Deep Learning-Based Data Analysis Method for Construction Economic Indicators
By analyzing the operating data of pavers and screeds using deep learning, the probability of damage levels is predicted and maintenance costs are output. This addresses the shortcomings of existing operation and maintenance management technologies and enables refined preventive maintenance and cost control of pavers.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA RAILWAY 17TH BUREAU GRP URBAN CONSTR CO LTD
- Filing Date
- 2026-03-25
- Publication Date
- 2026-05-26
Smart Images
Figure CN121901663B_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of data analysis, specifically a data analysis method for construction economic indicators based on deep learning. Background Technology
[0002] In highway asphalt pavement paving construction, the paver, as the core equipment, directly determines the uniformity and final smoothness of the paved surface; the screed, as a key working component of the paver, is fundamental to ensuring construction quality; however, there are significant technical bottlenecks in the current operation and maintenance management of pavers and screeds:
[0003] The industry generally adopts a preventive maintenance model based on fixed cycles or a passive maintenance model that repairs only when it breaks down. This model cannot accurately perceive the real-time health status of the equipment, which may lead to insufficient maintenance causing a decline in construction quality, or excessive maintenance causing waste of resources. The judgment of screed wear is highly dependent on the experience of operators and lacks quantitative and objective assessment basis, resulting in untimely wear warnings. Replacement is often carried out only after paving defects (such as excessive flatness or insufficient pre-compaction) have already occurred, which seriously affects the continuity of construction and cost control.
[0004] While modern pavers can generate a large amount of operational data (such as vibration, temperature, and material level), this data is usually only used for real-time monitoring of individual machine parameters and simple alarms. The data sources are isolated from each other and fail to establish a deep correlation model with the core performance indicators of the equipment (such as compaction efficiency) and the condition of core components (such as screed wear). For example, screed wear is a gradual process affected by comprehensive working conditions. Simply monitoring the wear thickness cannot predict its development trend, and early signs such as abnormal vibration, decreased heating efficiency, and uneven material conveying are not effectively integrated to predict its health decline trajectory.
[0005] Existing technologies for predicting ironing plate wear are mostly based on simple linear models or historical averages, failing to fully consider the nonlinear impact of dynamic changes in actual equipment operating conditions on the wear rate. Furthermore, damage assessment is disconnected from maintenance decisions, making it impossible to correlate specific damage states with expected maintenance costs and construction risks, thus hindering refined preventative maintenance and construction cost estimation. To address the problems raised in this background, this application designs a deep learning-based method for analyzing construction economic indicators. Summary of the Invention
[0006] To address the aforementioned technical shortcomings, this application proposes a deep learning-based method for analyzing construction economic indicators.
[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: This application provides a method for analyzing construction economic indicators based on deep learning, which includes the following specific steps:
[0008] S1. Obtain data on the operation of the paver during highway construction, historical wear data of the screed, and current wear thickness data;
[0009] S2. Analyze the health index of pavers based on data on their operation during highway construction.
[0010] S3. Based on the paver health index analysis results and the historical wear data of the paver screed, predict and analyze the wear rate of the paver screed.
[0011] S4. Construct a deep learning model for predicting and analyzing the damage of the paver screed. Input the current wear thickness data and wear rate prediction analysis results of the paver screed into the deep learning model for predicting and analyzing the damage of the paver screed, and output the probability prediction results of the damage level of the paver screed.
[0012] S5. Based on the probability prediction results of the damage level of the paver screed, output the probability prediction results of the damage level of the paver in the future working cycle, and output the relevant maintenance costs.
[0013] It should be noted that, as a preferred technical solution for the data analysis method of construction economic indicators based on deep learning, the specific steps of S1 are as follows:
[0014] S11. Acquire paver operating data through nuclear density meter, construction design, dynamic signal acquisition instrument, mechanical floating sensor, side box display screen, thermocouple and equipment manual. The paver operating data includes actual compaction data of paver screed, target pre-compaction data, standard deviation data and average value data of vibration acceleration during screed working cycle, material level height data on the left and right sides of paver conveyor, paver screed set temperature data, current actual temperature change data with heating time and initial standard heating time data.
[0015] S12. Historical wear data of the screed is obtained by fitting and inverting historical data of the paver screed and using the wear rate formula. The historical wear data of the screed includes the average initial wear rate of the paver screed and the fitting coefficient of the wear rate influence.
[0016] S13. Obtain the current wear thickness data of the paver screed through the bolts of the internally embedded sensor;
[0017] S14. Store the acquired data in the storage component for use in the analysis process.
[0018] It should be noted that, as a preferred technical solution for the data analysis method of construction economic indicators based on deep learning, step S2 includes the following specific steps:
[0019] S21. The paver screed efficiency index analysis results are obtained from the actual compaction data, target pre-compaction data, standard deviation data and average vibration acceleration data of the screed during the working cycle of the paver screed.
[0020] S22. The paver conveying uniformity index analysis results are obtained from the left and right material height data of the paver.
[0021] S23. The paver thermal decay index analysis results are obtained from the paver screed set temperature data, the current actual temperature change data with heating time data, and the initial standard heating time data.
[0022] S24. The paver health index analysis results are obtained from the paver screed efficiency index analysis results, material conveying uniformity index analysis results, and thermal decay index analysis results.
[0023] It should be noted that, as a preferred technical solution for the construction economic indicator data analysis method based on deep learning, the specific steps of S21 are as follows: Based on the actual compaction data of the paver screed, the target pre-compaction data, the standard deviation data of vibration acceleration during the screed's working cycle, and the average data of vibration acceleration, the paver screed's efficiency index is analyzed. The paver screed efficiency index analysis process is as follows: The actual compaction data of the screed is divided by the target pre-compaction data to obtain the screed compaction effect; the standard deviation data of vibration acceleration during the screed's working cycle is divided by the average data of vibration acceleration to obtain the screed's vibration instability; the value is subtracted from the screed's vibration instability to obtain the screed's vibration stability; the screed compaction effect and vibration stability are multiplied to obtain the paver screed efficiency index analysis result.
[0024] It should be noted that, as a preferred technical solution for the construction economic indicator data analysis method based on deep learning, the specific steps of S22 are as follows: The paver conveying uniformity index is analyzed based on the left and right material height data of the paver. The paver conveying uniformity index analysis process involves: processing the absolute value of the difference between the left and right material height data of the paver conveying, then converting it to a percentage value to obtain the paver conveying deviation; subtracting the paver conveying deviation from the numerical value yields the paver conveying uniformity index analysis result.
[0025] It should be noted that, as a preferred technical solution for the construction economic indicator data analysis method based on deep learning, the specific steps of S23 are as follows: Based on the paver screed set temperature data, the current actual temperature change over heating time data, and the initial standard heating time data, a paver thermal degradation index analysis is performed. The paver thermal degradation index analysis process is as follows: Based on the paver screed set temperature data and the current actual temperature change over heating time data, the time required for the paver screed to reach the set temperature is obtained; based on the initial standard heating time data and the paver screed set temperature data, the initial heating time required to reach the current set temperature is obtained; the initial heating time required to reach the current set temperature is divided by the current time required for the paver screed to reach the set temperature to obtain the paver thermal degradation index analysis result.
[0026] It should be noted that, as a preferred technical solution for the construction economic indicator data analysis method based on deep learning, the specific steps of S24 are as follows: Based on the analysis results of the paver screed efficiency index, the material conveying uniformity index, and the thermal decay index, a paver health index analysis is performed. The paver health index analysis process involves weighting and summing the paver screed efficiency index, material conveying uniformity index, and thermal decay index to obtain the paver health index analysis result.
[0027] It should be noted that, as a preferred technical solution for the construction economic indicator data analysis method based on deep learning, the specific steps of S3 are as follows: Based on the paver health index analysis results, the historical initial wear rate average data of the paver screed, and the wear rate influence fitting coefficient, a paver screed wear rate prediction analysis is performed. The formula for the paver screed wear rate prediction analysis is:
[0028]
[0029] ,in, The average initial wear rate of the paver screed throughout history. The wear rate affects the fitting coefficient. The results of the paver health index analysis.
[0030] It should be noted that, as a preferred technical solution for the construction economic indicator data analysis method based on deep learning, the specific steps of S4 are as follows: Constructing a deep learning model for predicting and analyzing the damage of the paver screed; inputting the current wear thickness data and wear rate prediction analysis results of the paver screed into the deep learning model for predicting and analyzing the damage of the paver screed; and outputting the probability prediction results of the paver screed damage level. The specific process involves: standardizing and preprocessing the data: regularly arranging 25 wear thickness measurement points on the screed; each time data is sampled from the measurement points, the collected current wear thickness data and wear rate prediction analysis results are... The data is reorganized into a 5x5 two-dimensional grid to preserve the relative positions of each measurement point. The thickness grid and the rate grid are stacked in the channel dimension to obtain a two-dimensional multi-channel feature map. This provides the first input sample data for the deep learning model of paver screed damage prediction and analysis. The total input sample data for the paver screed damage prediction and analysis deep learning model is obtained in this way. The collected and processed total input sample data of the paver screed damage prediction and analysis deep learning model is then input into the trained paver screed damage prediction and analysis deep learning model to output the probability prediction result of paver screed damage level.
[0031] It should be noted that, as a preferred technical solution for the construction economic indicator data analysis method based on deep learning, the specific steps of S5 are as follows: obtaining the probability prediction results of the paver screed damage level, associating the probability prediction results of the paver screed damage level with the corresponding set maintenance costs, obtaining the probability of the paver damage level and the corresponding maintenance costs in the future working cycle, and pushing the probability of the paver damage level and the corresponding maintenance costs in the future working cycle to relevant personnel for processing.
[0032] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention acquires data on the paver's operating status during highway construction, historical wear data of the screed, and current wear thickness data; performs paver health index analysis based on the paver's operating status data during highway construction; performs paver screed wear rate prediction analysis based on the paver health index analysis results and historical wear data of the paver screed; constructs a deep learning model for predicting paver screed damage, inputs the current wear thickness data and wear rate prediction analysis results of the paver screed into the deep learning model, and outputs a probability prediction result for the paver screed damage level; and outputs the paver screed damage level probability prediction result based on the paver screed damage level probability prediction result. This system predicts the probability of damage levels to the paver during its future working cycle and outputs related maintenance costs. It transforms discrete operating parameters into continuous quantitative indicators reflecting the paver's core working capabilities, enabling an objective and dynamic analysis of the paver's overall operating condition. By deeply fusing data, it dynamically predicts the paver's wear rate, constructs and utilizes a deep learning model to handle high-dimensional nonlinear relationships, and comprehensively considers wear thickness and rate to output accurate damage level probabilities. Simultaneously, it links these probabilities to relevant maintenance costs, allowing the damage level probability prediction results to directly serve the user, providing clear action guidelines and cost expectations, facilitating user management decisions, and thus achieving predictive maintenance of the paver equipment. This effectively avoids sudden failures and ensures the continuity and stability of construction. Attached Figure Description
[0033] Figure 1 This is a schematic diagram of the overall process of the deep learning-based construction economic indicator data analysis method proposed in this application.
[0034] Figure 2 This is a flowchart illustrating the S2 step of the deep learning-based construction economic indicator data analysis method of this application.
[0035] Figure 3 This is a schematic diagram illustrating the process of obtaining health index analysis results using the deep learning-based construction economic indicator data analysis method of this application.
[0036] Figure 4 This is a schematic diagram illustrating the process of obtaining the wear rate prediction and analysis results of the construction economic indicator data analysis method based on deep learning in this application.
[0037] Figure 5 This is a schematic diagram illustrating the process of obtaining the probability prediction results of damage level using the deep learning-based construction economic indicator data analysis method of this application. Detailed Implementation
[0038] To better understand this application, various aspects of this application will be described in more detail with reference to the accompanying drawings.
[0039] To address the technical problems raised in the background art, this application provides a preferred embodiment:
[0040] The specific content of this embodiment is as follows:
[0041] like Figure 1 As shown, the data analysis method for construction economic indicators based on deep learning includes the following specific steps:
[0042] S1. Obtain data on the operation of the paver during highway construction, historical wear data of the screed, and current wear thickness data;
[0043] In this embodiment, the specific steps of S1 are as follows:
[0044] S11. Paver operating data includes actual compaction data of the paver screed, target pre-compaction data, standard deviation and average vibration acceleration data of the screed during its working cycle, material level height data on the left and right sides of the paver's conveying surface, screed set temperature data, current actual temperature variation with heating time data, and initial standard heating time data. The actual compaction data of the paver screed is obtained by averaging the compaction at various points on the paver's operating location using a nuclear density meter. Target pre-compaction data is obtained through construction design. Time-domain data from the triaxial accelerometer installed on the screed is recorded using a dynamic signal acquisition device, and the standard deviation and average vibration acceleration data of the screed during its working cycle are processed and calculated. Material level height data on the left and right sides of the paver's conveying surface is obtained using a mechanical floating sensor. The paver screed set temperature data is directly read from the side box display screen. The current actual temperature variation with heating time is obtained using thermocouples placed near the screed's heating plate. Initial standard heating time data is obtained from the paver screed equipment user manual.
[0045] S12. Historical wear data of the screed includes the average initial wear rate of the paver screed and the fitting coefficient of wear rate influence. The process of obtaining the fitting coefficient of wear rate influence is as follows: The historical data of the paver screed equipment is organized into a two-dimensional array [health index, actual wear rate]; a wear model function is defined in the analysis software, with the health index and two parameters to be inverted (baseline wear rate, attenuation coefficient) as input variables, and the corresponding theoretical wear rate as output; reasonable initial guess values are set for the baseline wear rate and attenuation coefficient based on engineering experience (for example, the baseline wear rate is the average wear rate under healthy conditions, and the attenuation coefficient is 1.5); the linear least squares fitting algorithm is called to fit the model after taking the natural logarithm, and the residual sum of squares between the logarithmic sequence of theoretical calculation values and the logarithmic sequence of measured data is minimized by optimizing the parameter values; after the fitting converges, the goodness of fit index (i.e., the coefficient of determination R) is calculated. 2The residual distribution was analyzed to verify the reliability of the inversion results. Finally, the attenuation coefficient extracted from the optimal solution is the fitting coefficient of the wear rate influence. The process of obtaining the historical initial wear rate mean data of the paver screed is as follows: collect the historical maintenance data of the paver screed (the thickness of the bottom plate and the corresponding cumulative paving area for each maintenance), calculate the actual wear rate in each maintenance period according to the wear rate formula (wear rate = (the thickness of the previous maintenance - the thickness of this maintenance) / the paving area in the interval period), screen the data in the first fault-free cycle after the new screed is installed, and exclude the records under extreme working conditions such as paving special mixtures or ultra-thick layers. Calculate the arithmetic mean of the multiple selected wear rate data to obtain the historical initial wear rate mean data of the paver screed.
[0046] S13. Current wear thickness data of the paver screed is obtained by installing bolts with internally embedded sensors between the paver screed and the base (during installation, a specified preload is applied to the bolts with a torque wrench, causing the bolts to elongate elastically like a stretched spring). The reduction in bolt elongation is obtained to get the current wear thickness data of the paver screed (as the screed wears and thins, the total clamping thickness between the base plate and the base decreases, and the elastic elongation of the bolts is released, causing the stress and strain inside the bolts to decrease simultaneously, that is, at this time the wear of the screed is equal to the reduction in bolt elongation).
[0047] S14. Store the acquired data in the storage component for use in the analysis process.
[0048] S2, such as Figure 2 As shown, a health index analysis of pavers is conducted based on data on their operation during highway construction.
[0049] S21. The paver screed efficiency index analysis results are obtained from the actual compaction data, target pre-compaction data, standard deviation data and average vibration acceleration data of the screed during the working cycle of the paver screed.
[0050] In this embodiment, S21 includes the following specific steps: Based on the actual compaction data of the paver screed, the target pre-compaction data, the standard deviation data of vibration acceleration during the screed's working cycle, and the average vibration acceleration data, the paver screed efficiency index analysis is performed. The paver screed efficiency index analysis process is as follows: The actual compaction data of the screed is divided by the target pre-compaction data to obtain the screed compaction effect; the standard deviation data of vibration acceleration during the screed's working cycle is divided by the average vibration acceleration data to obtain the screed vibration instability; the value is subtracted from the screed vibration instability to obtain the screed vibration stability; the screed compaction effect and vibration stability are multiplied to obtain the paver screed efficiency index. Performance index analysis results; it should be noted that when compaction is insufficient or vibration is unstable, the performance index analysis results of the paver screed decrease, indicating that the screed performance is in a state of decline. The ratio of actual compaction degree to target pre-compaction degree is used as the compaction effect index, directly reflecting the paver's ability to compact the mixture. Vibration acceleration is used to measure the stability of the screed's operation; the higher the vibration stability, the better the smoothness and uniformity of the paved surface. Multiplying the compaction effect by the vibration stability yields a comprehensive index, avoiding the one-sidedness that a single index may bring, and allowing for a more comprehensive evaluation of the overall performance of the screed. A high performance index means that the screed is working in good condition, which can reduce repetitive work or repairs, saving material, labor and time costs.
[0051] S22. The paver conveying uniformity index analysis results are obtained from the left and right material height data of the paver.
[0052] In this embodiment, S22 includes the following specific steps: Analyzing the uniformity index of paver material feeding based on the left and right material height data. The analysis process involves: processing the absolute value of the difference between the left and right material height data, then converting it to a percentage value to obtain the paver material feeding deviation. Subtracting the deviation from the value yields the paver material feeding uniformity index analysis result. It should be noted that uniform material feeding is fundamental to ensuring road surface smoothness and compaction uniformity. Calculating the absolute value of the left and right material height difference directly reflects the balance of material feeding from the paver's auger distributor. A higher uniformity index indicates closer left and right material levels and more uniform mixture distribution, thus ensuring consistent pavement thickness, reducing the risk of material segregation, and preventing quality defects such as uneven longitudinal joints and transverse thickness fluctuations in highways.
[0053] S23. The paver thermal decay index analysis results are obtained from the paver screed set temperature data, the current actual temperature change data with heating time data, and the initial standard heating time data.
[0054] In this embodiment, S23 includes the following specific steps: performing a paver thermal degradation index analysis based on the paver screed's set temperature data, the current actual temperature change over heating time data, and the initial standard heating time data. The paver thermal degradation index analysis process is as follows: obtaining the time required for the paver screed to reach the set temperature based on the paver screed's set temperature data and the current actual temperature change over heating time data; obtaining the initial heating time required to reach the current set temperature based on the initial standard heating time data and the paver screed's set temperature data; and dividing the initial heating time required to reach the current set temperature by the current heating time required to reach the set temperature to obtain the paver thermal degradation index. Analysis results; it should be noted that thermal decay is a gradual process. Continuous monitoring of this index can detect problems in advance before the heating system completely fails or causes serious paving quality accidents (such as poor material adhesion due to insufficient temperature). Low heating efficiency means that more fuel or electricity needs to be burned to reach the same temperature. When the heating time increases, the paver's thermal decay index decreases, indicating that the paver's thermal efficiency has decreased. By comparing the current required heating time with the initial standard heating time, the degree of performance degradation of the screed heating system (such as burners, heat transfer oil system, electric heating elements) can be directly quantified. Using the same set temperature as a benchmark for comparison ensures the fairness and accuracy of the analysis, making the results purely reflect the performance changes of the equipment itself.
[0055] S24. The paver health index analysis results are obtained from the paver screed efficiency index analysis results, material conveying uniformity index analysis results, and thermal decay index analysis results.
[0056] like Figure 3 As shown, in this embodiment, S24 includes the following specific steps: A paver health index analysis is performed based on the paver screed efficiency index analysis results, material conveying uniformity index analysis results, and thermal decay index analysis results. The paver health index analysis process involves weighting and summing the paver screed efficiency index analysis results, material conveying uniformity index analysis results, and thermal decay index analysis results to obtain the paver health index analysis result. It should be noted that for the paver screed bottom plate, the health index analysis result comprehensively evaluates vibration stability, heating efficiency, and compaction uniformity. When the paver health index analysis result is 1, it indicates that there is no wear. The weighting method is as follows: the entropy weight method is used to determine the weights of the efficiency index analysis results, material conveying uniformity index analysis results, and thermal decay index analysis results (i.e., the weights are calculated using the dispersion of each indicator in historical data; the greater the change in the indicator data, the higher its corresponding weight).
[0057] S3. Based on the paver health index analysis results and the historical wear data of the paver screed, predict and analyze the wear rate of the paver screed.
[0058] like Figure 4 As shown, in this embodiment, the specific steps of S3 are as follows: Based on the paver health index analysis results, the historical initial wear rate average data of the paver screed, and the wear rate influence fitting coefficient, a paver screed wear rate prediction analysis is performed. The formula for the paver screed wear rate prediction analysis is:
[0059]
[0060] ,in, The average initial wear rate of the paver screed throughout history. The wear rate affects the fitting coefficient. This is the result of the paver health index analysis. It should be noted that the historical average initial wear rate of the paver screed represents the inherent wear level of the paver screed under optimal working conditions, and is the starting point for measuring the performance degradation of the paver screed. The wear rate affects the fitting coefficient in this formula. The time when the performance of the paver screed degrades (i.e. The result is increased. The exponential increase in the wear rate caused by the increased wear on the paver screed (resulting in increased wear) is significant. The larger the screed, the more sensitive it is to performance degradation. Part of this refers to the performance degradation of the current paver screed. The analysis indicates that the wear rate of the paver screed is non-linear. In the initial stages of wear, the screed shows a slightly accelerated wear rate. However, as paver problems accumulate and the precision of the various parts of the paver gradually decreases, the wear accelerates (for example, slight vibration of the screed causes uneven wear on the base plate, which in turn triggers more severe vibrations and heat spots, further accelerating the wear of the paver screed). The paver health index analysis results... When it is 1, When the paver health index analysis results When it begins to descend, Increase, at this time It increases exponentially (the worse the performance of the paver screed, the more severe the degradation and the faster the wear).
[0061] S4. Construct a deep learning model for predicting and analyzing the damage of the paver screed. Input the current wear thickness data and wear rate prediction analysis results of the paver screed into the deep learning model for predicting and analyzing the damage of the paver screed, and output the probability prediction results of the damage level of the paver screed.
[0062] like Figure 5As shown, in this embodiment, the specific steps of S4 are as follows: Constructing a deep learning model for predicting and analyzing paver screed damage; inputting the current wear thickness data and wear rate prediction analysis results of the paver screed into the deep learning model for predicting and analyzing paver screed damage; and outputting the probability prediction results of paver screed damage level. The specific process is as follows: Data standardization preprocessing: 25 wear thickness measurement points are regularly arranged on the screed. Each time data is sampled from the measurement points, the collected current wear thickness data and wear rate prediction analysis results are reorganized into a 5x5 two-dimensional grid to preserve the relative positional relationship between each measurement point; The thickness... The thickness grid and rate grid are stacked along the channel dimension to obtain a two-dimensional multi-channel feature map. This provides the initial input sample data for the deep learning model predicting paver screed damage. This process yields the total input sample data for the model. The collected and processed total input sample data is then fed into the trained deep learning model to output the probability prediction result of paver screed damage level. It should be noted that the stacking of the thickness grid and rate grid along the channel dimension results in a two-dimensional multi-channel feature map with a shape of (5, 5). (5, 5, 2): This two-dimensional grid has a height of 5 (i.e., 5 rows), a width of 5 (i.e., 5 columns), and 2 channels (i.e., two channels: channel 0: a 5x5 thickness layer, and channel 1: a 5x5 rate layer); The deep learning model architecture for predicting and analyzing damage to the screed of a paver (using a hybrid design combining convolutional neural networks and fully connected networks) consists of an input layer, a convolutional layer, a fully connected layer, and an output layer. The input layer receives the two-dimensional multi-channel feature map of (5, 5, 2) above; the convolutional layer consists of two sub-convolutional layers (the first sub-convolutional layer uses 32 3x3 convolutional kernels, extracts primary spatial features using same padding, and then...). The first sub-convolutional layer uses ReLU activation and batch normalization. The second sub-convolutional layer uses 64 3x3 convolutional kernels to further extract higher-order features, followed by a 2x2 max pooling layer to reduce the spatial dimension to (2, 2, 64), and a dropout rate of 0.25 is added to prevent overfitting. The fully connected layer flattens the pooled data features into a 256-dimensional vector, which is then input into a sub-network containing two fully connected layers (the first fully connected layer has 128 neurons, and the second layer has 64 neurons; each layer is batch normalized and activated with ReLU, and a dropout rate of 0 is set between the two layers).(5 random deactivation to enhance generalization ability); the output layer is a fully connected layer with 4 neurons, using the Softmax activation function, and the output represents the probability distribution of four damage levels. The entire deep learning model for predicting and analyzing paver screed damage contains approximately 66,000 trainable parameters. The training process and detailed configuration of the deep learning model for predicting and analyzing paver screed damage are as follows: the loss function is sparse classification cross-entropy suitable for integer labels, the optimizer is Adam, its initial learning rate is set to 0.001, and a dynamic decay strategy is used, that is, if the validation set loss does not decrease for 5 consecutive epochs, the learning rate will be multiplied by 0.1; the batch size is set to 32, the maximum number of training epochs is 150, and an early stopping method is used, which automatically terminates training and rolls back to the optimal weights when the validation set loss does not improve for 15 consecutive epochs; online data augmentation can also be performed on the input feature map during training, including Random 90-degree, 180-degree, and 270-degree rotations, as well as horizontal and vertical flips (to improve robustness); the dataset is divided into training, validation, and independent test sets at 70%, 15%, and 15% respectively. The training objective is to minimize the loss on the validation set. The final performance of the deep learning model for predicting paver screed damage will be comprehensively evaluated on the test set using metrics such as classification accuracy, macro-average F1 score, and confusion matrix; through a data-driven approach, the complex mapping relationship between paver screed data and damage state is automatically learned end-to-end; in the output paver screed damage level probability prediction results, the Softmax activation function is used, and the output represents the probability distribution of four damage levels. The specific process and calculation formula are as follows: the output layer is a fully connected layer containing 4 neurons. Before inputting into the Softmax function, the values output by these 4 neurons are called Logits (denoted as ). (where i is the number corresponding to each damage state level, i=1, 2, 3, 4). The function of Softmax is to... Transform the output into a probability distribution such that the sum of all output values is 1, and each value is between 0 and 1. For the i-th damage state level, the corresponding predicted probability is calculated as follows: , This represents the original value output by the neuron corresponding to the i-th damage state level. For the natural constant e The exponentiation is used to ensure that the result is positive. Let j be the exponent value of the j-th neuron, where j is the index variable used to iterate through the four neurons. This represents the sum of the values of the four neurons, used for normalization to ensure that the probability sum is 1. Damage status is divided into four levels: healthy, minor damage, requiring maintenance, and severe damage. This division is based on the spatial distribution features of thickness and rate automatically extracted by the convolutional neural network. The model learns a complex mapping relationship between these features and the four discrete levels (healthy, minor damage, requiring maintenance, and severe damage) through a data-driven approach, ultimately outputting a probability vector. The level with the highest probability is taken as the prediction result. Specifically, this includes three aspects: First, a multi-dimensional mapping of the feature space, with inputs being a 5×5 thickness grid and a rate grid containing spatial location information. The model learns local features (e.g., whether the thickness at a certain edge measurement point is abnormal), global features (e.g., whether the wear presents a uniform distribution as in a healthy state or a localized severe depression as in a requiring maintenance state), and dynamic features (the rate grid reflects the development trend of wear; for example, although the current thickness of the ironing plate is acceptable, if the rate is extremely high (i.e., the value of channel 1 is large), it will be judged as requiring maintenance or severely damaged). Second, the definition of training labels, which are assigned to each input in the training dataset. The input samples, i.e., 5×5×2 feature maps, each correspond to a real label annotated by professionals (labels are 0, 1, 2, 3, 0: healthy, wear thickness within the safe threshold, wear rate stable, no maintenance required; 1: slight damage, slight wear exists, but has not yet affected paving smoothness; 2: maintenance required, wear is close to the critical value or wear rate is accelerating, maintenance or component replacement needs to be arranged in the near future; 3: severe damage, wear exceeds the standard, may have caused paving quality defects, immediate shutdown and repair are required). The model learns a nonlinear mapping from the spatial distribution of thickness and rate to four discrete levels by minimizing the sparse classification cross-entropy loss. For example, the spatial pattern of thin thickness and fast rate in the upper left corner corresponds to the damage state level of needing maintenance. The third is the decision mechanism of probability output. The four probability values given by the output layer represent the model's confidence in the current input belonging to each damage state level. The final predicted damage state level is determined by taking the position of the maximum probability (for example, if the output result is [0.1, 0.7, 0.15, 0.05], the damage state level is determined as slight damage). The damage state level is determined in this way.
[0063] S5. Based on the probability prediction results of the damage level of the paver screed, output the probability prediction results of the damage level of the paver in the future working cycle, and output the relevant maintenance costs.
[0064] In this embodiment, the specific steps of S5 are as follows: obtain the probability prediction result of the damage level of the paver screed, associate the probability prediction result of the damage level of the paver screed with the corresponding set maintenance cost to obtain the damage level probability and corresponding maintenance cost of the paver in the future working cycle, and push the damage level probability and corresponding maintenance cost of the paver in the future working cycle to relevant personnel for processing; it should be noted that the damage level probability is the damage state level corresponding to the highest probability. Relevant personnel can scientifically formulate maintenance plans, optimize spare parts inventory, and accurately calculate the single machine construction cost based on the probability distribution of different damage levels and the corresponding cost estimates.
[0065] Based on the above implementation details, this embodiment has the following advantages over the prior art: This embodiment acquires data on the paver's operating status during highway construction, historical wear data of the screed, and current wear thickness data; performs paver health index analysis based on the paver's operating status data during highway construction; performs paver screed wear rate prediction analysis based on the paver health index analysis results and historical wear data of the paver screed; constructs a deep learning model for predicting paver screed damage, inputs the current wear thickness data and wear rate prediction analysis results of the paver screed into the deep learning model for predicting paver screed damage, and outputs the probability prediction result of paver screed damage level; based on the probability prediction result of paver screed damage level... The system outputs a probability prediction of the paver's damage level during future work cycles, along with related maintenance costs. It transforms discrete operating parameters into continuous quantitative indicators reflecting the paver's core working capabilities, enabling an objective and dynamic analysis of the paver's overall operating condition. Through deep data fusion, it dynamically predicts the paver's wear rate, constructs and utilizes a deep learning model to handle high-dimensional nonlinear relationships, and comprehensively considers wear thickness and rate to output accurate damage level probabilities. Simultaneously, it links these probabilities to maintenance costs, ensuring that the damage level probability predictions directly benefit users, providing clear action guidelines and cost expectations. This facilitates user management decisions, enabling predictive maintenance of the paver equipment, effectively preventing sudden failures, and ensuring the continuity and stability of construction.
[0066] The specific steps for each unit module to implement the corresponding function in the deep learning-based construction economic indicator data analysis method of this application can be found in the steps in the embodiments of the deep learning-based construction economic indicator data analysis method above, and will not be repeated here.
[0067] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the foregoing application concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions claimed in this application.
Claims
1. A method for analyzing construction economic indicators based on deep learning, characterized in that, include: S1. Obtain data on the operation of the paver during highway construction, historical wear data of the screed, and current wear thickness data; S2. Based on the paver operating data during highway construction, perform paver health index analysis. S2 includes the following specific steps: S21. The paver screed efficiency index analysis results are obtained from the actual compaction data, target pre-compaction data, standard deviation data and average vibration acceleration data of the screed during the working cycle of the paver screed. S22. The paver conveying uniformity index analysis results are obtained from the left and right material height data of the paver. S23. The paver thermal decay index analysis results are obtained from the paver screed set temperature data, the current actual temperature change data with heating time data, and the initial standard heating time data. S24. The paver health index analysis results are obtained from the paver screed efficiency index analysis results, material conveying uniformity index analysis results, and thermal degradation index analysis results. S3. Based on the paver health index analysis results and the historical wear data of the paver screed, predict and analyze the wear rate of the paver screed. S4. Construct a deep learning model for predicting and analyzing paver screed damage. Input the current wear thickness data and wear rate prediction analysis results of the paver screed into the deep learning model for predicting and analyzing paver screed damage, and output the probability prediction result of paver screed damage level. The specific steps of S4 are as follows: Standardize and preprocess the data: Arrange 25 wear thickness measurement points regularly on the screed. Each time the measurement points are sampled, the collected current wear thickness data and wear rate prediction analysis results are reorganized into a 5x5 two-dimensional grid to preserve the relative positional relationship between the measurement points. Stack the thickness grid and the rate grid in the channel dimension to obtain a two-dimensional multi-channel feature map. This yields the first input sample data of the deep learning model for predicting and analyzing paver screed damage. In this way, the total input sample data of the deep learning model for predicting and analyzing paver screed damage is obtained. Input the collected and processed total input sample data of the deep learning model for predicting and analyzing paver screed damage into the trained deep learning model for predicting and analyzing paver screed damage, and output the probability prediction result of paver screed damage level. S5. Based on the probability prediction results of the damage level of the paver screed, output the probability prediction results of the damage level of the paver in the future working cycle, and output the relevant maintenance costs.
2. The method for analyzing construction economic indicators based on deep learning as described in claim 1, characterized in that, The specific steps of S21 are as follows: Based on the actual compaction data, target pre-compaction data, standard deviation data, and average vibration acceleration data within the working cycle of the paver screed, the paver screed efficiency index analysis is performed. The paver screed efficiency index analysis process involves: dividing the actual compaction data by the target pre-compaction data to obtain the screed compaction effect; dividing the standard deviation data of vibration acceleration within the working cycle by the average vibration acceleration data to obtain the screed vibration instability; subtracting the screed vibration instability from the standard deviation data to obtain the screed vibration stability; and multiplying the screed compaction effect and vibration stability to obtain the paver screed efficiency index analysis result.
3. The method for analyzing construction economic indicators based on deep learning as described in claim 2, characterized in that, The specific steps of S22 are as follows: Analyze the uniformity index of paver material conveying based on the left and right material height data of the paver. The paver material conveying uniformity index analysis process is as follows: After processing the absolute value of the difference between the left and right material height data of the paver material conveying, convert it to a percentage value to obtain the paver material conveying deviation. Subtract the paver material conveying deviation from the numerical value to obtain the paver material conveying uniformity index analysis result.
4. The method for analyzing construction economic indicators based on deep learning as described in claim 3, characterized in that, The specific steps of S23 are as follows: Based on the paver screed set temperature data, the current actual temperature change over heating time data, and the initial standard heating time data, a paver thermal degradation index analysis is performed. The paver thermal degradation index analysis process is as follows: Based on the paver screed set temperature data and the current actual temperature change over heating time data, the time required for the paver screed to reach the set temperature is obtained; based on the initial standard heating time data and the paver screed set temperature data, the initial heating time required to reach the current set temperature is obtained; the initial heating time required to reach the current set temperature is divided by the current heating time required for the paver screed to reach the set temperature to obtain the paver thermal degradation index analysis result.
5. The method for analyzing construction economic indicators based on deep learning as described in claim 4, characterized in that, The specific steps of S24 are as follows: Based on the analysis results of the paver screed efficiency index, the material conveying uniformity index, and the thermal decay index, a paver health index analysis is performed. The paver health index analysis process is as follows: The paver screed efficiency index, material conveying uniformity index, and thermal decay index are weighted and summed to obtain the paver health index analysis result.
6. The method for analyzing construction economic indicators based on deep learning as described in claim 5, characterized in that, The specific steps of S3 are as follows: the paver screed wear rate prediction analysis results are obtained by using the paver health index analysis results, the historical initial wear rate average data of the paver screed, and the wear rate influence fitting coefficient.
7. The method for analyzing construction economic indicators based on deep learning as described in claim 6, characterized in that, The specific steps of S5 are as follows: obtain the probability prediction result of the damage level of the paver screed, associate the probability prediction result of the damage level of the paver screed with the corresponding set maintenance cost, obtain the probability of the damage level of the paver in the future working cycle and the corresponding maintenance cost, and push the probability of the damage level of the paver in the future working cycle and the corresponding maintenance cost to relevant personnel for processing.