Multi-sensor fused intelligent bituminous pavement compactness closed-loop control system

Abstract

The invention relates to the technical field of intelligent road construction control, and discloses a multi-sensor fusion asphalt pavement intelligent compactness closed-loop control system, which comprises a data acquisition module, a data processing module, a control algorithm module, a feedback control module and a user interaction interface, the data acquisition module is connected with the vibration acceleration sensor, the temperature sensor and the RTK differential GPS positioning terminal, and the vibration acceleration sensor is connected with the temperature sensor and the RTK differential GPS positioning terminal through coaxial cables; the feedback control module coordinates operation of each module through an embedded microcontroller, the microcontroller comprises an adaptive adjustment unit, the data processing module comprises a wavelet analysis unit, an IIR filtering unit and a feature extraction unit, and processed data is transmitted to the control algorithm module through a PCIe bus. According to the invention, the master sensor sends trigger signals to other slave sensors through the trigger signal lines, so that synchronous data acquisition can be realized; pID parameters are dynamically adjusted through a self-adaptive PID algorithm, and the method can adapt to complex working condition changes.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent road construction control, and in particular to a multi-sensor fusion intelligent compaction closed-loop control system for asphalt pavement. Background Art

[0002] The asphalt pavement compaction closed-loop control system refers to an automated system that uses sensors to collect vibration, temperature, displacement and other data during the construction process in real time, and dynamically adjusts roller parameters such as vibration frequency, amplitude, and rolling speed in combination with control algorithms to achieve precise control and optimization of the asphalt mixture density. The current asphalt pavement compaction control system has a single source of sensor data and only relies on vibration accelerometers or nuclear density meters. It lacks multi-source information fusion and dynamic modeling capabilities, resulting in insufficient compaction prediction accuracy and delayed control response. It is unable to achieve adaptive adjustment under complex working conditions such as temperature fluctuations and uneven materials, resulting in waste of resources and unstable construction quality. In addition, the existing multi-sensor fusion asphalt pavement intelligent compaction closed-loop control system collects data at different frequencies and times, and the data synchronization is poor, resulting in the fused data being unable to accurately reflect the compaction status, affecting the accuracy of control decisions.

[0003] Under different asphalt material properties (such as aggregate gradation and asphalt content), environmental conditions (temperature and humidity differences), and construction equipment (different roller models and performance), the fixed-parameter PID control algorithm cannot accurately adjust the compaction parameters. Summary of the Invention

[0004] In order to make up for the above shortcomings, the present invention provides a multi-sensor fusion asphalt pavement intelligent compaction closed-loop control system, which aims to improve the problem that the frequency and time of multi-sensor data collection are different, the data synchronization is poor, resulting in the fused data being unable to accurately reflect the compaction status, and the fixed parameter PID control algorithm being unable to accurately adjust the compaction parameters under different asphalt material properties, environmental conditions, and construction equipment.

[0005] To achieve the above objectives, the present invention adopts the following technical solutions: a multi-sensor fusion asphalt pavement intelligent compaction closed-loop control system, comprising a data acquisition module, a data processing module, a control algorithm module, a feedback control module, and a user interaction interface; the data acquisition module is connected to a vibration acceleration sensor, a temperature sensor, and an RTK differential GPS positioning terminal via a CAN bus protocol; the vibration acceleration sensor is connected to the temperature sensor and the RTK differential GPS positioning terminal via a coaxial cable; and the data acquisition module is connected to the data processing module via a CAN bus protocol;

[0006] The data processing module includes a wavelet analysis unit, an IIR filter unit and a feature extraction unit, and is connected to the control algorithm module via a PCIe bus;

[0007] The control algorithm module includes a support vector machine unit, a hidden Markov model unit and a BP neural network unit, which are parallel calculated by an FPGA accelerator and connected to a feedback control module through a serial communication interface;

[0008] The feedback control module includes an embedded microcontroller, which adjusts the vibration frequency, amplitude and rolling speed of the roller through the CANopen protocol, and the microcontroller includes an adaptive adjustment unit;

[0009] The user interaction interface communicates with other modules via the TCP / IP protocol.

[0010] A method for operating a closed-loop control system for intelligent compaction of asphalt pavement using multi-sensor fusion. The method for operating a data acquisition module process is as follows:

[0011] S1: The vibration acceleration sensor is used as the master sensor and has a BNC interface. The temperature sensor and the RTK differential GPS positioning terminal are installed with an ADC chip with an integrated trigger signal detection circuit as slave sensors. The master and slave sensors are connected via a coaxial cable.

[0012] S2: The master sensor sends a trigger signal at a fixed frequency. After receiving the signal, the slave sensors complete the acquisition within the same clock cycle to ensure that the acquisition time of all sensors is aligned.

[0013] S3: Measure the transmission delay of the trigger signal from the master sensor to the slave sensor, and compensate the timestamp of the slave sensor during data processing to eliminate the slight error caused by hardware delay.

[0014] Preferably, the method for running the data processing module process is as follows:

[0015] S1: The vibration acceleration signal is decomposed and reconstructed in five layers through the wavelet analysis unit, and the high-frequency coefficients are denoised by hard thresholding. The threshold formula is: Where N is the number of sampling points, and the low-frequency coefficients are filtered through a fourth-order elliptic filter to eliminate phase distortion;

[0016] S2: The feature extraction unit calculates the effective value RMS of the vibration acceleration, with a sliding window of 1 second and a step length of 0.5 seconds, and associates the temperature data through the Kalman filter. Its state equation is x k =Ax k-1 +Bu k +w k , the observation equation is z k =Hx k +v k , where A is the unit matrix, B = 1, H = 0.9, and the noise w k and u kObey Gaussian distribution with variances of 0.01 and 0.1 respectively. The covariance matrices Q and R are updated online by minimum mean square error;

[0017] S3: The processed composite feature vector is transmitted to the control algorithm module via the PCIe bus with a delay of ≤50ms.

[0018] Preferably, the control algorithm modules work together in the following ways:

[0019] S1: The support vector machine unit uses the radial basis kernel function, C = 10, γ = 0.1, to perform five-level classification on the input vibration parameter-compactness sample;

[0020] S2: The hidden Markov model unit decodes the optimal state sequence through the Viterbi algorithm, and the Baum-Welch algorithm iteratively optimizes the state transfer matrix A and the observation probability matrix B. (0) For uniform distribution, B (0) Based on the GMM model, the convergence threshold is the log-likelihood change rate ≤ 0.01%;

[0021] S3: The BP neural network unit contains 12 input nodes, a hidden layer of 20-15 nodes, and 3 output nodes. It uses an adaptive learning rate with an initial value of 0.01 and a momentum term of 0.9. It is optimized using 5000 sets of historical data through 5-fold cross validation.

[0022] S4: The three are calculated in parallel through the FPGA shared memory mechanism, the algorithm response time is ≤200ms, and the compaction prediction value and the optimal vibration parameter combination are output.

[0023] Preferably, the feedback control module adjusts the roller parameters through the following closed-loop process:

[0024] S1: The PWM signal generation circuit adjusts the vibration frequency, amplitude, and rolling speed through the H-bridge driver; the Hall effect sensor monitors the actual vibration parameters in real time, forming a closed-loop feedback loop;

[0025] S2: PID compensator K p =0.8,T i =0.5s,T d =0.1s, correct the control error, and the compensation signal is output through the FPGADSP module at a sampling rate of 200kHz;

[0026] S3: The adaptive unit uses the environmental conditions, construction equipment status and other data collected by sensors in real time to adjust the PID parameter K based on the fuzzy algorithm. p 、T i and T d Make online adjustments;

[0027] S4: Control cycle ≤ 100ms, CANopen protocol synchronizes actuator movements.

[0028] Preferably: the user interface realizes human-computer interaction in the following ways:

[0029] S1: The real-time data display area uses OpenGL ES 3.2 to render a 0.5m×0.5m resolution dynamic heat map with a color mapping range of 0-100% compaction and a refresh rate of 1Hz;

[0030] S2: The historical data analysis area uses the FFT algorithm and a 4096-point Hamming window to generate a spectrum diagram, supporting time domain / frequency domain comparative analysis;

[0031] S3: The action suggestion area is based on 128 preset rules and a reinforcement learning module, using the Q-learning algorithm. The state space includes temperature, RMS, and compaction level. The action space includes ±5% amplitude and ±3Hz frequency adjustment. The reward function is R = Δρ × t -1 ,The rule base is updated every minute through online learning.

[0032] Preferred: The embedded microcontroller coordinates module operation by:

[0033] S1: The ARM Cortex-A72 processor uses the SCHED_FIFO real-time scheduling strategy. Critical control instructions are assigned the highest priority: level 0, with a response time of ≤10μs; non-critical tasks are assigned the lowest priority: level 7, with a response time of ≤10ms. Time slices are scheduled in a round-robin manner: the time slice length is 1ms, and the latency is <50μs. The IEEE 802.1AS TSN protocol manages inter-module communication, with end-to-end latency of ≤100ms.

[0034] Preferably, the remote communication module realizes data synchronization by:

[0035] S1: The 5G NR-Uu interface supports dual-band 3.5GHz Sub-6GHz and 28GHz millimeter wave. The network slice includes a control plane with QoS level 9 and a bandwidth of 20Mbps, and a user plane with QoS level 5 and a bandwidth of 10Mbps. The MQTT protocol with QoS level 2 uses a three-way handshake retransmission mechanism: ACK / REACK / NACK. The UPF is deployed to the construction site to ensure a transmission rate of ≥100Mbps downlink and 50Mbps uplink. The AES-256-GCM encryption algorithm ensures data security.

[0036] The present invention has the following beneficial effects:

[0037] 1. In the present invention, a vibration acceleration sensor, a temperature sensor, and an RTK differential GPS positioning terminal are first integrated into a data acquisition module, and the multi-source data are transmitted to the data processing module using the CAN bus protocol. The wavelet analysis unit, the IIR filter unit, and the feature extraction unit in the data processing module preprocess and extract features from the data, thereby improving the accuracy and reliability of the data. In particular, the temperature data is correlated with the effective value (RMS) of the vibration acceleration through the Kalman filter to generate a composite feature vector for use in the subsequent control algorithm, thereby achieving high-precision feature extraction. This mechanism solves the problem of insufficient prediction accuracy caused by a single sensor data source, making the compaction degree prediction more accurate and adaptable to complex working conditions. A hardware trigger mechanism is used to synchronize data acquisition between sensors. When the master sensor acquires data, a trigger signal is sent to other slave sensors via a trigger signal line, causing them to synchronize data acquisition. Microsecond or even nanosecond synchronization can be achieved, which has the advantages of high accuracy and strong reliability. It solves the problem that the frequency and time of data acquisition by multiple sensors are different, the data synchronization is poor, and the fused data cannot accurately reflect the compaction state, affecting the accuracy of control decisions, thereby making the compaction degree prediction more accurate.

[0038] 2. In the control algorithm module, the support vector machine unit, hidden Markov model unit, and BP neural network unit work together to predict compaction and generate the optimal control strategy based on the processed composite feature vector. These three units are responsible for classification, sequence modeling, and nonlinear mapping tasks, respectively. The FPGA accelerator achieves millisecond-level response, ensuring the real-time and stability of the system. This multi-model collaborative decision-making approach not only improves the accuracy of compaction prediction but also enables rapid adjustment of roller parameters (such as vibration frequency, amplitude, and rolling speed), thereby achieving adaptive control. This overcomes the control response lag problem in existing technologies and greatly improves construction quality and resource utilization.

[0039] 3. In the feedback control module, the adaptive PID control algorithm is used to perceive the system status in real time and dynamically adjust the proportional (P), integral (I), and differential (D) parameters. This solves the problem that the fixed-parameter PID control algorithm cannot accurately adjust the compaction parameters under different asphalt material properties, environmental conditions, and construction equipment, and adapts to changes in complex working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is the overall flow chart of a multi-sensor fusion asphalt pavement intelligent compaction closed-loop control system proposed by the present invention;

[0041] Figure 2 This is a multi-sensor fusion asphalt pavement intelligent compaction closed-loop control system proposed by the present invention, which collects data from multiple sensors simultaneously;

[0042] Figure 3This is a diagram of the PID adaptive adjustment method of the multi-sensor fusion asphalt pavement intelligent compaction closed-loop control system proposed by the present invention. DETAILED DESCRIPTION

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0044] Example 1: Reference Figure 1-3 , a multi-sensor fusion asphalt pavement intelligent compaction closed-loop control system, including data acquisition module, data processing module, control algorithm module, feedback control module and user interaction interface. The data acquisition module is the basis of the entire system, responsible for obtaining real-time data from multiple sensors and transmitting this data to the data processing module through the CAN bus protocol.

[0045] Specifically, it includes vibration acceleration sensor, temperature sensor and RTK differential GPS positioning terminal. The vibration acceleration sensor adopts high-sensitivity MEMS technology and works based on piezoelectric effect or capacitance effect. When the vibration wheel is subjected to periodic excitation force, the crystal inside the sensor produces polarization, causing the charge to flow in a directional manner on the crystal surface, thereby generating a voltage signal. The IEPE piezoelectric acceleration sensor is selected, and its parameters are axial sensitivity 10.24mV / (m / s 2 ), measuring range ±50g, maximum lateral sensitivity <5%, weight about 28g, working voltage 18-30VDC, working current 2-10mA, installed at the axis position of the vibration wheel to ensure accurate capture of the dynamic changes of the vibration wheel.

[0046] The temperature sensor uses non-contact infrared temperature measurement technology to determine the temperature of the object being measured by measuring its thermal radiation. The captured infrared radiation energy is converted into an electrical signal and then calibrated to obtain the temperature value. The ABSD-01A series temperature sensor was selected. Its parameters are working power supply 24VDC, maximum current 30mA, output signal 0~5V, temperature measurement range 0~300℃, response time 150ms. It is installed in the key parts of the roller to ensure accurate monitoring of the working environment temperature.

[0047] The RTK differential GPS positioning terminal achieves centimeter-level precision positioning by receiving signals from multiple satellites and combining them with the differential correction information provided by the ground base station. The selected model uses the Zhongheda intelligent compaction system, which consists of a GNSS antenna, a 3G antenna, a differential antenna and a power adapter. It supports microelectronics technology, wireless communication technology and GNSS centimeter-level high-precision positioning. It is installed on the top of the roller to ensure that it can receive sufficient satellite signals and provide accurate position information.

[0048] In the data acquisition process, the vibration acceleration sensor (master sensor) is connected to the temperature sensor (slave sensor) and the RTK differential GPS positioning terminal (slave sensor) through a coaxial cable. The vibration acceleration sensor generates a TTL level signal when collecting data, which is transmitted to the slave sensor through the signal trigger line. The slave sensor starts to collect data synchronously. The master sensor and the slave sensor transmit the data to the data processing module in the same cycle; the vibration acceleration sensor collects the acceleration signal of the vibration wheel in real time and transmits it to the data processing module through the CAN bus protocol. The data acquisition frequency is set to 1kHz; the temperature sensor collects the working environment temperature in real time and transmits it to the data processing module through the CAN bus protocol. The data acquisition frequency is set to 1Hz; the RTK differential GPS positioning terminal collects the position information of the roller in real time and transmits it to the data processing module through the CAN bus protocol. The data acquisition frequency is set to 10Hz. The hardware trigger mechanism is used to enable each sensor to collect data synchronously. When the main sensor collects data, a trigger signal is sent to other slave sensors through the trigger signal line to enable them to collect data synchronously. It can achieve microsecond or even nanosecond synchronization, with the advantages of high precision and strong reliability. It solves the problem that the frequency and time of multi-sensor data collection are different, the data synchronization is poor, and the fused data cannot accurately reflect the compaction status, affecting the accuracy of control decisions, making the compaction degree prediction more accurate.

[0049] Example 2: The data processing module is responsible for preprocessing and feature extraction of the collected raw data for use by the subsequent control algorithm module.

[0050] This module includes a wavelet analysis unit, an IIR filter unit, and a feature extraction unit. The wavelet analysis unit effectively separates the different frequency components in the signal by performing multi-scale decomposition on the signal. The vibration acceleration signal is decomposed into five layers of wavelet to generate high-frequency coefficients and low-frequency coefficients. The high-frequency coefficient is denoised by hard thresholding using the formula:

[0051]

[0052] Where N is the number of sampling points, the low-frequency coefficient is eliminated by the fourth-order elliptic filter to eliminate the phase distortion transfer function:

[0053]

[0054] The input vibration acceleration signal is decomposed into five layers of wavelet, then hard threshold denoising and fourth-order elliptic filter are applied to reconstruct the signal and output to the IIR filter unit.

[0055] The IIR filter unit is an infinite impulse response filter with good frequency selection characteristics. The state equation is:

[0056]

[0057] A low-pass filter is designed to further remove noise, receive the signal output from the wavelet analysis unit, perform IIR filtering processing on it, and output the filtered signal to the feature extraction unit.

[0058] The feature extraction unit is responsible for extracting key features from the processed signal for use by subsequent algorithms. The formula for calculating the effective value RMS of vibration acceleration is:

[0059]

[0060] The state equation for correlating temperature data using Kalman filtering is:

[0061] x[k]=Ax[k-1]+w[k]

[0062] The observation equation is:

[0063] z[k]=Hx[k]+v[k]

[0064] The signal output from the IIR filter unit is received to calculate the effective value (RMS) of the vibration acceleration and the temperature data is correlated using the Kalman filter to generate a composite feature vector, which is then transmitted to the control algorithm module via the PCIe bus.

[0065] Example 3: The control algorithm module is responsible for predicting the degree of compaction based on the processed data and generating the optimal control strategy. The module includes a support vector machine unit, a hidden Markov model unit and a BP neural network unit.

[0066] The support vector machine unit is a supervised learning method suitable for classification and regression tasks, using the radial basis kernel function:

[0067] K(x,x')=exp(-γ||x-x'|| 2 )

[0068] The input vibration parameter-compactness sample is classified into five categories, and the cross-validation method is used to select the optimal kernel parameter γ and penalty coefficient C. The composite feature vector output from the data processing module is received as input data, and SVM is used to perform five-category classification to generate preliminary compaction prediction results and transmit the classification results to the hidden Markov model unit.

[0069] The hidden Markov model unit is a statistical model suitable for modeling and predicting sequence data. The optimal state sequence formula for decoding using the Viterbi algorithm is:

[0070]

[0071] Use the Baum-Welch algorithm to iteratively update the model parameters:

[0072] λ=(A,B,π)

[0073] The classification results output from the support vector machine unit are received as observation sequences, the optimal state sequence is decoded using the Viterbi algorithm, the model parameters are optimized using the Baum-Welch algorithm, and the optimized model parameters are transmitted to the BP neural network unit.

[0074] The BP neural network unit is a feedforward neural network suitable for complex nonlinear mapping tasks. The three-layer BP neural network input layer, hidden layer and output layer have several nodes respectively. The input layer receives the processed composite feature vector, the hidden layer uses the ReLU activation function, and the output layer uses the Sigmoid activation function. The network weights are updated through the back propagation algorithm, and the loss function is the mean square error. The model parameters output from the hidden Markov model unit are received as input data, and the BP neural network is used for training to generate the final compaction prediction result and transmit the prediction result to the feedback control module.

[0075] Example 4: The feedback control module is responsible for adjusting the operating parameters of the roller in real time according to the compaction prediction results provided by the control algorithm module to achieve the best compaction effect. The module adjusts the vibration frequency, amplitude and rolling speed of the roller through the CANopen protocol. The vibration frequency is adjusted by sending instructions to adjust the vibration frequency of the vibrating wheel through the CANopen protocol according to the compaction prediction results. The frequency range is usually 30Hz to 35Hz. The vibration frequency is adjusted by using a PWM signal generation circuit through an H-bridge drive. The required vibration frequency is calculated based on the compaction prediction results output from the control algorithm module and instructions are sent through the CANopen protocol to adjust the vibration frequency. The actual vibration frequency is monitored in real time and closed-loop feedback is performed through a Hall effect sensor.

[0076] Amplitude adjustment also uses the CANopen protocol to adjust the amplitude of the vibrating wheel to ensure effective energy utilization during the compaction process. The amplitude range should be optimized according to the specific working conditions and is usually ±8mm. The PWM signal generation circuit is used to adjust the amplitude through the H-bridge drive. The compaction degree prediction result output from the control algorithm module is received to calculate the required amplitude and send instructions to adjust the amplitude through the CANopen protocol. The actual amplitude is monitored in real time and closed-loop feedback is performed through the Hall effect sensor.

[0077] Rolling speed adjustment: The roller's rolling speed is maintained between 4 km / h and 5 km / h. A PID compensator is used to correct control errors. The compensation signal is output through the FPGA DSP module at a 200 kHz sampling rate. A control cycle of ≤100 ms ensures real-time and stability. The control algorithm module receives the compaction prediction results output by the control algorithm module, calculates the required rolling speed, and sends instructions to adjust the rolling speed via the CANopen protocol. The actual rolling speed is monitored in real time and closed-loop feedback is provided through a Hall effect sensor.

[0078] An adaptive adjustment unit is added to the microcontroller, and a fuzzy rule table is established through fuzzy logic. The parameters are adjusted in real time according to the system operating status. When it is detected that the compaction difficulty increases, the system error increases and converges slowly, then Kp is increased, the control quantity output speed is improved, and the error is quickly reduced; when it is detected that the compaction degree is close to the target value but there is a small error, Ki is increased to quickly eliminate the residual error. When there is a large error in the compaction degree for a long time, Ki is reduced to avoid overshoot; when it is detected that the error change rate becomes larger, Kd is increased to suppress overshoot in advance, thereby realizing adaptive control. The system status is perceived in real time through the adaptive PID control algorithm, and the proportional (P), integral (I), and differential (D) parameters are dynamically adjusted. This solves the problem that the fixed-parameter PID control algorithm cannot accurately adjust the compaction parameters under different asphalt material properties, environmental conditions, and construction equipment, and adapts to complex working conditions.

[0079] Example 5: The user interface communicates with other modules via the TCP / IP protocol, and displays compaction data, historical analysis charts, and parameter adjustment suggestions in real time. The interface provides intuitive visualization tools to facilitate operators to perform real-time monitoring and parameter adjustment.

[0080] The real-time data display area uses OpenGL ES 3.2 to render a 0.5m×0.5m resolution dynamic heat map with a color mapping range of 0-100% and a compaction refresh rate of 1Hz. The compaction data is obtained in real time and mapped to the color range. OpenGL ES3.2 is used for efficient rendering to ensure real-time and smoothness. The real-time compaction data output from the feedback control module is received for color mapping and OpenGL ES 3.2 is used to render the dynamic heat map to display the compaction distribution in real time.

[0081] The historical data analysis area uses the FFT algorithm to generate a spectrum diagram to help operators perform trend analysis and problem troubleshooting. The FFT algorithm is used to perform frequency domain analysis on historical compaction data to generate a spectrum diagram, and line graphs and bar graphs are provided to facilitate visual analysis by operators. The historical compaction data output from the feedback control module is received for FFT analysis to generate spectrum diagrams, line graphs, and bar graphs to display the historical data analysis results in real time.

[0082] Parameter adjustment suggestions are provided based on current compaction data and historical analysis results to help operators optimize the compaction process. Machine learning algorithms are used to predict the optimal parameter combination to generate adjustment suggestions, which provide detailed adjustment steps and expected effects for easy execution by operators. Current compaction data and historical analysis results are output from the feedback control module for prediction and parameter adjustment suggestions are generated and displayed in real time on the user interface for easy reference by operators.

[0083] Working principle: A multi-sensor fusion asphalt pavement intelligent compaction closed-loop control system, including data acquisition module, data processing module, control algorithm module, feedback control module and user interaction interface. The data acquisition module is connected to the vibration acceleration sensor, temperature sensor and RTK differential GPS positioning terminal through the CAN bus protocol. The vibration acceleration sensor is connected to the temperature sensor and RTK differential GPS positioning terminal through a coaxial cable to achieve multi-sensor synchronous data acquisition; the data processing module includes a wavelet analysis unit, an IIR filter unit and a feature extraction unit, and transmits the processed data to the control algorithm module through the PCIe bus; the control algorithm The algorithm module includes a support vector machine unit, a hidden Markov model unit, and a BP neural network unit. After parallel calculations using an FPGA accelerator, the compaction prediction results are transmitted to the feedback control module via a serial communication interface. The feedback control module adjusts the roller's vibration frequency, amplitude, and rolling speed using the CANopen protocol, and coordinates the operation of each module through an embedded microcontroller. The microcontroller includes an adaptive adjustment unit that establishes a fuzzy rule table using fuzzy logic to adjust parameters in real time based on the system's operating status. The user interface communicates with other modules via the TCP / IP protocol, displaying real-time compaction data, historical analysis charts, and parameter adjustment suggestions.

[0084] The multi-sensor fusion asphalt pavement intelligent compaction closed-loop control system integrates vibration acceleration sensor, temperature sensor and RTK differential GPS positioning terminal through the data acquisition module. The vibration acceleration sensor is connected to the temperature sensor and RTK differential GPS positioning terminal through a coaxial cable to realize multi-sensor simultaneous data acquisition, and prevent the different frequencies and times of data acquisition, which lead to the fused data not accurately reflecting the compaction status; the multi-source data is transmitted to the data processing module using the CAN bus protocol, and the wavelet analysis unit, IIR filter unit and feature extraction unit in the data processing module pre-process and extract features from these data, thereby improving the accuracy and reliability of the data. In particular, the temperature data is associated with the effective value RMS of the vibration acceleration through the Kalman filter to generate a composite feature vector for use by the subsequent control algorithm, thereby realizing high-precision feature extraction. This mechanism solves the problem of insufficient prediction accuracy caused by a single sensor data source, making the compaction prediction more accurate and adaptable to changes in complex working conditions; and the multi-sensor fusion asphalt pavement intelligent compaction closed-loop control system In the control algorithm module, the support vector machine unit, the hidden Markov model unit and the BP neural network unit work together to predict the compaction degree and generate the optimal control strategy based on the processed composite feature vector. These three units are responsible for classification, sequence modeling and nonlinear mapping tasks respectively, and achieve millisecond-level response through the FPGA accelerator, ensuring the real-time and stability of the system. This multi-model collaborative decision-making method not only improves the accuracy of compaction prediction, but also can quickly adjust the roller parameters (such as vibration frequency, amplitude and rolling speed). At the same time, an adaptive adjustment unit is added to the microcontroller, and a fuzzy rule table is established through fuzzy logic. The parameters are adjusted in real time according to the system operating status, thereby realizing adaptive control. It overcomes the problem of control response lag in the existing technology, greatly improves construction quality and resource utilization, and solves the problems of single sensor data source and lack of multi-source information fusion and dynamic modeling capabilities, resulting in insufficient compaction prediction accuracy and control response lag, and inability to achieve adaptive adjustment under complex working conditions, resulting in resource waste and unstable construction quality.

[0085] The relevant modules involved in this system are all hardware system modules or functional modules that combine computer software programs or protocols with hardware in the existing technology. The computer software programs or protocols involved in the functional modules are themselves technologies that are well known to those skilled in the art and are not improvements to this system. The improvements to this system are the interaction or connection relationships between the modules, that is, improvements to the overall structure of the system to solve the corresponding technical problems to be solved by this system.

[0086] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A multi-sensor fusion intelligent closed-loop control system for asphalt pavement compaction, characterized by: It includes a data acquisition module, a data processing module, a control algorithm module, a feedback control module and a user interaction interface; the data acquisition module is connected to the vibration acceleration sensor, the temperature sensor and the RTK differential GPS positioning terminal via the CAN bus protocol, the vibration acceleration sensor is connected to the temperature sensor and the RTK differential GPS positioning terminal via a coaxial cable, and the data acquisition module is connected to the data processing module via the CAN bus protocol; The data processing module includes a wavelet analysis unit, an IIR filter unit and a feature extraction unit, and is connected to the control algorithm module via a PCIe bus; The control algorithm module includes a support vector machine unit, a hidden Markov model unit and a BP neural network unit, which are parallel calculated by an FPGA accelerator and connected to a feedback control module through a serial communication interface; The feedback control module includes an embedded microcontroller, which adjusts the vibration frequency, amplitude and rolling speed of the roller through the CANopen protocol, and the microcontroller includes an adaptive adjustment unit; The user interaction interface communicates with other modules via the TCP / IP protocol.

2. A method for operating a process of a multi-sensor fusion asphalt pavement intelligent compaction closed-loop control system, comprising the multi-sensor fusion asphalt pavement intelligent compaction closed-loop control system according to claim 1, characterized in that: The data acquisition module process runs as follows: S1: The vibration acceleration sensor is used as the master sensor and has a BNC interface. The temperature sensor and the RTK differential GPS positioning terminal are installed with an ADC chip with an integrated trigger signal detection circuit as slave sensors. The master and slave sensors are connected via a coaxial cable. S2: The master sensor sends a trigger signal at a fixed frequency. After receiving the signal, the slave sensors complete the acquisition within the same clock cycle to ensure that the acquisition time of all sensors is aligned. S3: Measure the transmission delay of the trigger signal from the master sensor to the slave sensor, and compensate the timestamp of the slave sensor during data processing to eliminate the slight error caused by hardware delay.

3. The method for operating a closed-loop control system for intelligent compaction of asphalt pavement using multi-sensor fusion according to claim 2 is characterized by: The data processing module process runs as follows: S1: The vibration acceleration signal is decomposed and reconstructed in five layers using a wavelet analysis unit. The high-frequency coefficients are denoised using a hard threshold. The threshold formula is σ√(2lnN), where N is the number of sampling points. The low-frequency coefficients are filtered using a fourth-order elliptic filter to eliminate phase distortion. S2: The feature extraction unit calculates the effective value RMS of the vibration acceleration, with a sliding window of 1 second and a step length of 0.5 seconds, and associates the temperature data through the Kalman filter. Its state equation is x k =Ax k-1 +Bu k +w k , the observation equation is z k =hx k +v k , where A is the unit matrix, B = 1, H = 0.9, and the noise w k and u k Obey Gaussian distribution with variances of 0.01 and 0.1 respectively. The covariance matrices Q and R are updated online by minimum mean square error; S3: The processed composite feature vector is transmitted to the control algorithm module via the PCIe bus with a delay of ≤50ms.

4. The method for operating a closed-loop control system for intelligent compaction of asphalt pavement using multi-sensor fusion according to claim 2 is characterized by: The control algorithm modules work together in the following ways: S1: The support vector machine unit uses the radial basis kernel function with C = 10 and γ = 0.1 to perform five-class classification on the input vibration parameter-compactness sample; S2: The hidden Markov model unit decodes the optimal state sequence through the Viterbi algorithm, and the Baum-Welch algorithm iteratively optimizes the state transfer matrix A and the observation probability matrix B. (0) For uniform distribution, B (0) Based on the GMM model, the convergence threshold is the log-likelihood change rate ≤ 0.01%; S3: The BP neural network unit contains 12 input nodes, a hidden layer of 20-15 nodes, and 3 output nodes. It uses an adaptive learning rate with an initial value of 0.01 and a momentum term of 0.

9. It is optimized using 5000 sets of historical data through 5-fold cross validation. S4: The three are calculated in parallel through the FPGA shared memory mechanism, the algorithm response time is ≤200ms, and the compaction prediction value and the optimal vibration parameter combination are output.

5. The method for operating a closed-loop control system for intelligent compaction of asphalt pavement using multi-sensor fusion according to claim 2 is characterized by: The feedback control module adjusts the roller parameters through the following closed-loop process: S1: The PWM signal generation circuit adjusts the vibration frequency, amplitude, and rolling speed through the H-bridge driver; the Hall effect sensor monitors the actual vibration parameters in real time, forming a closed-loop feedback loop; S2: PID compensator K p =0.8,T i =0.5s,T d =0.1s, correct the control error, and the compensation signal is output through the FPGA DSP module at a sampling rate of 200kHz; S3: The adaptive unit uses the environmental conditions, construction equipment status and other data collected by sensors in real time to adjust the PID parameter K based on the fuzzy algorithm. p 、T i and T d Make online adjustments; S4: Control cycle ≤ 100ms, CANopen protocol synchronizes actuator movements.

6. The method for operating a closed-loop control system for intelligent compaction of asphalt pavement using multi-sensor fusion according to claim 2 is characterized by: The user interface realizes human-computer interaction through the following methods: S1: The real-time data display area uses OpenGL ES 3.2 to render a 0.5m×0.5m resolution dynamic heat map with a color mapping range of 0-100% compaction and a refresh rate of 1Hz; S2: The historical data analysis area uses the FFT algorithm to generate a spectrum diagram, Hamming window, 4096 points, and supports time domain / frequency domain comparative analysis; S3: The action suggestion area is based on 128 preset rules and a reinforcement learning module, using the Q-learning algorithm. The state space includes temperature, RMS, and compaction level. The action space includes ±5% amplitude and ±3Hz frequency adjustment. The reward function is R = Δρ × t -1 ,The rule base is updated every minute through online learning.

7. The method for operating a closed-loop control system for intelligent compaction of asphalt pavement using multi-sensor fusion according to claim 2 is characterized by: The embedded microcontroller coordinates module operations by: S1: The ARM Cortex-A72 processor uses the SCHED_FIFO real-time scheduling strategy. Critical control instructions are assigned the highest priority: level 0, with a response time of ≤10μs; non-critical tasks are assigned the lowest priority: level 7, with a response time of ≤10ms. Time slice round-robin scheduling: time slice length is 1ms, and latency is <50μs. The TSN protocol manages communication between modules, with end-to-end latency of ≤100ms.

8. The method for operating a closed-loop control system for intelligent compaction of asphalt pavement using multi-sensor fusion according to claim 2 is characterized by: The remote communication module achieves data synchronization through the following methods: S1: The 5G NR-Uu interface supports dual-band sub-6GHz and 28GHz millimeter wave (3.5GHz bandwidth). The network slice includes the control plane: QoS level 9, bandwidth 20Mbps, and the user plane: QoS level 5, bandwidth 10Mbps. The MQTT protocol with QoS level 2 uses a three-way handshake retransmission mechanism. The UPF is deployed to the construction site to ensure a transmission rate of ≥100Mbps downlink and 50Mbps uplink. AES-256-GCM encryption algorithm ensures data security.

Images