A multi-mode engineering construction compaction quality intelligent evaluation and precise pressure compensation method

Abstract

The application discloses a kind of multi-mode engineering construction compaction quality intelligent evaluation and accurate supplement pressure method, it is related to fill engineering quality evaluation technical field, comprising: S1, construction area standardization grid encoding;S2, multi-source data synchronous acquisition and space-time binding preprocessing;S3, multi-domain feature extraction and automatic dimension reduction;S4, multi-mode parallel training;S5, grid level compaction quality evaluation;S6, compaction quality grid visualization and weak grid accurate positioning;Can accurately find the position of supplement pressure and efficiently complete supplement pressure, avoid excessive compaction and supplement pressure omission, improve the overall compaction quality of fill engineering and construction efficiency.

Description

Technical Field

[0001] This invention relates to the field of embankment engineering quality assessment technology, specifically to a method for intelligent assessment and precise compaction quality assessment in multi-mode engineering construction. Background Technology

[0002] Compaction degree, as a core indicator for measuring the compaction quality of engineering construction, directly determines the workability and long-term stability of embankment projects. Currently, the methods for assessing the compaction quality of engineering construction are mainly divided into traditional testing methods and modern non-destructive testing methods. Traditional testing methods calculate compaction degree by excavating and sampling, which has destructive and time-sensitive problems. Modern non-destructive testing methods collect vibration signals by installing accelerometers on road rollers, or extract features and combine them with machine learning modeling, or directly perform deep learning processing on the signals. Although these methods achieve non-destructive testing, current technologies have not integrated compaction quality assessment with the gridded construction area, nor have they formed a closed-loop system of "assessment-location-compaction".

[0003] Some studies have attempted to divide the compacted area into simple zones, but these rely solely on manual marking or rough positioning without combining satellite positioning to achieve precise grid mapping. This makes it impossible to quantify the compaction quality of each grid or accurately identify the location of weak grids. At the same time, the compaction operation lacks specificity, often involving large-area repeated rolling, which results in low efficiency, over-compaction, and incomplete compaction. This can easily lead to uneven compaction of the subgrade, causing subsequent settlement and deformation, and increasing maintenance costs. Summary of the Invention

[0004] This invention provides a method for intelligent assessment and precise compaction quality of multi-mode engineering construction to solve the problems of lack of grid-based assessment, inaccurate positioning, lack of targeted compaction, and failure to form a closed-loop control in existing engineering construction compaction quality assessment.

[0005] This invention provides the following technical solution: a method for intelligent assessment and precise compaction quality in multi-mode engineering construction, comprising: S1. Divide the target filling project compaction construction area into grids, assign a unique identifier to each grid, establish a one-to-one mapping relationship between grid identifiers and geographic coordinates, and construct a grid-based construction basic database. S2. Collect the triaxial vibration acceleration signal of the road roller, the positioning and compaction quality data of the road roller, realize the real-time spatiotemporal binding of vibration acceleration signal, compaction quality data and grid coordinates, and construct a standardized time-series signal dataset; S3. Perform multi-domain feature extraction on the standardized time series signal to construct a complete artificial feature set. Then, perform automatic dimensionality reduction optimization on the complete artificial feature set to obtain the dimensionality-reduced core feature set. S4. Based on standardized time-series signal datasets, complete artificial feature sets, and dimensionality-reduced core feature sets, three independent compaction quality assessment models are constructed and trained in parallel: signal direct deep learning, feature extraction machine learning, and dimensionality-reduced feature fusion deep learning. S5. Based on the characteristics of each grid in the gridded construction foundation database, automatically match the optimal evaluation mode for each area grid, complete the grid-level compaction quality evaluation, and construct a gridded compaction quality database. S6. Based on the gridded compaction quality database, construct a gridded spatial visualization distribution map of the compaction quality of engineering construction, classify the grid according to the evaluation results, accurately locate the weak grids, and accurately reinforce them.

[0006] The present invention has the following beneficial effects: By standardizing the gridding of the construction area and real-time positioning of the geographic location, the compaction quality data is accurately bound to the grid coordinates; by integrating three evaluation modes—direct signal deep learning, feature extraction machine learning, and dimensionality reduction feature fusion deep learning—accurate grid-level compaction quality assessment is completed; the location of additional compaction is accurately located and additional compaction is completed efficiently, avoiding over-compaction and omissions in additional compaction, thereby improving the overall compaction quality and construction efficiency of the filling project. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of the overall process operation in this invention.

[0008] Figure 2 This is a schematic diagram showing the installation position of the three-dimensional acceleration sensor on the road roller in this invention.

[0009] Figure 3 This is a schematic diagram of the refined preprocessing of multi-source data in this invention.

[0010] Figure 4 This is a schematic diagram of the BiGRU-Casul CNN-Dilated CNN network model.

[0011] Figure 5 This refers to the compaction intelligent control system in this invention.

[0012] Figure 6 , 7 Figure 8 shows the spatial visualization distribution of the compaction quality during engineering construction. Detailed Implementation

[0013] The technical solutions of the embodiments of this specification will be explained and described below with reference to the accompanying drawings. However, the following embodiments are only preferred embodiments of this specification and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments in the implementation methods without creative effort are all within the protection scope of this specification.

[0014] Please see Figure 1-8 As shown, a method for intelligent assessment and precise compaction quality in multi-mode engineering construction is characterized by comprising: S1. Standardized grid coding of construction area: The construction area of ​​the target filling project is divided into grids, each grid is assigned a unique identifier, a one-to-one mapping relationship between grid identifier and geographic coordinates is established, and a grid-based construction basic database is constructed.

[0015] Specifically: S101. Based on the operating width of the road roller (e.g., 2-4m), construction accuracy requirements, and fill material type, divide the target construction area into equal-sized square grids (preferably with a grid side length of 1-2m, which can be adaptively adjusted according to project needs) or rectangular grids (preferably with a grid end side length of 1-2m and a long side length of 4-8m, ensuring that a single grid can be fully covered by a single compaction by the road roller, avoiding positioning deviations caused by cross-grid evaluation).

[0016] S102. Using a two-dimensional coding rule of "row number and column number" (e.g., R05C08 represents the 5th row and 8th column grid), a unique identifier is assigned to each grid. At the same time, a one-to-one mapping relationship between the grid code and GPS / BeiDou geographic coordinates is established to achieve accurate positioning of grid coordinates.

[0017] S103. Input basic information such as grid code, geographic coordinates, filler type, design compaction degree, and target number of compaction passes into the system according to region to form a gridded construction basic database, which provides a basis for subsequent targeted assessment and compaction.

[0018] By standardizing and gridding, the quality assessment of compaction in engineering construction is refined from "the entire road section" to "a single grid," achieving "one data point per grid and one assessment per grid." This accurately captures weak areas in local compaction, completely solving the shortcomings of traditional assessments that are "coarse-grained and difficult to detect local problems," and improving the overall uniformity of compaction in embankment projects. It also enables refined and spatial control of compaction quality.

[0019] S2. Collect the triaxial vibration acceleration signal of the road roller, the positioning and compaction quality data of the road roller, realize the real-time spatiotemporal binding of vibration acceleration signal, compaction quality data and grid coordinates, and construct a standardized time-series signal dataset.

[0020] Specifically: S201. Install a three-dimensional acceleration sensor calibrated with zero-point drift and gravity component on the upper plane of the front frame of the road roller to simultaneously collect vibration acceleration signals of the X, Y, and Z axes at a sampling frequency of 200Hz~1000Hz and collect signals during the stable driving phase of the road roller. S202. Install a GPS / BeiDou dual-mode positioning module on the road roller, with a positioning accuracy of ≤0.5m, and collect the road roller's geographical coordinates, travel speed, and compaction trajectory in real time. The positioning data acquisition frequency is synchronized with the acceleration signal. S203. Simultaneously record construction parameters (machinery model, vibration frequency, current number of compaction passes, etc.), and use the sand cone method to take the average of multiple points to obtain the true compaction value as a label. S204. The system determines the current grid location of the road roller in real time and automatically binds the acceleration signal, positioning coordinates, construction parameters and corresponding grid codes to achieve spatiotemporal unification of multi-source data.

[0021] Next, the above multi-source data undergoes refined preprocessing: 1. Format standardization: Convert the original CSV data into pure numeric data, delete invalid / blank rows, and ensure data consistency; 2. Outlier handling: A 512-row sliding window and a 128-row stride are used to remove outliers using the interquartile range (IQR). When the effective data percentage is ≥85%, linear interpolation is used to complete the data. If it is <85%, it is marked as invalid data and re-collected. 3. Filtering and noise reduction: A zero-phase FIR bandpass filter (5-95Hz) is used to suppress high and low frequency interference and normalize the signal frequency; 4. Data standardization: Perform mean-standard deviation standardization on each channel of the triaxial signal to eliminate dimensional differences; 5. Positioning data correction: Through a planar four-parameter coordinate system mapping model, GPS / BeiDou geodetic coordinates are converted into a construction local coordinate system, which is accurately matched with the grid coordinates to eliminate positioning deviations.

[0022] By binding GPS / BeiDou precise positioning with grid coding, centimeter-level positioning of weak grids is achieved, and compaction is only carried out on unqualified grids, avoiding large-area repeated compaction, improving construction efficiency by 20%-30%, significantly reducing the idle travel distance and energy consumption of road rollers, and avoiding excessive compaction that damages the soil structure; the positioning of weak grids is accurate, and the compaction is highly targeted.

[0023] S3. Perform multi-domain feature extraction on the standardized time series signal to construct a complete artificial feature set. Then, perform automatic dimensionality reduction and optimization on the complete artificial feature set to obtain the dimensionality-reduced core feature set.

[0024] Specifically: 1. Comprehensive Multi-Domain Feature Extraction: For the preprocessed unified signal, time-domain, power spectral density, and nonlinear features are extracted to form a complete artificial feature set, fully integrating prior engineering information. Time-domain characteristics: peak value, peak factor, kurtosis, skewness, mean, variance, standard deviation, waveform factor, impulse factor, 25% / 75% quantile, etc.; Power spectral density characteristics: After Fast Fourier Transform (FFT) transformation, the mean, standard deviation, centroid frequency, mean square frequency, frequency entropy, peak frequency, spectral kurtosis, etc. are extracted. Nonlinear characteristics: Absolute difference mean at time scales of 2, 5, 10, and 20.

[0025] 2. Automatic Feature Dimensionality Reduction and Optimization: Feature simplification is achieved based on Pearson correlation coefficient, eliminating redundancy and multicollinearity to obtain the core feature set after dimensionality reduction. Calculate the pairwise correlation coefficients between features and filter out highly redundant features with a correlation coefficient ≥ 0.8; Calculate the correlation coefficient between features and compaction degree, select core features by sorting them by correlation, maximize the retention of effective information and reduce model complexity.

[0026] S4. Based on standardized time-series signal datasets, complete artificial feature sets, and dimensionality-reduced core feature sets, three independent compaction quality assessment models are constructed and trained in parallel: direct signal deep learning, feature extraction machine learning, and dimensionality-reduced feature fusion deep learning.

[0027] Because the vibration signals during road roller operation are non-stationary and locally time-varying, direct processing of the entire signal is susceptible to interference from local anomalies. Therefore, a sliding window mechanism is used to segment the continuously acquired acceleration time series for processing.

[0028] The sliding window is segmented as follows: The window length is set to 512 sampling points. This length corresponds to a signal segment of approximately 2.56 seconds at a sampling frequency of 200Hz (512÷200=2.56s), which can cover multiple compaction vibration cycles (typical compaction excitation frequency is 5-30Hz, with a period of approximately 0.03-0.2s) while avoiding blurring of local features due to an excessively long window.

[0029] A sliding step size of 128 sampling points (0.64 seconds) was used to achieve 75% window overlap. This high overlap helps preserve signal continuity, reduces feature loss due to window boundary truncation, and improves the temporal resolution of subsequent feature extraction.

[0030] The output format divides the original N-point signal into M overlapping subsequences. Each subsequence contains 3×512-dimensional data along the X, Y, and Z axes, forming the basic unit for subsequent processing.

[0031] Outliers in the triaxial acceleration signals of each window are removed using a box plot method. An FIR filter is used to filter the acceleration signal. The sensor's sampling frequency is 200Hz, and according to the Nyquist theorem, its highest corresponding frequency is 100Hz. To suppress high-frequency noise and low-frequency interference, a zero-phase FIR bandpass filter is used to filter the original signal. The filter's passband range is set to 5-95Hz, and the signal frequency is normalized after filtering. Within each sliding window, outlier detection and removal are performed on the X, Y, and Z axis acceleration signals to eliminate spike interference caused by sudden road surface changes, mechanical collisions, or momentary sensor malfunctions. The box plot method works by calculating the first quartile (Q1), third quartile (Q3), and interquartile range (IQR = Q3 - Q1) for a single-axis acceleration data sequence a = [a1, a2, ..., a512]. The normal value range is defined as follows: Q1 - 1.5 × IQR, Q3 + 1.5 × IQR; Data points falling outside this interval are considered outliers.

[0032] The processing strategy involves replacing detected outliers with the median of the acceleration signal within the window (or using linear interpolation) instead of deleting them directly, in order to maintain a consistent time series length and facilitate subsequent frequency domain analysis and model input alignment. Compared to the 3σ criterion, the box plot method is more robust to non-Gaussian distributed signals and is suitable for the asymmetric, heavy-tailed distribution characteristics commonly found in compaction vibrations.

[0033] Zero-phase FIR bandpass filter: To extract the effective vibration components that are strongly correlated with the compaction state, it is necessary to filter out irrelevant frequency interference: Sampling frequency and Nyquist frequency, the sensor sampling frequency is F S =200Hz. According to the Nyquist sampling theorem, the highest signal frequency that can be reconstructed without distortion is FN=FS / 2=100Hz. Among them, low-frequency interference mainly comes from rigid body displacement or sensor baseline drift caused by the movement of the road roller itself (such as traveling and turning), and is unrelated to the material compaction state; High-frequency noise (>95Hz) mainly comes from the thermal noise of electronic devices, electromagnetic interference, or transient high-frequency components generated by the impact of small gravel on the road surface, which can easily mask the true compaction response.

[0034] The filter is designed as a finite impulse response (FIR) digital filter, which avoids signal waveform distortion due to its linear phase characteristics. Furthermore, a zero-phase filtering method is employed (i.e., filtering the signal once in both the forward and reverse directions) to completely eliminate phase delay and ensure that the physical meaning of time-domain characteristics (such as peak value and kurtosis) is not compromised. The bandpass range is set from 5Hz to 95Hz, covering the typical compaction excitation frequency (5-50Hz) and its harmonic components, while being strictly limited to the Nyquist frequency.

[0035] The filter order and window function can be designed using either the Kaiser window or the Hamming window. The order is determined based on the transition bandwidth requirements (for example, if the transition band is 2Hz, the order is approximately 200-300), ensuring a flat passband and a stopband attenuation greater than 60dB.

[0036] Although the filtered signal has removed invalid frequency bands, the signal amplitude may still differ in dimensions or scale under different construction sections or different roller operating conditions. Therefore, frequency normalization is performed: The power spectral density of each filtered signal is normalized to eliminate the influence of external factors such as equipment power and excitation intensity. This allows feature extraction to focus on the spectral shape rather than the absolute amplitude, improving the model's generalization ability and ensuring its total energy is 1. in, The original power spectral density function represents the signal at frequency . Energy distribution at the location; This is the normalized power spectral density, i.e., the spectrum per unit energy; The total integral of the original power spectrum from 5 Hz to 95 Hz represents the total energy (or total power) in that frequency band. Multi-domain features of preprocessed multi-source vibration signal parameters are extracted to obtain time-domain features, power spectral density features and nonlinear features. Based on feature correlation analysis, the multi-domain features are automatically reduced in dimensionality to reduce the risk of model overfitting. In this process, feature parameters are extracted from the preprocessed vibration signal from three dimensions: time domain, power spectral density, and nonlinearity. The time domain features reflect the amplitude variation and statistical characteristics of the signal over time. The extracted feature parameters include peak value, peak factor, kurtosis, skewness, mean, variance, standard deviation, waveform factor, impulse factor, 25th percentile, and 75th percentile. A Fast Fourier Transform (FFT) is performed on the time-domain signal to convert it into a frequency-domain signal. The extracted power spectral density features include mean, standard deviation, centroid frequency, mean square frequency, frequency variance, frequency entropy, peak frequency, bandwidth, spectral kurtosis, and spectral skewness. Nonlinear features are used to capture dynamic characteristics of the signal that do not conform to linear laws, and the absolute difference mean values ​​at time scales of 2, 5, 10, and 20 are extracted. Among them, the time-domain characteristics directly reflect the amplitude distribution, fluctuation intensity and statistical regularity of the vibration signal on the time axis, and are suitable for capturing transient response changes caused by compaction. Traditional linear features struggle to capture the complex dynamic behavior during compaction caused by material inhomogeneity and contact nonlinearity. Therefore, multi-scale difference analysis is introduced: For signal (t), calculated at time scales of 2, 5, 10, and 20; in, Time scale The mean of the absolute differences calculated below; It is a time-domain acceleration signal sequence; This is the signal length (number of sampling points). pass judge: Small scale (τ=2) to capture details of high-frequency micro-vibrations; Large-scale (τ=20) reflects low-frequency trend changes; When compaction is good, the signal is more stable, the MAD value is lower and increases slowly with τ; Loose regions exhibit high MAD values ​​and strong scale sensitivity due to irregular vibrations.

[0037] The dimensionality-reduced multi-domain features are trained using a machine learning model, and Bayesian optimization is used to search for the best combination of hyperparameters to output the optimal machine learning classification model. Specifically, automatic feature dimensionality reduction lowers the risk of model overfitting. According to Pearson correlation coefficient theory, a correlation coefficient of 0.8 or higher between features indicates a high linear correlation. The Pearson correlation coefficients between all pairs of features are calculated, and a threshold of 0.8 is set. Redundant features with correlation coefficients higher than this threshold are automatically identified and removed, thus eliminating multicollinearity among features and reducing model complexity. Furthermore, a higher correlation coefficient between a feature and the target variable indicates a greater contribution of that feature to the target variable. Therefore, features are sorted according to their correlation coefficients, and core features with higher correlation are automatically selected. By prioritizing the retention of features that contribute significantly to the target variable, the effective information of the features is maximized while compressing feature dimensionality. The initial total number of multi-domain features is 33 (time domain) + 30 (frequency domain) + 12 (nonlinear) = 75 dimensions; To avoid overfitting caused by the curse of dimensionality and multicollinearity, a two-stage automatic dimensionality reduction strategy based on the Pearson correlation coefficient is adopted: S1. Eliminate redundant features; Calculate the Pearson correlation coefficients between all pairwise features of the 75 dimensions: in, Features and The Pearson correlation coefficient between them; Features and covariance; Features Standard deviation; Features Standard deviation; Subsequently, a threshold of 0.8 was set, and if the calculated value was... >0.8, retain those features that are more correlated with the target variable (compaction degree label), and remove the other feature; S2. Filter core features; Calculate the Pearson correlation coefficient between the residual features and the target variable. ; Press | | Sort in descending order; Select the first n features (n=20-30) to ensure that the cumulative contribution is >90%.

[0038] Then, machine learning modeling and optimization are performed; The extracted features were normalized (min-max normalization to the [0, 1] interval), and the training and test sets were divided in a 7:3 ratio. The datasets were then input into machine learning models such as Support Vector Machine, Random Forest, XGBoost, and ANN for training. Bayesian optimization was used to search for the optimal combination of hyperparameters. Accuracy, precision, and recall were used to evaluate model performance, and the optimal machine learning classification model was selected for use in assessing the working status of road rollers or diagnosing faults. It should be noted that the candidate models are support vector machine, random forest, XGBoost, and artificial neural network. Hyperparameter optimization methods construct surrogate models (such as Gaussian processes) to approximate the objective function (such as validation set accuracy) for Bayesian optimization. Intelligent selection of the next set of hyperparameters based on expected improvement criteria; Compared to grid search / random search, it converges faster and is more efficient.

[0039] The optimization objective is to maximize the overall performance metric (such as F1-score) under cross-validation.

[0040] Among them, the evaluation indicators are: Accuracy rate, overall correctness rate; Accuracy rate, the proportion of predicted qualified results that are actually qualified; Recall rate is the percentage of genuine qualified products that are correctly identified. The final model is then output, and the one with the best overall performance on the test set is selected (e.g., XGBoost performs best in most scenarios).

[0041] The optimal machine learning classification model is used to analyze the dynamically acquired multi-source vibration signal parameters and output the compaction quality assessment results. The vibration signals collected and preprocessed in real time during construction are extracted according to the above-mentioned features and input into the trained intelligent evaluation model to obtain the real-time compaction quality evaluation results. If the evaluation result is unqualified, the system will automatically issue an early warning signal and provide optimization suggestions (such as increasing the number of compaction passes) based on the construction parameters. If the evaluation result is qualified or excellent, the evaluation data is recorded, a compaction quality evaluation report is generated, and the compaction quality is monitored in real time throughout the entire process.

[0042] The specific details of the three independent compaction quality assessment models are as follows: 1. Signal Direct Deep Learning Evaluation Mode No feature extraction is required. The pre-processed triaxial vibration time series signal is directly used as input to fully capture the original time series multi-scale features of the signal. It is suitable for grids with complex filling and variable working conditions (such as roadbed corners and grids in semi-fill and semi-cut areas). The core is the BiGRU-Causal CNN-Dilated CNN multi-scale time series model, which outputs the grid compaction degree prediction value.

[0043] 2. Feature Extraction Machine Learning Evaluation Mode Using the dimensionality-reduced multi-domain full feature set as input, a classic machine learning algorithm is used for modeling. It makes full use of the engineering prior information of artificial features, which has the advantages of high training efficiency, fast inference speed and low hardware requirements. It is suitable for conventional grids with uniform filling and simple working conditions (such as grids in large flat areas in the middle of the roadbed). Support Vector Machine (SVM), Random Forest (RF), XGBoost and Artificial Neural Network (ANN) are selected to construct the model group. The hyperparameters are optimized by Bayesian optimization, and the grid compaction degree prediction value is output.

[0044] 3. Dimensionality Reduction Feature Fusion Deep Learning Evaluation Model Using the fusion of the dimensionality-reduced core feature set and the preprocessed signal as input, and combining the engineering priors of artificial features with the temporal feature capture capability of deep learning, the synergistic effect of "artificial features and deep learning" is achieved. It is suitable for key grids with extremely high compaction accuracy requirements (such as grids that are prone to compaction blind spots, such as bridge abutment back, culvert side, and roadbed edge), and is the optimal accuracy scheme among the three modes. It outputs the grid compaction degree prediction value and compaction uniformity index.

[0045] For different fillers and operating conditions, a flexible matching evaluation mode is used for the grid. The key grid adopts a dimensionality reduction feature fusion deep learning mode. The test set R 2 It can reach above 0.90, with MSE as low as around 1.05, significantly improving the accuracy of grid-level compaction prediction, effectively avoiding compaction errors caused by assessment errors, and adapting to different grid conditions with multiple modes, resulting in high assessment accuracy.

[0046] S5. Based on the characteristics of each grid in the gridded construction foundation database, automatically match the optimal evaluation mode for each grid, complete the grid-level compaction quality evaluation, and construct a gridded compaction quality database.

[0047] The steps for assessing the quality of grid-level compaction are as follows: The system automatically matches the optimal evaluation mode for each grid based on the filler type and working condition label in the grid base database; After each grid is compacted, the system immediately calls the corresponding evaluation mode, outputs the predicted compaction degree and compaction uniformity index of the grid, and compares them with the grid design compaction degree and target uniformity index. The assessment results are linked to grid codes and geographic coordinates in real time and stored in the gridded compaction quality database, achieving "one grid, one data, one grid, one assessment".

[0048] S6. Based on the gridded compaction quality database, construct a gridded spatial visualization distribution map of the compaction quality of engineering construction, classify the grid according to the evaluation results, accurately locate the weak grids, and accurately reinforce them.

[0049] Among them, the spatially visualized distribution map of the compaction quality during engineering construction is a heat map or a three-color distribution map, with color mapping based on the grid compaction degree values: (1) Green grid: The compaction degree is greater than or equal to the design value and the uniformity index meets the standard, which is a qualified grid; (2) Yellow grid: The compaction degree is slightly lower than the design value (deviation ≤5%), or the uniformity index is slightly substandard, which is a grid to be re-inspected; (3) Red grid: The compaction degree is significantly lower than the design value (deviation > 5%), or the uniformity index is seriously substandard. It is a weak grid and needs to be compacted immediately.

[0050] The aforementioned visualization interface displays information such as grid code, geographic coordinates, compaction degree value, and number of compaction passes in real time, and supports zooming in, zooming out, and precise querying.

[0051] The grid-based spatial visualization distribution map of the compaction quality during engineering construction provides an intuitive display. The list, path, and parameters for the compaction of weak grids are automatically generated by the system and pushed to the vehicle terminal. On-site operators do not need to make manual judgments and can directly follow the instructions to operate, reducing human decision-making errors and promoting the transformation of roadbed construction from "experience-driven" to "data-driven". It has a high degree of visualization and intelligence and is easy to operate.

[0052] Among these, precise positioning and compaction of weak grids includes: Based on the compaction deviation value and uniformity index, the weak grids are divided into first-level weak grids (deviation > 10%, requiring key compaction) and second-level weak grids (5% < deviation ≤ 10%, requiring regular compaction). The system automatically extracts the unique code, precise geographic coordinates, compaction trajectory, and current number of compaction passes of the weak grid, and generates a list of weak grid reinforcement. The list includes the grid location, reinforcement priority, and suggested reinforcement parameters (excitation frequency, number of compaction passes), and pushes the reinforcement list to the roller's on-board terminal and the on-site management terminal in real time.

[0053] The aforementioned list of weak grid compaction also includes uncompacted blank grids. The method for identifying uncompacted blank grids is as follows: By comparing the roller compaction trajectory with the grid division map, the system automatically identifies the uncompacted blank grids (i.e., compaction blind spots) and includes them in the list of weak grids for compaction.

[0054] The specific steps for precise pressure replenishment are as follows: First, the system automatically plans the optimal compaction path for the road roller based on the geographical coordinates and distribution of the weak grid, following the principles of "proximity, continuous compaction, and no repetition or omission," thereby reducing the road roller's empty running distance and improving compaction efficiency. Second, based on the type of filler in the weak grid, the compaction deviation value, and the operating conditions, the system provides customized pressure compensation parameter suggestions: (1) First-level weak grid: It is recommended to increase the number of compaction passes by 2-3, appropriately increase the excitation frequency (e.g., 30-33Hz), and reduce the driving speed (1.5-2km / h). (2) Secondary weak grid: It is recommended to increase the number of compaction passes by 1-2, maintain the normal excitation frequency (28-30Hz), and drive at a speed of 2-2.5km / h; (3) Weak grids in narrow areas such as bridge abutment back and culvert side: Use special equipment such as automatic linkage hydraulic compactor to carry out deep and precise compaction, with a compaction depth of 3-5m.

[0055] Third, the road roller performs additional compaction according to the planned path and compaction parameters. The system monitors the roller's compaction trajectory in real time to ensure that weak grids are fully covered and compaction is performed according to parameters, avoiding missed compaction or insufficient compaction.

[0056] After compaction is completed, the original assessment mode of the corresponding grid is called to carry out a special re-inspection. The grid quality status is updated according to the re-inspection results. For grids that fail the re-inspection, the compaction plan is iteratively optimized and re-compacted until they pass. After all grids have completed assessment, compaction, and re-inspection, the system automatically generates a grid-based general report on the compaction quality of the project construction, which includes the compaction data, assessment records, compaction records, and re-inspection results of each grid, realizing full-process traceability of compaction quality.

[0057] The criteria for passing the re-inspection are: compaction degree ≥ design value, grid color changes from red / yellow to green, and it is included in the qualified grid library.

[0058] The above S1~S6 form a closed-loop system of "grid assessment - weak grid location - precise compaction - post-compaction re-inspection". Grids that do not meet the standards after compaction need to be re-compacted until they are qualified. At the same time, the system automatically identifies compaction blind spots, achieving full coverage of compaction operations. The qualified rate of roadbed compaction has increased from 85% to over 98%, the rework rate has decreased by 90%, and the construction rework cost has been greatly reduced.

[0059] This invention only requires the installation of a three-dimensional acceleration sensor and a GPS / BeiDou positioning module on the road roller, without the need for additional complex equipment. It can be linked with existing hydraulic compactors and other dedicated compaction equipment, and is suitable for different types of road rollers and filling construction scenarios. The equipment modification cost is low and the project is highly practical.

[0060] Compared with existing methods, this method enables rapid, non-destructive testing of compaction quality during construction with high accuracy, thus allowing for precise and comprehensive assessment of compaction status. It eliminates the need to add additional construction and soil parameters to the model; high-precision prediction of compaction degree can be achieved directly using only triaxial vibration signals.

[0061] This specification also provides a computer-readable storage medium storing instructions that, when executed on a computer or processor, cause the computer or processor to perform the multiple steps described in the above embodiments. If the constituent modules of the above-described electronic device are implemented as software functional units and sold or used as independent products, they can be stored in the computer-readable storage medium.

[0062] This specification also provides a computer program product, including a computer program that, when executed by a processor, implements the multiple steps described in the above embodiments.

[0063] Where there is no conflict, the technical features in this embodiment and implementation scheme can be combined arbitrarily.

[0064] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes multiple computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this specification are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center integrating multiple available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., digital versatile discs (DVDs)), or semiconductor media (e.g., solid state disks (SSDs)).

[0065] When implemented through hardware or firmware, the aforementioned method flow is programmed into the hardware circuit to obtain the corresponding hardware circuit structure and achieve the corresponding function. For example, a Programmable Logic Device (PLD) (such as a Field Programmable Gate Array (FPGA)) is such an integrated circuit, whose logic function is determined by the user programming the device. Designers can program a digital system onto a PLD themselves, eliminating the need for chip manufacturers to design and fabricate dedicated integrated circuit chips. Furthermore, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly implemented using "logic compiler" software, similar to the software compiler used in program development. The original code before compilation must also be written in a specific programming language, called a Hardware Description Language (HDL). There is not just one HDL, but many. Those skilled in the art should understand that by simply performing some logic programming on the method flow using one of the aforementioned hardware description languages ​​and programming it into an integrated circuit, the hardware circuit implementing the logic method flow can be easily obtained.

[0066] The embodiments described above are merely preferred embodiments of this specification and are not intended to limit the scope of this specification. Any modifications and improvements made by those skilled in the art to the technical solutions of this specification without departing from the spirit of this specification should fall within the protection scope defined by the claims of this specification.

Claims

1. A method for intelligent assessment and precise compaction quality evaluation in multi-mode engineering construction, characterized in that, include: S1. Divide the target filling project compaction construction area into grids, assign a unique identifier to each grid, establish a one-to-one mapping relationship between grid identifiers and geographic coordinates, and construct a grid-based construction basic database. S2. Collect the triaxial vibration acceleration signal of the road roller, the positioning and compaction quality data of the road roller, realize the real-time spatiotemporal binding of vibration acceleration signal, compaction quality data and grid coordinates, and construct a standardized time-series signal dataset; S3. Perform multi-domain feature extraction on the standardized time series signal to construct a complete artificial feature set. Then, perform automatic dimensionality reduction optimization on the complete artificial feature set to obtain the dimensionality-reduced core feature set. S4. Based on standardized time-series signal datasets, complete artificial feature sets, and dimensionality-reduced core feature sets, three independent compaction quality assessment models are constructed and trained in parallel: signal direct deep learning, feature extraction machine learning, and dimensionality-reduced feature fusion deep learning. S5. Based on the characteristics of each grid in the gridded construction foundation database, automatically match the optimal evaluation mode for each grid, complete the grid-level compaction quality evaluation, and construct a gridded compaction quality database. S6. Based on the gridded compaction quality database, construct a gridded spatial visualization distribution map of the compaction quality of engineering construction, classify the grid according to the evaluation results, accurately locate the weak grids, and accurately reinforce them.

2. The intelligent assessment and precise compaction method for multi-mode engineering construction compaction quality according to claim 1, characterized in that, In S1, the steps for constructing the gridded construction foundation database are as follows: Based on the operating width of the road roller, the construction accuracy requirements, and the type of fill material in the embankment project, the target construction area is divided into square grids or rectangular grids of equal size; A two-dimensional coding rule of "row number and column number" is adopted to assign a unique identifier to each grid, and at the same time, a one-to-one mapping relationship between grid code and GPS / BeiDou geographic coordinates is established; The basic information of the grid, such as grid code, geographic coordinates, filler type, design compaction degree, and target number of compaction passes, is entered into the system according to the region to form a grid-based construction foundation database.

3. The intelligent assessment and precise compaction method for multi-mode engineering construction compaction quality according to claim 1, characterized in that, In S2, the steps for constructing a standardized time-series signal dataset are as follows: The three-dimensional acceleration sensor of the road roller is used to simultaneously collect vibration acceleration signals of the X, Y, and Z axes, and to collect signals during the stable driving phase of the road roller. The system collects the geographical coordinates, travel speed, and compaction trajectory of the road roller in real time, and the location data acquisition frequency is synchronized with the acceleration signal. Simultaneously record construction parameters, and use the sand filling method to take the average of multiple points to obtain the true compaction value as a label; The system determines the current grid position of the road roller in real time and automatically binds the acceleration signal, positioning coordinates, construction parameters and the corresponding grid code.

4. The intelligent assessment and precise compaction method for multi-mode engineering construction compaction quality according to claim 1, characterized in that, In S3, the steps for constructing the core feature set after dimensionality reduction are as follows: For the preprocessed unified signal, three types of features are extracted: time domain, power spectral density, and nonlinearity, forming a complete artificial feature set and fully integrating prior engineering information; Calculate the pairwise correlation coefficients between features and filter out highly redundant features with a correlation coefficient ≥0.8; Calculate the correlation coefficient between features and compaction degree, select core features by sorting them by correlation, maximize the retention of effective information and reduce model complexity.

5. The intelligent assessment and precise compaction method for multi-mode engineering construction compaction quality according to claim 1, characterized in that, In S5, the steps for assessing grid-level compaction quality are as follows: The system automatically matches the optimal evaluation mode for each grid based on the filler type and working condition label in the grid base database; After each grid is compacted, the system immediately calls the corresponding evaluation mode, outputs the predicted compaction degree and compaction uniformity index of the grid, and compares them with the grid design compaction degree and target uniformity index. The assessment results are linked to grid codes and geographic coordinates in real time and stored in a gridded compaction quality database.

6. The intelligent assessment and precise compaction method for multi-mode engineering construction compaction quality according to claim 1, characterized in that, In step S6, the spatial visualization distribution map of the compaction quality of the engineering construction grid is a heat map or a three-color distribution map, and color mapping and quality classification are performed based on the grid compaction degree value.

7. The intelligent assessment and precise compaction method for multi-mode engineering construction compaction quality according to claim 1, characterized in that, In step S6, accurately locating and compacting the weak grid includes: Based on the compaction deviation value and uniformity index, the weak grid is divided into first-level weak grid and second-level weak grid; The system automatically extracts the unique code, precise geographic coordinates, compaction trajectory, and current number of compaction passes of the weak grid, generates a list of weak grids to be reinforced, and pushes the list to the roller's on-board terminal and the on-site management terminal in real time.

8. The intelligent assessment and precise compaction method for multi-mode engineering construction compaction quality according to claim 7, characterized in that, The list of weak mesh compaction also includes uncompacted blank meshes. The method for identifying uncompacted blank meshes is as follows: By comparing the roller compaction trajectory with the grid division map, the system automatically identifies the uncompacted blank grids and includes them in the list of weak grids to be compacted.

9. The intelligent assessment and precise compaction method for multi-mode engineering construction compaction quality according to claim 1, characterized in that, In S6, the specific steps for precise pressure compensation are as follows: The system automatically plans the optimal compaction path for the road roller based on the geographical coordinates and distribution of the weak grid. Based on the type of filler, compaction deviation, and operating conditions of the weak grid, the system provides customized pressure compensation parameter suggestions. The road roller performs additional compaction according to the planned path and compaction parameters, and the system monitors the compaction trajectory of the road roller in real time.

10. The intelligent assessment and precise compaction method for multi-mode engineering construction compaction quality according to claim 9, characterized in that, In S6, after the compaction is completed, the original evaluation mode of the corresponding grid is called to carry out a special re-inspection. The grid quality status is updated according to the re-inspection results. For grids that fail the re-inspection, the compaction scheme is iteratively optimized and compaction is carried out again until they pass the test. After all grids have completed evaluation, compaction, and re-inspection, the system automatically generates a grid-based general report on the compaction quality of the engineering construction.

Images