CN112252292B - Real-time highway compactness monitoring method based on artificial neural network - Google Patents
Real-time highway compactness monitoring method based on artificial neural network Download PDFInfo
- Publication number
- CN112252292B CN112252292B CN202011103702.4A CN202011103702A CN112252292B CN 112252292 B CN112252292 B CN 112252292B CN 202011103702 A CN202011103702 A CN 202011103702A CN 112252292 B CN112252292 B CN 112252292B
- Authority
- CN
- China
- Prior art keywords
- compaction
- data
- parameters
- road roller
- real
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- E—FIXED CONSTRUCTIONS
- E02—HYDRAULIC ENGINEERING; FOUNDATIONS; SOIL SHIFTING
- E02D—FOUNDATIONS; EXCAVATIONS; EMBANKMENTS; UNDERGROUND OR UNDERWATER STRUCTURES
- E02D1/00—Investigation of foundation soil in situ
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Soil Sciences (AREA)
- General Life Sciences & Earth Sciences (AREA)
- Mining & Mineral Resources (AREA)
- Paleontology (AREA)
- Civil Engineering (AREA)
- Structural Engineering (AREA)
- Road Paving Machines (AREA)
Abstract
The invention relates to a real-time monitoring method for highway compactness based on an artificial neural network, which trains the neural network by using historical data to establish a compactness calculation model, considers filling body parameters, road roller control parameters and sensor acquisition parameters, is applied to the real-time calculation and real-time display of the compactness in compaction construction, and improves the prediction precision and the calculation efficiency. Obtaining historical data according to a data platform, training a compaction degree calculation model, transferring the trained compaction degree calculation model into a road roller control system, inputting measured parameters of a filling body before compaction and road roller control parameters into the compaction degree calculation model transferred by the road roller control system before the road roller works, and feeding back the compaction degree in real time by combining the collected parameters of the road roller control and the acquired parameters of a sensor in the compaction process. The method considers different working conditions of various application scenes, creates conditions for establishing a monitoring system of the whole compaction process, and has universality.
Description
Technical Field
The invention relates to the technical field of roadbed and pavement engineering, in particular to a method for displaying compaction degree of an artificial intelligence method applied to a compaction process of an expressway, which can be used for monitoring compaction quality of roadbed and pavement of the expressway.
Background
The compaction of the highway is one of the most important construction links in the construction of the highway, and the compaction directly has essential influence on the use and maintenance of roads in the later period, so that the control of the compaction degree in the compaction process is particularly important, wherein the calculation of the compaction degree is the basis of the control of the compaction quality. The traditional road rolling quality control mainly adjusts various parameters in the road rolling process by means of experience and experience of a driver, is high in quality control difficulty and strong in subjectivity, can cause a series of problems of insufficient compaction, excessive compaction and the like, meanwhile, the whole compaction process cannot record various control parameters and compaction in real time, and quality tracing of a highway is difficult to carry out, so that resource waste is caused. The quality sampling inspection after compaction belongs to hysteresis control, and cannot play a guiding role in the quality control of the compaction process.
Traditional artificial neural networks calculate compaction in two forms: (1) the method can only be used as rough pre-construction guidance to give a general compaction strategy; an asphalt pavement compactness prediction model based on a BP artificial neural network is researched [ J ] traffic world, 2020, No.536(14):31-34 ] to construct the BP neural network, and the compactness of a target area is predicted according to paving temperature, compaction pass, compaction temperature and compaction speed, so that the relation between the influencing factors and the compactness is represented by small errors. However, the characteristic values of the method belong to a construction plan before construction or a construction general situation after construction, no real-time parameter is used as the characteristic value, the method cannot be applied to real-time compaction degree display in the compaction process, and the method cannot guide the quality control (2) of the compaction to predict the compaction degree by using various parameters of the compacted filling body, belongs to post prediction, searches for the relation among various indexes, and can be used as an evaluation method of the compaction degree after construction. The compaction degree of the filling body is predicted by utilizing the rolling thickness, the wet density, the dry density and the water content of the filling material, so that the defects of time waste, labor waste, damage and the like caused by the traditional compaction degree detection are avoided. However, the characteristic value selected by the method is an index of the compacted filling body, can only be used as the engineering acceptance or quality evaluation of the compaction degree after construction, and cannot provide information support for real-time display and real-time monitoring of the compaction degree in intelligent compaction control. The two forms of compaction degree calculation can not provide direct technical guidance for the compaction process, can not improve the compaction quality of the compaction process in the real sense, and can not be used as a data material for intelligent compaction.
Disclosure of Invention
In view of the defects of the current situation, the technical problem to be solved by the invention is to establish an artificial neural network-based real-time monitoring method for the compaction degree of the expressway, the method can be applied to real-time monitoring of the compaction degree in actual construction of the expressway, based on various real-time monitoring rolling parameters, the historical data is used for training the neural network to establish a correlation analysis model between the compaction measurement index and the compaction degree, namely a compaction degree calculation model, the filling body parameters, the road roller control parameters and the sensor acquisition parameters are considered, the filling body parameters, namely the intrinsic parameters, the road roller control parameters and the sensor acquisition parameters are real-time parameters, the method is applied to real-time calculation and real-time display of the compaction degree in compaction construction, and the prediction precision and the calculation efficiency are improved. The method considers various factors such as different working conditions, fillers and control parameters of various application scenes, has independent characteristic values, can display the compaction degree in real time in the compaction process of a roadbed, a water stabilization layer or a road surface, creates conditions for building a monitoring system in the whole compaction process, and has universality. Meanwhile, the limitation of hardware facilities such as road roller manufacturers and models is not considered, the compaction data in the compaction process is effectively utilized, and the self-optimization and self-correction of the prediction model can be realized. The method can provide data guidance for the traditional rolling process, and can also provide decision-making materials for forming intelligent compaction components.
In order to solve the technical problems, the technical scheme of the invention is as follows:
a real-time monitoring method for highway compactness based on an artificial neural network comprises the following steps:
1) establishing a compaction degree calculation model by utilizing a neural network:
the input to the neural network takes into account three factors: the method comprises the following steps that road roller control parameters, pre-compaction filling body parameters and sensor acquisition parameters are obtained, and the road roller control parameters and the pre-compaction filling body parameters represent difference information of a scene and a road roller to be utilized so as to correspond to different application environments; the sensor obtains parameters reflecting compaction state information in the compaction process; the output of the neural network is a mechanical property index capable of representing compaction;
establishing a correlation analysis model between the influence factors in the three aspects and mechanical performance indexes capable of representing compaction by using a neural network, namely a compaction degree calculation model;
2) constructing a compaction data platform:
the data platform has the functions of importing, exporting, optimizing, modifying and calling; the data source used by the compaction degree calculation model is a data platform, the data after compaction construction is finished is stored and uploaded to the data platform, and the data platform continuously utilizes gradually expanded data to optimize and update the compaction degree calculation model; the data platform calls the latest compaction degree calculation model to a road roller control system;
3) and (3) calculating the real-time compactness:
and acquiring historical data according to the data platform, training a compaction degree calculation model, transferring the trained compaction degree calculation model into a road roller control system, inputting the measured parameters of the filling body before compaction and the road roller control parameters into the compaction degree calculation model transferred by the road roller control system before the road roller works, and feeding back the compaction degree in real time by combining the collected parameters of the road roller control and the acquired parameters of the sensor in the compaction process.
A system used in the monitoring method, characterized in that: the road roller comprises a road roller, a data platform and a filling body; the road roller is provided with a recorder, a display screen, a manual input platform and a sensor; the data platform comprises a database and a computing center;
the database is used for collecting and storing historical data related to compaction of each expressway, is used as a data source of a calculation center and is also used as a data receiving position after each compaction is finished;
the calculation center can train a neural network by using the imported data, and finally, a neural network model is optimized to obtain a compactness calculation model for compactness display; meanwhile, the calculation center can call the compaction degree calculation model to a road roller control system to serve as a front end to guide the whole compaction process; the calculation center also receives road roller control parameters and filling body parameters;
the sensor receives real-time sensing information in the compaction process, the sensor comprises a temperature sensor and an acceleration sensor, the temperature sensor collects temperature data in real time, the acceleration sensor collects acceleration information in real time, and the collected information is arranged into a data group and transmitted to the computing center;
the display screen receives the calculation result data from the calculation center, can display the compaction degree of the current area in real time and can visually image;
the recorder is used for temporarily storing data information, temporarily storing the control parameters of the road roller, the filling body parameters, the real-time target values calculated by the compaction degree calculation model and the sensor acquisition parameters obtained by the sensor when the road roller works, and uploading the compaction degree data of the final spot check to the data platform together.
Compared with the prior art, the invention has the beneficial effects that:
1. the method is based on real-time compaction measurement data, utilizes an artificial neural network to analyze mechanical indexes after compaction, namely compaction degree, utilizes a historical data training network to establish a correlation analysis model, replaces the traditional calibration method of rolling times and the like by utilizing a test road section, solves the defects that the existing neural network cannot monitor in real time and is not suitable for different construction scenes, and obviously improves the calculation precision.
2. The method is suitable for road pressing scenes in various environments, has comprehensive characteristic values and universality, and solves the problem that how to display the compaction degree in real time and the target value of the model is different.
3. The existing data of the data platform are effectively utilized, the current compaction data are recorded, the construction data are guaranteed not to be lost, the construction data can be traced, and the method is an important basis for establishing a highway data system. The method can be used for construction of various highway sections and has universality, data universality and sharing under different conditions are realized due to the universality of the method, and rich data platforms can be continuously expanded during each construction, so that the method has great significance for constructing highway large data platforms.
4. The method has the characteristics of intelligence, real time, high precision and the like, continuously receives the information of compaction sensing in the compaction construction, displays the compaction degree in real time, is beneficial to the following control feedback, and is an indispensable component of an intelligent compaction control system.
In a word, the method can efficiently and conveniently acquire the compaction degree in the compaction process, provide accurate, real-time and reliable data materials for monitoring the compaction quality, and reduce the human subjectivity. The method constructs and maintains a data platform, realizes automatic display of the compaction degree of the road roller in the whole process, continuously upgrades the compaction data platform, and promotes the forward development of intelligent compaction.
Drawings
FIG. 1 is a diagram showing the relationship between the components of the present invention;
FIG. 2 is a diagram of a hybrid neural network architecture according to the present invention;
FIG. 3 is a step chart showing the real-time compaction degree display according to the present invention.
Detailed Description
Specific examples of the present invention are given below. The specific examples are only for illustrating the present invention in further detail and do not limit the scope of protection of the present application.
The invention discloses a real-time highway compactness monitoring method based on an artificial neural network, which comprises the following steps:
1. and establishing a compaction degree calculation model by utilizing a neural network.
The input to the neural network takes into account three factors: road roller control parameters, filling body parameters before compaction and sensor acquisition parameters. The road roller control parameters and the pre-compaction filling body parameters represent the difference information of the scene and the road roller used, so as to correspond to different application environments, different road roller models and the like. The parameters acquired by the sensors reflect the information of the compaction condition in the compaction process and are the most direct factors influencing the current compaction. The output of the neural network is a mechanical property index capable of representing compaction, and the embodiment takes the degree of compaction as an example.
And establishing a correlation analysis model between the compaction measurement index and the compaction degree, namely a compaction degree calculation model.
The measurement index refers to all parameters of the input neural network, each parameter is a characteristic value, and the parameters are divided into three types.
The intrinsic parameters are parameters which do not change in the characteristic values of the calculation model for calculating the compaction degree, namely parameters which do not change in the compaction process after the high-speed road section and the filling body are selected, and the intrinsic parameters can enable the calculation model for the compaction degree to be applied to different construction scenes; the real-time parameters can reflect the degree of compaction in real time, and can change in real time in the compaction process, such as road roller control parameters, sensor acquisition parameters and the like.
2. And constructing a data platform. The data platform has functions of importing, exporting, optimizing, modifying, calling and the like. The data source used by the compaction degree calculation model is a data platform, the data after compaction construction is finished is stored and uploaded to the data platform, and the data platform continuously utilizes gradually expanded data to optimize and update the compaction degree calculation model. The data platform can conveniently call the latest compaction degree calculation model into the compaction control system of the road roller.
3. And calculating the compaction degree in real time. And acquiring historical data according to the data platform, training a compaction degree calculation model, transferring the trained compaction degree calculation model into a road roller control system, inputting the measured parameters of the filling body before compaction into the compaction degree calculation model transferred into the road roller control system before the road roller works, and feeding back the compaction degree in real time by combining the collected parameters of the road roller control and the acquired parameters of the sensor in the compaction process.
The invention constructs a method for displaying the compaction degree in real time by using a neural network, which can be used for the front end of intelligent compaction and provides stable condition parameters to guide the control feedback of the intelligent compaction. As shown in fig. 1, the system used in the monitoring method comprises three components in the highway compaction process: the system comprises a road roller, a data platform and a filling body; the road roller is provided with a recorder, a display screen, a manual input platform, a sensor, a control system and the like; the data platform comprises a database and a computing center; the data platform can be a central server and the like, a display screen in the road roller is a liquid crystal display screen and the like, the manual input platform can be a mouse and a pad disc and is connected with a road roller control system, and input parameters of the manual input platform enter a currently called compaction degree calculation model in the road roller control system; the recorder can be a storage device, temporarily stores the control parameters of the road roller, the real-time target values calculated by the filling body parameters and the compaction degree calculation model, and the sensor acquisition parameters obtained by the sensor when the road roller works, and simultaneously records the real compaction degree data of the final spot check of the current construction road section and the data of the corresponding three types of parameters as historical data for training to be sent to a database; the sensors are a temperature sensor and an acceleration sensor and are connected with a road roller control system.
In fig. 1, reference numerals denote relationships between components.
And part 1, the data platform comprises a database and a computing center, and the database can derive data and transmit the data to the computing center. Corresponding to the number 1 in FIG. 1
The database is used for collecting data, storing the data and simply processing the data, is used as a data source of a calculation center and is also used as a data receiving position after each compaction is finished. The database collects and stores historical data related to compaction from various highways, the characteristic value part input by the neural network comprises information difference between different highways, and the data of the database is sent to the computing center, so that the data between different highways can be shared and interacted.
The form of the database needs to be interfaced with information required by a computing center, namely, the database needs to be acquired according to the data and form required by the computing center, and the parameter types and the parameter styles need to be kept consistent, namely, the computing center establishes a compaction degree computing model by using a neural network to compute the compaction degree, the data needs to be collected according to the parameter types of the figure 2 to obtain characteristic values and target values, and the data needs to be acquired according to a unified unit, a unified format and a text type required by a computing center interface. Including the fill parameters: 1. and (3) carrying out statistics on aggregate gradation of each expressway according to d10-d100, and regarding the aggregate gradation as ten characteristic values, wherein d eta is the particle size corresponding to the fact that the cumulative percentage content of soil with less than a certain particle size is eta%. 2. The glue-sand ratio of the water stable layer of the corresponding highway. 3. The water content of the highway filling material. 4. Porosity of highway filler material. And for the pile number sections with filling body parameter difference, respectively collecting data according to the pile number sections, and otherwise, uniformly collecting and storing the data according to the whole road.
The method comprises the following steps of: 1. and (3) the pre-paving thickness of the road roller is the expected paving thickness in the compaction scheme before compaction. 2. The pre-vibration frequency of the road roller is generally divided into two states: big shake and little shake, data acquisition method mark is "0" and "1" respectively. For pile number sections with different compaction vibration schemes, data are collected respectively according to the pile number sections, and otherwise, data are collected and stored uniformly according to the whole road; the difference corresponds to a change in paving thickness and vibration frequency. 3. The road roller driving direction is divided into a positive direction, the data acquisition method is marked as '1', a negative direction and a data acquisition method is marked as '0'. 4. The running speed of the road roller, which is measured in real time in the compaction process, is a continuous variable with uniform units.
The method comprises the following steps of acquiring parameters by a real-time sensor: 1. and the compaction temperature of the road roller comprises the compaction temperature measured at each measuring point. 2. The Intelligent Compaction Measurement Value (ICMV) is a characteristic quantity of vibration response in a compaction process and is measured by statistics of vibration acceleration vibration waves of the road roller. Such as may be represented by a Compaction Measure (CMV), a compaction control measure (CCV), a compaction run out (BV), or the like.
In this embodiment, a CMV value is used as a vibration response index, that is, the ICMV is characterized by CMV, and the calculation method is as follows:
CMV＝CA2Ω/A1Ω
wherein A is1Ω、A2ΩBase amplitude after low pass filtering for acceleration signalValue and second harmonic amplitude; c is a constant.
The method comprises direct compaction measurement indexes, namely mechanical property indexes capable of representing compaction, and parameters capable of representing the mechanical property of the compacted filling, such as the optional compactness, the elastic modulus and the like. The present embodiment uses the degree of compaction as a characterizing parameter of the compaction mechanics index. When the historical data is subjected to feature storage, because two parameters of the sensor acquisition parameters belong to continuous measurement values, the real compactness belongs to index values obtained by post sampling inspection, and the measurement units of the sensor acquisition parameter characteristic values and the compactness target values for neural network training are required to be consistent, the measurement units of the sensor acquisition parameter characteristic values are obtained according to the real acquisition measurement units of the compactness, and if the measurement units do not correspond to each other, the temperature and the sensor acquisition parameters need to be subjected to interval processing. If the measured value of the compactness is obtained every one hundred meters, positioning a sensor corresponding to the compactness measuring point to obtain parameters according to the real-time position recorded in the compaction process of the road roller, wherein the measuring unit is still one hundred meters; if the single-point pile number compaction degree is measured, the measurement units of the parameters acquired by the sensors are unified into single pile number points. If the parameters obtained by the real-time sensor have missing items, the missing item parameters can be changed into inherent parameters, and the inherent parameters are changed into uniform values when corresponding to other parameters. The intrinsic parameters are parameters which do not change in the characteristic values of the calculation model for calculating the compaction degree, the relative sensor acquisition parameters are parameters which change all the time in the compaction process, and if some sensor acquisition parameters cannot be acquired, the parameters can be changed into uniform constants to be input into the neural network.
The GPS position information is the corresponding position information when various data are collected, and is also the basis for matching various data into a group. The position information is not used as a characteristic value, but the position information needs to be collected, the characteristic value is corresponding to a target value through the position information, for example, a sensor with a pile number K2+000 obtains a parameter, and a road roller control parameter corresponds to the compactness of a pile number point.
Therefore, the database has a data export function, and historical data are imported into the computing center according to a fixed form required by the computing center.
The computing center can train the neural network by using the imported data
In the process, a neural network framework needs to be constructed, the imported data in the database is processed, a neural network model is trained, and finally the neural network model is optimized to obtain a compactness calculation model for compactness display.
1) Building neural network framework
As shown in fig. 2, the characteristic values of the neural network include the aforementioned filling body parameters, road roller control parameters and sensor acquisition parameters. The filling parameters include: gradation, rubber-sand ratio, water content and porosity. The road roller control parameters include: paving thickness, vibration frequency, driving speed and driving direction. The sensor acquiring parameters comprises: compaction temperature and ICMV. The target value is the degree of compaction.
The neural network adopts a network structure of mixing a long-short term memory neural network (LSTM-NN) and a BP neural network (BP-NN), and comprises an input layer, a hidden layer and an output layer, wherein the hidden layer comprises an LSTM layer and a plurality of BP layers. The LSTM layer characterizes the near-region synergy of compaction, and the BP characterizes the non-linear behavior of static prediction. In order to prevent the dilemma of trapping the local minimum of the cost function caused by the excessive BP layers, a neuron discarding layer (Dropout layer) is added between the LSTM layer and the BP layer, and meanwhile, weight regularization and bias regularization are carried out on each hidden layer, so that the generalization capability of the hybrid neural network is ensured.
2) Training neural networks
After the database data is obtained, processing the data, including: and eliminating noise values and interpolating vacancy values. After the available data set is obtained, a multi-layer verification method is used for dividing the training set, the testing set and the verification set. The multi-layer verification method is that the verification set comprises a plurality of data subsets, each data subset needs to comprise data combinations with different characteristic values, each data subset is a series of data combinations with only one characteristic value different from other characteristic values and the same characteristic value, or each data subset is a series of data combinations with one type of characteristic value different from other types of characteristic values and the same characteristic value, for example, the data of the first subset is the data of the same highway with the same filling body parameters and different road roller setting parameters, the data of the second subset is the data of the road roller with the same parameters and different rolling properties of the filling body, and the data of the third subset is the data of different highways. The method can effectively avoid the defect of narrow applicable scene of the neural network. The training set and the test set do not contain the same type of data as the validation set, and the training set and the validation set are as follows, 0.8: the scale of 0.2 is divided laterally. And then vertically dividing the data according to the characteristic value and the target value.
Training a neural network, and optimizing the neural network by using a grid search method. The parameters needing to be manually set in the neural network training process are hyper-parameters, and the hyper-parameters included in the invention are as follows: the method comprises the steps of learning rate, iteration times, batch processing data size, regularization options and Dropout layer setting, the number of neurons in each layer, the batch processing data size refers to the data volume of each iteration, the regularization options are classified into the presence, absence and types, and the Dropout layer setting is the presence, absence or discarding rate. The method process of grid search is as follows: randomly combining the super parameters according to a certain gradient, wherein each combination is a network structure, each structure is evaluated in a cross validation mode, namely, a training set is averagely divided, the division number is a fold number, the average of precision obtained by multi-fold data training is taken as the precision of the combination under the super parameters, and a group of super parameters with the minimum error is selected; and finally obtaining the neural network which requires precision control.
3) Saving compaction calculation model and embedding into road roller
The trained neural network is a compaction degree calculation model, and the compaction degree calculation model is stored in a programming environment to be used as a backup. And simultaneously, calling the compaction degree calculation model as a subsystem into a control system of the road roller, connecting the front end of the subsystem with a sensor and a manual input platform of the road roller, and connecting the terminal of the subsystem with a display screen of the road roller.
Thus, the compute farm has write, compute, optimize, and call functions.
The road roller comprises a sensor, a display screen and a recorder. The sensor receives sensing information in the compaction process, corresponding to a serial number 4 in fig. 1, wherein the sensing information comprises real-time collection of filling body temperature data by the temperature sensor, real-time collection of acceleration information by the acceleration sensor, processing and conversion of the acceleration information into an ICMV value, arrangement of the collected information into a data set and transmission of the data set to a calculation center, corresponding to a serial number 5 in fig. 1. The calculation center also receives the control parameters and the filling body parameters of the road roller, and continuously calculates the real-time compaction degree in the compaction process through the trained and optimized neural network corresponding to the sequence number 2 and the sequence number 3 in the figure 1. The calculation center transfers the compaction degree calculation model to the road roller control system, a display screen can receive calculation result data of the compaction degree calculation model self-adjusted to the road roller control system corresponding to the figure 1 serial number 6, and meanwhile, the display screen has a visualization function, can display the compaction degree of the current area in real time, and can visually image after processing. The road roller receives real-time compaction degree information predicted by a compaction degree calculation model according to the current condition, and feeds the real-time compaction degree information back to a control system of the road roller according to the current compaction state, and the real-time compaction degree information is used for adjusting various control parameters of compaction in real time at the later stage, and belongs to the field of intelligent compaction. The recorder is a part for temporarily storing data information, and is used for correspondingly and temporarily storing the control parameters of the road roller, the filling body parameters, the compaction degree data of each moment predicted by the compaction degree calculation model and the sensor acquisition parameters obtained by the sensor at each moment in the working process of the road roller, uploading the instantaneous data to a database, and correspondingly storing the instantaneous data according to the pile number section or the whole expressway, so that the transient data in the later compaction process can be conveniently traced; and performing spot check after compaction, inputting and storing a spot check result in a recorder through a manual input platform, and uploading the compaction degree data of the spot check and the interval real-time parameters and inherent parameters corresponding to the section to a database as a whole to serve as new historical data for training the neural network model. Therefore, the recorder can record the control parameters of the road roller, the parameters of the filling body and the acquisition parameters of the sensor, the compaction values of the later-stage spot check are transmitted to the database together, and the database receives data corresponding to the serial number 8 in the figure 1.
The data platform has the functions of data import, storage, modification, calculation, calling, export, visualization and the like. The compaction data recorded by the recorder and the sampling inspection data after compaction are imported into a data platform, and the data platform allows an import data interface. The data required in the compaction process can be imported into the database according to requirements, the database completes the data processing work in the previous stage, the data processing work comprises data cleaning, dimensionless, dimension unification, data enhancement, dimension reduction and the like, the data processing method is fixed, and the data processing method can be realized by code programming. The obtained data can be stored according to a certain mode, the storage position can be searched and can be modified, the mode in the storage according to the certain mode is a mode acceptable by the compactness calculation model, for example, the interface requirement for calculating the compactness calculation model is in a table form, the type requirement is in a floating point type, the characteristic value and the target value need to be longitudinally arranged, the index is added, and the data type is set to be the floating point type. And the calculation center calls the compaction degree calculation model into a road roller control system under a certain construction environment, and the road roller can calculate the data in real time by using the compaction degree calculation model to obtain the compaction degree under the current state. Meanwhile, the calculation center can call the compaction degree calculation model based on the neural network into a compaction control system of the road roller to serve as a front end to guide the whole compaction process. Data in the database can be exported at any time to be used for quality evaluation and data tracing during maintenance of the highway, the data in the database are stored according to the number of highway piles, the highway corresponding to the data, time and the like are marked during data exporting, and tracing is facilitated.
The road roller has the functions of manual input, information acquisition, calculation, display, data recording, uploading and the like. Before compaction, parameter values can be manually input according to a pre-construction scheme and a filling body state of the road roller, in the compaction process, sensing signals are collected in real time and converted into the parameter values, real-time calculation is carried out according to a compaction degree calculation model based on a neural network called by a data platform, and the parameter values are displayed in a display screen. And finally, uploading the data of the compaction process and the actual compaction degree data of the spot check after compaction to a data platform, reusing the data of the compaction after the compaction is finished, uploading the data to the data platform, updating and training the compaction degree calculation model again, and finishing self-correction of the compaction degree calculation model for construction of the next high-speed road section to form closed-loop transmission of data flow.
The specific operation flow of the compaction construction is shown in figure 3.
Step 1, extracting historical data to train a neural network to obtain a trained compactness calculation model, wherein each hyper-parameter and weighted value of the neural network are fixed and unchangeable at the moment;
In the embodiment, the compactness is used as a characterization parameter of the compaction mechanics index, the sensor acquisition parameter input during the neural network model training must be consistent with the measurement unit of the compactness, two types of parameters of the real-time sensor acquisition parameter actually obtained belong to continuous measurement values, and the compactness belongs to an index value obtained by post spot check, so that the measurement unit of the sensor acquisition parameter input during the training is obtained according to the acquisition measurement unit of the compactness. If the measured value of the compactness is obtained every one hundred meters, positioning a sensor acquisition parameter corresponding to the compactness measuring point according to the real-time position recorded in the compaction process of the road roller, and carrying out interval processing on the temperature and the sensor acquisition parameter, wherein the measuring unit is still one hundred meters; if the single-point pile number compaction degree is measured, the measurement units of the sensor acquisition parameters input during training are unified into single pile number points. If the parameters acquired by the sensors have missing items, the missing item parameters can be changed into inherent parameters, and the inherent parameters are changed into uniform values when corresponding to other parameters. The intrinsic parameters are parameters which do not change in the characteristic values of the calculation model for calculating the compaction degree, the relative sensor acquisition parameters are parameters which change all the time in the compaction process, and if some sensor acquisition parameters cannot be acquired, the parameters can be changed into uniform constants to be input into the neural network.
And 5, importing data into the database, accumulating the newly uploaded data to historical data by the computing center, and training the compaction degree computing model again to realize self-updating of the compaction degree computing model and reuse of the data.
In the method, multiple factors are considered, the characteristic values considered by the neural network comprise comprehensive types and comprise filling body parameters representing construction scenes and road roller control parameters, the sensors acquire the parameters, and the verification set adopts a multi-layer verification method, so that the compaction degree of the dynamic compaction process can be displayed in real time, the dynamic compaction process can be applied to various scenes, the dynamic compaction process can be used without marking sections in advance on new scenes, and the precision of compaction control is improved. Meanwhile, the data platform accepts various data of various application scenes, the data is many and comprehensive, the data constructed each time can be reused and transmitted to the data platform to carry out self-correction updating on the neural network, and therefore the method has universality.
Nothing in this specification is said to apply to the prior art.
Claims (5)
1. A real-time monitoring method for highway compactness based on an artificial neural network comprises the following steps:
1) establishing a compaction degree calculation model by utilizing a neural network:
the input to the neural network takes into account three factors: the method comprises the following steps that road roller control parameters, filling body parameters and sensor acquisition parameters are obtained before compaction, and the road roller control parameters and the filling body parameters represent difference information of a scene and a road roller to be utilized so as to correspond to different application environments; the sensor obtains parameters reflecting compaction state information in the compaction process; the output of the neural network is a mechanical property index capable of representing compaction,
establishing a correlation analysis model between the influence factors in the three aspects and mechanical performance indexes capable of representing compaction by using a neural network, namely a compaction degree calculation model;
2) constructing a compaction data platform:
the data platform has the functions of importing, exporting, optimizing, modifying and calling; the data source used by the compaction degree calculation model is a data platform, the data after compaction construction is finished is stored and uploaded to the data platform, and the data platform continuously utilizes gradually expanded data to optimize and update the compaction degree calculation model; the data platform calls the latest compaction degree calculation model to a road roller control system;
3) and (3) calculating the real-time compactness:
acquiring historical data according to the data platform, training a compaction degree calculation model, transferring the trained compaction degree calculation model into a road roller control system, inputting measured parameters of the filling body before compaction into the compaction degree calculation model transferred into the road roller control system before the road roller works, and feeding back the compaction degree in real time by combining the collected parameters of the road roller control and the acquired parameters of the sensor in the compaction process;
the construction process of the monitoring method comprises the following steps:
step 1, extracting historical data to train a neural network, wherein parameters needing to be set manually in the training process of the neural network are hyper-parameters, a trained compaction degree calculation model is obtained, and all hyper-parameters and weight values of the neural network are fixed;
step 2, collecting parameters of the filling body before compaction, inputting the parameters into a compaction degree calculation model, and inputting the parameters obtained by a current compaction sensor and control parameters of the road roller into the compaction degree calculation model in the compaction process to obtain the current compaction degree;
step 3, displaying the current compaction degree on a display screen of the road roller, and judging whether the compaction of the road roller is finished at the moment according to the comparison between the compaction degree of the current road section and the standard index; if not, inputting the data acquired in the current compaction process into the compaction degree calculation model for recalculation and displaying to obtain a continuous value of the compaction degree;
step 4, when the compaction degree of the compacted road section reaches the standard index requirement, finishing compaction; uploading the compacted spot check result as a compaction degree real value to a database, and simultaneously uploading road roller control parameters, filling body parameters and sensor acquisition parameters corresponding to the compaction degree real value to the database;
and 5, the data platform comprises a database and a calculation center, the database imports data, the calculation center accumulates newly uploaded data into historical data, and the compaction degree calculation model is trained again to realize self-updating of the compaction degree calculation model and data reuse.
2. The real-time highway compactness monitoring method based on the artificial neural network as claimed in claim 1, wherein the parameter types and data pattern rules of the data in the data platform are as follows:
filling body parameters: (1) aggregate gradation of each expressway is counted according to d10-d100 and is regarded as ten characteristic values, wherein d eta is a particle size corresponding to the fact that the cumulative percentage content of soil with a particle size smaller than certain weight is eta%; (2) the glue-sand ratio of the water stable layer of the corresponding expressway; (3) the water content of the highway filling material; (4) porosity of highway filler material; (5) for the pile number sections with filling body parameter difference, data are collected respectively according to the pile number sections, otherwise, data are collected and stored uniformly according to the whole road;
controlling parameters of the road roller: (1) pre-paving thickness of the road roller, and predicting the paving thickness in a compaction scheme before compaction; (2) the pre-vibration frequency of the road roller is divided into two states: big vibration and small vibration, the data acquisition method is marked as '0' and '1' respectively; for pile number sections with different compaction vibration schemes, data are collected respectively according to the pile number sections, otherwise, data are collected and stored uniformly according to the whole road, and the difference corresponds to the change of paving thickness and vibration frequency; (3) the road roller driving direction is divided into a positive driving direction, a data acquisition mode is marked as '1', a negative driving direction and a data acquisition mode is marked as '0'; (4) the running speed of the road roller is measured in real time in the compaction process and is a continuous variable with uniform units;
acquiring parameters by a sensor: (1) the compaction temperature of the road roller comprises the compaction temperature measured at each measuring point; (2) the intelligent compaction measurement value ICMV is a characterization quantity of vibration response in the compaction process, and is measured by statistics after vibration acceleration vibration waves of the road roller are processed, wherein the statistics comprises at least one of a compaction degree measurement value CMV, a compaction control measurement value CCV and a compaction jump value BV; both parameters of the sensor acquisition parameter belong to continuous measurement values,
the mechanical property index of compaction can be represented, and the mechanical property index comprises at least one of compactness and elastic modulus; the acquisition parameters of the sensor for model training and the measurement units of the real compactness are required to be consistent, and the real compactness belongs to the index values obtained by post spot inspection, so that the measurement units of the acquisition parameters of the sensor for model training are obtained according to the acquisition measurement units of the real compactness; if the parameters obtained by the sensor used for model training have missing items, the missing item parameters are changed into inherent parameters, and the inherent parameters are changed into uniform values when corresponding to other parameters;
the GPS position information is the corresponding position information when various data are collected, and is also the basis for matching various data into a group.
3. The method for real-time monitoring of highway compactness based on artificial neural network as claimed in claim 1, wherein said neural network adopts a mixed network structure of long and short term memory neural networks LSTM-NN and BP-NN, comprising an input layer, a hidden layer and an output layer, the hidden layer comprises an LSTM layer and a plurality of BP layers; the LSTM layer represents the cooperative characteristic of the similar areas of the compactness, and the BP represents the nonlinear characteristic of static prediction; in order to prevent the dilemma of trapping the local minimum value of the cost function caused by the excessive BP layers, a neuron discarding layer is added between the LSTM layer and the BP layer, and meanwhile, weight regularization and bias regularization are carried out on each hidden layer, so that the generalization capability of the hybrid neural network is ensured.
4. The real-time highway compactness monitoring method based on the artificial neural network as claimed in claim 3, wherein the historical data is obtained according to the data platform, and the process of training the compactness calculation model is as follows:
after obtaining historical highway compaction related data, processing the data, including: noise value elimination and vacancy value interpolation processing are carried out to obtain a data set, and then a multi-layer verification method is used for dividing a training set, a testing set and a verification set;
the multi-layer verification method is that a verification set comprises a plurality of data subsets, each data subset needs to comprise data combinations with different characteristic values, namely each data subset is a series of data combinations with only one characteristic value different from other characteristic values, or each data subset is a series of data combinations with one type of characteristic value different from other types of characteristic values; training and validation sets were as per 0.8: performing horizontal division according to the proportion of 0.2, and then performing vertical division on the data according to the characteristic value and the target value;
finally, optimizing the neural network by using a grid search method to obtain a trained compactness calculation model;
and storing the trained compaction degree calculation model in a programming environment as a backup, calling the trained compaction degree calculation model into a control system of the road roller as a subsystem, connecting the front end of the subsystem with a sensor and a manual input platform of the road roller, and connecting the terminal of the subsystem with a display screen of the road roller.
5. The system for the real-time monitoring method of the highway compactness based on the artificial neural network as claimed in claim 1, is characterized in that: the road roller comprises a road roller, a data platform and a filling body; the road roller is provided with a recorder, a display screen, a manual input platform and a sensor; the data platform comprises a database and a computing center;
the database is used for collecting and storing historical data related to compaction of each expressway, is used as a data source of a calculation center and is also used as a data receiving position after each compaction is finished;
the calculation center can train a neural network by using the imported data, and finally, a neural network model is optimized to obtain a compactness calculation model for compactness display; meanwhile, the calculation center can call the compaction degree calculation model to a road roller control system to serve as a front end to guide the whole compaction process; the calculation center also receives road roller control parameters and filling body parameters;
the sensor receives real-time sensing information in the compaction process, the sensor comprises a temperature sensor and an acceleration sensor, the temperature sensor collects temperature data in real time, the acceleration sensor collects acceleration information in real time, and the collected information is arranged into a data group and transmitted to the computing center;
the display screen receives the calculation result data from the calculation center, can display the compaction degree of the current area in real time and can visually image;
the recorder is used for temporarily storing data information, temporarily storing the control parameters of the road roller, the filling body parameters, the real-time target values calculated by the compaction degree calculation model and the sensor acquisition parameters obtained by the sensor when the road roller works, and uploading the compaction degree data of the final spot check to the data platform together.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011103702.4A CN112252292B (en) | 2020-10-15 | 2020-10-15 | Real-time highway compactness monitoring method based on artificial neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011103702.4A CN112252292B (en) | 2020-10-15 | 2020-10-15 | Real-time highway compactness monitoring method based on artificial neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112252292A CN112252292A (en) | 2021-01-22 |
| CN112252292B true CN112252292B (en) | 2021-05-28 |
Family
ID=74243444
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011103702.4A Active CN112252292B (en) | 2020-10-15 | 2020-10-15 | Real-time highway compactness monitoring method based on artificial neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112252292B (en) |
Families Citing this family (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112985608B (en) * | 2021-02-01 | 2022-08-02 | 河北工业大学 | Method and system for monitoring temperature in asphalt conveying process |
| CN112862200B (en) * | 2021-02-20 | 2022-08-02 | 河北工业大学 | Intelligent feedback optimization method for whole process of highway construction quality |
| CN112734305A (en) * | 2021-02-24 | 2021-04-30 | 西南交通大学 | Roadbed compaction quality evaluation method, device and equipment and readable storage medium |
| CN113063930B (en) * | 2021-03-22 | 2024-01-23 | 上海建工建材科技集团股份有限公司 | Neural network-based 3D printed concrete mechanical property online monitoring method |
| CN113174918B (en) * | 2021-04-14 | 2022-07-15 | 中交四航工程研究院有限公司 | Method for detecting transverse compaction degree and uniformity of roadbed |
| CN113129283B (en) * | 2021-04-19 | 2022-11-29 | 河北工业大学 | Digital image-based filler grading intelligent detection compaction control method |
| CN113534667B (en) * | 2021-07-30 | 2023-07-04 | 清华大学 | Method and device for adjusting vibration compaction parameters of rock-fill material |
| CN113821861A (en) * | 2021-10-14 | 2021-12-21 | 北京壹陆科技发展有限公司 | Roadbed and pavement compaction quality measurement method and system based on multiple regression |
| CN114323853A (en) * | 2021-12-27 | 2022-04-12 | 广州城投路桥工程有限公司 | Intelligent monitoring method for asphalt pavement compactness based on decision tree model |
| CN114493291B (en) * | 2022-01-28 | 2022-11-01 | 中铁北京工程局集团有限公司 | Intelligent detection method and system for high fill quality |
| CN115034697B (en) * | 2022-08-12 | 2022-11-08 | 河北工业大学 | Multi-domain analysis-based multi-element intelligent compaction index grading optimization method and system |
| CN116383697B (en) * | 2023-03-29 | 2024-02-06 | 四川省交通建设集团有限责任公司 | Bituminous pavement compaction degree prediction method based on double-gate circulation network |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102289716A (en) * | 2011-06-18 | 2011-12-21 | 合肥工业大学 | Method for modeling neural network of optimum working parameters of intelligent road roller |
| CN108764540A (en) * | 2018-05-16 | 2018-11-06 | 杭州电子科技大学 | Water supply network pressure prediction method based on parallel LSTM series connection DNN |
| CN109356003A (en) * | 2018-09-18 | 2019-02-19 | 北京龙马智行科技有限公司 | A kind of subgrade and pavement intelligence rcc system |
| CN109468918A (en) * | 2018-10-31 | 2019-03-15 | 北京龙马智行科技有限公司 | A kind of subgrade and pavement intelligence compacting decision system |
| CN109783988A (en) * | 2019-03-01 | 2019-05-21 | 中国水利水电第七工程局有限公司 | A kind of grinding coagulation soil compaction evaluation method based on GA-BP network |
| CN109919481A (en) * | 2019-03-01 | 2019-06-21 | 河海大学 | A kind of grinding coagulation soil compaction multi-parameter joint real-time estimating method |
| CN110453572A (en) * | 2019-07-30 | 2019-11-15 | 阳光学院 | Electric-controlled intelligent hydraulic vibration road roller system and its control method |
| CN111105332A (en) * | 2019-12-19 | 2020-05-05 | 河北工业大学 | Highway intelligent pre-maintenance method and system based on artificial neural network |
| CN111707265A (en) * | 2020-06-02 | 2020-09-25 | 天津大学 | Compaction quality evaluation-based roller-repairing path planning system and method |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11071450B2 (en) * | 2016-06-29 | 2021-07-27 | Ace Vision Group, Inc. | System and methods using real-time predictive virtual 3D eye finite element modeling for simulation of ocular structure biomechanics |
| WO2018034745A1 (en) * | 2016-08-18 | 2018-02-22 | The Regents Of The University Of California | Nanopore sequencing base calling |
-
2020
- 2020-10-15 CN CN202011103702.4A patent/CN112252292B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102289716A (en) * | 2011-06-18 | 2011-12-21 | 合肥工业大学 | Method for modeling neural network of optimum working parameters of intelligent road roller |
| CN108764540A (en) * | 2018-05-16 | 2018-11-06 | 杭州电子科技大学 | Water supply network pressure prediction method based on parallel LSTM series connection DNN |
| CN109356003A (en) * | 2018-09-18 | 2019-02-19 | 北京龙马智行科技有限公司 | A kind of subgrade and pavement intelligence rcc system |
| CN109468918A (en) * | 2018-10-31 | 2019-03-15 | 北京龙马智行科技有限公司 | A kind of subgrade and pavement intelligence compacting decision system |
| CN109783988A (en) * | 2019-03-01 | 2019-05-21 | 中国水利水电第七工程局有限公司 | A kind of grinding coagulation soil compaction evaluation method based on GA-BP network |
| CN109919481A (en) * | 2019-03-01 | 2019-06-21 | 河海大学 | A kind of grinding coagulation soil compaction multi-parameter joint real-time estimating method |
| CN110453572A (en) * | 2019-07-30 | 2019-11-15 | 阳光学院 | Electric-controlled intelligent hydraulic vibration road roller system and its control method |
| CN111105332A (en) * | 2019-12-19 | 2020-05-05 | 河北工业大学 | Highway intelligent pre-maintenance method and system based on artificial neural network |
| CN111707265A (en) * | 2020-06-02 | 2020-09-25 | 天津大学 | Compaction quality evaluation-based roller-repairing path planning system and method |
Non-Patent Citations (1)
| Title |
|---|
| 碾压混凝土施工热层实时压实度监控模型应用研究;郑祥等;《人民长江》;20200128;第51卷(第1期);第160-165、179页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112252292A (en) | 2021-01-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112252292B (en) | Real-time highway compactness monitoring method based on artificial neural network | |
| Xu et al. | Adaptive quality control and acceptance of pavement material density for intelligent road construction | |
| CN109584510B (en) | Road high slope landslide hazard early warning method based on evaluation function training | |
| CN109468918B (en) | Intelligent compaction decision-making system for roadbed and pavement | |
| CN109753745B (en) | Road detection and evaluation method based on BIM and Internet of things technology | |
| CN115034697B (en) | Multi-domain analysis-based multi-element intelligent compaction index grading optimization method and system | |
| CN108396723A (en) | A kind of highway subgrade compaction quality rapid detection method | |
| CN109919481A (en) | A kind of grinding coagulation soil compaction multi-parameter joint real-time estimating method | |
| CN111444560A (en) | High-fill compaction quality real-time evaluation method considering influence of underlying surface | |
| CN113640175A (en) | Earth and rockfill dam material compaction quality evaluation method | |
| Xu et al. | The pioneer of intelligent construction—An overview of the development of intelligent compaction | |
| CN113821861A (en) | Roadbed and pavement compaction quality measurement method and system based on multiple regression | |
| CN113470008B (en) | Method and system for intelligently monitoring construction quality of asphalt pavement | |
| Yao et al. | Intelligent compaction methods and quality control | |
| CN112435058A (en) | Intelligent engineering cost management system based on cloud computing | |
| Karakikes et al. | Evaluation of freight measures by integrating simulation tools: the case of Volos Port, Greece | |
| JP6965489B2 (en) | Rolling count prediction system and rolling method | |
| CN116402951A (en) | Digital twinning-based side slope safety management method and system | |
| CN114819506A (en) | Online intelligent management and control platform and method for roadbed layered compaction quality | |
| CN116430739B (en) | Whole-process intelligent compaction system based on digital twin technology and control method | |
| Abdelmuhsen et al. | On the variants of SVM method for the estimation of soil elastic modulus from TSD model: Numerical parametric study | |
| Castro Illusanguil | Do we build what we design? Using data-driven approaches to couple the construction process into air void ratio regression | |
| CN116579520A (en) | Digital twin modeling method and device for carbon emission calculation | |
| CN117648746A (en) | Digital twinning-based tunnel construction concrete super-consumption data statistics method and system | |
| Zhang et al. | Comprehensive Evaluation Method for Asphalt Pavement Construction Quality Based on Internet of Things Technology |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |