Concrete mix proportion cross-batch closed-loop optimization method, device and storage medium

Abstract

This invention discloses a method, equipment, and storage medium for cross-batch closed-loop optimization of concrete mix proportions, relating to the field of concrete production technology. The optimization method includes collecting characteristic data and constructing a hardening feature vector during the early hardening stage of the current batch of concrete; inputting the hardening feature vector into a pre-trained performance prediction model to obtain the predicted performance value of the batch of concrete; calculating the cross-batch performance deviation between the predicted performance value and the target performance value; calculating the cross-batch performance deviation with an existing cross-batch adjustment gain to generate a mix proportion correction amount for the next batch of concrete; repeating the above steps to form a closed-loop optimization at multiple batch production scales. This invention upgrades early hardening monitoring from passive evaluation to active optimization, using early information to adjust subsequent production mix proportions, significantly reducing the batch-to-batch dispersion of key performance characteristics and improving the consistency and reliability of continuous material supply.

Description

Technical Field

[0001] This invention belongs to the field of concrete production technology, and in particular relates to a method, equipment and storage medium for cross-batch closed-loop optimization of concrete mix proportions based on real-time monitoring of hardening characteristics. Background Technology

[0002] As a primary material in modern building structures, the uniformity and stability of ready-mixed concrete directly affect engineering safety and structural durability. Currently, the industry generally adopts a traditional management model of "static mix design + random sampling quality evaluation." Specifically, the initial mix proportion is determined before production based on raw material inspection reports and empirical parameters; during production, rapid release control is mainly achieved through fresh mix performance indicators such as slump and spread; while the core properties of concrete (such as compressive strength and durability) rely on destructive testing of laboratory specimens at long ages (7 days, 28 days, etc.) for post-production evaluation and acceptance.

[0003] This management model suffers from two prominent systemic flaws: First, there is no definite, quantifiable correlation between the fresh mix performance (such as workability) and the final mechanical properties and durability of concrete; simply meeting slump requirements cannot guarantee long-term performance standards. Second, the test results for key indicators such as 28-day strength lag significantly behind the production process. By the time performance deviations are detected, concrete produced with the same mix proportion may have already been poured into multiple project batches or even critical structural components, leading to a continuous accumulation of quality risks over time. Subsequent problems arising from this include: insufficient strength in already formed structural components requiring reinforcement or demolition; project claims; and shortened structural service life due to potential durability defects.

[0004] To overcome the aforementioned lag issues, in recent years, non-destructive or semi-non-destructive sensing technologies, such as ultrasonic propagation velocity, concrete resistivity, and hydration heat monitoring, have been applied to track and characterize the early hardening process of concrete. However, the application of existing technologies mostly remains at the level of "passive quality judgment," that is, using monitoring data to determine whether the current batch of concrete has reached the demolding strength, prestressing tensile strength, or meets a certain single acceptance threshold. This monitoring data has not yet been systematically used to guide the mix proportion optimization decision for the next batch or subsequent production, failing to form an executable feedback loop from "monitoring-analysis-decision-execution." Therefore, even with process monitoring, the batch-to-batch dispersion of key concrete properties remains significant when faced with real-world disturbances such as fluctuations in raw material performance, changes in the compatibility of admixtures and cementitious materials, differences in environmental temperature and humidity, and inconsistencies in transportation and waiting times. The deeper value of monitoring information has not been fully explored.

[0005] In summary, the current technological system has significant shortcomings in terms of real-time performance, predictability, and closed-loop control capabilities. The industry urgently needs an innovative method that can transform rapidly obtainable monitoring information from the early stages of hardening into accurate predictions of the final concrete performance, and further generate quantifiable and executable mix design optimization instructions. This would enable intelligent, closed-loop quality control with early warning, proactive correction, and continuous convergence across multiple batches of continuous production, fundamentally improving the uniformity, reliability, and engineering safety of concrete production. Summary of the Invention

[0006] To address the aforementioned deficiencies in existing technologies, the present invention aims to provide a method, equipment, and storage medium for batch-to-batch closed-loop optimization of concrete mix proportions. This addresses the problems of large cumulative risks caused by performance feedback lag in traditional quality management models and the large batch-to-batch dispersion in long-term concrete performance resulting from existing early monitoring technologies being used only for passive judgment and unable to achieve proactive and forward-looking control.

[0007] This invention solves the above-mentioned technical problems through the following technical solution: a method for cross-batch closed-loop optimization of concrete mix proportions, comprising:

[0008] Step S1: After the i-th batch of concrete is poured, collect characteristic data reflecting the concrete hardening process in its early hardening stage.

[0009] Step S2: Based on the feature data, construct a hardened feature vector;

[0010] Step S3: Input the hardening feature vector into the pre-trained performance prediction model to obtain the performance prediction value of the i-th batch of concrete;

[0011] Step S4: Calculate the difference between the predicted performance value and the target performance value to obtain the cross-batch performance deviation of the i-th batch;

[0012] Step S5: Calculate the cross-batch performance deviation of the i-th batch with the existing cross-batch control gain to generate the mix proportion correction amount for the (i+1)-th batch of concrete.

[0013] Step S6: Repeat steps S1 to S5 to regulate the (i+1)th batch based on the predicted evaluation of the hardening process of the i-th batch, forming a closed-loop optimization at the multi-batch production scale to reduce the batch-to-batch dispersion of key concrete properties.

[0014] Traditional methods rely on fresh concrete indicators such as slump, which cannot accurately predict long-term strength and durability. This invention, by collecting and analyzing characteristic data from the early stages of hardening, directly captures key signals of internal concrete structure formation, thereby constructing a performance prediction model strongly correlated with final performance. This shifts the core of quality evaluation from apparent workability to intrinsic hardening quality, fundamentally improving the scientific rigor and accuracy of quality prediction.

[0015] Traditional methods relying on destructive testing of 28-day-old specimens result in a feedback cycle of up to one month, leading to cumulative risks of quality deviations. This invention utilizes rapidly obtainable early hardening characteristics to predict performance at 28 days or longer within 24-48 hours after casting, shortening the quality feedback cycle from months to days or even hours. This enables timely intervention, achieving a fundamental shift from post-cast assessment to pre-cast warning.

[0016] Existing monitoring technologies such as ultrasonic and resistivity are mostly used for single-batch quality compliance determination, and the value of the data remains largely untapped. This invention calculates cross-batch performance deviations and combines them with control gain to generate precise mix ratio correction instructions for the next production batch, establishing a complete control loop of "monitoring-prediction-decision-execution." This allows monitoring data to directly drive the optimization of production parameters, achieving a leap from perceiving the current situation to shaping the future.

[0017] Even with monitoring, existing technologies still exhibit significant batch-to-batch dispersion. This invention, by repeatedly executing the closed-loop process described above (step S6), transforms the deviation of each batch into optimization adjustments for the next batch, achieving dynamic convergence and stabilization of performance output across multiple batch scales. This is equivalent to endowing the concrete production system with self-learning and adaptive capabilities, enabling it to proactively counteract the effects of various disturbances and significantly improve the reliability and consistency of engineering supply.

[0018] Furthermore, the early hardening stage is defined as the time period from the initial setting of the concrete to 24 or 48 hours.

[0019] Alternatively, the early hardening stage is defined as the time period during which the ultrasonic wave propagation velocity in the concrete first enters and remains within a preset characterization range.

[0020] By providing clear and operational objective criteria for determining the "early hardening stage," the ambiguity and uncertainty that may arise from relying solely on experience or subjective judgment are effectively overcome. This dual-definition approach not only makes the determination of the monitoring window highly repeatable but also provides flexibility to adapt to different material systems, environmental conditions, or engineering needs, ensuring the universality and feasibility of the method in different scenarios.

[0021] Furthermore, the feature data includes at least two of the following categories:

[0022] The propagation speed of ultrasound and its first derivative with time;

[0023] Concrete resistivity and its first derivative with time;

[0024] Cumulative heat release during hydration or heat release power;

[0025] Condensation time;

[0026] The elastic modulus measured during the early stage of hardening.

[0027] Ultrasonic propagation speed, resistivity, heat of hydration, and other physical or chemical characteristics reflect the microscopic processes of early concrete hardening from different perspectives (such as structural densification, changes in pore solution ion concentration, and the progress of chemical reactions). Using at least two types of data avoids the limitations or misjudgments that may arise from single monitoring methods. By utilizing cross-validation and fusion of multi-source information, the information richness and representativeness of the hardening feature vector are significantly improved, thus providing a more reliable and robust input foundation for subsequent performance prediction models.

[0028] Furthermore, the construction of the hardened feature vector includes:

[0029] The feature data of each category are subjected to denoising, smoothing, missing data completion and normalization processing;

[0030] For each type of processed feature data, extract one or more of the following key features: peak value, inflection point, rate of change, and integral area;

[0031] The hardened feature vector is constructed by concatenating the key features extracted from various feature data.

[0032] By performing preprocessing such as denoising, smoothing, and normalization on each type of feature data, measurement errors and environmental interference were effectively eliminated. Furthermore, key features with physical or statistical significance were extracted for each data type, rather than simply using the original data sequence, achieving data dimensionality reduction and information purification. This step transforms massive amounts of noisy time-series data into a set of refined, highly representative feature values, greatly improving the training efficiency, inference speed, and prediction accuracy of subsequent prediction models.

[0033] Furthermore, the performance prediction model is a machine learning-based regression model used to establish a mapping relationship between the hardening feature vector and at least one of the concrete's 28-day compressive strength, durability index, shrinkage, or creep index.

[0034] By employing a machine learning-based regression model, this invention can automatically learn and capture the implicit, non-linear mapping between early hardening feature vectors and complex indicators such as 28-day compressive strength and durability. This method not only overcomes the shortcomings of traditional empirical formulas, such as insufficient accuracy and poor generalization ability, but also provides reliable performance predictions in a very short time after pouring, compressing the long 28-day quality waiting period into the early hardening stage, thus gaining valuable time for proactive quality control. Simultaneously, the model can flexibly accommodate unified or individual predictions of multiple performance indicators such as strength, chloride ion diffusion coefficient, frost resistance grade, shrinkage, and creep, achieving an upgrade from single strength control to comprehensive performance synergistic optimization, providing an unprecedented digital decision-making tool for the long-term durability design and assurance of concrete engineering.

[0035] Furthermore, the existing cross-batch control gain is maintained through the following dynamic process:

[0036] Based on the historical series of actual mix ratio corrections from multiple batches and the corresponding cross-batch performance deviation series, the initial cross-batch regulation gain is determined by least squares regression.

[0037] In a continuous production process, a recursive relationship is established for the i-th batch, where i ≥ 1:

[0038] Based on the cross-batch performance deviation of the i-th batch and the mix proportion correction amount for the (i+1)-th batch of concrete, the existing cross-batch control gain is updated using an adaptive update algorithm to obtain the updated cross-batch control gain.

[0039] The updated cross-batch control gain will be used as the existing cross-batch control gain for the (i+1)th batch of concrete; when i=1, the existing cross-batch control gain is the initial cross-batch control gain.

[0040] The dynamic maintenance of cross-batch control gain includes two stages: initial determination and recursive update. The initial value is derived from historical data regression, laying the scientific foundation for optimization. The recursive update mechanism allows the gain to be continuously fine-tuned based on the latest production feedback (performance deviations and corrections). This design endows the system with the ability to learn from experience and adapt to changes, enabling the control strategy to dynamically evolve with real-world conditions such as raw material fluctuations and environmental changes, thereby ensuring the long-term effectiveness and robustness of the closed-loop optimization system.

[0041] Furthermore, the existing cross-batch control gain is updated using an adaptive update algorithm, specifically including:

[0042] Based on the inter-batch performance deviation of batch i and the mix proportion correction for batch i+1 concrete, the instantaneous gain estimate is obtained by solving the following least squares problem:

[0043] ;

[0044] in, This represents the instantaneous gain estimate; This represents the mix proportion correction amount used for the (i+1)th batch of concrete; K represents the variables in the optimization process. This represents the cross-batch performance deviation of the i-th batch; Represents the square of the norm;

[0045] The instantaneous gain estimate and the existing cross-batch control gain are fused according to the following formula to obtain the updated cross-batch control gain:

[0046] ;

[0047] in, This indicates the updated cross-batch adjustment gain; This indicates the existing cross-batch adjustment gain; This represents the forgetting factor, which is between 0 and 1 and is used to control the weight of historical information and new observation information.

[0048] A weighted fusion approach balances old and new knowledge: the immediate gain estimate is calculated based on the latest batch of data, representing the latest experience; the forgetting factor is used to control the weight of historical accumulated experience. This update method can quickly respond to the latest changes in production status, smooth out random fluctuations, prevent drastic gain oscillations caused by abnormal data in a single batch, and ensure the stability and convergence of the control process.

[0049] Furthermore, in step S5, the specific process for generating the mix proportion correction amount for the (i+1)th batch of concrete includes:

[0050] Multiply the cross-batch performance deviation of the i-th batch by the existing cross-batch adjustment gain to obtain the initial correction amount;

[0051] A continuity constraint is applied to the initial correction amount to obtain the mix proportion correction amount for the (i+1)th batch of concrete; wherein, the continuity constraint is to limit the variation range of the initial correction amount to not exceed a preset limit value.

[0052] By introducing continuity constraints, the variation range of the mix ratio correction between adjacent batches is limited, effectively preventing sudden changes in instructions that may occur due to algorithm optimization. Such sudden changes could lead to process risks such as uneven mixing and loss of operational control in actual production. Continuity constraints force a smooth transition in the optimization path, ensuring the stability and controllability of the production process. This allows advanced closed-loop optimization algorithms to be safely and robustly integrated into existing production lines, improving the engineering applicability and acceptability of this invention.

[0053] Furthermore, after obtaining the mix proportion correction amount for the (i+1)th batch of concrete, a safety judgment measurement is also performed: if the mix proportion correction amount for the (i+1)th batch of concrete causes the water-cement ratio to exceed the corresponding preset upper limit, the total amount of cementitious materials to be lower than the corresponding preset lower limit, or the predicted value of the durability index to not meet the corresponding preset threshold, then the correction item for directly adjusting the water content is frozen, and performance compensation is achieved preferentially by adjusting the admixture dosage, cementitious material composition, or sand ratio.

[0054] When key parameters such as the water-cement ratio and the amount of cementitious materials reach preset safety thresholds, the system will automatically trigger and switch to a preset, safe alternative adjustment scheme (such as adjusting the admixture instead of adding water directly). This is equivalent to adding a set of mandatory safety rules to the automatic operation that pursues optimal performance, fundamentally eliminating the risk of sacrificing the basic durability and structural safety of concrete in pursuit of performance goals, and reflecting the core engineering ethics of "optimization without forgetting safety" in this invention.

[0055] Based on the same concept, the present invention also provides an electronic device, including a memory, a processor, and a computer program or instructions stored in the memory, wherein the processor executes the computer program or instructions to implement the concrete mix proportion cross-batch closed-loop optimization method as described above.

[0056] Based on the same concept, the present invention also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implements the cross-batch closed-loop optimization method for concrete mix proportions as described above.

[0057] Compared with existing technologies, this invention constructs a cross-batch closed-loop optimization mechanism of "monitoring-prediction-correction" through real-time monitoring of early hardening features and machine learning prediction, achieving significant beneficial effects:

[0058] First, it achieves a leap in quality evaluation from apparent workability to intrinsic hardening potential, solving the defect that the performance of fresh mix cannot reflect the final quality. Second, it shortens the quality feedback cycle from 28 days to the early stage of hardening, changing post-event assessment to pre-event early warning, eliminating the risk of delayed corrective action. Third, it transforms passive monitoring data into proactive optimization instructions, forming an executable adaptive control closed loop. Fourth, it significantly reduces the batch-to-batch performance dispersion caused by fluctuations in raw materials and the environment, improving the homogeneity and reliability of continuous production. Simultaneously, through continuity constraints and safety adjudication strategies, it ensures the smoothness of the optimization process and engineering safety. Ultimately, this invention upgrades concrete production from a static, experience-based model to dynamic, intelligent optimization. Attached Figure Description

[0059] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only one embodiment of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0060] Figure 1 This is the concrete mix proportion cross-batch closed-loop optimization method in the embodiments of the present invention;

[0061] Figure 2 This is a schematic diagram of the monitoring setup for the early stage of hardening characteristics in an embodiment of the present invention;

[0062] Figure 3 This is a schematic diagram of hardening characteristic vector construction and feature extraction in an embodiment of the present invention;

[0063] Figure 4 This is a schematic diagram comparing the dispersion of compressive strength before and after closed-loop optimization in an embodiment of the present invention.

[0064] Explanation of reference numerals in the attached figures: 1-Resistivity probe, 2-Ultrasonic transmitter, 3-Ultrasonic receiver, 4-Thermocouple, 5-Concrete specimen. Detailed Implementation

[0065] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0067] Example 1

[0068] This invention addresses the long-standing engineering challenges of delayed quality feedback and large batch-to-batch performance dispersion in ready-mixed concrete production by proposing a closed-loop optimization method for concrete mix proportions across batches based on real-time monitoring of hardening characteristics.

[0069] like Figure 1 As shown, the core of this invention lies in a closed-loop process of "monitoring-prediction-decision-execution" across production batches, specifically including the following steps executed sequentially:

[0070] Step S1, Early Hardening Monitoring: After the i-th batch of concrete is poured, characteristic data reflecting the hardening process of the concrete are collected during its early hardening stage.

[0071] Step S2, Feature Vector Construction: Based on the feature data collected in step S1, construct a feature vector representing the hardening behavior of the i-th batch, i.e., the hardening feature vector.

[0072] Step S3, Early Performance Prediction: Input the hardening feature vector constructed in step S2 into the pre-trained performance prediction model to obtain the long-term performance prediction value of the i-th batch of concrete.

[0073] Step S4, cross-batch deviation calculation: Calculate the difference between the performance prediction value obtained in step S3 and the predetermined target performance value to obtain the cross-batch performance deviation of the i-th batch.

[0074] Step S5, Correction Quantity Generation and Constraints: The cross-batch performance deviation of the i-th batch obtained in step S4 is calculated with the existing cross-batch control gain to generate the mix proportion correction quantity for the (i+1)-th batch of concrete, and the necessary engineering constraints are applied to the mix proportion correction quantity.

[0075] Step S6: Apply the mix proportion correction output in step S5 to the production of the i-th batch of concrete, and repeat steps S1 to S5 to achieve continuous optimization and stabilization of performance on a multi-batch production scale.

[0076] The steps S1 to S6 above form an adaptive closed loop, which uses early information from the current batch (i) to optimize the production of the next batch (i+1). This iterative cycle aims to proactively reduce the batch-to-batch dispersion of key concrete properties.

[0077] The implementation of the method of this invention relies on an intelligent monitoring and control system deeply integrated with the concrete production process. For example... Figure 2 As shown, the system mainly consists of a field monitoring unit, a data acquisition and transmission unit, and a central processing and decision-making unit in terms of physical structure.

[0078] The on-site monitoring unit is deployed in a representative concrete test block 5 or a solid structure to perform step S1. Specifically, it includes:

[0079] Resistivity probe 1: Inserted into the concrete specimen 5 and connected to the resistivity tester to measure the resistivity of the concrete.

[0080] Ultrasonic transducer pair: includes an ultrasonic transmitter 2 and an ultrasonic receiver 3, connected to an ultrasonic detector for measuring the propagation speed of ultrasonic waves.

[0081] Temperature sensor (such as thermocouple 4): implanted inside concrete block 5 and connected to temperature recorder to monitor temperature changes in concrete block 5, which can help analyze hydration heat release.

[0082] The data acquisition and transmission unit (such as the data acquisition system DAS) is mainly responsible for synchronously receiving raw signals from resistivity testers, ultrasonic detectors and temperature recorders, performing analog-to-digital conversion and packaging, and transmitting them to the central processing and decision-making unit via wired or wireless networks.

[0083] The central processing and decision-making unit receives data from the data acquisition and transmission unit and carries the core algorithm software modules for steps S2 to S5. The central processing and decision-making unit can be deployed on a local server at the mixing plant, a cloud computing platform, or a combination of both. Its main software modules include:

[0084] Feature preprocessing and vector construction module: corresponding to step S2, responsible for cleaning and transforming the original feature data and constructing hardened feature vectors.

[0085] Performance prediction module: corresponding to step S3, it has a built-in pre-trained performance prediction model, which is used to input the hardening feature vector into the pre-trained performance prediction model to obtain the performance prediction value of the i-th batch of concrete.

[0086] Cross-batch control module: corresponding to steps S4 and S5, responsible for calculating cross-batch performance deviation, generating a mix ratio correction amount based on the adaptive cross-batch control gain, and integrating constraint processing logic.

[0087] Execution and Recording Module: Responsible for issuing the final optimization instructions to the existing mixing plant production control system and recording data throughout the process.

[0088] Step S1 is completed collaboratively by the on-site monitoring unit and the data acquisition system. Specifically, monitoring is initiated immediately after the i-th batch of concrete is poured. The boundary of the early hardening stage can be determined using a combination of the following two parallel standards, with the earlier one prevailing:

[0089] Physical quantity interval standard: Continuous monitoring of ultrasonic wave propagation speed. The starting point is when the ultrasonic wave propagation speed first reaches and exceeds a preset lower limit (e.g., 2000 m / s), and the ending point is when it first exceeds a preset upper limit (e.g., 4000 m / s). The interval between this starting point and the ending point usually corresponds to the core stage of rapid structural formation.

[0090] Maximum time limit standard: An absolute time limit is set, such as 48 hours after the completion of pouring. Regardless of whether the physical quantity standard is met, this time is the latest monitoring endpoint.

[0091] In this embodiment, the effective monitoring period is typically about 8 to 36 hours after pouring.

[0092] During the monitoring period, the system synchronously collects the following three types of characteristic data at fixed intervals (e.g., every 30 minutes):

[0093] Ultrasonic propagation speed The first-order numerical derivative is directly measured by an ultrasonic testing instrument and calculated in real time, reflecting the rate of structural densification.

[0094] concrete resistivity The resistivity is measured by a resistivity meter, and its first numerical derivative is calculated in real time to reflect the change in ion concentration in the pore solution.

[0095] Hydration temperature rise curve: Recorded by temperature sensors, the cumulative heat release during hydration is calculated using an adiabatic temperature rise model. The adiabatic temperature rise model is based on the fundamental assumption that all the heat released during cement hydration is used to raise the temperature of the concrete itself, with no heat lost to the environment. Under these conditions, the cumulative heat release during hydration can be directly calculated by monitoring the temperature change of the concrete specimen in an adiabatic environment. .

[0096] Setting time: This is not a directly monitored time point, but rather obtained through analysis of the resistivity curve or the first derivative curve of the ultrasonic wave propagation velocity. Specifically, the system automatically identifies the time corresponding to the first sharp increase in the resistivity curve, or the time when the first derivative curve of the ultrasonic wave propagation velocity first reaches its peak value, and records this as the setting time. Setting time is a key node characterizing the early hardening process.

[0097] Elastic modulus E(t) measured in the early hardening stage: Directly measuring the elastic modulus of concrete in the early hardening stage is difficult. This embodiment uses an indirect estimation method: utilizing the synchronously acquired ultrasonic wave propagation velocity... With concrete density Through empirical formulas A dynamic estimation is performed, where k is an empirical coefficient related to Poisson's ratio. This estimated value can serve as an equivalent index reflecting early stiffness development. In another implementation, discrete early elastic modulus data points can also be obtained on standard cured specimens using specialized equipment such as a micro-penetrator.

[0098] In this embodiment, the propagation speed of ultrasound is selected. Concrete resistivity Cumulative heat release from hydration To construct hardened feature vectors.

[0099] Step S2 is performed by the feature preprocessing and vector construction module in the central processing and decision-making unit. Specifically, it involves the feature preprocessing and vector construction module for the ultrasonic wave propagation speed. Concrete resistivity Cumulative heat release from hydration These three types of time series data are denoised and smoothed respectively. Then, after missing data completion, the data of each sequence is mapped to the [0,1] interval.

[0100] From the normalized data of each category, extract the most representative key features (such as...). Figure 3 (as shown)

[0101] Extract the following from the ultrasonic propagation speed (i.e., ultrasonic velocity) curve: ① Velocity values ​​at times t1 and t2 , ②Instantaneous rate of change of velocity at time t1 ③ The integral area A under the curve from time t1 to the end time (e.g., 48 hours) u .

[0102] Here, time t1 is the moment when the curve first inflects (used to characterize condensation time), and time t2 is the moment corresponding to the peak value. If there is no peak value within the termination time, only the value at time t1 is extracted.

[0103] Extract the following from the concrete resistivity curve: ① Resistivity values ​​at times t1 and t2 , ② The instantaneous rate of change of resistivity at time t1 ③ Peak resistivity ④ The area under the curve A from time t1 to the end time (e.g., 48 hours) e .

[0104] Extract the cumulative heat release at the termination time from the cumulative heat release curve of hydration.

[0105] The eight key feature values ​​extracted above are arranged in a fixed order [ , A u , , , , A e , The components are concatenated to form a hardening feature vector H representing the hardening behavior of the i-th batch. i .

[0106] Step S3 is executed by the performance prediction module, the core of which is the application of a pre-trained performance prediction model. The key function of the performance prediction model is to establish a nonlinear mapping relationship between the hardening feature vector and the final key engineering performance indicators of concrete. The input to the performance prediction model is the hardening feature vector constructed in step S2, and the output is one or more long-term performance prediction values ​​for the i-th batch of concrete. These long-term performance prediction values ​​may include 28-day compressive strength, 56-day or 90-day compressive strength, chloride ion diffusion coefficient, freeze-thaw resistance grade, etc.

[0107] Performance prediction models can be implemented using various machine learning-based regression models, and those skilled in the art can choose the appropriate model based on data conditions and engineering requirements.

[0108] Deep learning neural networks, such as fully connected networks and wide-depth networks, are suitable for scenarios with large amounts of data and complex relationships between features and performance, and can automatically mine deep features.

[0109] Support Vector Machine Regression: It can provide robust predictive performance when the number of samples is limited or the monitoring data is noisy.

[0110] Gradient boosting tree regression: It has a good ability to capture nonlinear relationships and interactions in features and usually has good interpretability.

[0111] In a preferred embodiment, a fully connected neural network (BP neural network) is used to implement the performance prediction model. Specifically, a 4-8-8-1 four-layer BP neural network is designed. The input layer has four nodes, each selecting four key input features from the hardening feature vector: the ultrasonic wave propagation speed at 10 hours. The rate of change of ultrasonic propagation velocity over 10 hours resistivity at 10 hours Rate of change of resistivity over 10 hours .

[0112] The system uses two hidden layers, each containing 8 neurons, activated by the ReLU function. The output layer has one neuron, activated by a linear activation function, and outputs the predicted stress level over 28 days.

[0113] Five hundred sets of valid data from the historical production of the mixing plant were collected as training samples. Each set of valid data included the four key input features mentioned above and the corresponding measured compressive strength values ​​over 28 days. During training, mean squared error (MSE) was used as the loss function, and the Adam optimizer was used for parameter optimization. The learning rate was set to 0.001, and a total of 200 training epochs were conducted. Performance was monitored on the validation set to prevent overfitting.

[0114] After the model is trained, it is deployed in the performance prediction module. In actual operation, for the i-th batch, the system will construct the hardened feature vector H. i The data is input into a trained performance prediction model. The model performs forward propagation calculations and ultimately outputs the predicted 28-day compressive strength of this batch of concrete, which serves as a core component of the performance prediction. This prediction process can be completed within approximately 24 hours after pouring, enabling early and quantitative assessment of long-term performance.

[0115] Step S4 is executed by the cross-batch control module. The target performance value is determined based on the concrete design strength grade, including necessary margins. For example, for C30 concrete, the target 28-day compressive strength is set at 38 MPa.

[0116] The performance deviation across batches is equal to the difference between the predicted performance value and the target performance value. If the difference is greater than 0, it means that the predicted performance value is higher than the target performance value and needs to be adjusted downward; if the difference is less than 0, it means that the predicted performance value is lower than the target performance value and needs to be adjusted upward.

[0117] Step S5 is completed collaboratively by the cross-batch control module and the constraint processing logic.

[0118] For the i-th batch of concrete, the existing cross-batch regulation gain Used to quantify the cross-batch performance deviation of the i-th batch. Mix proportion correction amount for the (i+1)th batch of concrete The transformation relationship is maintained through the following dynamic process:

[0119] When the system is first run, the correction amount is based on the historical sequence of actual mix proportions from multiple batches. With the corresponding cross-batch performance deviation sequence The initial cross-batch control gain was determined through statistical regression analysis. Taking the least squares method as an example, that is, solving... , where K represents the variable in the optimization process; Represents the square norm.

[0120] In continuous production, based on cross-batch performance deviations With the correction amount of the mix ratio The existing cross-batch adjustment gain is adjusted through an adaptive update algorithm. Updates are performed to track changes in raw materials and the environment. In this embodiment, the specific update formula is:

[0121] (1)

[0122] in, This represents the updated cross-batch control gain, which is the existing cross-batch control gain for the next batch. The forgetting factor is used to control the weighting of historical information and new observation information. ; Represents the instantaneous gain estimate, based on the data pair of the i-th batch ( , The calculation yields the result, i.e., the solution. .

[0123] When i=1, the existing cross-batch control gain of the i-th batch. .

[0124] For the i-th batch, the mix proportion correction amount used for the (i+1)-th batch of concrete. Equal to the existing cross-batch control gain Performance deviation across batches Multiplication, the specific formula is:

[0125] (2)

[0126] For example, for the 12th batch (i=12), it was detected that , , resistivity change rate over 10 hours .

[0127] The predicted 28-day compressive strength for batch 12 is 52.5 MPa, and the target 28-day compressive strength is 55 MPa. Therefore, the inter-batch performance deviation (i.e., the 28-day compressive strength deviation) for batch 12 is -2.5 MPa. For the 28-day compressive strength deviation, the inter-batch control gains for the two corrections—driving the water-cement ratio and the admixture dosage—were fitted. Where k1 represents the control coefficient of the 28-day compressive strength deviation on the water-cement ratio of the next batch of concrete, and k2 represents the control coefficient of the 28-day compressive strength deviation on the admixture dosage of the next batch of concrete. In this example, k1 = −0.0032 (MPa) -1 k2 = 0.012 (% / MPa).

[0128] in accordance with Calculate using formula (2):

[0129] (3)

[0130] , (4)

[0131] in, This indicates the correction amount for the water-cement ratio. This indicates the adjustment amount for the admixture dosage. Specifically, it suggests that for the next batch of concrete, the water-cement ratio should be increased by 0.008, while the admixture dosage should be reduced by 0.030%.

[0132] To ensure the smoothness and safety of the optimization adjustment, dual constraints are imposed on the generated correction amount:

[0133] Continuity constraint: Correction amount for mix proportion The range of change is limited, that is ,in, To set limit values. For example, Set to 0.015, if the calculated water-to-binder ratio correction is... If the condition is met, then the continuity constraint is applied; if it is exceeded, the water-to-glue ratio correction amount is adjusted accordingly. Within the range.

[0134] Safety adjudication strategy: Conduct a safety boundary review of the mix proportion corrections after continuous constraint treatment. Preset upper limits for the water-cement ratio (e.g., 0.5) and lower limits for the total amount of cementitious materials. When the water-cement ratio correction exceeds its upper limit or the total amount of cementitious materials falls below its lower limit in the mix proportion corrections after continuous constraint treatment, execute a safety adjudication: freeze the correction item that directly adjusts the water content, and prioritize fine-tuning by adjusting the admixture dosage, cementitious material composition (such as the proportion of mineral admixtures), or sand ratio to achieve equivalent performance compensation. Equivalence can be determined by maintaining the target range of yield stress and plastic viscosity of the fresh concrete.

[0135] When performance is measured by multiple indicators, such as 28-day compressive strength, chloride ion diffusion coefficient, shrinkage index, creep index, etc., there may be batch-to-batch performance deviations. For multi-index deviation vectors, cross-batch adjustment gain For a matrix that matches its dimension, the generation and constraint principles of the correction are the same as above, achieving collaborative optimization of multiple objectives.

[0136] Step S6 completes the closed loop from decision-making to execution and evaluates the overall optimization effect of the system.

[0137] Command execution: The final mix proportion correction command (e.g., adjusted water-cement ratio, admixture dosage, etc.) after all constraint checks is issued to the batching plant production control system for the initial mix proportion setting and production of the i+1 batch of concrete.

[0138] Iterative optimization: For the (i+1)th batch, repeat steps S1 to S5, which involves monitoring, predicting, and calculating deviations in the early hardening stage and generating the mix proportion correction for the (i+2)th batch. This process is repeated continuously to achieve continuous optimization across batches.

[0139] The system continuously tracks batch-to-batch fluctuations in key performance indicators (such as 28-day compressive strength). Convergence criteria can be based on the batch-to-batch coefficient of variation (CV) or mean squared error (MSE). When the predicted or measured performance CV of m consecutive batches falls below a preset threshold (e.g., 4.5%), the system determines that closed-loop optimization has entered the steady-state control phase. In this phase, the system automatically reduces the learning intensity of the adaptive update algorithm (e.g., by decreasing the forgetting factor) to maintain high-quality production in a more stable state, signifying that performance at multiple batch scales has converged and stabilized.

[0140] Concrete of the same project and strength grade was selected, and the measured compressive strength at 28 days of age for 30 consecutive production batches before and after applying the cross-batch closed-loop optimization method of this invention was statistically analyzed. By calculating and comparing their coefficient of variation (CV), the degree of improvement in performance uniformity can be quantitatively assessed. Figure 4 As shown, after implementing closed-loop optimization, the batch-to-batch coefficient of variation (CV) of the 28-day compressive strength of concrete is significantly reduced compared to before optimization. For example, in a typical embodiment, the CV value can be reduced from about 8.5% before optimization to about 4.1% after optimization. This quantitative result directly proves that the method of the present invention can effectively suppress performance dispersion caused by factors such as raw material fluctuations and environmental changes, and greatly improve the quality consistency and stability of concrete production.

[0141] The effectiveness of this invention is not only reflected in the optimization of statistical indicators but also in the assurance of engineering safety. Throughout the optimization cycle, the system automatically constrains all adjustment commands through a built-in safety adjudication strategy, ensuring that there are no engineering risk events such as temporary on-site water addition, uncontrolled water-cement ratio, or insufficient cementitious materials caused by optimization algorithm suggestions. This proves that this invention can reliably safeguard the core durability and structural safety of concrete while actively and dynamically optimizing the mix proportion.

[0142] Quantitative comparison through the coefficient of variation (CV) (e.g.) Figure 4 The dual verification of the batch closed-loop optimization method (as shown) and the engineering safety record fully demonstrates that the method provided by this invention has significant effectiveness and practical engineering value in improving the uniformity of concrete quality and achieving forward-looking quality control.

[0143] Example 2

[0144] This invention also provides an electronic device, which includes a memory, a processor, and a computer program or instructions stored in the memory. The processor executes the computer program or instructions to implement the concrete mix proportion cross-batch closed-loop optimization method in Embodiment 1 of this invention.

[0145] Although not shown, the electronic device includes a processor that can perform various appropriate operations and processes based on programs and / or data stored in read-only memory (ROM) or loaded from a storage portion into random access memory (RAM). The processor can be a multi-core processor or may contain multiple processors. In some embodiments, the processor may include a general-purpose main processor and one or more specialized coprocessors, such as a central processing unit, graphics processing unit (GPU), neural network processor (NPU), digital signal processor (DSP), etc. Various programs and data required for device operation are also stored in RAM. The processor, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0146] The processor and memory described above are used together to execute programs / instructions stored in the memory. When the program / instructions are executed by the computer, they can implement the methods, steps, or functions described in the above embodiments.

[0147] Although not shown, embodiments of the present invention also provide a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implements the cross-batch closed-loop optimization method for concrete mix proportions in Embodiment 1 of the present invention.

[0148] Readable storage media include both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0149] The above description only discloses specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or modifications that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A closed-loop optimization method for concrete mix design across batches, characterized in that, The optimization method includes: Step S1: After the i-th batch of concrete is poured, collect characteristic data reflecting the concrete hardening process in its early hardening stage. Step S2: Based on the feature data, construct a hardened feature vector; Step S3: Input the hardening feature vector into the pre-trained performance prediction model to obtain the performance prediction value of the i-th batch of concrete; Step S4: Calculate the difference between the predicted performance value and the target performance value to obtain the cross-batch performance deviation of the i-th batch; Step S5: Calculate the cross-batch performance deviation of the i-th batch with the existing cross-batch control gain to generate the mix proportion correction amount for the (i+1)-th batch of concrete. Step S6: Repeat steps S1 to S5 to regulate the (i+1)th batch based on the predicted evaluation of the hardening process of the i-th batch, forming a closed-loop optimization at the multi-batch production scale to reduce the batch-to-batch dispersion of key concrete properties.

2. The method for cross-batch closed-loop optimization of concrete mix proportions according to claim 1, characterized in that, The early hardening stage is defined as the period from the initial setting of the concrete to 24 or 48 hours. Alternatively, the early hardening stage is defined as the time period during which the ultrasonic wave propagation velocity in the concrete first enters and remains within a preset characterization range.

3. The method for cross-batch closed-loop optimization of concrete mix proportions according to claim 1, characterized in that, The feature data includes at least two of the following categories: The propagation speed of ultrasound and its first derivative with time; Concrete resistivity and its first derivative with time; Cumulative heat release during hydration or heat release power; Condensation time; The elastic modulus measured during the early stage of hardening.

4. The method for cross-batch closed-loop optimization of concrete mix proportions according to claim 1 or 3, characterized in that, The construction of the hardened feature vector includes: The feature data of each category are subjected to denoising, smoothing, missing data completion and normalization processing; For each type of processed feature data, extract one or more of the following key features: peak value, inflection point, rate of change, and integral area; The hardened feature vector is constructed by concatenating the key features extracted from various feature data.

5. The method for cross-batch closed-loop optimization of concrete mix proportions according to claim 1, characterized in that, The existing cross-batch control gain is maintained through the following dynamic process: Based on the historical series of actual mix ratio corrections from multiple batches and the corresponding cross-batch performance deviation series, the initial cross-batch regulation gain is determined by least squares regression. In a continuous production process, a recursive relationship is established for the i-th batch, where i ≥ 1: Based on the cross-batch performance deviation of the i-th batch and the mix proportion correction amount for the (i+1)-th batch of concrete, the existing cross-batch control gain is updated using an adaptive update algorithm to obtain the updated cross-batch control gain. The updated cross-batch control gain will be used as the existing cross-batch control gain for the (i+1)th batch of concrete; when i=1, the existing cross-batch control gain is the initial cross-batch control gain.

6. The method for cross-batch closed-loop optimization of concrete mix proportions according to claim 5, characterized in that, The existing cross-batch control gain is updated using an adaptive update algorithm, specifically including: Based on the inter-batch performance deviation of batch i and the mix proportion correction for batch i+1 concrete, the instantaneous gain estimate is obtained by solving the following least squares problem: ； in, Represents the instantaneous gain estimate; This represents the mix proportion correction amount used for the (i+1)th batch of concrete; K represents the variables in the optimization process. This represents the cross-batch performance deviation of the i-th batch; Represents the square of the norm; The instantaneous gain estimate and the existing cross-batch control gain are fused according to the following formula to obtain the updated cross-batch control gain: ； in, This indicates the updated cross-batch adjustment gain; This indicates the existing cross-batch adjustment gain; This represents the forgetting factor, which is between 0 and 1 and is used to control the weight of historical information and new observation information.

7. The method for cross-batch closed-loop optimization of concrete mix proportions according to claim 1, characterized in that, In step S5, the specific process for generating the mix proportion correction amount for the (i+1)th batch of concrete includes: Multiply the cross-batch performance deviation of the i-th batch by the existing cross-batch adjustment gain to obtain the initial correction amount; A continuity constraint is applied to the initial correction amount to obtain the mix proportion correction amount for the (i+1)th batch of concrete; wherein, the continuity constraint is to limit the variation range of the initial correction amount to not exceed a preset limit value.

8. The method for cross-batch closed-loop optimization of concrete mix proportions according to claim 7, characterized in that, Step S5 further includes implementing a safety adjudication strategy: when the water-cement ratio calculated based on the mix proportion correction amount for the (i+1)th batch of concrete exceeds the corresponding preset upper limit, the total amount of cementitious materials is lower than the corresponding preset lower limit, or the predicted value of the durability index does not meet the corresponding preset threshold, the correction item for directly adjusting the water content is frozen, and performance compensation is achieved preferentially by adjusting the admixture dosage, cementitious material composition, or sand ratio.

9. An electronic device comprising a memory, a processor, and a computer program or instructions stored in the memory, characterized in that, The processor executes the computer program or instructions to implement the cross-batch closed-loop optimization method for concrete mix proportions as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by the processor, they implement the batch-wise closed-loop optimization method for concrete mix proportions as described in any one of claims 1 to 8.

Images