Intelligent concrete vibrating device and control method thereof

Abstract

This invention discloses an intelligent concrete vibration device, comprising: a vibration execution unit, including a motor and a vibrator; a torque sensor connected in series at the transmission connection between the motor and the vibrator; an intelligent sensing and acquisition unit, including a torque sensor, a power sensor, and an inclination sensor, wherein the power sensor is connected in parallel to the mains power supply circuit of the motor, the torque sensor is connected in series in the power transmission link, and the inclination sensor independently acquires spatial attitude signals to achieve synchronous data acquisition; a signal processing unit, used to receive signals from the torque sensor, power sensor, and inclination sensor; the signal processing unit transmits the processed standardized data to the intelligent control unit; the intelligent control unit, including a PLC, a frequency converter, and a control cabinet, used to receive the processed data and human operating parameters and send control commands and judgment results; and a human-machine interaction unit, bidirectionally connected to the PLC via a signal bus. This invention achieves real-time and accurate judgment of concrete vibration quality and early warning of abnormal attitudes.

Description

Technical Field

[0001] This invention relates to the field of building engineering technology, and in particular to an intelligent concrete vibration device and its control method. Background Technology

[0002] Concrete vibration is a core process in concrete structure construction. It uses high-frequency vibration of the vibration equipment to liquefy the concrete mixture, expel internal air bubbles, and reduce aggregate gaps, thereby improving the density, strength, and durability of the concrete, which directly determines the construction quality and service life of the concrete structure.

[0003] Existing intelligent vibration compaction equipment mainly falls into two categories. The first is vision-based equipment, which uses a camera mounted on the vibrator to capture images of the concrete surface and classifies the compaction quality based on these images. However, this type of equipment has a simplistic judgment method with poor adaptability. Relying on camera-based image acquisition of the concrete surface fails to consider the actual conditions of construction sites, such as dust, water mist, and concrete obstruction, easily leading to image acquisition failures, distorted detection data, and poor on-site adaptability. The second type is sensor-based monitoring equipment, which integrates one or more sensors into the vibrator or main unit to achieve real-time acquisition of some parameters during the compaction process. Some devices are equipped with simple controllers and early warning components, such as monitoring energy data through a power meter and providing audible and visual alerts when the energy reaches a preset threshold, thus determining whether the compaction is qualified. This type of equipment lacks integrated and systematic hardware design. Compaction quality is affected by multiple factors, including equipment status, material characteristics, and construction conditions. Existing sensors are often single or simply combined, unable to effectively identify complex conditions. Furthermore, current research lacks a unified architecture for receiving and processing multi-channel sensor (tilt angle, torque, power) data; and it has not designed an automatic identification method for the stages of vibration operation, making it impossible to distinguish between the four operation stages of moving, inserting, liquefying, and pulling out. Energy calculation does not limit the core liquefying stage, resulting in low quantification accuracy. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an intelligent concrete vibration device and its control method, which can overcome the shortcomings of the prior art and realize real-time and accurate judgment of concrete vibration quality and abnormal posture early warning.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows.

[0006] A smart concrete vibrating device, comprising, The vibration actuator includes a motor and a vibrating rod. The motor output shaft is rigidly connected to the eccentric vibration component inside the vibrating rod. The power supply circuit of the motor is electrically connected to the inverter and the mains interface of the control cabinet. A torque sensor is connected in series at the transmission connection between the motor and the vibrating rod to realize the integration of power transmission and parameter acquisition. The intelligent sensing and acquisition unit includes a torque sensor, a power sensor, and a tilt sensor. The torque sensor, power sensor, and tilt sensor are connected to the signal processor via shielded signal lines to achieve multi-channel bidirectional signal connection. The power supply terminals of the sensors are all electrically connected to the power supply. The power sensor is connected in parallel to the mains power supply circuit of the motor, the torque sensor is connected in series to the power transmission link, and the tilt sensor independently acquires spatial attitude signals to achieve interference-free synchronous data acquisition. The signal processing unit has its input terminals connected to the shielded signal lines of the torque sensor, power sensor, and tilt sensor to receive multi-channel raw sensor signals. The output terminals of the signal processing unit are connected to the PLC via an industrial bus to achieve high-speed signal transmission, transmitting the processed standardized data to the intelligent control unit. The power supply terminal of the signal processing unit is electrically connected to the power supply. The intelligent control unit includes a PLC, a frequency converter, and a control cabinet. The input terminals of the PLC are connected to the signal processing unit and the human-machine interaction unit to receive processed data and manual operating parameters. The output terminals of the PLC are connected to the frequency converter, the human-machine interaction unit, and the memory to send control commands and judgment results. The human-machine interface unit is bidirectionally connected to the PLC via a signal bus and is also electrically connected to a 24V power supply.

[0007] A control method for the above-mentioned intelligent concrete vibrating equipment includes the following steps: S1. Working condition parameter setting: Receive and store one or more on-site working condition parameters input by the operator through the human-machine interaction unit, including parameters characterizing the fluidity of concrete and / or parameters characterizing the complexity of the construction environment. S2. Synchronous acquisition of multi-sensor signals: The intelligent sensing and acquisition unit continuously and synchronously acquires data on rotational speed, torque, electric power, and spatial tilt angle; S3. Sensor signal processing: Automatic identification of vibration stage, calculation of vibration energy, and calculation of vibration tilt angle using raw signal data; S4. Vibration quality judgment: Set up a machine learning model, integrate the working condition parameters obtained in step S1 and the signal processing data obtained in step S3 into a multi-dimensional input vector, input it into the machine learning model for inference calculation, and output the vibration quality status and abnormal status identification results. S5. Visual feedback and storage of judgment results: The PLC synchronously outputs the judgment results of step S4 and the signal processing data of step S3 to the human-machine interaction unit for visualization. The PLC stores the entire data for subsequent retraining and optimization of the machine learning model.

[0008] Preferably, in step S3, the automatic identification of the vibration stage includes the following steps: The system acquires real-time rotational speed signals and monitors their changes relative to a preset no-load operating speed threshold. When the rotational speed signal remains stably above the no-load operating speed threshold, the vibrator is determined to be in the moving phase. When the rotational speed signal continuously decreases from a stable state, the vibrator is determined to have entered the insertion phase. When the rotational speed signal decreases below a preset decrease threshold and remains relatively stable within a rotational speed range with fluctuation amplitude less than a preset range, accompanied by an increase in the torque signal, the vibrator is determined to be in the liquefaction phase, which effectively compacts the concrete, and the start time of the liquefaction phase is marked. When the rotational speed signal changes from a relatively stable state to a continuous increase until it returns to the no-load operating speed threshold level, while the torque signal decreases, the vibrator is determined to have entered the withdrawal phase, and the end time of the liquefaction phase is marked, thereby determining the duration of the liquefaction phase.

[0009] Preferably, in step S3, the method for calculating the vibration energy is as follows: ,in, Energy consumed by the electrical power of concrete (J), , These are the start and end times of the selected stage, and the time interval is determined by the vibration stage identification method.

[0010] Preferably, in step S3, the calculation of the vibration tilt angle includes the following steps: A Butterworth low-pass filter with a cutoff frequency of 5Hz was applied to the raw data from the tilt sensor to remove high-frequency vibration noise. .

[0011] As a preferred option, machine learning models include, The input layer includes repeat nodes with a clone function, used for dimension alignment of the input data; Stacked MLP-RF layers, consisting of several MLP-RF layers, each with the same weight matrix. The non-shared parameters of each MLP-RF are The implicit weight matrix formed is Achieving multi-model prediction with a single forward propagation ; Mean pooling output layer; used to output prediction results. .

[0012] Preferably, the MLP-RF layer includes an input layer, a hidden layer, and an output layer. Each neuron in the hidden layer is encapsulated by a decision tree. The mathematical expression of the MLP-RF layer is y=Dropout(ReLU(Wx+b)). The MLP-RF layer is trained in two rounds. In the first round of training, the decision tree is treated as a neuron for training. In the second round of training, the input and output data of each neuron after the first round of training are used as the training dataset to train the decision tree. Before the second round of training, the input and output data of the neurons are filtered to remove redundant features with low importance, and the importance of each dimension of the remaining features is weighted.

[0013] Preferably, the loss function of MLP-RF is: LMSE is the MSE loss, N is the decision tree structure function, x is the decision tree input, and q is the smoothed output probability of the decision tree. E is the information entropy of the decision tree input. L i This is the path probability output of the neuron for sample i.

[0014] Preferably, in step S5, the visualization includes, The interface displays the current quality status of the vibration operation in real time through the core area, and uses differentiated colors and text to mark the three core statuses: under-vibration, qualified, and over-vibration. For the two types of abnormal states identified, namely local over-vibration and excessive operating angle of the vibrator, a synchronized sound and light alarm is triggered. The trigger threshold, alarm volume, and light frequency of the sound and light alarms can all be manually preset and adjusted through the human-computer interaction unit. The system generates and displays dynamic curves and charts for the entire vibration operation process in real time, including dynamic curves of rotation speed marked with operation stages, real-time curves of changes in ground angle, real-time dynamic curves of power, cumulative effective energy growth curves, and real-time timing curves of liquefaction time. All curves are updated in real time as the operation progresses. The interface provides a historical data query entry, allowing operators to query historical parameters and judgment results for the current operation.

[0015] The beneficial effects of adopting the above technical solution are as follows: This invention automatically identifies the four stages of vibration operation based on rotation speed threshold and load changes, and only performs net power integral energy calculation and filtered spatial tilt angle conversion for the "liquefaction stage" of effective vibration, eliminating parameter interference from invalid operation stages and improving the quantitative accuracy of core vibration parameters. A multi-dimensional machine learning model for judging vibration quality is designed, integrating construction condition parameters. This model combines vibration process parameters such as concrete slump, rebar density, liquefaction time, cumulative effective energy, and average tilt angle as model inputs, replacing the single fixed threshold judgment logic of existing technologies, and achieving accurate judgment of vibration quality adaptability to different working conditions. Attached Figure Description

[0016] Figure 1 This is the system hardware schematic diagram of the present invention.

[0017] Figure 2 This is a diagram showing the rotational speed of the vibrator during the compaction process.

[0018] Figure 3 This is a schematic diagram of the machine learning model of the present invention. Detailed Implementation

[0019] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0020] Reference Figure 1 The hardware components of this invention are standardized and integrated according to a configuration list. Each component belongs to one of five functional units, physically concentrated in the rod-shaped vibrator and the box-type intelligent vibrator main unit / control cabinet. All electrical and signal connections use standardized interfaces and shielded signal cables to ensure transmission stability. Mechanical connections employ rigid transmission, fixed connections, or rail-mounted fixation to meet the vibration and mobile operation requirements of construction scenarios. The component composition, installation location, connection relationship, and function of each unit are as follows, and the component numbers are completely consistent with the configuration list: (a) Vibration execution unit Components: Motor (220V single-phase / 380V three-phase optional), vibrator (50×6000mm).

[0021] Shape and position: The vibratory rod is a long rod-shaped structure with an eccentric vibration component embedded inside, and the whole is a handheld operating end; the motor is a vertical drive structure, installed at the front of the intelligent vibratory main unit box, and is the direct power source for the vibratory rod.

[0022] Connection relationship: The motor output shaft is rigidly connected to the eccentric vibration component inside the vibrator, and the motor power supply circuit is electrically connected to the inverter and the mains interface of the control cabinet; a torque sensor is connected in series at the transmission connection between the motor and the vibrator to realize the integration of power transmission and parameter acquisition.

[0023] (II) Intelligent Sensing and Acquisition Unit Components: Torque sensor, power meter (power sensor), tilt sensor.

[0024] Shape and position: The torque sensor is a ring-shaped sensing structure, nested in the transmission end of the motor and the vibrator; the power meter is a modular circuit structure, built into the intelligent vibrator main unit box; the tilt sensor is a miniature patch structure, fixed to the outer side of the middle section of the vibrator (no interference position for handheld operation).

[0025] Connection relationship: All three sensors are connected to the signal processor through shielded signal lines to achieve multi-channel bidirectional signal connection, and the power supply terminals of the sensors are all electrically connected to the power supply; the power meter is connected in parallel to the mains power supply circuit of the motor, the torque sensor is connected in series to the power transmission link, and the tilt sensor independently collects spatial attitude signals. The three achieve synchronous data acquisition without interference.

[0026] Core functions: Provides real-time and continuous acquisition of three core parameters for equipment: speed / torque, electric power, and spatial tilt angle. These parameters serve as the basic data source for intelligent judgment, which is different from the single / simple sensor combination configuration of existing technologies.

[0027] (III) Signal Processing Unit Components: 5# Signal Processor.

[0028] Shape and position: It has a modular circuit board structure and is installed in the center of the intelligent vibratory main unit box, forming the shortest signal transmission path with each sensor and intelligent control unit.

[0029] Connection relationship: The input end is connected to the shielded signal line of the torque sensor, power meter, and tilt sensor to receive multi-channel raw sensor signals; the output end is connected to the PLC through the industrial bus to realize high-speed signal connection and transmit the processed standardized data to the intelligent control unit; the power supply end is electrically connected to the power supply.

[0030] Core functions: As a unified processing carrier for multi-channel sensor signals, it performs professional processing such as filtering of raw signals, stage identification, energy calculation, and tilt angle conversion.

[0031] (iv) Intelligent control unit Components: PLC, frequency converter, control cabinet.

[0032] Shape and location: The control cabinet is a closed box structure with an operation panel on the outside and standardized mounting rails and fixing holes inside, serving as an integrated carrier for all electrical control components; the PLC and frequency converter are modular structures, fixed inside the control cabinet by rails, and seamlessly connected to the main unit; the control cabinet panel is equipped with start / stop buttons and heat dissipation modules to meet the needs of construction operation and equipment heat dissipation.

[0033] Connection relationship: The PLC is the core control hub. Its input terminals are connected to the signal processor and HMI to receive processed data and manual operating parameters. Its output terminals are connected to the frequency converter, HMI, and memory to send control commands and judgment results. The PLC control program is pre-programmed inside the PLC to provide program support for the entire process of equipment operation logic.

[0034] Core functions: Realizes the logic control of equipment, motor speed regulation, data transfer and program execution, which is different from the simple controllers of existing technology and forms a standardized intelligent control architecture of "PLC + frequency converter + control cabinet".

[0035] (v) Human-computer interaction unit Components: Human-computer interaction interface.

[0036] Shape and position: It is an embedded touch screen structure, installed in a prominent position on the operation panel on the outside of the control cabinet, and adapted to the manual operation perspective of handheld operation.

[0037] Connection relationship: It is bidirectionally connected to the PLC via a signal bus, and is also electrically connected to a 24V power supply; Core functions: Enables input of operating parameters, visualization of judgment results, abnormal alarms, and data display. It serves as an interactive platform between humans and equipment, distinct from simple audio-visual prompts in existing technologies.

[0038] This invention addresses the professional data processing and intelligent judgment method for the entire process of concrete vibration operation. By embedding a PLC control program into the equipment, it achieves automated execution of "data acquisition-processing-judgment-feedback-storage".

[0039] Step 1: Equipment startup and manual setting of operating parameters Operators input two types of working condition parameters for this construction project—concrete slump and rebar density—in preset input boxes through the HMI touch interface. The parameters are transmitted to the PLC in real time via the signal bus and stored, providing a basis for working condition adaptation for subsequent machine learning judgment. The PLC preloads the pre-trained machine learning model to complete the parameter and model preparation before judgment.

[0040] Step 2: Real-time acquisition of multi-channel raw sensor signals The torque sensor, power meter, and tilt sensor operate synchronously in continuous acquisition mode, collecting data in real time. Step 3: Professional processing of sensor signals After receiving the raw signal, the signal processor completes professional processing in three sub-steps: automatic identification of vibration stage, calculation of vibration energy, and calculation of vibration tilt angle.

[0041] Logical Description: Based on the speed signal from the torque sensor, and by monitoring the no-load threshold and load changes, the workflow is divided into four stages: movement, insertion, liquefaction, and withdrawal. Figure 2 As shown.

[0042] Key output: Real-time status flag: Which stage is the current stage? Liquefaction time: The cumulative time spent in the "liquefaction stage".

[0043] Baseline capture: Automatically records the average power during the "moving phase (no load)".

[0044] Vibration energy calculation Logical description: Calculations are performed only during the "liquefaction phase". Energy input is quantified by integrating the net power.

[0045] Calculation formula: ,in, Energy consumed by the electrical power of concrete (J), , These are the start and end times of the selected stage, and the time interval is determined by the vibration stage identification method.

[0046] Processing requirements: The power data must be averaged over seconds and then accumulated, with the energy statistics updated every second.

[0047] Vibration tilt angle calculation Preprocessing: A Butterworth low-pass filter with a cutoff frequency of 5Hz was applied to the raw GyroX and GyroY data to remove high-frequency vibration noise.

[0048] Calculation of the angle relative to the ground: .

[0049] Key output: Real-time display of the current angle to the ground. Calculate the average tilt angle during the liquefaction stage.

[0050] Step 4: Intelligent judgment of vibration quality based on pre-trained machine learning model Input parameters: The PLC integrates the working condition parameters (slump, rebar density) from step 1 with the core processing data (liquefaction time, cumulative effective energy in the liquefaction stage, average tilt angle in the liquefaction stage) from step 3 as input parameters for the pre-trained machine learning model; Execution logic: The model triggers an inference calculation every 1~2 seconds, and achieves a refined judgment of the vibration quality through multi-parameter fusion analysis, which is different from the single-parameter fixed threshold judgment of the existing technology; Core outputs: ① Vibration quality status (under-vibration / qualified / over-vibration); ② Abnormal status identification (local over-vibration, excessive operating angle of vibrator).

[0051] See Figure 3 The machine learning model of this invention consists of an input layer, a stacked MLP-RF layer, and a mean pooling output layer.

[0052] The input layer includes repeat nodes with a clone function, used for dimension alignment of the input data.

[0053] Traditional Multilayer Perceptron (MLP) is a standard feedforward neural network, which suffers from drawbacks such as being prone to getting trapped in local minima during training and having poor generalization ability on small datasets. It is also inadequate for processing large volumes of time-series data, as in this invention. This invention reconstructs the MLP by introducing the Random Forest (RF) model, designing a novel MLP-RF model. The MLP-RF layer consists of an input layer, hidden layers, and an output layer. Each neuron in the hidden layer is encapsulated by a decision tree. The mathematical expression for the MLP-RF layer is y = Dropout(ReLU(Wx + b)). The MLP-RF layer is trained in two rounds. In the first round, the decision tree is treated as a neuron for training. In the second round, the input and output data of each neuron after the first round of training are used as the training dataset to train the decision tree. Before the second round of training, the input and output data of the neurons are filtered to remove redundant features with low importance, and the remaining features are weighted according to their importance for each dimension.

[0054] The loss function of the MLP-RF layer is: LMSE is the MSE loss, N is the decision tree structure function, x is the decision tree input, and q is the smoothed output probability of the decision tree. E is the information entropy of the decision tree input. L i This is the path probability output of the neuron for sample i.

[0055] By modifying neurons in the hidden layer using decision trees, generalization error can be significantly reduced, comprehensively improving the prediction performance of tabular data tasks. First, a self-supervised pre-training method using an MLP is employed to learn general feature representations. Then, a pre-trained feature extractor is used to generate features, which are then used to train a random forest. This addresses the insufficient generalization ability of the two models on time-series datasets and fully utilizes unlabeled data.

[0056] Stacked MLP-RF layers consist of several MLP-RF layers, each with the same weight matrix. The non-shared parameters of each MLP-RF are The implicit weight matrix formed is Achieving multi-model prediction with a single forward propagation .

[0057] Mean pooling output layer is used to output prediction results. .

[0058] Traditional models (such as SVM) are easily affected by sample distribution. This invention improves generalization ability and reduces overfitting through implicit ensemble. Traditional ensemble models (such as Random Forest) have high computational costs, and deep ensemble requires training multiple networks. This invention completes ensemble prediction in a single inference, significantly improving inference efficiency. Traditional models struggle to stably handle multi-condition data. Under different slump and rebar conditions, feature distribution changes significantly. This invention provides stronger robustness and cross-condition recognition capability through a multi-sub-model structure. Experimental results show that this invention achieves an accuracy of 91.8% in determining five types of vibration conditions, outperforming traditional models.

[0059] Step 5: Real-time feedback and visualization of the judgment results The PLC synchronously transmits the judgment results of the machine learning model, along with all the processing data from step 3 (real-time stage, liquefaction time, cumulative energy, and real-time tilt angle), to the HMI. The HMI provides multi-dimensional visualization: ① Real-time display of quality status (under-vibration / qualified / over-vibration); ② Triggering abnormal alarms (when local over-vibration / excessive angle is detected, the HMI issues a simultaneous audible and visual alarm); ③ Plotting dynamic curves and charts (speed curve with stage markers, real-time angle curve, real-time power curve, cumulative energy curve, and liquefaction time). Operators can monitor the vibration quality in real time through the HMI and adjust their work methods accordingly, forming a closed-loop operation of "intelligent equipment judgment - precise manual adjustment." This equipment integrates non-volatile memory (expanded hardware) within the PLC to store all data in real time: including multi-channel raw sensor signals, professionally processed data from step 3, operating parameters from step 1, and judgment results from step 4. The stored historical data can be exported to an external terminal via the HMI's USB interface for subsequent retraining and optimization of the machine learning model, continuously improving the model's judgment accuracy under different construction conditions.

[0060] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent concrete vibration device, characterized in that: include, The vibration actuator includes a motor and a vibrating rod. The motor output shaft is rigidly connected to the eccentric vibration component inside the vibrating rod. The power supply circuit of the motor is electrically connected to the inverter and the mains interface of the control cabinet. A torque sensor is connected in series at the transmission connection between the motor and the vibrating rod to realize the integration of power transmission and parameter acquisition. The intelligent sensing and acquisition unit includes a torque sensor, a power sensor, and a tilt sensor. The torque sensor, power sensor, and tilt sensor are connected to the signal processor via shielded signal lines to achieve multi-channel bidirectional signal connection. The power supply terminals of the sensors are all electrically connected to the power supply. The power sensor is connected in parallel to the mains power supply circuit of the motor, the torque sensor is connected in series to the power transmission link, and the tilt sensor independently acquires spatial attitude signals to achieve interference-free synchronous data acquisition. The signal processing unit has its input terminals connected to the shielded signal lines of the torque sensor, power sensor, and tilt sensor to receive multi-channel raw sensor signals. The output terminals of the signal processing unit are connected to the PLC via an industrial bus to achieve high-speed signal transmission, transmitting the processed standardized data to the intelligent control unit. The power supply terminal of the signal processing unit is electrically connected to the power supply. The intelligent control unit includes a PLC, a frequency converter, and a control cabinet. The input terminals of the PLC are connected to the signal processing unit and the human-machine interaction unit to receive processed data and manual operating parameters. The output terminals of the PLC are connected to the frequency converter, the human-machine interaction unit, and the memory to send control commands and judgment results. The human-machine interface unit is bidirectionally connected to the PLC via a signal bus and is also electrically connected to a 24V power supply.

2. A control method for the intelligent concrete vibrating equipment according to claim 1, characterized in that... Includes the following steps: S1. Working condition parameter setting: Receive and store one or more on-site working condition parameters input by the operator through the human-machine interaction unit, including parameters characterizing the fluidity of concrete and / or parameters characterizing the complexity of the construction environment. S2. Synchronous acquisition of multi-sensor signals: The intelligent sensing and acquisition unit continuously and synchronously acquires data on rotational speed, torque, electric power, and spatial tilt angle; S3. Sensor signal processing: Automatic identification of vibration stage, calculation of vibration energy, and calculation of vibration tilt angle using raw signal data; S4. Vibration quality judgment: Set up a machine learning model, integrate the working condition parameters obtained in step S1 and the signal processing data obtained in step S3 into a multi-dimensional input vector, input it into the machine learning model for inference calculation, and output the vibration quality status and abnormal status identification results. S5. Visual feedback and storage of judgment results: The PLC synchronously outputs the judgment results of step S4 and the signal processing data of step S3 to the human-machine interaction unit for visualization. The PLC stores the entire data for subsequent retraining and optimization of the machine learning model.

3. The control method for the intelligent concrete vibrating equipment according to claim 2, characterized in that: In step S3, the automatic identification of the vibration stage includes the following steps: The system acquires real-time rotational speed signals and monitors their changes relative to a preset no-load operating speed threshold. When the rotational speed signal remains stably above the no-load operating speed threshold, the vibrator is determined to be in the moving phase. When the rotational speed signal continuously decreases from a stable state, the vibrator is determined to have entered the insertion phase. When the rotational speed signal decreases below a preset decrease threshold and remains relatively stable within a rotational speed range with fluctuation amplitude less than a preset range, accompanied by an increase in the torque signal, the vibrator is determined to be in the liquefaction phase, which effectively compacts the concrete, and the start time of the liquefaction phase is marked. When the rotational speed signal changes from a relatively stable state to a continuous increase until it returns to the no-load operating speed threshold level, while the torque signal decreases, the vibrator is determined to have entered the withdrawal phase, and the end time of the liquefaction phase is marked, thereby determining the duration of the liquefaction phase.

4. The control method for the intelligent concrete vibrating equipment according to claim 2, characterized in that: In step S3, the method for calculating the vibration energy is as follows: ,in, Energy consumed by the electrical power of concrete (J), , These are the start and end times of the selected stage, and the time interval is determined by the vibration stage identification method.

5. The control method for the intelligent concrete vibrating equipment according to claim 2, characterized in that: In step S3, the calculation of the vibration tilt angle includes the following steps: A Butterworth low-pass filter with a cutoff frequency of 5Hz was applied to the raw data from the tilt sensor to remove high-frequency vibration noise. .

6. The control method for the intelligent concrete vibrating equipment according to claim 2, characterized in that: Machine learning models include, The input layer includes repeat nodes with a clone function, used for dimension alignment of the input data; Stacked MLP-RF layers, consisting of several MLP-RF layers, each with the same weight matrix. The non-shared parameters of each MLP-RF are The implicit weight matrix formed is Achieving multi-model prediction with a single forward propagation ; Mean pooling output layer; used to output prediction results. .

7. The control method for the intelligent concrete vibrating equipment according to claim 6, characterized in that: The MLP-RF layer consists of an input layer, a hidden layer, and an output layer. Each neuron in the hidden layer is encapsulated by a decision tree. The mathematical expression for the MLP-RF layer is y=Dropout(ReLU(Wx+b)). The MLP-RF layer is trained in two rounds. In the first round, the decision tree is treated as a neuron for training. In the second round, the input and output data of each neuron after the first round of training are used as the training dataset to train the decision tree. Before the second round of training, the input and output data of the neurons are filtered to remove redundant features with low importance, and the importance of each dimension of the remaining features is weighted.

8. The control method for the intelligent concrete vibrating equipment according to claim 7, characterized in that: The loss function of the MLP-RF layer is: LMSE is the MSE loss, N is the decision tree structure function, x is the decision tree input, and q is the smoothed output probability of the decision tree. E is the information entropy of the decision tree input. L i This is the path probability output of the neuron for sample i.

9. The control method for the intelligent concrete vibrating equipment according to claim 2, characterized in that: In step S5, the visualization includes, The interface displays the current quality status of the vibration operation in real time through the core area, and uses differentiated colors and text to mark the three core statuses: under-vibration, qualified, and over-vibration. For the two types of abnormal states identified, namely local over-vibration and excessive operating angle of the vibrator, a synchronized sound and light alarm is triggered. The trigger threshold, alarm volume, and light frequency of the sound and light alarms can all be manually preset and adjusted through the human-computer interaction unit. The system generates and displays dynamic curves and charts for the entire vibration operation process in real time, including dynamic curves of rotation speed marked with operation stages, real-time curves of changes in ground angle, real-time dynamic curves of power, curves of cumulative effective energy growth, and real-time timing curves of liquefaction time. All curves are updated in real time as the operation progresses. The interface provides a historical data query entry, allowing operators to query historical parameters and judgment results for the current operation.

Images