Fresh concrete vibration apparent quality identification method based on convolutional neural network

Abstract

The invention discloses a fresh concrete vibration apparent quality identification method based on a convolutional neural network, and relates to the technical field of concrete apparent quality identification, and the method comprises the steps: obtaining vibration track data, vibration frequency parameters and a concrete surface texture image of vibration equipment in an operation process, and generating an original data set; performing model training optimization based on the initial convolutional neural network model to obtain a depth feature extraction network adaptive to the vibration process; analyzing time sequence detection data in the vibration process, extracting vibration energy density distribution characteristics, aggregate distribution uniformity indexes and slurry wrapping degree parameters, and constructing a vibration quality correlation characteristic matrix; in the depth feature extraction network, the vibration quality correlation feature matrix and vibration sensing data collected in real time are fused, and a quality evaluation system is established; according to the quality evaluation system, the vibration operation process is detected, and a visual evaluation result is obtained; the fresh concrete vibration apparent quality identification method provided by the invention is more intelligent and accurate.

Description

Technical Field

[0001] The present application relates to the technical field of concrete apparent quality identification, and in particular to a method for identifying the apparent quality of fresh concrete vibration based on a convolutional neural network. Background Art

[0002] In construction projects, vibrating fresh concrete is a critical process for ensuring its density, strength, and durability. However, traditional manual or semi-automated vibration methods struggle to maintain consistent quality standards, and subtle defects introduced during the vibration process are difficult to detect. This directly impacts the safety and service life of concrete structures.

[0003] Traditionally, the quality inspection of fresh concrete vibration is mostly done through manual visual inspection, which has low detection efficiency and a high rate of missed detection of hidden defects. Although image recognition technology that uses a single visual sensor to capture surface images has emerged, there are still recognition errors, making it difficult to accurately reflect the dynamic changes in the vibration process and its impact on the final concrete appearance quality. Therefore, it is particularly important to develop a more intelligent and accurate method for identifying the apparent quality of fresh concrete vibration. Summary of the Invention

[0004] In order to address the shortcomings of the existing technology and provide a more intelligent and accurate method for identifying the apparent quality of fresh concrete vibrated, the present application provides a method for identifying the apparent quality of fresh concrete vibrated based on a convolutional neural network.

[0005] In the first aspect, the invention objectives of this application are achieved by adopting the following technical solutions: The method for identifying the apparent quality of fresh concrete vibration based on convolutional neural network includes: Acquire the vibration trajectory data, vibration frequency parameters, and concrete surface texture images of the vibrating equipment during operation, and generate a raw data set containing vibration intensity distribution characteristics and apparent quality labels; Constructing an initial convolutional neural network model, performing transfer learning on the network structure based on the collected vibration equipment operating parameters and the original data set, completing model training optimization under a predefined concrete quality evaluation index system, and obtaining a deep feature extraction network adapted to the vibration process; Analyze the time series detection data of the vibration process, calculate the correlation between the vibration energy density distribution characteristics and the concrete surface texture characteristics, extract the vibration energy density distribution characteristics, aggregate distribution uniformity index and paddle body wrapping parameters, and construct the vibration quality correlation feature matrix; In the deep feature extraction network, the vibration quality correlation feature matrix and the real-time collected vibration sensor data are integrated to establish a quality assessment system including vibration defect location and quality score prediction; The vibration operation process is inspected according to the quality assessment system to obtain a visual assessment result.

[0006] By adopting the above technical solution, the apparent quality label includes the honeycomb surface grade, bubble distribution density and surface flatness index; the visual evaluation results include defect type annotation, quality grade assessment and process parameter optimization suggestions; this application integrates multi-source heterogeneous detection data to improve the accuracy of identifying the apparent quality of fresh concrete vibration. By integrating the dynamic parameters of the vibration trajectory, the time series characteristics of the vibration frequency and the multimodal data of the concrete surface texture, the apparent quality feature recognition accuracy is higher than the traditional single sensor detection method, especially the error in detecting the distribution of micron-level bubbles can be reduced; and based on the closed-loop feedback mechanism of the deep feature extraction network and real-time sensor data, the adaptive adjustment of process parameters such as the vibration time and the vibration head trajectory is realized to dynamically optimize the process control parameters of the vibration equipment during the vibration operation. By constructing a system including The quality assessment system for vibration defect location and quality score prediction realizes defect visualization and defect area location of vibration apparent quality. Compared with traditional image processing, the defect location accuracy of this application is further improved; based on the constructed vibration quality correlation feature matrix, the influence correlation of the dynamic process parameters of the vibration equipment on key quality indicators such as slurry penetration depth and aggregate distribution uniformity in the apparent quality of fresh concrete is established, so as to establish a quantitative mapping relationship between the formation mechanism of fresh concrete defects and the process parameters of the vibration equipment, providing data support for future intelligent optimization of the vibration process; at the same time, it can also adapt to different types of vibration equipment through transfer learning strategies, which is conducive to meeting the industrial application needs of complex scenarios such as tunnel projects and high-rise buildings; thus, the fresh concrete vibration apparent quality identification method provided by this application is more intelligent and refined.

[0007] In a preferred example of the present application, the step of obtaining the vibration trajectory data, vibration frequency parameters, and concrete surface texture image of the vibrating equipment during operation includes: The distributed sensor network collects the three-dimensional positioning trajectory data of the vibrating equipment in real time, and simultaneously obtains the pressure time domain signal and vibration frequency response curve of the vibrating rod contact surface; The machine vision system captures the concrete surface micro-texture image, and combines it with the surface slurry distribution morphology data obtained by the laser scanner to construct an original monitoring data set containing spatiotemporal characteristic parameters. The time window threshold is set according to the vibration process specification, and the continuously collected time series data is divided into multiple monitoring periods. Each monitoring period generates a sub-dataset containing the vibration intensity distribution characteristics and apparent quality labels. The time series feature segmentation algorithm is used to preprocess the raw data of the original monitoring data set, extract the vibration energy density distribution characteristics, aggregate distribution uniformity index and slurry wrapping parameters, and establish the original data set containing the multidimensional feature matrix of vibration quality.

[0008] By adopting the above technical solution, a digital and spatial data synchronization acquisition method combining a distributed sensor data acquisition network and a machine vision system is adopted to finely process and analyze the time series data of fresh concrete during the vibration process. The data is divided by adopting a time window threshold (such as every 5 minutes as a monitoring cycle), and the vibration energy density distribution characteristics are extracted through a time series feature segmentation algorithm to analyze the time series correlation of key process parameters of the vibration equipment and the corresponding data analysis error. This is conducive to solving the data alignment problem under non-steady-state conditions during the vibration process, facilitating the proposal of process parameter optimization suggestions for the apparent quality problems of fresh concrete, and improving the product quality of fresh concrete. The construction of a multi-dimensional feature matrix reduces the redundancy of data features.

[0009] In a preferred example of the present application, the vibration quality correlation characteristic matrix includes the vibration equipment energy consumption factor, the aggregate distribution loss factor, and the slurry wrapping loss factor; and constructing the vibration quality correlation characteristic matrix includes: Based on the motor current waveform and oil pressure sensor data of the vibrating equipment, the correlation coefficient between the energy consumption factor of the vibrating equipment and the vibration density is calculated, and the performance attenuation coefficient of the energy consumption factor of the vibrating equipment is determined using the grey correlation analysis model. Based on the porosity test results of concrete surface texture images and the aggregate distribution uniformity index, a dynamic evaluation model of aggregate distribution loss factor was constructed, and the nonlinear coupling coefficient between aggregate distribution loss factor and surface smoothness was calculated. Using the temporal variation trends of the paste encapsulation parameters and the bubble distribution density, a transfer learning model for the paste encapsulation loss factor is established, and the correlation weight between the paste encapsulation loss factor and the concrete strength development is calculated; A dynamic weight allocation mechanism is adopted to integrate the performance attenuation coefficient, nonlinear coupling coefficient and correlation weight of each factor in the vibration quality correlation characteristic matrix to generate the vibration quality correlation characteristic matrix.

[0010] By adopting the above technical solution, a multi-factor dynamic correlation model is constructed. Through the grey correlation analysis model (GRM) and transfer learning model (TLM), the correlation coefficient between the energy consumption factor of the vibration equipment and the density (such as the correlation coefficient ρ = 0.76), and the correlation weight between the slurry encapsulation loss factor and the strength development are calculated, thereby realizing the quantitative coupling of multiple physical field parameters and solving the pain point that the traditional method cannot quantify the correlation between process parameters and quality indicators. By dynamically allocating the weights of each factor (energy consumption factor weight 0.32, aggregate factor weight 0.41), a vibration quality correlation feature matrix containing spatial position information is generated to clarify the contribution of each factor to the formation of defects.

[0011] In a preferred example of the present application, the establishment of a quality assessment system including vibration defect location and quality score prediction includes: Based on the vibration sensor data of the vibrating equipment, the Pearson correlation coefficient between the vibration energy density distribution characteristics and the bubble distribution density on the concrete surface is calculated; The mutual information value of aggregate distribution uniformity and surface roughness is extracted through the gray-level co-occurrence matrix of the concrete surface texture image. Using the slurry encapsulation parameters and the vibration energy density distribution characteristics, a support vector machine regression model of slurry penetration depth and concrete density is established; A weighted tensor fusion algorithm is used to integrate the correlation degree of each feature in the vibration quality correlation feature matrix to generate a quality assessment system including vibration defect positioning and quality score prediction.

[0012] By adopting the above technical solution, the present application adopts the weighted tensor fusion algorithm (WTF) to integrate the vibration energy density distribution characteristics (such as Pearson correlation coefficient r = 0.82) and the slurry penetration depth (such as SVM regression R 2 =0.91), a dual-branch evaluation system including defect coordinate positioning and quality scoring was constructed; and a complete quality assessment system was formed by integrating the correlation of each feature using a weighted tensor fusion algorithm.

[0013] In a preferred embodiment of the present application, the method further includes: The hazard index H of the vibration defect location area is calculated based on formula (1) i : Where i is the defect area identifier; S i is the defect area ratio; D i is the defect depth; C i is the defect type risk factor; A i is the texture complexity of the defect neighborhood; The K-means++ clustering algorithm is used to classify the hazard index and generate a three-dimensional decision tree that includes warning level, disposal priority and process correction suggestions.

[0014] By adopting the above technical solution, the risk index H is calculated based on formula (1). i To quantitatively grade the severity of defects, this application incorporates the texture complexity of the defect neighborhood into the evaluation system for the first time, solving the problem that traditional methods ignore the impact of the defect surrounding environment (such as the adjacent bubble-dense area that easily causes chain damage); then use the K-means++ clustering algorithm to classify the H i The system is divided into different levels to generate a three-dimensional decision tree including warning levels (red / yellow / green), disposal priorities (emergency / general / observation) and process correction suggestions (re-vibration / slump adjustment / equipment replacement).

[0015] In a preferred example of the present application, the process correction suggestion includes a vibration time adjustment suggestion and a vibration head trajectory planning suggestion, and the method further includes: Dynamically adjust the vibration time T according to formula (2) opt : T opt =α×H max / v concrete +β×ln(N defect +1) Among them, H max is the maximum defect depth; v concrete is the slump of concrete; N defect is the total number of defects; α and β are empirical coefficients; The improved A* algorithm is used to plan the trajectory of the vibrating head. The constraints include: Among them, d jk represents the actual path distance between node j and node k; ∠jk represents the angle between the path direction from node j to node k and the direction of the previous path; L max and θ min These are pre-set parameters, representing the maximum stroke and minimum turning angle of the vibrating head.

[0016] By adopting the above technical solution, formula (2) dynamically adjusts the vibration time T opt , so that the vibration time is accurately matched with the actual working conditions, and the improved A* algorithm improves the planning efficiency of the vibration head movement trajectory.

[0017] In a preferred embodiment of the present application, the method includes: Based on the real-time collected vibration energy density distribution characteristics and ultrasonic echo delay data, a dynamic evaluation model for vibration density is constructed. The calculation formula is: in, is the dynamic evaluation value of real-time vibration density; P wave is the power spectrum density of pressure wave propagation; is the time derivative of the power spectrum density of pressure wave propagation, reflecting the vibration energy transmission efficiency; f ultra is the main frequency of ultrasonic echo; λ, μ are the equipment type correction coefficients; when Less than D threshold When the vibration time compensation mechanism is triggered: T compensate =T opt ×(1+|D gap | / D avg ) Among them, Tcompensate The vibration time for compensation; D gap is the real-time density deviation value; D avg is the historical average density.

[0018] By adopting the above technical solution, including the pressure wave propagation power spectrum density P wave and the ultrasonic echo main frequency f ultra Dynamic evaluation model In order to realize real-time calculation of vibration density, the parameters of different types of vibration equipment can be adapted through the equipment correction coefficients of λ and μ, avoiding energy waste caused by the "one-size-fits-all" strategy.

[0019] In the second aspect, the invention objective of this application is achieved by adopting the following technical solutions: A system for identifying the apparent quality of fresh concrete vibrated by a convolutional neural network is applied to the above-mentioned method for identifying the apparent quality of fresh concrete vibrated by a convolutional neural network. The system comprises: The data acquisition and processing module is used to obtain the vibration trajectory data, vibration frequency parameters and concrete surface texture images of the vibration equipment during operation, and correspondingly generate the original data set containing the vibration intensity distribution characteristics and apparent quality labels; The feature analysis module is used to analyze the time series detection data of the vibration process, calculate the correlation between the vibration energy density distribution characteristics and the concrete surface texture characteristics, extract the vibration energy density distribution characteristics, aggregate distribution uniformity index and paddle body wrapping parameters, and construct the vibration quality correlation feature matrix; A quality assessment system establishment module is used to integrate the vibration quality correlation feature matrix and the real-time collected vibration sensor data in the deep feature extraction network to establish a quality assessment system including vibration defect location and quality score prediction; The visual display module is used to detect the vibration operation process according to the quality assessment system and obtain visual assessment results.

[0020] In a third aspect, the invention objective of this application is achieved by adopting the following technical solutions: A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned method for identifying the apparent quality of fresh concrete vibration based on a convolutional neural network.

[0021] Fourthly, the invention objectives of this application are achieved by adopting the following technical solutions: A computer program product includes a computer program / instruction, which, when executed by a processor, implements the steps of the method for identifying the apparent quality of fresh concrete vibration based on a convolutional neural network as described above.

[0022] In summary, this application includes at least one of the following beneficial technical effects: 1. This application integrates three types of heterogeneous data: vibration trajectory, frequency parameters, and texture images, to construct an original data set containing vibration intensity distribution characteristics and apparent quality labels; improves the accuracy of apparent quality evaluation of fresh concrete through multimodal data fusion; and performs transfer learning on the pre-trained CNN model based on the operating parameters of the vibration equipment (such as motor current and oil pressure), thereby improving the feature extraction accuracy of the deep feature extraction network in the vibration process scenario; and also integrates the quality correlation feature matrix with real-time vibration sensor data to establish a joint evaluation system including defect location and quality scoring, realizing end-to-end closed-loop control from data acquisition to process optimization.

[0023] 2. Incorporate the texture complexity of the defect neighborhood into the evaluation system to solve the problem that the traditional method ignores the influence of the defect surrounding environment (such as the adjacent bubble dense area is prone to chain damage); then use the K-means++ clustering algorithm to H i The system is divided into different levels to generate a three-dimensional decision tree including warning levels (red / yellow / green), disposal priorities (emergency / general / observation) and process correction suggestions (re-vibration / slump adjustment / equipment replacement). BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a flow chart of a method for identifying the apparent quality of fresh concrete vibration based on a convolutional neural network in one embodiment of the present application; Figure 2 This is a flowchart of step S3 in the method for identifying the apparent quality of fresh concrete vibrated based on a convolutional neural network in one embodiment of the present application; Figure 3 This is a flowchart of step S4 in the method for identifying the apparent quality of fresh concrete vibration based on a convolutional neural network in one embodiment of the present application. DETAILED DESCRIPTION

[0025] The present application is further described in detail below with reference to the accompanying drawings.

[0026] In one embodiment, if Figure 1 As shown, the present application discloses a method for identifying the apparent quality of fresh concrete vibration based on a convolutional neural network, which specifically includes the following steps: S1: Obtain the vibration trajectory data, vibration frequency parameters, and concrete surface texture images of the vibration equipment during operation, and generate a corresponding raw data set containing vibration intensity distribution characteristics and apparent quality labels.

[0027] In this embodiment, the vibration trajectory data refers to the three-dimensional spatial coordinate sequence (x, y, z, t) of the vibrating rod in the concrete operation; the vibration frequency parameters include the vibration frequency (Hz) and amplitude (mm) when the vibrating rod contacts the concrete; the concrete surface texture image is a microscopic morphology image of the concrete surface captured by an industrial camera (resolution ≥ 0.1mm / pixel); the original data set is a labeled data set containing vibration intensity distribution characteristics (such as vibration energy density) and apparent quality labels (such as honeycomb surface grade).

[0028] Specifically, a distributed sensor network is installed on the vibrating equipment: ① A three-axis accelerometer (sampling rate ≥ 1kHz) and a pressure sensor (range 0-50MPa) are embedded in the end of the vibrating rod. ② A laser scanner (accuracy ±0.02mm) and a high-frame-rate industrial camera (≥100fps) are deployed to synchronously collect surface texture and slurry distribution data. Sensor data are aligned through timestamps to ensure temporal and spatial consistency of trajectory, frequency, and image data.

[0029] Furthermore, step S1 includes: S11: The three-dimensional positioning trajectory data of the vibrating equipment is collected in real time through a distributed sensor network, and the pressure time domain signal and vibration frequency response curve of the vibrating rod contact surface are obtained synchronously.

[0030] In this embodiment, the distributed sensing network is a network composed of multiple sensor nodes, which is used to collect multi-dimensional data of the vibration equipment; the three-dimensional positioning trajectory data refers to the position coordinate sequence (t, x, y, z) of the vibrating rod in three-dimensional space, which can be obtained through an inertial navigation module or a laser positioning system; the pressure time domain signal is a waveform signal of the pressure change generated when the vibrating rod contacts the concrete that changes with time; the vibration frequency response curve is a spectral characteristic curve of the vibration frequency of the vibration equipment that changes with time or excitation.

[0031] Specifically, based on the GPS timestamp, the sensor data is time-aligned at the nanosecond level. The trajectory data is saved in the [timestamp, x, y, z] format, and the pressure and vibration signals are saved in the [timestamp, voltage value] format, which are subsequently converted into physical quantities through calibration.

[0032] S12: Based on the micro-texture image of the concrete surface captured by the machine vision system and combined with the surface slurry distribution morphology data obtained by the laser scanner, an original monitoring data set containing spatiotemporal characteristic parameters is constructed.

[0033] In this embodiment, the machine vision system refers to a visual inspection system composed of an industrial camera, a light source, and image processing software; the micro-texture image refers to a high-resolution image of the microscopic morphology of the concrete surface; the laser scanner obtains the three-dimensional morphological data of the thickness of the concrete surface slurry through the laser ranging principle; the spatiotemporal characteristic parameters refer to data that simultaneously contains spatial position and time change characteristics (such as the change of texture roughness with vibration time).

[0034] Specifically, the surface texture image and laser scanning data are aligned using the ICP algorithm to establish a spatial correspondence; typical defect areas (such as bubbles and honeycombs) in the surface texture image are marked; and the Mask R-CNN algorithm is used to automatically segment the slurry coverage area.

[0035] S13: A time window threshold is set according to the vibration process specification, and the continuously collected time series data is divided into multiple monitoring periods. A sub-dataset containing vibration intensity distribution characteristics and apparent quality labels is generated for each monitoring period.

[0036] In this embodiment, the time window threshold is based on the data segmentation time length set by the vibration process specification (such as a single vibration cycle is 30 seconds); the monitoring period refers to the independent data segments divided by time windows in the continuously collected time series data; the apparent quality label refers to the defect type or quality grade label generated by artificial or machine learning models (such as "no defects", "slight honeycomb", "severe pitting").

[0037] Specifically, the size of the time window is adaptively adjusted according to the vibration frequency parameters (such as vibration duration). For example, if the vibration time is ≤20 seconds, the window length is set to 25 seconds; if the vibration time is >20 seconds, the window length is set to 1.2 times the vibration time.

[0038] Vibration energy density distribution features (such as vibration energy density mean and energy attenuation coefficient) are extracted from the raw data. The defect areas are annotated using a support vector machine (SVM) classifier to generate apparent quality labels. When storing data, each sub-dataset is saved as [time window ID, trajectory data, pressure signal, vibration signal, texture image, slurry data, quality label].

[0039] S14: The original data of the original monitoring data set is preprocessed using a time series feature segmentation algorithm to extract the vibration energy density distribution characteristics, aggregate distribution uniformity index and slurry wrapping parameters, and establish an original data set containing a multidimensional feature matrix of vibration quality.

[0040] In this embodiment, the time series feature segmentation algorithm is used to extract key feature segments from continuous time series data (such as sliding window + statistical feature extraction); data preprocessing includes denoising filtering and normalization processing, and denoising filtering includes Savitzky-Golay filtering denoising on the pressure signal and wavelet denoising on the vibration signal.

[0041] Specifically, the vibration energy density distribution refers to the extraction of the mean and variance of the energy spectrum density in the 20-100 Hz frequency band by calculating the short-time Fourier transform of the vibration signal. The aggregate distribution uniformity is calculated by using the gray-level co-occurrence matrix (GLCM) to calculate the contrast, correlation, energy and uniformity of the texture image. The slurry coverage is calculated using laser scanning data to calculate the standard deviation of the slurry thickness and coverage, where the coverage is equal to the slurry coverage area / total surface area × 100%.

[0042] Specifically, when constructing the multidimensional feature matrix of vibration quality, the mutual information method is used to screen the 12 features with the highest correlation with the quality label. A feature matrix of [number of samples × 12] is constructed. Each row corresponds to the multidimensional features of a single vibration cycle. The storage format is CSV format, containing columns named [sample ID, mean energy density, aggregate contrast, slurry coverage, ..., quality label]. S2: Construct an initial convolutional neural network model, perform transfer learning on the network structure based on the collected vibration equipment operating parameters and original data set, complete model training and optimization under the predefined concrete quality evaluation index system, and obtain a deep feature extraction network adapted to the vibration process.

[0043] In this embodiment, the initial convolutional neural network model is a pre-trained CNN model (such as ResNet50); the operating parameters include the operating status data of the vibration equipment (such as motor current, oil pressure, and temperature); and the concrete quality evaluation index system includes a standardized evaluation system for indicators such as honeycomb surface grade, bubble density, and surface flatness.

[0044] Specifically, we use the pre-trained ResNet50 as the backbone network and replace the last fully connected layer with a multi-task output layer: Branch 1: Vibration defect location (output is defect area coordinate probability map); Branch 2: Quality score prediction (output is a continuous value from 0 to 100).

[0045] Freezing strategy: Freeze the parameters of the first 10 convolutional layers and only train the last 3 fully connected layers (to prevent overfitting).

[0046] Transfer learning optimizations for the initial convolutional neural network model include: ① Data augmentation: random rotation (±15°), Gaussian noise injection (σ=0.03), and size scaling (0.8-1.2 times) of the original image; ② Design a multi-task joint loss function: L All =α L ×L Dice +β L ×L CrossEntropy +γ L ×L Sooth , where L All is the total loss; L Dice is the defect segmentation loss; L CrossEntropy is the quality score loss; L Sooth is the smoothness constraint; α L , β L , γ L is the corresponding loss weight coefficient.

[0047] ③ The training configuration is: using the Adam optimizer (learning rate = 0.001, batch size = 32), training for 120 epochs on an NVIDIA V100 GPU; in the performance verification phase, the defect segmentation mIoU = 89.2% and the quality score prediction RMSE = 4.2 (out of 100) were achieved on the test set.

[0048] S3: Analyze the time series detection data of the vibration process, calculate the correlation between the vibration energy density distribution characteristics and the concrete surface texture characteristics, extract the vibration energy density distribution characteristics, aggregate distribution uniformity index and paddle wrapping parameters, and construct the vibration quality correlation feature matrix.

[0049] In this embodiment, the vibration energy density distribution characteristic refers to the spatial distribution density of the vibration energy in the concrete (unit: J / m 3 ); the aggregate distribution uniformity index refers to the degree of spatial dispersion of aggregate particles in concrete (dimensionless); the paste coverage parameter refers to the coverage rate of cement paste on aggregate (%).

[0050] Specifically, if Figure 2 As shown, step S3 includes: S31: Based on the motor current waveform and oil pressure sensor data of the vibrating equipment, the correlation coefficient between the energy consumption factor of the vibrating equipment and the vibration density is calculated, and the performance attenuation coefficient of the energy consumption factor of the vibrating equipment is determined using a grey correlation analysis model.

[0051] In this embodiment, the motor current waveform is a curve of the current change over time when the motor of the vibrating equipment is running, reflecting the load status of the equipment; the oil pressure sensor data refers to the real-time data for monitoring the pressure change of the oil pressure cylinder of the vibrating rod, which characterizes the intensity of the vibration impact force; the energy consumption factor is a parameter used to quantify the relationship between the energy consumption of the vibrating equipment and the vibration density; the gray correlation analysis model is an algorithm for analyzing the degree of correlation of factors by calculating the geometric similarity between sequences; the performance attenuation coefficient characterizes the degradation index of the energy consumption factor as the vibration time or number of times changes; this embodiment collects the three-phase current waveform through the motor controller (sampling rate ≥1kHz); uses an oil pressure sensor (range 0-20MPa) to synchronously record the oil pressure data, and normalizes the oil pressure data.

[0052] Specifically, the vibration density test value (such as ultrasonic echo delay) is used as the reference sequence X0; the motor current effective value I rms and oil pressure mean P mean As comparison sequences X1 and X2, the correlation coefficient is calculated based on the following formula: Among them, s is the influencing factor identifier; ξ s (k) is the sth comparison sequence X s The correlation coefficient with the reference sequence X0 at the kth moment; X0(k) is the reference sequence, which represents the detection value of the vibration density (such as ultrasonic echo delay); X s (k) is the time series data of the influencing factors (such as the effective value of the motor current I rms , oil pressure mean P mean ; min s min k |X0(k)-X s (k)| is the minimum difference between the two levels, which means the minimum value of the absolute difference between all comparison sequences and the reference sequence at each time point k. For example, the first level: for each time point k, find all comparison sequences X s The minimum difference between (k) and X0(k), the second level: among all time points, find the minimum of these minimum differences; ρ = 0.5 is the resolution coefficient; max s max k |X0(k)-X s (k) | Maximum difference between the two levels, which represents the maximum absolute difference between all comparison sequences and the reference sequence at each time point k.

[0053] Performance attenuation coefficient = ρ initial -ρ current / ρ initial × 100%, where ρ initialis the initial correlation coefficient, which indicates the correlation between the energy consumption factor (such as motor current, oil pressure) and the vibration density when the vibration equipment is new or in the initial state; ρ current is the current correlation coefficient, which indicates the correlation between the energy consumption factor and the vibration density under the current state of the vibration equipment.

[0054] S32: Based on the porosity detection results of the concrete surface texture image and the aggregate distribution uniformity index, a dynamic evaluation model of the aggregate distribution loss factor is constructed, and the nonlinear coupling coefficient between the aggregate distribution loss factor and the surface flatness is calculated.

[0055] In this embodiment, porosity detection is to measure the pore volume ratio inside the concrete through CT scanning or mercury intrusion testing; the dynamic evaluation model is a machine learning model that predicts the aggregate distribution state based on time series data; the nonlinear coupling coefficient is an indicator used to measure the nonlinear relationship between aggregate distribution and surface flatness.

[0056] Specifically, a long short-term memory network (LSTM) is used to build a dynamic evaluation model, with the input being time series detection data and the output being the aggregate distribution loss factor. in, is the model prediction value, that is, the aggregate distribution loss factor prediction value output by the LSTM network; N is the total number of training samples; y ig is the actual value, which is the actual value of the aggregate distribution loss factor obtained through experiments or manual annotation; g is the regularization coefficient; is the L2 norm square of the model parameters; R on the test set 2 =0.87, RMSE=0.03 (based on the uniformity of aggregate distribution).

[0057] S33: Using the temporal variation trend of the paste encapsulation parameters and the bubble distribution density, a transfer learning model of the paste encapsulation loss factor is established, and the correlation weight between the paste encapsulation loss factor and the concrete strength development is calculated.

[0058] 1. In this embodiment, the transfer learning model is an algorithm that migrates the pre-trained model to a new task (paste encapsulation prediction); the association weight is a parameter used to quantify the correlation between paste encapsulation and concrete strength development.

[0059] Specifically, transfer learning implementation includes: Pre-trained model: ResNet50 pre-trained on ImageNet is used, and the last fully connected layer is replaced with the intensity prediction output layer.

[0060] Fine-tuning strategy: freeze the first 10 convolutional layers and only train the last 3 layers with a learning rate of 10^-4.

[0061] The correlation weight calculation = strength prediction error reduction rate / slurry encapsulation change rate. When verifying the model, 5-fold cross validation was used, and the mean absolute error MAE = 2.1 MPa.

[0062] S34: A dynamic weight allocation mechanism is used to integrate the performance attenuation coefficient, nonlinear coupling coefficient and correlation weight of each factor in the vibration quality correlation characteristic matrix to generate the vibration quality correlation characteristic matrix.

[0063] In this embodiment, the dynamic weight allocation mechanism is an algorithm that adjusts weights in real time according to the importance of factors.

[0064] Specifically, the entropy weight method is used to calculate the information entropy H of each factor k′ , the formula is: Where i is the sample index; x′ is the identifier of each factor in the vibration quality correlation characteristic matrix; p ix′ is the normalized value of the x′th factor in the i-th sample; an expert scoring mechanism is used to invite five engineers to score the importance of the factors (1-10 points), and the average value is taken as the subjective weight; the calculation formula of the comprehensive weight is: Then the energy consumption factor of the vibrating equipment η energy , aggregate distribution loss factor η aggregate and slurry wrapping loss factor η paste Integration by weight: M quality =ω1×η energy +ω2×η aggregate +ω3×η paste , Min-Max normalization is used to map the matrix values ​​to the [0, 1] interval.

[0065] S4: In the deep feature extraction network, the vibration quality-related feature matrix and the real-time collected vibration sensor data are integrated to establish a quality assessment system that includes vibration defect location and quality score prediction.

[0066] In this embodiment, the quality assessment system includes a defect localization subnetwork and a quality score prediction subnetwork; the defect localization subnetwork is a defect localization subnetwork of a semantic segmentation network based on the U-Net architecture, which adopts an improved U-Net architecture and introduces an attention gating mechanism (AG) to enhance the feature response of key areas; the quality score prediction subnetwork is a regression network based on the Softmax classifier, which extracts global features based on ResNet18 and generates quality scores through global average pooling (GAP).

[0067] Specifically, if Figure 3 As shown, step S4 includes: Establish a quality assessment system that includes vibration defect location and quality score prediction, including: S41: Based on the vibration sensor data of the vibrating equipment, the Pearson correlation coefficient between the vibration energy density distribution characteristics and the bubble distribution density on the concrete surface is calculated.

[0068] In this embodiment, the Pearson correlation coefficient is an indicator for measuring the degree of linear correlation between two variables, and its value range is [-1, 1].

[0069] Specifically, vibration signals are collected by the acceleration sensor built into the vibrating rod, and bubble volume distribution data is obtained through ultrasonic CT scanning. The acceleration signal is analyzed in the time-frequency domain to calculate the mean and variance of the energy spectrum density in the 20-100 Hz frequency band to obtain information on the vibration energy density distribution. Based on the binarization processing of the CT scan image, the number of bubbles per unit volume is counted to obtain the bubble distribution density.

[0070] The calculation formula of Pearson correlation coefficient r is: Among them, E i is the vibration energy density; B i is the bubble density; is the mean.

[0071] S42: The mutual information value of aggregate distribution uniformity characteristics and surface roughness is extracted through the gray level co-occurrence matrix of the concrete surface texture image.

[0072] In this embodiment, the aggregate distribution uniformity characteristic refers to a parameter that quantifies the degree of discreteness of the spatial distribution of aggregate in concrete (such as fractal dimension); surface roughness refers to the degree of irregularity of the microscopic morphology of the concrete surface (unit: μm); and the mutual information value (MI) is an indicator used to measure the amount of information shared by two variables, with a larger value indicating a stronger correlation.

[0073] Specifically, the fractal dimension D is calculated based on the fractal box counting method. f The surface roughness characteristics are extracted by calculating the contrast, correlation, energy, and uniformity of the gray level co-occurrence matrix (ASM). When calculating the mutual information value, the co-occurrence probability P(I′, J′) of the aggregate distribution uniformity characteristics and the surface roughness characteristics is statistically calculated, and then the following mutual information formula is used: Among them, X is the aggregate distribution characteristics (such as “uniform distribution”, “local aggregation”, “dispersed”), and Y is the surface roughness characteristics (such as “smooth”, “medium rough”, “highly rough”); is the marginal probability distribution of X, which indicates the probability that the aggregate distribution feature X alone takes the value i (without considering Y); is the marginal probability distribution of Y, which indicates the probability that the surface roughness feature Y alone takes the value j (without considering X); is the “information gain” term of the mutual information, which measures the difference between the joint distribution of X and Y and their independent distributions.

[0074] S43: Using the paste encapsulation parameters and vibration energy density distribution characteristics, a support vector machine regression model of paste penetration depth and concrete density is established.

[0075] In this embodiment, support vector machine regression (SVR) maps nonlinear problems to a regression model in high-dimensional space through kernel function; the coefficient of determination (R 2 ) is an indicator used to measure the degree of fit between the model prediction value and the true value (the value range is [0, 1]).

[0076] Specifically, the function selection of the support vector machine regression model is: using the radial basis function (RBF) kernel: K(x i , x j )=exp(-γ‖x i -x j ‖ 2 ) Among them, x i , x j is the eigenvector of the input sample (such as the eigenvalue in the concrete vibration quality correlation characteristic matrix); ‖x i -x j ‖ 2 For sample x i with x j The square of the Euclidean distance; γ is the kernel parameter, which controls the width of the kernel function. The larger γ is, the smaller the range of the kernel function is. Then, C (penalty coefficient) and γ are optimized through grid search.

[0077] S44: A weighted tensor fusion algorithm is used to integrate the correlation of each feature in the vibration quality correlation feature matrix to generate a quality assessment system that includes vibration defect location and quality score prediction.

[0078] In this embodiment, the weighted tensor fusion algorithm realizes the joint modeling of defect location and quality score through the weighted fusion of multi-dimensional feature tensors; the quality assessment system includes a semantic segmentation network based on the U-Net architecture (defect location subnetwork, outputs vibration defect location results) and (quality score prediction subnetwork, outputs quality score prediction value).

[0079] Specifically, a vibration quality correlation feature matrix is ​​constructed, wherein the input features are the vibration energy density distribution feature in step S41, the aggregate distribution uniformity feature in step S42, and the slurry penetration depth prediction value in step S43.

[0080] The matrix form is: Where E is the energy density; B is the bubble density; P is the slurry penetration depth; combining the spatial characteristics (image resolution) and temporal characteristics (time series change rate), a tensor fusion strategy is used to perform weighted calculation: W = Softmax (W·F T +b), where W is the weight matrix; b is the bias term; and the loss function L of the vibration quality correlation feature matrix is ​​established ju : L ju =α′L seg +β′L score , where L seg is the defect segmentation loss; L score is the quality score loss (MSE); Adam optimizer is used for training optimization (learning rate 0.0001).

[0081] S5: The vibration operation process is inspected according to the quality assessment system to obtain visual assessment results.

[0082] In this embodiment, the visual evaluation results include a heat map of defect types and quality score prediction results. In actual applications, the quality score prediction results will divide the vibration process into green (high quality), yellow (qualified), and red (rework required) warning areas; the visual evaluation results are displayed through an interactive visualization terminal, such as the development of a Web-based visualization platform, which supports: real-time data stream access (such as using the WebSocket protocol for communication interaction), three-dimensional scene roaming and defect annotation, partner process parameter playback, and process optimization suggestion query.

[0083] Specifically, the defect probability map is superimposed on the three-dimensional model of the concrete surface, and the red area represents high-risk defects (probability > 0.8) to achieve defect location visualization; a color score map is generated based on the quality score obtained in the quality assessment system, and visual rendering is performed. K-means++ is used to cluster the hazard index to generate three levels of warning (such as red, yellow, and green). The specific number of warning levels can be customized according to actual display needs. This application takes the three-level warning as an example.

[0084] For example, the process correction suggestions include triggering the "supplementary vibration + slump adjustment" instruction when the red warning level is met; recommending "optimizing the vibration trajectory" when the yellow warning level is met; and recommending maintaining the current process parameters when the green warning level is met.

[0085] In one embodiment, in step S4, a density monitoring submodule is established to jointly model the defect location and quality score prediction in the dynamic assessment model and the quality assessment system. The method for identifying the apparent quality of fresh concrete vibrated based on a convolutional neural network also includes: Based on the real-time collected vibration energy density distribution characteristics and ultrasonic echo delay data, a dynamic evaluation model for vibration density is constructed. The calculation formula is: in, is the dynamic evaluation value of real-time vibration density; P wave is the power spectrum density of pressure wave propagation; is the time derivative of the power spectrum density of pressure wave propagation, reflecting the vibration energy transmission efficiency; f ultra is the main frequency of ultrasonic echo; λ, μ are the equipment type correction coefficients; when Less than D threshold When the vibration time compensation mechanism is triggered: T compensate =T opt ×(1+|D gap | / D avg ) Among them, T compensate The vibration time for compensation; D gap is the real-time density deviation value; D avg is the historical average density.

[0086] In one embodiment, after step S4, the defect hazard index is further calculated and graded, that is, the method for identifying the apparent quality of fresh concrete vibrated based on a convolutional neural network further includes: The hazard index H of the vibration defect location area is calculated based on formula (1) i : Where i is the defect area identifier; S i is the defect area ratio; D i is the defect depth; C i is the defect type risk factor; A i is the texture complexity of the defect neighborhood; The K-means++ clustering algorithm is used to classify the hazard index and generate a three-dimensional decision tree that includes warning level, disposal priority and process correction suggestions.

[0087] In this embodiment, based on the above-mentioned hazard index classification results, process parameters are dynamically optimized. Process correction suggestions include vibration time adjustment suggestions and vibration head trajectory planning suggestions. The convolutional neural network-based method for identifying the apparent quality of fresh concrete vibration also includes: Dynamically adjust the vibration time T according to formula (2) opt : T opt =α×H max / v concrete+β×ln(N defect +1) Among them, H max is the maximum defect depth; v concrete is the slump of concrete; N defect is the total number of defects; α and β are empirical coefficients; The improved A* algorithm is used to plan the trajectory of the vibrating head. The constraints include: Among them, d jk represents the actual path distance between node j and node k; ∠jk represents the angle between the path direction from node j to node k and the direction of the previous path; L max and θ min These are pre-set parameters, representing the maximum stroke and minimum turning angle of the vibrating head.

[0088] It should be understood that the serial numbers of the steps in the above embodiments do not imply the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0089] In one embodiment, a system for identifying the apparent quality of fresh concrete by vibration based on a convolutional neural network is provided. The system for identifying the apparent quality of fresh concrete by vibration based on a convolutional neural network corresponds to the method for identifying the apparent quality of fresh concrete by vibration based on a convolutional neural network in the above-mentioned embodiment.

[0090] The convolutional neural network-based system for identifying the apparent quality of fresh concrete vibrated concrete consists of a data acquisition and processing module, a feature analysis module, a quality assessment system establishment module, and a visualization module. A detailed description of each functional module is as follows: The data acquisition and processing module is used to obtain vibration trajectory data, vibration frequency parameters, and concrete surface texture images during operation of the vibrating equipment, and correspondingly generates a raw data set containing vibration intensity distribution characteristics and apparent quality labels; The feature analysis module is used to analyze the time series detection data of the vibration process, calculate the correlation between the vibration energy density distribution characteristics and the concrete surface texture characteristics, extract the vibration energy density distribution characteristics, aggregate distribution uniformity index and paddle body wrapping parameters, and construct the vibration quality correlation feature matrix; A quality assessment system establishment module is used to integrate the vibration quality correlation feature matrix with real-time vibration sensor data in a deep feature extraction network to establish a quality assessment system that includes vibration defect location and quality score prediction; The visual display module is used to detect the vibration operation process according to the quality assessment system and obtain visual assessment results.

[0091] For the specific limitations of the fresh concrete vibration apparent quality identification system based on convolutional neural networks, please refer to the limitations of the fresh concrete vibration apparent quality identification method based on convolutional neural networks in the above text, which will not be repeated here; the various modules in the above-mentioned fresh concrete vibration apparent quality identification system based on convolutional neural networks can be implemented in whole or in part through software, hardware and their combination; the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0092] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented: S1: Obtain the vibration trajectory data, vibration frequency parameters and concrete surface texture images of the vibration equipment during operation, and generate a raw data set containing vibration intensity distribution characteristics and apparent quality labels; S2: Construct an initial convolutional neural network model, perform transfer learning on the network structure based on the collected vibration equipment operating parameters and original data set, complete model training and optimization under a predefined concrete quality evaluation index system, and obtain a deep feature extraction network adapted to the vibration process; S3: Analyze the time series detection data of the vibration process, calculate the correlation between the vibration energy density distribution characteristics and the concrete surface texture characteristics, extract the vibration energy density distribution characteristics, aggregate distribution uniformity index and paddle body wrapping parameters, and construct the vibration quality correlation feature matrix; S4: In the deep feature extraction network, the vibration quality correlation feature matrix is ​​integrated with the real-time vibration sensor data to establish a quality assessment system that includes vibration defect location and quality score prediction; S5: The vibration operation process is inspected according to the quality assessment system to obtain visual assessment results.

[0093] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0094] In one embodiment, in particular, according to an embodiment of the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product comprising a computer program / instructions that, when executed by a processor, implement the steps of the method for identifying the apparent quality of fresh concrete vibration based on a convolutional neural network. In such an embodiment, the computer program can be downloaded and installed from a network via a communication module and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), the various functions defined in the present invention are performed.

[0095] Those skilled in the art will clearly understand that for the sake of convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0096] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, it should be understood by those skilled in the art that the technical solutions described in the aforementioned embodiments may still be modified, or some of the features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. A method for identifying the apparent quality of fresh concrete vibration based on convolutional neural networks, characterized in that: include: Acquire the vibration trajectory data, vibration frequency parameters, and concrete surface texture images of the vibrating equipment during operation, and generate a raw data set containing vibration intensity distribution characteristics and apparent quality labels; Constructing an initial convolutional neural network model, performing transfer learning on the network structure based on the collected vibration equipment operating parameters and the original data set, completing model training optimization under a predefined concrete quality evaluation index system, and obtaining a deep feature extraction network adapted to the vibration process; Analyze the time series detection data of the vibration process, calculate the correlation between the vibration energy density distribution characteristics and the concrete surface texture characteristics, extract the vibration energy density distribution characteristics, aggregate distribution uniformity index and paddle body wrapping parameters, and construct the vibration quality correlation feature matrix; In the deep feature extraction network, the vibration quality correlation feature matrix and the real-time collected vibration sensor data are integrated to establish a quality assessment system including vibration defect location and quality score prediction; The vibration operation process is inspected according to the quality assessment system to obtain a visual assessment result.

2. The method for identifying the apparent quality of fresh concrete vibration based on a convolutional neural network according to claim 1, characterized in that: The method of obtaining the vibration trajectory data, vibration frequency parameters and concrete surface texture image of the vibration equipment during operation includes: The distributed sensor network collects the three-dimensional positioning trajectory data of the vibrating equipment in real time, and simultaneously obtains the pressure time domain signal and vibration frequency response curve of the vibrating rod contact surface; The machine vision system captures the concrete surface micro-texture image, and combines it with the surface slurry distribution morphology data obtained by the laser scanner to construct an original monitoring data set containing spatiotemporal characteristic parameters. The time window threshold is set according to the vibration process specification, and the continuously collected time series data is divided into multiple monitoring periods. Each monitoring period generates a sub-dataset containing the vibration intensity distribution characteristics and apparent quality labels. The time series feature segmentation algorithm is used to preprocess the raw data of the original monitoring data set, extract the vibration energy density distribution characteristics, aggregate distribution uniformity index and slurry wrapping parameters, and establish the original data set containing the multidimensional feature matrix of vibration quality.

3. The method for identifying the apparent quality of fresh concrete vibration based on convolutional neural network according to claim 2, characterized in that: The vibration quality correlation characteristic matrix includes the vibration equipment energy consumption factor, the aggregate distribution loss factor and the slurry wrapping loss factor; The construction of the vibration quality correlation feature matrix includes: Based on the motor current waveform and oil pressure sensor data of the vibrating equipment, the correlation coefficient between the energy consumption factor of the vibrating equipment and the vibration density is calculated, and the performance attenuation coefficient of the energy consumption factor of the vibrating equipment is determined using the grey correlation analysis model. Using the porosity test results of concrete surface texture images and the aggregate distribution uniformity index, a dynamic evaluation model for the aggregate distribution loss factor was constructed, and the nonlinear coupling coefficient between the aggregate distribution loss factor and surface smoothness was calculated. A transfer learning model for the paste encapsulation loss factor was established using the temporal variation trend of the paste encapsulation parameter and the bubble distribution density, and the correlation weight between the paste encapsulation loss factor and the concrete strength development was calculated. A dynamic weight allocation mechanism is adopted to integrate the performance attenuation coefficient, nonlinear coupling coefficient and correlation weight of each factor in the vibration quality correlation characteristic matrix to generate the vibration quality correlation characteristic matrix.

4. The method for identifying the apparent quality of fresh concrete vibration based on convolutional neural network according to claim 1, characterized in that: The establishment of a quality assessment system including vibration defect location and quality score prediction includes: Based on the vibration sensor data of the vibrating equipment, the Pearson correlation coefficient between the vibration energy density distribution characteristics and the bubble distribution density on the concrete surface is calculated; The mutual information value of aggregate distribution uniformity and surface roughness is extracted through the gray-level co-occurrence matrix of the concrete surface texture image. A support vector machine regression model of slurry penetration depth and concrete density is established using the slurry encapsulation parameters and the vibration energy density distribution characteristics; A weighted tensor fusion algorithm is used to integrate the correlation degree of each feature in the vibration quality correlation feature matrix to generate a quality assessment system including vibration defect positioning and quality score prediction.

5. The method for identifying the apparent quality of fresh concrete vibration based on convolutional neural network according to claim 1, characterized in that: The method further comprises: The hazard index H of the vibration defect location area is calculated based on formula (1) i : Where i is the defect area identifier; S i is the defect area ratio; D i is the defect depth; C i is the defect type risk factor; A i is the texture complexity of the defect neighborhood; The K-means++ clustering algorithm is used to classify the hazard index and generate a three-dimensional decision tree that includes warning level, disposal priority and process correction suggestions.

6. The method for identifying the apparent quality of fresh concrete vibration based on convolutional neural network according to claim 5, characterized in that: The process correction suggestion includes a vibration time adjustment suggestion and a vibration head trajectory planning suggestion, and the method further includes: Dynamically adjust the vibration time T according to formula (2) opt : T opt =α×H max / v concrete +β×ln(N defect +1) Among them, H max is the maximum defect depth; v concrete is the slump of concrete; N defect is the total number of defects; α and β are empirical coefficients; The improved A* algorithm is used to plan the trajectory of the vibrating head. The constraints include: Among them, d jk represents the actual path distance between node j and node k; ∠jk represents the angle between the path direction from node j to node k and the direction of the previous path; L max and θ min These are pre-set parameters, representing the maximum stroke and minimum turning angle of the vibrating head.

7. The method for identifying the apparent quality of fresh concrete vibration based on convolutional neural network according to claim 6, characterized in that: The method comprises: Based on the real-time collected vibration energy density distribution characteristics and ultrasonic echo delay data, a dynamic evaluation model for vibration density is constructed. The calculation formula is: in, is the dynamic evaluation value of real-time vibration density; P wave is the power spectrum density of pressure wave propagation; is the time derivative of the power spectrum density of pressure wave propagation, reflecting the vibration energy transmission efficiency; f ultra is the main frequency of ultrasonic echo; λ, μ are the equipment type correction coefficients; when Less than D threshold When the vibration time compensation mechanism is triggered: Among them, T compensate The vibration time for compensation; D gap is the real-time density deviation value; D avg is the historical average density.

8. A fresh concrete vibration apparent quality recognition system based on convolutional neural network, characterized by: The method for identifying the apparent quality of fresh concrete vibration based on a convolutional neural network according to any one of claims 1 to 7 comprises: a data acquisition and processing module for acquiring vibration trajectory data, vibration frequency parameters, and concrete surface texture images of a vibrating device during operation, and correspondingly generating an original data set including vibration intensity distribution characteristics and apparent quality labels; The feature analysis module is used to analyze the time series detection data of the vibration process, calculate the correlation between the vibration energy density distribution characteristics and the concrete surface texture characteristics, extract the vibration energy density distribution characteristics, aggregate distribution uniformity index and paddle body wrapping parameters, and construct the vibration quality correlation feature matrix; A quality assessment system establishment module is used to integrate the vibration quality correlation feature matrix and the real-time collected vibration sensor data in the deep feature extraction network to establish a quality assessment system including vibration defect location and quality score prediction; The visual display module is used to detect the vibration operation process according to the quality assessment system and obtain visual assessment results.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for identifying the apparent quality of fresh concrete vibration based on a convolutional neural network as described in any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method for identifying the apparent quality of fresh concrete vibration based on a convolutional neural network as described in any one of claims 1 to 7 are implemented.

Images