24 is used as the model's judgment standard, specifically including indicator thresholds and indicator weights. The model needs to compare the machine learning prediction results with the thresholds of the indicator system, and calculate a comprehensive quality score by combining the indicator weights. The calculation formula is as follows:

[0079] Z = ∑α × G;

[0080] Where Z represents the overall quality score, α represents the weight, and G represents the indicator compliance rate.

[0081] The interpretability of the model is enhanced by incorporating association rule results. The factor-quality association rules mined in S21 are embedded into the model as auxiliary judgment criteria. The model stratification is refined using clustering results. Clusters of similar processes obtained from S22 are used as stratification variables in the model. For processes in different clusters, the model adopts targeted prediction parameters to make the predictions more closely match the characteristics of the processes.

[0082] Specifically, after standardizing and preprocessing the input on-site data, the data is mapped to the corresponding indicator dimensions based on the quality monitoring indicator system. The predicted values ​​of each quality indicator are then calculated and output through a machine learning model. Each predicted value is compared with the corresponding threshold in the indicator system for single-indicator judgment. The comprehensive quality score is calculated by combining the indicator weights. If the score is less than the preset pass line, the overall quality is judged to be unqualified.

[0083] A dynamic adjustment mechanism is introduced to dynamically optimize model parameters based on real-time collected data and historical feedback. For example, if the prediction deviation of a certain type of process is consistently greater than 10%, the clustering results of S22 are used to re-divide the same type of process clusters and adjust the prediction parameters of the clusters. If a new factor-quality correlation is discovered, the input features of the model are updated through the association rule mining of S21.

[0084] S26: Input the current highway construction site data and environmental data into the quality monitoring model to predict the current project quality acceptance results, and automatically generate a quality inspection report.

[0085] The prediction results include qualitative results that clearly indicate whether the quality is up to standard and quantitative results that compare key indicators. The predicted values ​​of key quality indicators are compared with standard thresholds. For example, in the concrete pouring process, the predicted concrete strength value (e.g., the actual predicted strength of C30 concrete is 28MPa, which is lower than the design standard of 30MPa) and whether there is a risk of cracking are considered.

[0086] The quality inspection report includes basic information about the process, such as process name, construction time, construction team, supervisor, and data collection time; comparative analysis of key quality indicators, i.e., the deviation between the measured / predicted values ​​of physical and process parameters during construction and the corresponding standard data. If the deviation is too large, it does not meet the specifications; explanation of quality prediction results, i.e., the conclusion of quality being qualified / unqualified and the basis for obtaining the conclusion; potential risk warning, analyzing the potential quality hazards for indicators that are predicted to be unqualified or fluctuate; and data collected on site, including pictures, videos, sensor data, and equipment operation status records.

[0087] Furthermore, step S22 includes the following steps:

[0088] S221: Collect process quality data and extract quality features, including numerical quality features and indirect quality features, and use correlation analysis to retain core features that are highly correlated with quality.

[0089] Numerical quality features include concrete strength, slump, and curing temperature. Indirect quality features refer to auxiliary features that have a potential impact on quality, such as construction process duration, equipment operating parameters (e.g., vibration frequency), and environmental factors (e.g., temperature and humidity during construction). If the extracted quality features have too high a dimensionality, dimensionality reduction can be achieved through principal component analysis (PCA).

[0090] S222: Combining the differences in quality characteristics of highway engineering processes, the elbow method is used to determine the number of clusters, and clustering is performed based on the K-Means algorithm;

[0091] Specifically, K samples are randomly selected from the dataset as initial cluster centers. The Euclidean distance between each sample and the K centers is calculated, and the sample is assigned to the nearest cluster. The mean of features of all samples in each cluster is calculated and used as the new cluster center. The process of assigning samples to clusters and updating cluster centers is repeated until the cluster centers no longer change or the maximum number of iterations is reached, ensuring that the clustering results are stable.

[0092] S223: Perform statistical analysis on the quality data of each cluster, extract common characteristics, and analyze common influencing factors;

[0093] S224: Iterate the clustering process continuously and update the cluster features.

[0094] Furthermore, step S223 includes the following steps:

[0095] S2231: Align the quality data of all processes within the same cluster in terms of time and space, and perform a second screening of the data within the cluster to remove outliers;

[0096] S2232: Calculate descriptive statistics for each quality index within the cluster, and extract common quality characteristics based on the statistical results;

[0097] Descriptive statistics include central tendency (mean, median), dispersion (variance / standard deviation, maximum / minimum) and frequency distribution.

[0098] S2233: By combining the influencing factor data corresponding to the processes within the cluster, we can identify the common causes that lead to common quality characteristics;

[0099] Specifically, data on influencing factors (including personnel, equipment, raw materials, and environmental factors) are retrieved, and statistical tests and association analyses are used to locate common influencing factors within clusters. Within-group difference analysis is performed on the influencing factors; if the values ​​of a certain factor are highly concentrated within a cluster, it is considered a potential common factor. The correlation strength between the influencing factors and common quality characteristics is calculated using chi-square tests (for categorical variables) or Pearson correlation coefficients (for continuous variables).

[0100] S2234: Verify the correlation between the identified common features and influencing factors;

[0101] The rationality of the data is verified by combining it with highway engineering construction specifications, eliminating accidental correlations that are related in data but not logically related, and the authenticity can be confirmed by on-site traceability for key common influencing factors.

[0102] S2235: Extract cluster feature labels, sort them by influence intensity, and generate a list of common influences.

[0103] For example, the list of influencing factors for low-strength concrete clusters is as follows: the main factors are that the strength of cement from Supplier A's March batch did not meet the standard (correlation degree 0.85) and the vibration time was insufficient (<30 seconds, correlation degree 0.73); the secondary factors are that the construction environment temperature is >38℃ (correlation degree 0.52).

[0104] Suppose that 20 concrete pouring processes in a highway project are clustered into 3 clusters using K-Means, where cluster 3 contains 6 pouring processes. The analysis process is as follows:

[0105] Cluster quality data statistics: The average strength of the 6 pours was 26MPa (design 30MPa), the average slump was 150mm (standard 180±20mm), the average curing temperature was 39℃, and 100% of the pours showed slight cracks.

[0106] Common quality characteristics: low strength (average <28MPa), low slump, high curing temperature, and common minor cracks.

[0107] Correlation of influencing factors: All six pours used cement from Supplier B's April batch (with an incoming strength deviation of -8%), the vibration equipment frequency was consistently <50Hz (standard 60-80Hz), and the ambient temperature during construction was consistently >37℃. On-site investigation confirmed that the April batch of cement from Supplier B did indeed exhibit quality fluctuations, the vibration equipment's aging resulted in insufficient frequency, and the high temperature accelerated concrete moisture evaporation. The common influencing factors were ultimately determined to be substandard cement batch quality, insufficient vibration equipment parameters, and the high-temperature environment.

[0108] Through the above process, the commonalities and causes of quality issues in similar processes can be accurately identified, providing data support for subsequent targeted rectification and iterative updates of cluster features.

[0109] S3: Execute the corresponding strategy based on the quality prediction results. If the predicted quality is qualified, fill in the relevant data in the process inspection application form. The relevant data includes physical parameters and process parameters during construction. If the predicted quality is unqualified, conduct a risk assessment in conjunction with the quality monitoring report, score the risk assessment results, and trigger an early warning based on the score.

[0110] Once the quality monitoring model predicts that the construction quality is up to standard, it automatically extracts physical parameters (such as concrete pouring pressure and temperature, roadbed compaction, etc.) and process parameters (such as rebar tying spacing, welding length, asphalt paving temperature and speed, etc.) from the construction process and fills this data into the process inspection application form. Construction personnel check and confirm the data in the application form on a mobile app, and if the check is correct, submit it to the supervision unit for review.

[0111] Upon receiving the application for inspection of a construction process, the supervising engineer first conducts a detailed review of the data in the application form, comparing it with the design documents and quality acceptance specifications to check the completeness and accuracy of the data and whether the construction process meets the requirements. Simultaneously, the supervising engineer can further understand the on-site construction situation by reviewing photos, videos, and other materials documenting the construction process. If the review is approved, the supervising engineer signs off in the system, and the construction unit can proceed to the next construction process. If problems are found with the data or flaws in the construction process, the supervising engineer notes the issues and returns the application form, requiring the construction unit to rectify them.

[0112] When the quality monitoring model predicts that the construction quality is substandard, a risk assessment is conducted based on the quality inspection report and from three dimensions: severity, scope of impact, and probability of occurrence. Each dimension is quantified into a score, and the weight of each dimension is determined using the analytic hierarchy process (AHP). The comprehensive risk score is calculated using the following formula:

[0113] M = Attention(S)×S+Attention(R)×R+Attention(P)×P;

[0114] Where Attention(S), Attention(R), and Attention(P) represent the weights of each item, M represents the risk score, S represents the severity, R represents the scope of influence, and P represents the probability of occurrence.

[0115] Severity (S) measures the impact of quality problems on the safety, function, and durability of an engineering structure. 1 point indicates a minor deviation (such as surface flatness slightly exceeding the standard, which does not affect structural safety); 3 points indicate a moderate defect (such as concrete strength deviation of 10%-15%, which requires local repair); and 5 points indicate a serious hidden danger (such as concrete strength being more than 20% lower than the design value, which may lead to insufficient structural bearing capacity).

[0116] The scope of impact (R) is used to assess the scope of the project involved in the quality problem and its impact on subsequent processes. 1 point indicates that it only affects a part of the current process (e.g., the compaction degree of a certain section of the roadbed is not up to standard, and the area is <10㎡); 3 points indicate that it affects the current process as a whole and 1-2 subsequent processes (e.g., the unqualified concrete pouring leads to the delay of pile acceptance); 5 points indicate that it affects multiple key processes or the entire project (e.g., the pile foundation quality problem leads to the rework of the bridge substructure).

[0117] The probability of occurrence (P) is based on the frequency of similar quality problems in historical data (similar projects and processes). 1 point indicates rare (historical occurrence rate <5%); 3 points indicate occasional (historical occurrence rate 10%-30%); 5 points indicate high incidence (historical occurrence rate >50%, such as the probability of insufficient strength due to improper concrete curing in high temperature environment).

[0118] An attention mechanism is introduced to dynamically calculate the weights of severity, scope of impact, and probability of occurrence, including the following steps:

[0119] Collect historical risk assessment data that does not meet quality standards. Each data point contains quantitative values ​​for three dimensions: severity, scope of impact, and probability of occurrence, as well as corresponding risk consequence labels. The features are then preprocessed.

[0120] The quantitative values ​​of severity, scope of impact, and probability of occurrence are standardized to eliminate the influence of units, and the labels of actual risk consequences are normalized as target values ​​for model training.

[0121] We construct a lightweight model that includes a feature embedding layer, an attention computation layer, and a weight output layer, and map features into high-dimensional feature vectors to enhance the model's ability to capture the correlation between features.

[0122] For example, a fully connected layer maps each 1-dimensional feature to an 8-dimensional embedding vector:

[0123] Embedding(S) = W1 × S norm +b1;

[0124] Embedding(R) = W² × R norm +b2;

[0125] Embedding(P) = W3 × P norm +b3;

[0126] Where W1, W2, and W3 are learnable weight matrices, and S norm R norm P norm These are the feature vectors after standardization of each dimension, with b1, b2, and b3 being bias terms.

[0127] Global features strongly correlated with risk consequences are used as query vectors to calculate attention scores. The scores are then normalized using the Softmax function to obtain attention weights.

[0128] The similarity between the embedding vector and the query vector, i.e., the attention score, is calculated using the following formula:

[0129] Score(S) = Q × Embedding(S) T ;

[0130] Score(R) = Q × Embedding(R) T ;

[0131] Score(P) = Q × Embedding(P) T ;

[0132] The scores are normalized using the Softmax function to obtain the attention weights (the sum of the weights is 1), as shown in the formula:

[0133] ;

[0134] ;

[0135] ;

[0136] in, These are the attention scores for each dimension.

[0137] Attention weights are combined with original features to predict risk consequences through a fully connected layer and output weight values.

[0138] This approach involves training a model based on historical data. In each iteration of the training process, the model calculates predicted risk consequences based on the current attention weights, compares these predictions with actual consequences, and adjusts the parameters to tilt the attention weights towards dimensions that have a greater impact on risk consequences. The trained attention mechanism model is then deployed to the risk assessment system, enabling real-time dynamic weight allocation and a dynamic adjustment mechanism. For example, after accumulating 100 new cases, the model is retrained with the new data (incremental training), updating the attention weights to adapt to changes in the engineering scenario. Determining weights through the attention mechanism allows for more accurate capture of the dynamic impact of different dimensions in actual risk assessment, particularly suitable for scenarios with complex quality issues and variable influencing factors in highway engineering, improving the accuracy of risk scoring and the targeted nature of early warnings.

[0139] Based on the risk score, it is divided into three levels of warning thresholds, each triggering corresponding warning measures:

[0140] M≥4 indicates a high risk and triggers a red alert. In this case, an alert should be immediately sent to the project manager, chief supervising engineer, and general manager of the construction unit via SMS, APP push, or email. The alert should include the type of quality problem, its specific location, risk level and results, and emergency handling suggestions.

[0141] If 2 ≤ M < 4, it is considered a medium risk and a yellow alert is issued. At this time, an alert is sent to the on-site supervising engineer and construction team leader, including details of the problem and the rectification deadline, and requires real-time uploading of rectification progress.

[0142] M < 2 indicates low risk and a blue alert is issued. At this time, a reminder is sent to the construction team, requiring them to record the problem and provide supplementary rectification explanations during the next inspection.

[0143] S4: Based on correlation analysis, identify the influencing factors of non-compliance and carry out rectification and prevention based on the specific factors identified.

[0144] S4 includes the following steps:

[0145] S41: Identify the dimensions of potential influencing factors and refine specific indicators, linking the specific manifestations of quality non-compliance with the influencing factor data to form an analysis dataset;

[0146] For example, the strength test values ​​of each batch of concrete are matched one by one with the test data of the corresponding batch of cement, the parameters of the vibration equipment, and the information of the construction personnel to ensure that the data correspondence is accurate. Specifically, the influencing factors include construction personnel factors (operating skills, number of training sessions, attendance records, number of violations, etc.), equipment operation factors (equipment model, operating parameters, fault records, maintenance cycle, etc.), raw material quality factors (material batch, incoming inspection data, storage conditions, etc.), and environmental factors (temperature and humidity, rainfall, wind force, duration of sunlight, etc. during construction).

[0147] S42: Using the quantitative indicators of quality non-compliance as the dependent variable and the indicators of each factor as independent variables, construct a multiple linear regression model, and judge the intensity of influence by the magnitude of the absolute value of the regression coefficients;

[0148] S43: Calculate and rank the correlations of each factor based on SHAP technology;

[0149] For example, after analyzing "unqualified concrete strength", the correlation ranking was obtained as follows: cement moisture content (0.85) > vibration time (0.72) > ambient temperature (0.58) > operator qualification (0.41), indicating that cement moisture content and vibration time are the core influencing factors.

[0150] S44: Based on the results of correlation analysis and combined with on-site investigation, accurately locate the root cause of the problem;

[0151] For factors ranking highly, verification was conducted through on-site inspections and data tracing. For example, if cement moisture content had the highest correlation, the arrival records and storage environment (such as whether it was exposed to rain) of that batch of cement were retrieved, and the moisture content was retested to confirm whether improper storage caused the excessive moisture content. "False correlation" factors were eliminated (e.g., data analysis showed a correlation between "wind force" and quality issues, but on-site verification found that wind force did not affect construction, so this factor was excluded), ultimately pinpointing the specific cause. For example, it was confirmed that the specific factors for "unqualified concrete strength" were excessive moisture content in a certain batch of cement (due to rain during storage) and insufficient vibration time caused by a malfunctioning vibration equipment.

[0152] S45: Based on the specific factors of the positioning, formulate an executable and traceable rectification plan and clarify the rectification elements;

[0153] If the problem stems from raw material quality (e.g., excessive cement moisture content), immediately discontinue the use of that batch of materials, replace with a qualified supplier, and require the supplier to provide the material's factory inspection report. For the parts of the project where the material has already been used, conduct a strength retest and develop a reinforcement plan. If the problem is due to equipment issues (e.g., vibrating equipment malfunction), immediately repair or replace the equipment, calibrate the equipment parameters (e.g., set a minimum threshold for vibration time), and establish a "daily equipment inspection - maintenance - calibration" log. If the problem is due to personnel operation (e.g., insufficient vibration time), provide specialized training to the operators (only those who pass the assessment can work), clarify the vibration process standards (e.g., 30 seconds of vibration per area), and arrange for on-site supervision by a supervisor. The plan must include the person responsible for rectification (e.g., construction team leader, material handler), the completion deadline (e.g., replace cement within 48 hours), the acceptance criteria (e.g., cement moisture content ≤ 0.5%), and the required supporting materials (e.g., re-inspection report, photos after rectification).

[0154] S46: The construction unit shall implement the rectification plan and upload the progress in real time.

[0155] The supervision unit conducts regular on-site inspections (such as randomly checking vibration time and reviewing cement test reports) and signs off on its supervision. After rectification is completed, the construction unit conducts a self-inspection and submits a re-inspection application (attached with data after rectification, such as a concrete strength re-inspection report). The supervision unit organizes a re-inspection; if the standards are met (e.g., strength meets design values), the acceptance is confirmed and the process for addressing the quality issue is closed; if the standards are not met, the causes are re-analyzed, and rectification measures are adjusted until the standards are met. Through these steps, a full-process management system is achieved, from data correlation analysis to on-site location and closed-loop rectification, ensuring that the root cause of quality problems is thoroughly resolved, while also forming prevention and control experience to prevent similar problems from recurring.

[0156] The method further includes:

[0157] S5. Based on BIM technology, a visualization platform is built to display the overall construction status and key indicators in a visual way in real time, and to provide data interaction and sharing functions between different user roles.

[0158] On the visualization platform, users can view project progress, process inspection status, quality testing data, and safety management information in real time. Visualized charts and graphs intuitively present the overall project situation and key indicators. Construction units can submit process inspection applications and related documents to the supervision unit, which will review and provide feedback on the acceptance results to the construction unit. Project management personnel can view the work progress and data of all participating parties in real time, enabling unified coordination and management. Through data interaction and sharing, information silos are broken down, improving the efficiency of collaborative project management.

[0159] The above embodiments are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for on-site management of highway engineering based on process inspection, characterized in that, Includes the following steps: S1: Obtain highway construction site data, standards and technical specifications through multiple channels, and preprocess the collected data. The construction site data includes construction data, process flow data, equipment data and personnel data. S2: Utilize big data analytics to analyze the collected data, construct a quality monitoring model to monitor the construction process, compare it with standard data, predict whether the quality is up to standard, and generate a quality monitoring report. S3: Execute the corresponding strategy based on the quality prediction results. If the predicted quality is qualified, fill in the relevant data in the process inspection application form. The relevant data includes physical parameters and process parameters during construction. If the predicted quality is unqualified, conduct a risk assessment in conjunction with the quality monitoring report, score the risk assessment results, and trigger an early warning based on the score. S4: Based on correlation analysis, identify the influencing factors of non-compliance and implement rectification and control measures according to the specific factors identified. S2 includes the following steps: S21: Based on the association rule mining algorithm, analyze the potential associations between highway construction site data; S22: Use clustering algorithms to classify quality data during construction and group processes with similar quality characteristics into one category; S23: Using highway construction site data and environmental data as input, and engineering quality acceptance results as output labels, construct a machine learning model and train it using historical data; S24: Construct a comprehensive quality monitoring indicator system based on the highway engineering quality acceptance specifications and design requirements; S25: Combine big data analytics with a quality monitoring indicator system to build a quality monitoring model, and incorporate the association rules mined in S21 and the cluster features of clustering in S22; S26: Input the current highway construction site data and environmental data into the quality monitoring model to predict the current project quality acceptance results and automatically generate a quality inspection report; The quality monitoring indicator system constructed by S24 is used as the judgment standard for the model. Specifically, it includes indicator thresholds and indicator weights. The model needs to compare the machine learning prediction results with the thresholds of the indicator system, and calculate the comprehensive quality score by combining the indicator weights. The calculation formula is as follows: Z = ∑α × G; Where Z represents the overall quality score, α represents the weight, and G represents the indicator compliance rate; S4 includes the following steps: S41: Identify the dimensions of potential influencing factors and refine specific indicators, linking the specific manifestations of quality non-compliance with the influencing factor data to form an analysis dataset; S42: Using the quantitative indicators of quality non-compliance as the dependent variable and the indicators of each factor as independent variables, construct a multiple linear regression model, and judge the intensity of influence by the magnitude of the absolute value of the regression coefficients; S43: Calculate and rank the correlations of each factor based on SHAP technology; S44: Based on the results of correlation analysis and combined with on-site investigation, accurately locate the root cause of the problem; S45: Based on the specific factors of the positioning, formulate an executable and traceable rectification plan and clarify the rectification elements; S46: The construction unit shall implement the rectification plan and upload the progress in real time.

2. The method for on-site management of highway engineering based on process inspection as described in claim 1, characterized in that, Build a cloud-based data collaboration platform to receive collected data over the network, store the data using relational and non-relational databases, establish relationships between data, and perform cross-regional synchronization.

3. The method for on-site management of highway engineering based on process inspection as described in claim 1, characterized in that, in, Step S22 includes the following steps: S221: Collect process quality data and extract quality features, including numerical quality features and indirect quality features, and use correlation analysis to retain core features that are highly correlated with quality. S222: Combining the differences in quality characteristics of highway engineering processes, the elbow method is used to determine the number of clusters, and clustering is performed based on the K-Means algorithm; S223: Perform statistical analysis on the quality data of each cluster, extract common characteristics, and analyze common influencing factors; S224: Iterate the clustering process continuously and update the cluster features.

4. A method for on-site management of highway engineering based on process inspection as described in claim 3, characterized in that, Step S223 includes the following steps: S2231: Align the quality data of all processes within the same cluster in terms of time and space, and perform a second screening of the data within the cluster to remove outliers; S2232: Calculate descriptive statistics for each quality index within the cluster, and extract common quality characteristics based on the statistical results; S2233: By combining the influencing factor data corresponding to the processes within the cluster, we can identify the common causes that lead to common quality characteristics; S2234: Verify the correlation between the identified common features and influencing factors; S2235: Extract cluster feature labels, sort them by influence intensity, and generate a list of common influences.

5. A method for on-site management of highway engineering based on process inspection as described in claim 1, characterized in that, When the quality monitoring model predicts that the construction quality is substandard, a risk assessment is conducted based on the quality inspection report and from three dimensions: severity, scope of impact, and probability of occurrence. Each dimension is quantified into a score, and the weight of each dimension is determined using the analytic hierarchy process (AHP). The comprehensive risk score is calculated using the following formula: M = Attention(S)×S+Attention(R)×R+Attention(P)×P; Where Attention(S), Attention(R), and Attention(P) represent the weights of each item, M represents the risk score, S represents the severity, R represents the scope of influence, and P represents the probability of occurrence.

6. A method for on-site management of highway engineering based on process inspection as described in claim 5, characterized in that, An attention mechanism is introduced to dynamically calculate the weights of severity, scope of impact, and probability of occurrence, including the following steps: Collect historical risk assessment data that does not meet quality standards. Each data point contains quantitative values ​​for three dimensions: severity, scope of impact, and probability of occurrence, as well as corresponding risk consequence labels. The features are then preprocessed. We construct a lightweight model that includes a feature embedding layer, an attention computation layer, and a weight output layer, and map features into high-dimensional feature vectors to enhance the model's ability to capture the correlation between features. Global features strongly correlated with risk consequences are used as query vectors to calculate attention scores. The scores are then normalized using the Softmax function to obtain attention weights. Attention weights are combined with original features to predict risk consequences through a fully connected layer and output weight values.

7. A method for on-site management of highway engineering based on process inspection as described in claim 1, characterized in that, The method further includes: S5. Based on BIM technology, a visualization platform is built to display the overall construction status and key indicators in a visual way in real time, and to provide data interaction and sharing functions between different user roles.

Images