A fine aggregate moisture content real-time detection method for concrete production

Abstract

This invention discloses a real-time detection method for the moisture content of fine aggregates in concrete production, belonging to the field of fine aggregate moisture content detection. The method includes: acquiring several sand samples with different moisture contents for different fine aggregates; extracting stable feature values ​​of the signal from the original signal sequence acquired by a microwave moisture content sensor for each sand sample using the sliding window density peak method; calculating the average thickness, average flow velocity, and average temperature; constructing a first prediction model based on an enhanced multiple regression model; constructing a second prediction model based on a tree model; determining the optimal moisture content prediction model from the first and second prediction models; and in concrete production, acquiring the stable feature values, thickness, flow velocity, and temperature of the signal in real time, and calling the optimal moisture content prediction model for the corresponding fine aggregate to predict the moisture content. This invention solves the problem of measurement inaccuracy caused by random fluctuations in the original signal and variable material flow conditions when using microwave methods to detect the moisture content of sand and gravel online.

Description

Technical Field

[0001] This invention belongs to the field of fine aggregate moisture content detection, and particularly relates to a method for real-time detection of fine aggregate moisture content in concrete production. Background Technology

[0002] The moisture content of fine aggregates is a key parameter in concrete mix design, and its accuracy directly affects the water-cement ratio control and the final quality of concrete. Currently, methods for testing the moisture content of fine aggregates mainly fall into two categories: (1) Traditional direct methods: such as drying method, alcohol combustion method, volumetric flask method, etc. Although these methods have high accuracy (such as standard drying method), they generally have disadvantages such as long testing time, delayed results, and inability to measure online in real time, which makes it difficult to meet the needs of concrete production line to adjust the mix ratio in real time.

[0003] (2) Novel indirect method: The moisture content is indirectly calculated by measuring physical quantities related to moisture content, mainly including: Resistance method and capacitance method: simple structure and fast response, but low measurement accuracy and greatly affected by the shape, density and temperature of the material, making them unsuitable for online precision measurement.

[0004] Infrared method: non-contact and fast, but expensive and with poor penetration. It can only measure surface moisture and is greatly affected by the color and shape of the material.

[0005] Neutron method: High precision and strong penetration, but it has radioactive safety risks, the equipment is expensive, and it cannot distinguish between water and other hydrogen-containing substances.

[0006] Microwave method: Based on the difference in dielectric constant between moisture and fine aggregate, this method offers advantages such as fast response, penetrating measurement (penetration depth up to 75-100mm), and relatively low cost, making it the mainstream technology for online moisture content detection. A common microwave sensor for detecting fine aggregate moisture content is the HM100 intelligent moisture sensor. This sensor is a contact-type microwave moisture content measurement device that also measures and outputs real-time temperature values. It contacts the fine aggregate flow through a ceramic panel placed along the sand conveyor, outputting a 0-20mA current signal (corresponding to 0-100% humidity) based on changes in the dielectric constant. After simple linear calibration, the moisture content value is detected and output.

[0007] In practical engineering applications, existing microwave moisture content sensors, represented by HM100, often have a significant error of more than 1% between their measurement results and the actual measured values ​​obtained by the drying method, which seriously interferes with the quality control of concrete production. After analysis, the root cause of the technical problem is: (1) The equipment directly performs an arithmetic average calculation on the collected original fluctuation signal values, resulting in poor representativeness of the "average humidity value" and unstable test results; (2) The equipment's binary simple linear fitting model of "moisture content - average humidity value" fits the true data distribution pattern, and the selection of calibration points has a huge impact on the overall error, resulting in insufficient model prediction ability; (3) In actual production, factors such as the thickness of sand on the conveyor belt (H), the speed of the conveyor belt (S, reflecting the sand flow rate), and the ambient temperature (T) are constantly changing, and different working conditions affect the final output results. Therefore, solving the common industry problem of measurement inaccuracy caused by random fluctuations of the original signal and the changing working conditions of the material flow when using microwave method to detect the moisture content of sand and gravel has strong practical application value. Summary of the Invention

[0008] To address the aforementioned shortcomings in the existing technology, this invention provides a real-time detection method for the moisture content of fine aggregates in concrete production, which solves the problem of measurement inaccuracy caused by random fluctuations in the original signal and variable material flow conditions when using microwave methods to detect the moisture content of sand and gravel online.

[0009] To achieve the aforementioned objectives, the technical solution adopted by this invention is: a method for real-time detection of the moisture content of fine aggregates in concrete production, comprising: Determine the optimal moisture content prediction model for fine aggregates of different types and sizes, and integrate a library of aggregate moisture content calculation models: For fine aggregates of the current type and particle size, several sand samples with different moisture contents are obtained. Thickness and flow velocity sequences of each sand sample during transport, as well as the original signal and temperature sequences of the sand samples collected by a microwave moisture content sensor, are collected to obtain a sand sample dataset. For the original signal sequences corresponding to each sand sample in the dataset, stable signal features are extracted using the sliding window density peak method. Based on the thickness, flow velocity, and temperature sequences of each sand sample, the average thickness, average flow velocity, and average temperature during transport are calculated. Based on the stable signal features, average thickness, average flow velocity, average temperature, and moisture content of each sand sample, a first prediction model for moisture content prediction is constructed using an enhanced multiple regression model with interaction and higher-order terms. Based on the stable signal features, average thickness, average flow velocity, average temperature, moisture content, and density fractions corresponding to the stable signal features of each sand sample, a second prediction model for moisture content prediction is constructed using a tree model. Based on the goodness of fit, the optimal moisture content prediction model for the current type and particle size of fine aggregate is determined from the first and second prediction models. During the concrete production process, the signal stability characteristics, average thickness, average flow rate, and average temperature of the fine aggregate conveying process are collected in real time. The optimal moisture content prediction model for the corresponding fine aggregate is called from the aggregate moisture content calculation model library to predict the moisture content.

[0010] Furthermore, the microwave moisture content sensor is installed on the fine aggregate conveyor belt.

[0011] Furthermore, the extraction of stable signal features specifically includes: Null values ​​and invalid zero values ​​in the original signal sequence of the current sand sample are removed to obtain a number of remaining valid signals; The remaining valid signals are sorted by time to obtain the valid signal sequence; Set a sliding window with a fixed width; The sliding window moves sequentially over the effective signal sequence with a step size of 1 to form several windows, and the density fraction of the effective signal at the center of each window is calculated. The value of the effective signal corresponding to the maximum density fraction is used as the signal stability feature value of the current sand sample.

[0012] Furthermore, the expression for the density fraction is:

[0013]

[0014] in, For the first Density fraction of each effective signal; and All are valid signal indices; n The total number of valid sequences; The width of the sliding window; For kernel functions; For the first The value of each valid signal; For the first The value of each valid signal; This is the bandwidth parameter.

[0015] Furthermore, the expression for the first prediction model is:

[0016] in, Moisture content; For constant terms; , , and All are fitting coefficients for the first-order term; These are the stable characteristic values ​​of the signal; Average thickness; The average flow velocity; Average temperature; , , and All are quadratic fitting coefficients; , , , , and All are interaction term fitting coefficients.

[0017] The beneficial effects of this invention are as follows: A preprocessing method for microwave signals with moisture content is proposed, fundamentally improving the quality and stability of the input data; applying sliding window kernel density estimation to the processing of the original fluctuation signal can intelligently identify and extract the most stable and concentrated data points in the signal sequence as signal feature values ​​D, effectively filtering out outliers caused by material fluctuations and noise, fundamentally reducing the first bias. Experiments show that the coefficient of variation of the data processed by this method is significantly reduced, laying a solid foundation for subsequent accurate modeling. A multivariate nonlinear prediction model capable of capturing key interaction effects is constructed, achieving systematic and synergistic correction of multiple biases: abandoning the simple binary linear model of "moisture content - mean humidity", an enhanced prediction model is established that integrates four variables (D, H, S, T) and forcibly introduces key interaction terms (D×H, D×S, H×S, etc.) and higher-order terms; in addition, the data stability index (maximum density score) generated in the preprocessing stage is used as an additional feature input tree model, which can simultaneously overcome the second and third biases. This study not only describes the nonlinear relationship between moisture content and the core signal value D, but also eliminates the interference caused by these key operating conditions by introducing thickness H, flow velocity S, and temperature T as compensation variables. The model fit R² can be improved from less than 0.75 in traditional methods to over 0.93, and the root mean square error (RMSE) is reduced to about 0.35%. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method of the present invention.

[0019] Figure 2 This is a general flowchart of the overall operation of the present invention.

[0020] Figure 3 This is a graph showing the fitting effect between the predicted and actual moisture content values ​​in an embodiment of the present invention. Detailed Implementation

[0021] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0022] Example 1 like Figure 1 As shown, in one embodiment of the present invention, a method for real-time detection of the moisture content of fine aggregates in concrete production includes: Determine the optimal moisture content prediction model for fine aggregates of different types and sizes, and integrate a library of aggregate moisture content calculation models: For fine aggregates of the current type and particle size, several sand samples with different moisture contents are obtained. Thickness and flow velocity sequences of each sand sample during transport, as well as the original signal and temperature sequences of the sand samples collected by a microwave moisture content sensor, are collected to obtain a sand sample dataset. For the original signal sequences corresponding to each sand sample in the dataset, stable signal features are extracted using the sliding window density peak method. Based on the thickness, flow velocity, and temperature sequences of each sand sample, the average thickness, average flow velocity, and average temperature during transport are calculated. Based on the stable signal features, average thickness, average flow velocity, average temperature, and moisture content of each sand sample, a first prediction model for moisture content prediction is constructed using an enhanced multiple regression model with interaction and higher-order terms. Based on the stable signal features, average thickness, average flow velocity, average temperature, moisture content, and density fractions corresponding to the stable signal features of each sand sample, a second prediction model for moisture content prediction is constructed using a tree model. Based on the goodness of fit, the optimal moisture content prediction model for the current type and particle size of fine aggregate is determined from the first and second prediction models. During the concrete production process, the signal stability characteristics, average thickness, average flow rate, and average temperature of the fine aggregate conveying process are collected in real time. The optimal moisture content prediction model for the corresponding fine aggregate is called from the aggregate moisture content calculation model library to predict the moisture content.

[0023] In this embodiment, the method of the present invention pre-embeds the moisture content calculation models of various types and particle sizes of fine aggregates required at the engineering site into the data processing module, and calls the corresponding fine aggregate models during actual application. The establishment process of the moisture content calculation models for various fine aggregates follows the... Figure 1 The process is shown below.

[0024] The microwave moisture content sensor is installed on the fine aggregate conveyor belt.

[0025] The extraction of stable feature values ​​of the signal is specifically as follows: Null values ​​and invalid zero values ​​in the original signal sequence of the current sand sample are removed to obtain a number of remaining valid signals; The remaining valid signals are sorted by time to obtain the valid signal sequence; Set a sliding window with a fixed width; The sliding window moves sequentially over the effective signal sequence with a step size of 1 to form several windows, and the density fraction of the effective signal at the center of each window is calculated. The value of the effective signal corresponding to the maximum density fraction is used as the signal stability feature value of the current sand sample.

[0026] The expression for the density fraction is:

[0027]

[0028] in, For the first Density fraction of each effective signal; and All are valid signal indices; n The total number of valid sequences; The width of the sliding window; For kernel functions; For the first The value of each valid signal; For the first The value of each valid signal; This is the bandwidth parameter.

[0029] In this embodiment, step S1: Laboratory fine aggregate multi-element signal acquisition and data processing: Under laboratory conditions, for fine aggregates of the same type and particle size, at least 60 groups covering different moisture contents were set up. Y (Obtained by drying method test), thickness H Flow rate S ,temperature T The sand samples were simultaneously monitored, and raw electrical signal values ​​and temperature values ​​were collected from microwave sensors. T and the average thickness from the auxiliary sensing unit. H and flow rate S The collected data is then organized to obtain a dataset for this type of fine aggregate.

[0030] Step S2: First Deviation Correction – Signal Stability Feature Extraction Based on Sliding Window Density Peak Method This step involves preprocessing the original, time-series acquired microwave signal value sequences for each data set. Traditional methods directly calculate the arithmetic mean of the microwave signal value sequences, which can lead to inaccurate results due to factors such as uneven material flow, instantaneous changes in sensor surface contact conditions, and electronic noise interference. To overcome the random fluctuations (first deviation) in the original microwave signal sequences, this invention identifies the regions with the densest data distribution in the signal value sequences and extracts their representative values, thereby automatically resisting interference from abnormal jump points. This invention proposes an improved sliding window processing method, aiming to extract the stable signal feature values ​​that best represent the overall dielectric properties of the current material from the fluctuation sequence. D The specific sub-steps are as follows: S2.1 Data Cleaning and Preparation: First, remove null values ​​and invalid zero values ​​from the sequence (usually caused by no data passing through). Arrange the remaining valid signal values ​​in chronological order of acquisition time to preserve their fluctuation patterns.

[0031] S2.2 Sliding window kernel density estimation: Window definition: Set a fixed width. W A sliding window (e.g., W=21 or other data points, or covering a 1-2 second acquisition duration, or taking a fixed percentage of the total effective data) slides along the time series with a step size of 1 data point.

[0032] Core calculations: For the window W For each data point, instead of using a simple arithmetic mean, a kernel density estimate is calculated. Specifically, for the window center point (the... i (points), the formula for calculating its density fraction is:

[0033] in, K This refers to a kernel function (such as the Gaussian kernel function or the Epanechnikov kernel function). h This is the bandwidth parameter (which can be pre-calibrated based on the signal fluctuation range). The essence of this step is to evaluate the bandwidth parameter. i The density of local data distribution centered on a point. The higher the density score, the more data points are clustered around the signal value, and the more likely the value is to represent the steady state of the material rather than transient noise.

[0034] S2.3 Peak Identification and Feature Extraction: Calculate the density score for each valid data point in the complete sequence to obtain the density score sequence, and then search for the global maximum in this sequence. Unlike simply finding the maximum signal value, this step searches for the signal level corresponding to the "denseest data distribution".

[0035] Maximum density fraction The corresponding original signal value is extracted and used as a representative stable feature value for this acquisition cycle. D .

[0036] Compared to the traditional arithmetic average method, which treats all data points equally, the method in this invention uses sliding window kernel density estimation to assign higher weights to signal values ​​in "data-dense regions," automatically reducing the impact of sparse, discrete, and abnormal fluctuations. This is equivalent to performing "denoising" and "steady-state extraction" at the signal level, providing higher-quality and more stable signal feature values ​​for subsequent models. D The input value effectively overcomes the "first deviation" caused by random fluctuations in the original signal.

[0037] Alternative Solution 1: A symmetrical truncated mean method can be used as an alternative to step S2. That is, the first m% and last m% of the sorted data are removed, and the arithmetic mean of the middle (100-2m)% of the data is calculated as the mean. D This method is relatively simple, but its stability and adaptability to operating conditions are inferior to the sliding window density peak method.

[0038] Alternative Solution 2: Machine learning clustering algorithms (such as K-means) can be used to cluster the original signal, and the centroid or mean of the largest cluster can be used as the... D .

[0039] The expression for the first prediction model is:

[0040] in, Moisture content; For constant terms; , , and All are fitting coefficients for the first-order term; These are the stable characteristic values ​​of the signal; Average thickness; The average flow velocity; Average temperature; , , and All are quadratic fitting coefficients; , , , , and All are interaction term fitting coefficients.

[0041] In this embodiment, the stable signal feature value generated in step S2 DThis is the cornerstone of the high-precision prediction achieved by the model in step S3. If the mean of the wildly fluctuating original signal is used directly, even with a complex nonlinear model, the prediction results will still be inaccurate due to excessive input noise. Conversely, the multivariate interactive model in step S3 can fully utilize the reliable signal provided by S2. D and at the same time compensate H , S , T This has enabled the systematic and coordinated correction of multiple biases.

[0042] Step S3: Second and Third Deviation Co-correction – Moisture Content Prediction Based on Nonlinear Model of Multi-Source Information Fusion The stable eigenvalues ​​obtained in step S2 D The average thickness of the sand layer collected simultaneously H Average flow velocity S Average temperature T Together they are used as input features. Actual moisture content Y The relationships between these features are not simple independence or linear superposition, but rather involve complex coupling and nonlinear interactions. Therefore, this invention constructs a multivariate nonlinear prediction model. Y = F ( D , H , S , T The core of this approach lies in explicitly modeling these key interaction effects. The model preferably employs one of the following structures or a fusion thereof: 1. Enhanced multiple regression model with interaction terms and higher-order terms: The model includes not only D , H , S , T The first-order terms must also include the interaction terms of the key variables and the quadratic terms.

[0043] This model uses interaction items D × H To quantify "the difference in moisture content caused by different thicknesses under the same signal value"; through D × S This is used to quantify the "modulation effect of flow velocity on microwave measurement stability." This allows the model to collaboratively correct for variations in operating conditions (…). H , S Changes and core signals D The "second and third deviations" are caused by mutual coupling.

[0044] 2. Nonlinear fitting based on tree model: Algorithms such as Random Forest, XGBoost, LightGBM, and Gradient Boosting Decision Tree are employed. During model training, not only the input features [ D , H , S , T The maximum density score for each sample calculated in step S2 will also be used. As additional feature input. This characterizes the stability of the sample data itself, which can guide the tree model to perform differentiated fitting of data with different confidence levels, thereby further improving the robustness and accuracy of the model.

[0045] Collaborative design description of S2 and S3: Step S2 provides D and It has a deep synergy with the model structure in step S3. D As a core input, its stability directly determines the baseline accuracy of the model's predictions. As a "confidence" feature input to the model, it enables the model to know the current input. D The reliability of the prediction is determined, thereby dynamically adjusting its output weights during prediction.

[0046] Step S4: After voting and integration, obtain the optimal moisture content calculation model for fine aggregate of the current type and particle size: The results of the above multiple models are integrated, and a certain rule is adopted (such as comprehensively comparing the final goodness of fit R² and root mean square error RMSE of each model) to select the optimal moisture content calculation model for the current type and particle size of fine aggregate. The parameters and structure of this model are then pre-fixed and embedded into the memory of the data processing module in the system to form a fine aggregate moisture content calculation model library, so that it can be called up in time during production.

[0047] Example 2 like Figure 2 As shown, a method for real-time detection of the moisture content of fine aggregates used in concrete production is implemented as follows: (1) Preset the weight ratio of each ingredient in the industrial control computer according to the actual situation of the concrete mixing site; (2) According to the preset weight of fine aggregate, the industrial control system controls the opening of the batching silo. With the cooperation of the weight sensing module, the preset weight of fine aggregate is unloaded and the silo door is closed. At the same time, the fine aggregate enters the conveyor belt for conveying. (3) Fine aggregate flows over the surface of the microwave moisture content sensor via a conveyor belt, and the sensor collects the signal; Preferably, a threshold is set for the number of empty values ​​collected by the sensor. When the number of empty or zero values ​​collected is within the threshold, it is considered that the same batch of sand has passed through the microwave moisture content sensor. When the number of empty or zero values ​​collected exceeds the threshold, it is considered that all the sand needed has flowed through the sensor panel. (4) The microwave moisture content sensor, thickness detection / control unit, and flow rate detection unit transmit their respective detection data, such as the original signal value and sand thickness. H ), flow rate ( S ),temperature( T The data is output to the data processing module until the fine aggregate is completely unloaded; (5) The data processing module first uses the sliding window density peak method to process the original electrical signal value to obtain a representative signal value. D Then, it calls the moisture content calculation model for the corresponding fine aggregate from the moisture content calculation model library, calculates the overall moisture content of the current batch of fine aggregate based on the input multivariate data, and outputs the overall moisture content of the fine aggregate. Y To industrial control systems; (6) The industrial control system will receive the moisture content Y Other characteristic variables are displayed on the data monitoring page, along with the moisture content. Y The data is input into the built-in automatic concrete production mix design model, which calculates the adjusted amounts of fine aggregate and water. The material silo door is then reopened to replenish the fine aggregate to the adjusted level, and then the silo door is closed. At the same time, the water valve is opened to add water to the required amount before the water valve is closed.

[0048] Preferably, conventional concrete mix design models adjust material quantities according to the principle of "increasing sand and reducing water," but they overlook the fact that "increasing sand" introduces new moisture. Therefore, it is necessary to further adjust the amounts of fine aggregate and water using a "sand-water coupling" approach. For example, if the preset concrete mix proportion is cement:sand:water = Akg:Bkg:Ckg, and the moisture content of the sand is obtained through testing and data processing... Y Then the adjusted sand dosage B' = B(1- Y )+B Y / (1- Y The adjusted water usage is C' = C - B'. Y -SSD), where SSD is the saturated surface dry water absorption rate of sand. Fine aggregates with different water absorption capacities will have different SSD values. The corresponding values ​​can be obtained through experiments or by consulting sand manufacturers and can be pre-configured into the automatic mix design calculation model for concrete production.

[0049] In this way, the precise dosage of each material during concrete mixing ensures the quality of the finished concrete product.

[0050] We used manufactured river sand with a fineness modulus of 3.04. In the laboratory, we systematically designed multiple groups of samples with different moisture contents to be placed under sensors for detection, and obtained their datasets. Y , D , H , S , T The dataset was processed using the modeling method described in this invention, resulting in a quadratic multivariate polynomial regression model as the optimal moisture content calculation model. ; The model's predicted moisture content values ​​are compared with the actual values. The fitting effect of "predicted moisture content - actual value" is as follows: Figure 3 As shown; Its goodness of fit R 2 The accuracy can reach 0.9351, with a root mean square error of only about 0.35% (meaning the average error between the moisture content model prediction and the actual value is only about 0.35%, within 0.5%). The model introduces... SH , ST , HD , HT Interactive items such as these effectively compensate for deviations caused by changes in working conditions. They are pre-embedded in the memory of the data processing module in the system, forming part of the fine aggregate moisture content calculation model library, which can be called up in time during production.

[0051] The current concrete mixing task requires a mix ratio of cement:sand:water = 500kg:1000kg:200kg, which is pre-set in the industrial control computer. The sand is the aforementioned manufactured river sand with a fineness modulus of 3.04 (SSD = 3.1%). The industrial control system controls the opening of the batching silo. With the assistance of the weight sensing module, 200kg of fine aggregate is unloaded, and then the silo door is closed. Simultaneously, the fine aggregate enters the conveyor belt for transportation. The fine aggregate flows over the surface of the microwave moisture content sensor, and the sensor collects the signal. The microwave moisture content sensor, thickness detection / control unit, and flow rate detection unit transmit their respective detection data, such as the original signal value and sand thickness (…). H ), flow rate ( S ),temperature( T The output is sent to the data processing module until the fine aggregate is completely unloaded; the data processing module first uses the sliding window density peak method to process the original electrical signal value to obtain a representative signal value. D =2252, in addition, sand thickness H =4.5cm, flow rate S =40r / m, temperature T =23.4℃, then call the moisture content calculation model for the corresponding fine aggregate from the moisture content calculation model library:

[0052] The overall moisture content of this batch of sand is calculated and output from the input multivariate data. Y =5.09% (actual drying test moisture content is 5.1%, error is only 0.01%) is sent to the industrial control system; the industrial control system displays the received moisture content Y and other characteristic variables on the data monitoring page, and simultaneously inputs the moisture content Y=5.09% into the built-in automatic concrete production mix proportion calculation model to calculate the adjusted sand usage B'=B(1- Y )+B Y / (1- Y ) = 1000 × (1 - 5.09%) + 1000 × 5.09% / (1 - 5.09%) = 1002.73 kg, and the adjusted water usage is C' = C - B'. Y -SSD)=200-1002.73×(5.09%-3.1%)=180.05kg; then the industrial control computer controls the material silo door to reopen, unload 2.73kg of sand, and then close the silo door. At the same time, it controls the water valve to open and add water until 180.05kg is reached, and then closes the water valve. In this way, the concrete mixing plant achieves real-time dynamic and high-precision material feeding control, improving production quality.

Claims

1. A method for real-time detection of moisture content in fine aggregates used in concrete production, characterized in that, include: Determine the optimal moisture content prediction model for fine aggregates of different types and sizes, and integrate a library of aggregate moisture content calculation models: For fine aggregates of the current type and particle size, several sand samples with different moisture contents are obtained. Thickness and flow velocity sequences of each sand sample during transport, as well as the original signal and temperature sequences of the sand samples collected by a microwave moisture content sensor, are collected to obtain a sand sample dataset. For the original signal sequences corresponding to each sand sample in the dataset, a sliding window density peak method is used to extract stable signal features. Based on the thickness, flow velocity, and temperature sequences of each sand sample, the average thickness, average flow velocity, and average temperature during transport are calculated. Based on the stable signal features, average thickness, average flow velocity, average temperature, and moisture content of each sand sample, a first prediction model for moisture content prediction is constructed based on an enhanced multiple regression model with interaction and higher-order terms. Based on the stable signal features, average thickness, average flow velocity, average temperature, moisture content, and density fractions corresponding to the stable signal features of each sand sample, a second prediction model for moisture content prediction is constructed based on a tree model. Based on the goodness of fit, the optimal moisture content prediction model for the current type and particle size of fine aggregate is determined from the first and second prediction models. During the concrete production process, the signal stability characteristics, average thickness, average flow rate, and average temperature of the fine aggregate conveying process are collected in real time. The optimal moisture content prediction model for the corresponding fine aggregate is called from the aggregate moisture content calculation model library to predict the moisture content.

2. The method for real-time detection of fine aggregate moisture content in concrete production according to claim 1, characterized in that, The microwave moisture content sensor is installed on the fine aggregate conveyor belt.

3. The method for real-time detection of fine aggregate moisture content in concrete production according to claim 1, characterized in that, The extraction of stable feature values ​​of the signal is specifically as follows: Null values ​​and invalid zero values ​​in the original signal sequence of the current sand sample are removed to obtain a number of remaining valid signals; The remaining valid signals are sorted by time to obtain the valid signal sequence; Set a sliding window with a fixed width; The sliding window moves sequentially over the effective signal sequence with a step size of 1 to form several windows, and the density fraction of the effective signal at the center of each window is calculated. The value of the effective signal corresponding to the maximum density fraction is used as the signal stability feature value of the current sand sample.

4. The method for real-time detection of fine aggregate moisture content in concrete production according to claim 3, characterized in that, The expression for the density fraction is: in, For the first Density fraction of each effective signal; and All are valid signal indices; n The total number of valid sequences; The width of the sliding window; For kernel functions; For the first The value of each valid signal; For the first The value of each valid signal; This is the bandwidth parameter.

5. The method for real-time detection of fine aggregate moisture content in concrete production according to claim 1, characterized in that, The expression for the first prediction model is: in, Moisture content; For constant terms; , , and All are fitting coefficients for the first-order term; These are the stable characteristic values ​​of the signal; Average thickness; The average flow velocity; Average temperature; , , and All are quadratic fitting coefficients; , , , , and All are interaction term fitting coefficients.

Images