A ready-mixed concrete intelligent production and quality monitoring system and method

Abstract

The application discloses a kind of pre-mixed concrete intelligent production and quality monitoring system and method, specifically related to concrete production quality monitoring technical field, including the following steps: constructing initial control parameter set;Collect environmental data and act on initial control parameter set to generate applicable control parameter set;Based on applicable control parameter set, perform mixing production and collect process behavior data, input to machine learning model to predict compressive strength and setting time;Compare the prediction result with performance target, if it exceeds tolerance, it is determined as low quality and correct control parameter, if it does not exceed, it is high quality and allows discharge;The application uses historical weighing deviation and strength measured results to construct initial control parameter set, and dynamically generates applicable control parameter set in combination with real-time environmental data, compares the preset target by model prediction performance result, forms quality judgment and parameter correction closed loop, improves the rationality of concrete production parameters, environmental adaptability and quality control intelligent level.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of concrete production quality monitoring, and more particularly to an intelligent production and quality monitoring system and method for ready-mixed concrete. BACKGROUND

[0002] Ready-mixed concrete is one of the most widely used structural materials in modern civil engineering and construction, and its quality stability and construction adaptability are directly related to structural safety, project progress and resource utilization efficiency. In actual production process, concrete performance indicators such as compressive strength and setting time are usually affected by multiple variables coupling, including raw material component accuracy, environmental temperature and humidity conditions, mixing process rhythm and additive reaction characteristics. However, the existing concrete production mode still has significant technical bottlenecks in control mode and quality assurance mechanism.

[0003] On the one hand, the traditional production process highly depends on manual experience setting, and typical parameters such as feeding time, mixing time and additive adding method are mostly initially set by human experience, lacking the response ability to complex multi-variable dynamic changes, and cannot effectively adapt to the composition fluctuations between raw material batches or random disturbances of environmental conditions. On the other hand, although some enterprises have deployed digital acquisition equipment to obtain raw material weighing data and environmental parameters, these acquisition behaviors and actual control strategies lack deep coupling, and the system cannot adaptively adjust the parameter set based on the acquisition data, nor has performance prediction and feedback correction mechanism. Especially in actual production, the interaction between multiple control variables is strongly nonlinear, and traditional linear models or rule engines are difficult to achieve effective modeling, resulting in that even if data acquisition is implemented, a closed-loop control path cannot be formed. In addition, the quality inspection process commonly used in the current industry still mainly relies on post-strength detection, that is, the strength result of the finished concrete is detected after standard curing to determine whether it meets the standard. This way of post-feedback not only has strong lag, but also cannot prevent unqualified batches from flowing out. Therefore, an intelligent production and quality monitoring system and method for ready-mixed concrete are proposed to solve the above problems. SUMMARY

[0004] To achieve the above purpose, the present application provides the following technical scheme:

[0005] An intelligent production and quality monitoring method for ready-mixed concrete, comprising the following steps:

[0006] Acquiring actual weighing data of multiple batches of raw materials, calculating the difference between each raw material and the target ratio, and combining the measured strength results of the corresponding batches of concrete to construct an initial control parameter set for the current production batch;

[0007] Obtain temperature and humidity data of the current production environment, and apply the obtained environmental data as influencing factors to the initial control parameter set to construct a suitable control parameter set for the current batch;

[0008] Perform concrete mixing production operation according to the suitable control parameter set for the current batch, collect preset process behavior data during production, extract features from all collected process behavior data to obtain a process behavior feature group, and then input the process behavior feature group into a preset machine learning model to output the compressive strength and setting time of the batch of concrete under standard curing conditions;

[0009] Compare the output result of the machine learning model with the preset performance target, if it exceeds the preset standard range, determine that the batch is a low-quality production batch, and determine the control parameter combination that needs to be corrected in the suitable control parameter set based on the output result of the machine learning model, and regenerate an updated suitable control parameter set according to the preset correction rule; if it does not exceed the preset standard range, it is determined to be a high-quality production batch, and the batch of concrete is allowed to be discharged.

[0010] In a preferred embodiment, the construction process of the initial control parameter set includes introducing an error weight function, which takes the proportion of each raw material in the proportioning as the weight basis, and weights and accumulates the deviation value between the actual weighing data and the theoretical target proportioning of each batch to obtain a deviation influence value representing the fluctuation law between batches;

[0011] The deviation influence value is combined with the strength measurement result of the corresponding batch of concrete to determine the correlation degree of the raw material type and the strength response through a mapping relationship, and the raw material type with a correlation degree exceeding a preset correlation threshold is included in the initial control parameter set.

[0012] In a preferred embodiment, constructing a suitable control parameter set for the current batch includes the following steps:

[0013] Perform first layer clustering processing on the initial control parameter set, and use the deviation value between the actual weighing data and the target proportioning of the raw material as the input variable to form a plurality of deviation behavior groups;

[0014] Construct a joint environmental feature vector from the obtained temperature and humidity data as an input variable for second layer clustering processing to obtain a plurality of environmental response groups;

[0015] Cross-combine the deviation behavior groups and the environmental response groups corresponding to the batch numbers to obtain their corresponding control parameter adjustment rules under the clustering label index, and generate a suitable control parameter set for the current batch based on the control parameter adjustment rules.

[0016] In a preferred embodiment, the first layer clustering processing and the second layer clustering processing both use K-means clustering or Gaussian mixture model.

[0017] In a preferred embodiment, the applicable set of control parameters for the current batch contains the following four executable control parameters: feeding time window, feeding sequence priority, stirring rhythm variation rate, and liquid material premixing ratio.

[0018] In a preferred embodiment, the extraction of the process behavior feature set employs a time series-based behavior analysis strategy, and the process behavior feature set includes stirring load variation rate, slurry temperature variation rate, humidity response delay time, coefficient of variation of viscosity estimation value, stirring current fluctuation amplitude, and vibration frequency stabilization time period.

[0019] Among them, the stirring load variation rate is the slope value of the load rising trend in the preset time interval after stirring starts, the slurry temperature variation rate is the average speed of temperature rise in the time interval, the humidity response delay time is the time delay corresponding to the first time the humidity change amplitude exceeds the set threshold, the coefficient of variation of viscosity estimation value is the ratio of the standard deviation to the mean of load fluctuation in the stirring stable stage, the stirring current fluctuation amplitude is the difference between the maximum and minimum values of the stirring current in the specified time period, and the vibration frequency stabilization time period is the time length experienced from the start of vibration to the time when the frequency curve change rate is lower than the set threshold. Each item in the process behavior feature set is input as a time series vector to the preset machine learning model.

[0020] In a preferred embodiment, the machine learning model adopts a structure based on regression output, and is constructed using any one of a gradient boosting regression tree model, a support vector regression model, or a deep neural network structure. The compressive strength and setting time of the current batch of concrete under standard curing conditions output by the machine learning model are dual-channel results, reflecting the comprehensive performance of the concrete in terms of structural stability and construction continuity.

[0021] In a preferred embodiment, in the process of comparing the machine learning model output results with the preset performance target, a double-threshold judgment mechanism is used for quality classification judgment. The judgment mechanism sets a first performance tolerance range and a second performance warning range. When the compressive strength and setting time in the output results are both within the first performance tolerance range, the current batch is judged to be a high-quality production batch. When the compressive strength and setting time in the output results both exceed the second performance warning range, the current batch is judged to be a low-quality production batch.

[0022] In a preferred embodiment, the combination of control parameters in the applicable set of control parameters that need to be corrected based on the machine learning model output results refers to:

[0023] When the output result of the machine learning model is compared with the preset performance target, the output deviation of the output result is decomposed into two components of the compressive strength deviation value and the setting time deviation value, the compressive strength deviation value is obtained by calculating the difference between the compressive strength in the output result and the median of the corresponding compressive standard range in the first performance tolerance range, and the setting time deviation value is obtained by calculating the difference between the setting time in the output result and the median of the corresponding setting standard range in the first performance tolerance range;

[0024] Based on the Pearson correlation coefficient analysis, the correlation coefficient G1 of each control parameter in the applicable control parameter set with the compressive strength and the correlation coefficient G2 with the setting time are obtained, then the comprehensive correlation strength S = |G1* KY| + |G2* NJ| is calculated, KY represents the compressive strength deviation value, and NJ represents the setting time deviation value, then the comprehensive correlation strength S is arranged in descending order, the top m control parameters are taken, and the control parameter combination needing to be corrected in the applicable control parameter set is obtained.

[0025] In a preferred embodiment, a ready-mixed concrete intelligent production and quality monitoring system specifically comprises:

[0026] The raw material modeling module is configured to collect actual weighing data of a plurality of batches of raw materials, calculate the difference between each raw material and the target ratio, and combine the measured results of the concrete strength of the corresponding batches to construct an initial control parameter set for the current production batch.

[0027] The environment adaptation module is configured to obtain temperature and humidity data of the current production environment, and apply the obtained environmental data as an influencing factor to the initial control parameter set to construct an applicable control parameter set for the current batch.

[0028] The process prediction module is configured to perform concrete mixing production operations according to the applicable control parameter set of the current batch, collect a plurality of preset process behavior data during the production process, extract features from all the collected process behavior data to obtain a process behavior feature group, and then input the process behavior feature group into a preset machine learning model to output the compressive strength and setting time of the batch of concrete under standard curing conditions.

[0029] The quality control module is configured to compare the output result of the machine learning model with the preset performance target, if the output result exceeds the preset standard range, the batch is determined as a low-quality production batch, the control parameter combination needing to be corrected in the applicable control parameter set is determined based on the output result of the machine learning model, and the updated applicable control parameter set is regenerated according to the preset correction rule; if the output result does not exceed the preset standard range, the batch is determined as a high-quality production batch, and the batch of concrete is allowed to be discharged.

[0030] The technical effects and advantages of the present application are as follows:

[0031] The application can effectively utilize the real data in the historical production process to initialize the production control parameters in a data-driven manner by collecting the actual weighing data of multiple batches of raw materials, calculating the difference between each raw material and the target ratio, and combining the measured strength of the corresponding batch of concrete to construct the initial control parameter set of the current production batch. Compared with the traditional method of relying on experience setting or single theoretical ratio, this method introduces the correlation between actual deviation and performance response, ensuring the basic accuracy of the current batch in the ratio setting, which helps to improve the starting quality and parameter rationality of the overall mixing process.

[0032] The application constructs the applicable control parameter set of the current batch by obtaining the temperature and humidity data of the current production environment and using the environmental data as an influencing factor acting on the initial control parameter set. This method fully considers the actual influence of environmental variables on the performance fluctuation of concrete without changing the original ratio structure, so that the control parameter set has the response ability to real-time external environmental changes. By dynamically adapting the control setting under different temperature and humidity conditions, the consistency and quality controllability of concrete production under varying conditions are improved.

[0033] The application collects process behavior data during production, forms a process behavior feature group through feature extraction, and inputs it into a preset machine learning model to output the compressive strength and setting time of the batch of concrete. Then, the model output result is compared with the preset performance target, and the current batch quality state is identified according to the comparison result. If it is determined as a low-quality production batch, the system determines the control parameter combination that needs to be corrected based on the model result, and generates an updated control parameter set according to the preset correction rule; if it is determined as a high-quality production batch, it is directly released. This feedback mechanism effectively constructs a closed-loop logic from prediction, comparison, judgment to correction, improving the intelligent level of concrete quality monitoring and the timeliness of responding to abnormalities. BRIEF DESCRIPTION OF DRAWINGS

[0034] In order to facilitate the understanding of those skilled in the art, the application will be further described below with reference to the accompanying drawings;

[0035] Figure 1 A schematic diagram of the pre-mixed concrete intelligent production and quality monitoring method in the application.

[0036] Figure 2 A schematic diagram of the pre-mixed concrete intelligent production and quality monitoring system in the application. DETAILED DESCRIPTION

[0037] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort are within the scope of the present application.

[0038] With reference to Figure 1 Figure 2 The following examples are obtained:

[0039] Embodiment 1: A pre-mixed concrete intelligent production and quality monitoring method, comprising the following steps:

[0040] Actual weighing data of a plurality of batches of raw materials is collected, the difference between each raw material and the target ratio is calculated, and the initial control parameter set for the current production batch is constructed in combination with the measured results of the concrete strength of the corresponding batch; the core of this step is to use the deviation between the actual feeding data of the raw materials in the historical batches and the theoretical design ratio as the input basis to extract the fluctuation characteristics of each raw material in the actual production process. By collecting the deviation data of multiple batches, the unstable factors of different raw materials in the actual control can be captured. In combination with the measured results of the strength of these batches, that is, the measured compressive strength of each batch of concrete under standard curing conditions, the mapping relationship between the raw material deviation and the performance output is established. Based on this, the initial control parameter set of the current production batch is constructed, so that the parameter set not only reflects the design target of the current ratio, but also reflects the historical response of the influence of previous production behavior on the actual strength, providing the initial value of the parameters for the subsequent control strategy.

[0041] Temperature and humidity data of the current production environment are obtained, and the obtained environmental data are used as influencing factors acting on the initial control parameter set to construct the applicable control parameter set of the current batch; this step aims to introduce the key external factors of the actual production environment, that is, temperature and humidity, which affect the performance fluctuation of concrete, as dynamic correction parameters to participate in the control parameter calculation. Since the hydration reaction and initial and final setting behaviors of concrete production are highly dependent on environmental conditions, the current environmental temperature and humidity are added as external variables to the parameter derivation process, which helps to improve the adaptability of the parameter set to the current working conditions. By using the environmental data as influencing factors acting on the initial control parameter set, the transition from static empirical parameters to dynamic adjustable parameters is realized, and the applicable control parameter set of the current batch that is truly applicable to the current site conditions is constructed, providing precise control logic matched with the working conditions for subsequent actual mixing operations.

[0042] ​The concrete mixing production operation is performed according to the applicable control parameter set of the current batch, a plurality of process behavior data are collected in the production process, all the collected process behavior data are subjected to feature extraction to obtain a process behavior feature group, and then input to a preset machine learning model to output the compressive strength and setting time of the batch of concrete under standard curing conditions; this step reflects the connection between the execution layer and the intelligent analysis layer of the method. According to the control variables defined by the applicable control parameter set, such as feeding rhythm, mixing time, etc., the concrete mixing process of the current batch is performed. In this process, dynamic process behavior data in multiple dimensions including mixing load, current fluctuation, material temperature change, etc. are obtained through sensor devices and collection mechanisms. Then, through behavior feature extraction algorithms, these time series data are processed into structured process behavior feature groups, and the core features reflecting the change law of the physical state of concrete are extracted. These feature groups are sent as input into the machine learning model, and the model predicts the compressive strength and setting time that the current batch of concrete can reach under standard curing conditions according to the existing training mechanism, forming the basis for predicting the performance output.

[0043] The machine learning model output result is compared with the preset performance target, if it exceeds the preset standard range, it is determined that the batch is a low-quality production batch, and the control parameter combination that needs to be corrected in the applicable control parameter set is determined based on the machine learning model output result, and the updated applicable control parameter set is regenerated according to the preset correction rule; if it does not exceed the preset standard range, it is determined to be a high-quality production batch, and the batch of concrete is allowed to be discharged. This step is used to establish the automatic judgment logic between the prediction result and the engineering quality standard. By comparing the output values (compressive strength and setting time) of the machine learning model with the performance target range preset in the system, it can be quickly determined whether the predicted performance of the batch of concrete meets the engineering requirements. If all the predicted results are within the standard range, it is determined to be a high-quality production batch and is allowed to be discharged; otherwise, it is determined to be a low-quality production batch. For low-quality batches, the performance deviation result output by the model is further used to trace back and deduce the control parameter items in the current applicable control parameter set that may cause performance deviation. According to the historical correlation between the parameter items and the performance deviation, the control parameter combination that needs to be corrected is identified, and the correction rule is called to update it, and the applicable control parameter set for correcting the deviation is regenerated to support the re-production of the batch until the model prediction meets the discharge conditions.

[0044] In the method for intelligent production and quality monitoring of ready-mixed concrete, the construction of the initial control parameter set is one of the core steps, which is used to provide targeted control variable configuration for subsequent environment adaptation, process execution and prediction model. The construction process includes the introduction of an error weight function to realize the structured quantification of the actual deviation values of multiple batches of raw materials, and the completion of preliminary attribution analysis combined with performance results. The error weight function is a numerical function, whose input is the difference between the actual weight value of each raw material in each batch and its theoretical target proportion, which is called "proportion deviation value" and usually expressed in kilograms or percentage. The function takes the proportion of each raw material in the target proportion as the basic weight, i.e. the proportion value of each material in the total mass or total volume of concrete.

[0045] For example, in a standard proportion, cement accounts for 15% of the total mass, sand accounts for 30%, gravel accounts for 40%, water accounts for 10%, and additive accounts for 5%. These proportions are respectively taken as the weight coefficients of the corresponding materials. Multiply the weight value by its actual deviation value, and then sum the weighted deviation values of all raw materials within the same batch to obtain a numerical value representing the overall fluctuation level of the batch, called "deviation influence value". The deviation influence value is used to quantify the overall severity of the proportion deviation of the batch of raw materials. In order to use this data structure for control parameter item screening, the system performs paired analysis of the deviation influence values of multiple batches and their corresponding concrete strength measured results. The concrete strength measured result refers to the compressive strength value measured by pressure test after each batch of concrete is cured for 28 days under standard curing conditions (e.g. temperature 20±2℃, humidity above 95%).

[0046] After combining the deviation values of each raw material in all batches with the corresponding compressive strength measured values of the batches, a multivariate sample set is formed. On this basis, the system performs "mapping relationship analysis", i.e. uses correlation analysis algorithm (such as Pearson correlation coefficient analysis) to calculate the statistical correlation between the deviation value sequence of each raw material and the compressive strength sequence. The result is a real number between -1 and 1, and the closer the value is to 1 or -1, the greater the linear influence of the deviation of the raw material on the strength. Further, the system compares the correlation value of each raw material with a set of preset correlation thresholds, which can be set based on historical engineering experience or model training results, such as the default setting of zero point five. If the correlation degree (i.e. the absolute correlation coefficient between the deviation and the compressive strength) of a raw material is higher than the threshold, it means that its proportion deviation has a significant impact on the strength of the concrete, and the system will include the type of the raw material in the initial control parameter set.

[0047] For example, if the analysis result shows that the cement deviation has a Pearson correlation coefficient of 0.85 with the measured strength value, the sand has a Pearson correlation coefficient of 0.63, and the additive has a Pearson correlation coefficient of 0.35, the system will filter the first two (cement and sand) into the initial control parameter set, and the additive will not be included temporarily because it does not meet the threshold value. The initial control parameter set is the earliest constructed control variable set in this method, and the types of raw materials and their deviation characteristics included will be used in subsequent steps with environmental factors to generate an applicable control parameter set, and further participate in the whole process links such as control strategy implementation, machine learning prediction and quality judgment. Therefore, its construction method directly affects the response ability and accuracy of the subsequent intelligent adjustment mechanism.

[0048] In the ready-mixed concrete intelligent production and quality monitoring method of the application, constructing the applicable control parameter set of the current batch is a key process in the control logic, and its main goal is to realize dynamic adaptive adjustment of production parameters according to the raw material deviation characteristics and environmental response characteristics under the current production conditions. This process includes the following three specific steps:

[0049] The first step is to perform first-layer clustering processing on the initial control parameter set. The initial control parameter set refers to the key raw material parameter set selected by the difference between the actual weighing data and the target ratio of the raw materials in multiple batches, combined with the measured strength of the concrete. Its content has been disclosed in the foregoing. In this step, the ratio deviation value corresponding to each control parameter item in the initial control parameter set is taken as the input variable and input into the clustering algorithm to automatically divide the raw material deviation behavior of multiple batches. The clustering input format is a two-dimensional numerical matrix, where each row represents a historical batch and each column represents the deviation value corresponding to a raw material control parameter item. The matrix is input into the first-layer clustering algorithm for processing. The clustering method used is K-means clustering or Gaussian mixture model. K-means clustering is to divide the data into a predetermined number of clusters so that the sum of the squared distances between samples within each cluster is minimized. Gaussian mixture model is a probabilistic clustering method that assumes that samples are composed of multiple Gaussian distributions, and the distribution parameters of the clusters are iteratively solved by the expectation maximization algorithm. After clustering processing, all batches will be divided into several "deviation behavior groups", i.e. batch sets with similar ratio deviation characteristics. The batches in each group share similar raw material ratio control error patterns, which serve as the basis for subsequent parameter correction.

[0050] Second step: build joint environmental feature vector for the acquired temperature and humidity data, and perform second layer clustering processing. In the actual production process of concrete, environmental factors (especially temperature and humidity) have a significant impact on the hydration reaction, setting time and final strength. Therefore, in order to improve the adaptability of the control parameters, the present application introduces the environmental features formed by the temperature data and humidity data of the current batch as the second layer input. The environmental feature vector of each batch is a binary vector composed of the measured environmental temperature value (unit: ℃) and the environmental relative humidity value (unit: %). For example, the environment of a certain batch is 26℃ and 85% relative humidity, and its environmental feature vector is (26, 85). The environmental feature vectors of multiple batches form an environmental feature matrix, which is input into the second layer clustering algorithm. This algorithm can also use K-means clustering or Gaussian mixture model to form several "environmental response groups". Each group reflects a set of batches that produce similar strength fluctuations under specific environmental conditions, providing environmental reference for control parameter adaptation.

[0051] Third step: cross combination of deviation behavior groups and environmental response groups to determine the control parameter adjustment rule. In this step, the clustering results of the previous two layers are cross-mapped through batch numbers to realize composite label generation. That is, for each batch, the system records its belonging category in the deviation behavior group and the environmental response group, forms a joint label index by combining the category labels, for example, (P3, E1) represents that the batch belongs to the 3rd deviation behavior group and the 1st environmental response group. The system queries the corresponding adjustment rule in the preset "control parameter adjustment rule table" according to the joint label. The rule table is generated by experience rules and historical data training results, which specifies which control parameters need to be increased, decreased or kept unchanged under the combination of specific deviation behavior mode and environmental response mode. For example, the rule table may specify that under the condition of (P3, E1), the mixing rhythm change rate should be increased by 5% and the liquid material premixing ratio should be reduced by 10%. Finally, according to the joint rule set, the specific values of the original control parameters are adjusted to obtain the applicable control parameter set of the current batch. The parameter set will be input as a production instruction to guide the actual concrete mixing process.

[0052] K-means clustering is suitable for small data dimension and cluster distribution, with fast convergence speed and strong controllability; Gaussian mixture model is suitable for scenarios where data exists partially overlapping and cluster boundary is not obvious, and can provide classification explanation in the sense of probability. In the actual system, any of the above clustering methods can be selected flexibly according to the data distribution type, batch size and system response time limit, and the number of clusters K or the number of mixed components can be set through cross-validation method.

[0053] K-means clustering realizes the clustering process in the following way:

[0054] Input Form: The input is a two-dimensional numerical matrix, where each row represents a sample (e.g., a concrete production batch), and each column represents a feature variable (e.g., cement deviation, sand deviation, etc.);

[0055] Initial Stage: Randomly select K sample points as initial cluster centers, i.e., the representative centers of each category;

[0056] Sample Assignment Calculation: For each sample, calculate its Euclidean distance from the K cluster centers, and assign the sample to the cluster with the smallest distance;

[0057] Cluster Center Update: For all samples within each cluster, recalculate their average position (mean vector) as the new cluster center of the cluster;

[0058] Repeat Iteration: Repeat the sample assignment and cluster center update until all cluster center positions no longer change or reach the set maximum number of iterations;

[0059] Output Result: Each sample is finally assigned a cluster label, identifying its belonging category, representing the classification result of behavior differences between different labels.

[0060] The implementation process of Gaussian Mixture Model includes the following stages:

[0061] Input Form: The input is also a two-dimensional numerical matrix, where each row represents a sample, and each column is a feature item of the sample;

[0062] Modeling Assumption: Assume there are K different Gaussian distributions, each with its own mean vector, covariance matrix, and prior probability weight, and the overall model is composed of these K distributions weighted together;

[0063] Model Initialization: Initially set the parameters of each Gaussian distribution, usually provided with initial values from random initialization or K-means results;

[0064] Expectation Maximization Algorithm (EM Algorithm): E-step (Expectation Step): According to the current parameter estimation, calculate the probability of each sample belonging to each Gaussian distribution; M-step (Maximization Step): According to the probability calculated in the E-step, re-estimate the parameters of each Gaussian distribution (i.e., maximum likelihood estimation), update the mean, covariance, and weight; repeat the E-step and M-step until the model parameters converge or meet the termination condition;

[0065] Output Result: Each sample is assigned a label of the most likely Gaussian distribution it belongs to, while retaining its probability value of belonging to each distribution, achieving a soft classification result.

[0066] The applicable control parameter set is a dynamic parameter set constructed according to the linkage relationship between the initial control parameter set and the current batch production environment, and its main function is to guide the whole process control strategy execution of the current batch concrete production process. The generation mode of the parameter set is based on the cross mapping result of the first layer clustering and the second layer clustering processing, and is generated through the control parameter adjustment rule system, which is disclosed in the foregoing. In terms of parameter structure, the present application particularly points out that the applicable control parameter set under the current batch contains the following four executable control parameters, and supports the expansion of more control variables according to the actual scene, which are described as follows:

[0067] The feeding time window refers to the allowable time interval boundary of feeding various raw materials into the mixing tank, which is expressed in seconds or minutes. This parameter is used to limit the start and end time period of feeding different materials to avoid the occurrence of adverse process behaviors such as material accumulation, bonding and early hydration. For example, the feeding window of powder materials (such as cement) and water should be set in the first half, while the feeding time window of admixtures may be set in the later stage of mixing (such as 30 to 60 seconds after the start of mixing) to achieve the best dispersion effect. This parameter is determined comprehensively according to the material category, temperature and humidity environment response group label, historical batch effect, etc., and its adjustment will directly affect the uniformity of early reaction and the stability of strength development.

[0068] The feeding sequence priority refers to the order in which each type of material should be fed among all the raw materials participating in feeding, which is represented by a priority value (such as an integer from 1 to N). The higher the priority, the earlier the material enters the mixing process. For example, cement priority = 1; fly ash priority = 2; water priority = 3; admixture priority = 4. This parameter is represented as an ordered list in structure, and its setting depends on the target proportion, environmental temperature and humidity state and the response effect of priority change on performance in the previous batch. In actual production, adjusting the priority can be used to avoid local agglomeration, uneven condensation, early layering and other phenomena, thereby stabilizing the mixing quality.

[0069] The mixing rhythm change rate refers to the adjustment rate of the motor speed or mixing load during the whole concrete mixing process, reflecting the transition speed of the mixing process from rough mixing to fine mixing stage. This parameter is expressed in percentage change rate, for example, the mixing speed is 100% in the first 20 seconds, and then decreases to 80% in the next 10 seconds, so the change rate is 1% per second. Setting the rhythm change rate too fast can lead to insufficient concrete agglomeration, and setting it too slow can cause excessive energy consumption and low efficiency, so in different deviation behavior groups and environmental response groups, the system will give the adaptive setting value of this parameter to ensure the homogeneity of the final slurry structure and the dispersion efficiency.

[0070] Liquid material premixing ratio: The liquid material premixing ratio refers to the proportion setting of mixing the liquid materials (such as water, water reducing agent, retarder, etc.) in advance before all solid-liquid materials enter the main mixing stage. This parameter is defined as the mass or volume proportion of the total amount of liquid materials participating in the premixing stage. For example, if the liquid material premixing ratio is set to 60%, it means that 60% of the liquid has been uniformly mixed in the premixing tank before the main feeding. Adjusting this parameter can realize the intervention of the activity release degree of the admixture and the early hydration kinetics, thereby optimizing the paste cohesiveness, delaying the agglomeration, and improving the mixing efficiency, especially in the case of severe temperature and humidity fluctuations.

[0071] The above four parameters constitute the core content of the applicable control parameter set for the current batch, and are all parameters that can be actually executed and directly applied by the digital control system. This parameter set can be called and dynamically updated in real time according to the automatic control program of the production platform. Although the present application explicitly lists the above four control parameters as the core implementation content, it does not exclude the expansion of more control parameter items according to the production scene, such as: total mixing time, material feeding interval, paste temperature control target, etc., to form a more extensive applicable control parameter set, which still belongs to the technical protection scope of the present application.

[0072] By monitoring the time series of the changes of key physical quantities in the whole process of concrete production, a number of behavior characteristic variables with representativeness can be extracted, thereby providing reliable input basis for subsequent performance prediction. The present application uses a behavior analysis strategy based on time series to process the collected process behavior data. This strategy refers to setting a window interval according to the time variation law of the physical quantity during the concrete mixing process, calculating the change characteristics of the specific indicators over time, and outputting a set of structured behavior characteristic values, called process behavior characteristic group. This characteristic group includes the following six specific variables, which are described as follows:

[0073] Mixing load change rate refers to the trend of the load value of the mixing equipment changing with time within a predetermined time interval after the start of mixing. The specific calculation method is as follows: select the first few seconds (such as the first 30 seconds) after the start of mixing as the analysis interval; record the real-time load (unit: Newton-meter or current equivalent value) of the mixing equipment in this interval; take time as the horizontal axis and load as the vertical axis, perform linear fitting, and extract the slope value, which is the load change rate. This indicator reflects the resistance trend of the material to the mixing equipment in the initial mixing process, which is related to the material cohesiveness, fluidity, and particle size distribution.

[0074] The slurry temperature change rate refers to the average temperature rise speed of the slurry in the same preset time interval. The parameter is measured in units of degrees Celsius per second. Set the same time window as the stirring load; record the slurry temperature sensor readings during this period; calculate the difference between the end temperature and the initial temperature, divided by the time length, to obtain the average temperature rise rate. This parameter reflects the degree of early hydration reaction, and is related to the raw material temperature, water-cement ratio and the reaction rate of admixtures.

[0075] The humidity response delay time refers to the time taken from the start of stirring to the first time the humidity sensor detects a value that exceeds the set change threshold. The humidity change threshold is set (e.g. 10% relative humidity); when the monitoring value first meets the condition "current value - initial value ≥ threshold", the current time is recorded; the delay time is obtained by subtracting the stirring start time from the current time. This index reflects the water release or moisture absorption process in the slurry, and is related to the water absorption of the powder, the environmental humidity response and the diffusion rate of the liquid mixing material.

[0076] The coefficient of variation of the estimated value of the consistency is an index to measure the stability of the viscosity in the later stage of the stirring process. Usually, the fluctuation amplitude of the stirring load signal in the "stable stage" is used as a substitute variable. A fixed length of the stable window (e.g. 60 seconds) is set in the middle or end of the main stirring stage; record the stirring load value in this interval; calculate the ratio of the standard deviation to the mean as the coefficient of variation. The larger the coefficient of variation, the more uneven the consistency of the slurry, and the physical reactions in the system are still not sufficient; the smaller the coefficient of variation, the more stable the stirring state and the better the mixing uniformity.

[0077] The stirring current fluctuation amplitude is used to measure the fluctuation degree of the stirring motor load in a specified time. The calculation method is: set an analysis time period (e.g. 30-60 seconds after the start of stirring); record the current change curve in this time period; find the difference between the maximum and minimum values as the current fluctuation amplitude. This index is closely related to the material mixing uniformity, the agglomeration and loose behavior, and the distribution of admixtures.

[0078] The vibration frequency stabilization time period refers to the length of time experienced from the start of the vibration device to the time when the vibration frequency change rate first stabilizes below a certain threshold value, in seconds. A frequency change rate threshold value (e.g., a change rate less than 0.5 Hz / s) is set; the derivative of the vibration frequency is calculated in real time; when the change rate is stable for N consecutive seconds (e.g., 5 seconds) below the threshold value, it is considered that the frequency has entered a stable state; the required time period is obtained by subtracting the vibration start time from the stable start time, which reflects the dynamic response time required for the slurry to tend to structural stability under the action of vibration disturbance, and indirectly reveals its configuration consistency and anti-separation ability. The six process behavior characteristic parameters form a complete process behavior characteristic group, and each data is collected and extracted in time series. The results are input into the preset machine learning model used in the present application as performance prediction input variables in the form of a multi-dimensional vector in a unified format. Through the introduction of the characteristic group, the system can fully depict the dynamic characteristics of concrete during the mixing process, thereby significantly improving the accuracy and stability of the prediction model for the final compressive strength and setting time.

[0079] The machine learning model is constructed based on a regression output structure, and is constructed using any one of a gradient boosting regression tree model, a support vector regression model, or a deep neural network structure. The current batch of concrete under standard curing conditions output by the machine learning model is a double-channel result, reflecting the comprehensive performance of the concrete in terms of structural stability and construction continuity. The "regression output structure" refers to the model output result being a continuous numerical value rather than a classification label, which specifically reflects the numerical change trend of the prediction object in a certain physical quantity dimension, and is used to measure the quantitative characteristics of the target variable.

[0080] Gradient boosting regression tree is an ensemble learning algorithm that constructs a nonlinear prediction model based on the step-by-step iteration combination of multiple regression decision trees. The basic principle is that a weak predictor (such as a shallow regression tree) is initially used to fit the target variable residual error, and each subsequent training round targets the fitting of the previous round error, continuously iterating to optimize prediction accuracy. The final model is the weighted cumulative result of all tree outputs, with strong generalization ability, suitable for nonlinear mapping modeling between complex behavior characteristics and compressive strength.

[0081] Support vector regression is based on the support vector machine theory, which is characterized by mapping the original features to a high-dimensional space through the introduction of a kernel function, and constructing an optimal regression hyperplane in this space with a fitting error not exceeding ε. This method is suitable for small samples, high-dimensional features, or clear boundary scenarios, and can provide strong robustness for accurate fitting of compressive strength and setting time, especially when there is strong noise interference or uneven feature data distribution.

[0082] The deep neural network realizes end-to-end mapping of high-dimensional input data through a multi-layer nonlinear neuron structure, and is suitable for a situation where nonlinear correlation is strong and there is a complex cross influence between variables. In the present application, the model can include an input layer, a plurality of hidden layers (such as a full connection layer, an activation function layer, a normalization layer, etc.) and an output layer, and the training process adopts a back propagation algorithm to optimize the model parameters, which is suitable for concrete performance learning under large-scale data driving. The input of the machine learning model is a process behavior feature group, which includes six dynamic time series variables, comprehensively reflecting the physical behavior changes in the concrete mixing process. After all the variables are standardized, they are combined into an input vector in a unified format for the model to learn the internal relationship between them and the final performance of the concrete.

[0083] The output result of the machine learning model is a double-channel result, that is, the compressive strength prediction value: indicating the compressive strength value that the current batch of concrete can reach under standard curing conditions, usually defined as temperature 20±2℃, relative humidity not less than 95%, and curing for 28 days, with the unit of megapascal (MPa). The setting time prediction value: indicating the time required for the batch of concrete to reach initial setting and final setting from the completion of mixing, with the unit of minutes or hours. It is generally defined as the construction operable time index determined by the penetration instrument or temperature rise inflection point detection. The significance of this double-channel structure design lies in that the former (compressive strength) is used to measure the structural stability of the concrete, and the latter (setting time) is used to judge the construction continuity. Both of them together constitute a double-dimensional criterion for the performance of the concrete, covering the complete performance goals from structural safety to process adaptability.

[0084] In a specific embodiment, 1000 batches of historical process behavior feature group data and their corresponding measured values of compressive strength and setting time are used to construct a training data set. Gradient boosting regression tree and support vector regression model are used for training respectively, and cross-validation method is used to evaluate the model error in the verification stage to determine that the prediction error range is controlled within the accuracy threshold of ±5%. In the production process, the real-time process behavior feature group of the new batch of concrete is collected and sent to the trained model to output the double-channel prediction value, which is compared with the performance tolerance standard to realize intelligent judgment of high and low quality batches.

[0085] To ensure that the final performance of the production batch can meet the performance requirements of structural stability and construction continuity, the system needs to make a clear classification judgment on the quality of the current batch of concrete after the model prediction result is output to guide whether to release the discharge or to make parameter correction. To realize the dynamic balance of stability and sensitivity of the judgment standard, the present application adopts a double-threshold judgment mechanism for quality classification judgment, which realizes the detailed division of the batch quality state without affecting the prediction stability by setting two levels of performance tolerance limits.

[0086] The double-threshold judgment mechanism refers to that, when predicting the quality of concrete, a single performance range is not used as the basis for judgment, but two intervals are set, which are: the first performance tolerance range: representing the ideal performance target interval, which is the allowed deviation range of the target performance value (such as ±5%). The second performance warning range: representing the outermost tolerance of acceptable performance, which is wider than the first performance tolerance range, serving as a warning boundary (such as ±10%). This mechanism acts on the two prediction channel results of the model output, namely: the compressive strength of the current batch of concrete under standard curing conditions; and the setting time of the current batch of concrete.

[0087] The first case: completely qualified, when the compressive strength and setting time in the model output result both fall within the first performance tolerance range, that is, both indicators fluctuate within the optimal tolerance interval, indicating that the performance of the current batch of concrete is stable and reliable. In this case, the system judges the batch as a high-quality production batch, allowing it to be directly discharged and archived as a high-reliability sample without any correction operation.

[0088] The second case: borderline qualified, if one or both of the compressive strength and setting time in the model output result do not fall within the first performance tolerance range, but one or both are still within the second performance warning range, it indicates that the performance of the current batch of concrete does not fully meet the standards but is still within the acceptable range. In this case, the system still marks the current batch as a high-quality production batch, but sets its status as "to be verified" to distinguish it from the completely qualified batch. To achieve continuous optimization and parameter backtracking of the model, the applicable control parameter set and process behavior characteristic group of the batch are retained in the feature sample database for subsequent retraining and experience rule adjustment of the model. This mechanism allows boundary samples to enter the positive sample pool and provides real and boundary experience feedback for continuous learning of the model.

[0089] The third case: unqualified batch, when the compressive strength and setting time in the output result both exceed the second performance warning range, that is, both indicators deviate from the performance target boundary, indicating that the current batch of concrete has significant performance degradation or serious deviation of raw material parameters. At this time, the system judges the batch as a low-quality production batch, automatically prohibits the batch from being discharged, and calls the control parameter correction mechanism disclosed in the claims to perform the adjustment process of the applicable control parameter set.

[0090] For example, set the compressive strength target value to 42 MPa, the first performance tolerance range to ±5% (i.e. 39.9 MPa to 44.1 MPa), and the second performance warning range to ±10% (i.e. 37.8 MPa to 46.2 MPa); the setting time target is 180 minutes, and the corresponding tolerance ranges are ±5% (171 to 189 minutes) and ±10% (162 to 198 minutes), respectively. If the predicted value of a batch is: compressive strength 41 MPa, setting time 178 minutes → falls within the first performance tolerance range → high quality; if the predicted value is: compressive strength 40 MPa, setting time 194 minutes → setting time exceeds the first performance tolerance range but does not exceed the second performance warning range → marked as high quality, attributed to be verified; if the predicted value is: compressive strength 36 MPa, setting time 202 minutes → both are within the second performance warning range → determined to be low quality, which can ensure the stability of the prediction process and the safety of the release, avoid misjudging potential qualified batches, provide "critical state" samples for the model, help boundary learning and anomaly identification, and provide clear trigger conditions for control parameter modification, forming a quality closed-loop system.

[0091] After the performance output (including compressive strength and setting time) of the current batch of concrete is predicted by the machine learning model, the prediction result needs to be compared with the performance target to determine whether there is a performance deviation. Once a low-quality production batch is identified, the system needs to further locate the source of the control variable that caused the deviation, i.e. from the current batch of applicable type control parameter set, filter out the control parameter item most relevant to the deviation, and form the parameter combination that needs to be modified first, to update the control parameter set in the next round. The present application introduces a coupling calculation method based on Pearson correlation coefficient and performance deviation for the priority modification of control parameters to quantify the judgment of the priority modification degree of control parameters. The specific process is as follows:

[0092] After the machine learning model outputs the compressive strength prediction value and the setting time prediction value of the current batch of concrete, the system needs to calculate the difference between the two values and the median of the first performance tolerance range, which is defined as the performance deviation.

[0093] Compressive strength deviation value (KY): within the first performance tolerance range, take the arithmetic mean of the upper and lower limits of the corresponding compressive standard range as the median (for example, the compressive standard range is 39.9-44.1 MPa, and the median is 42 MPa); calculate the difference between the compressive strength prediction value of the current batch of model output and the median, denoted as KY.

[0094] The coagulation time deviation value (NJ): the median of the coagulation standard range corresponding to the first performance tolerance range is determined in the same way (for example, 180 minutes); the difference between the current predicted coagulation time and the median is NJ. This deviation resolution method can decompose the performance deviation behavior into two independent influencing factors, laying the foundation for subsequent correlation coefficient calculation.

[0095] The "applicable control parameter set" mentioned in the present application includes a plurality of control parameter variables with adjustability, such as the feeding time window, the feeding sequence priority, the stirring rhythm change rate, the liquid material premixing ratio, etc. To judge the influence degree of these parameter items on the performance output, the Pearson correlation coefficient analysis method is used to calculate the following two indexes respectively: using historical batch data samples, the historical set value of each control parameter and the corresponding measured value of the compressive strength of the batch are collected; the Pearson correlation coefficient between each control parameter and the compressive strength is calculated, denoted as G1. Similarly, the Pearson correlation coefficient between each control parameter and the coagulation time is calculated, denoted as G2. The Pearson correlation coefficient has a value range of [-1, 1], the greater the absolute value, the stronger the linear correlation, and the positive and negative directions represent positive and negative correlations respectively. The comprehensive correlation strength S = |G1x KY| + |G2x NJ| is calculated, KY represents the compressive strength deviation value, and NJ represents the coagulation time deviation value. All applicable control parameters are sorted in descending order according to the calculated S value, and the first m control parameters are taken, m is an empirical value or a set threshold, which depends on the number of control parameter variables, for example, if there are 10 control parameter variables, the first 3 to 5 control parameters can be taken, forming the control parameter combination to be corrected for the current batch, and the updated applicable control parameter set is regenerated according to the preset correction rule, to realize directional self-adaptive optimization from low-quality production batches to high-quality production batches.

[0096] If the current batch is determined as a low-quality production batch, the system will identify the source of deviation according to the model prediction results and make targeted corrections to the control parameters. The correction operation is based on the "preset correction rule" to execute the generation process of the "updated applicable control parameter set". The "preset correction rule" refers to a set of strategy rules defined by engineering experience, historical data analysis or expert strategy before the system runs, which is used to guide how to adjust the control parameters according to the type and degree of deviation. The rule set can include the following contents: determining the adjustment direction of the control parameter according to the sign (positive or negative) of the deviation: if the compressive strength deviation KY is negative (i.e. the predicted value is lower than the target), and a parameter G1 is positive, the parameter set value should be adjusted in the increasing direction; if G1 is negative, the parameter set value should be adjusted in the decreasing direction; the same logic can be used for the setting time deviation NJ and G2. Set the maximum adjustment amplitude and step unit of the control parameter: the maximum allowed adjustment percentage corresponding to each control parameter item; the step coefficient a, which can be set by experience, such as a = 0.05, indicating a maximum adjustment of 5% per round. Parameter boundary rule: ensure that the control parameters after correction do not exceed the physical or process allowed range: such as the liquid material premixing ratio should not exceed 80%; the feeding sequence priority should not be lower than the key path defined by the system; these rules ensure the controllability and physical reasonableness of the adjustment process. After the correction is completed, the system will update the set values of each control parameter to form a new set of parameters valid for the current batch, which is the updated applicable control parameter set. This set will be used to: re-execute the production of the current batch (if allowed to remix) or directly use the updated applicable control parameter set under the same conditions.

[0097] Embodiment 2: An intelligent production and quality monitoring system for ready-mixed concrete, specifically comprising:

[0098] A raw material modeling module for collecting actual weighing data of multiple batches of raw materials, calculating the difference between each raw material and the target ratio, and combining the measured results of the concrete strength of the corresponding batch to construct an initial control parameter set for the current production batch.

[0099] An environment adaptation module for obtaining temperature and humidity data of the current production environment, and applying the obtained environmental data as influencing factors to the initial control parameter set to construct an applicable control parameter set for the current batch.

[0100] A process prediction module for performing concrete mixing production operation according to the applicable control parameter set of the current batch, collecting a plurality of preset process behavior data during production, extracting features from all collected process behavior data to obtain a process behavior feature group, and then inputting the process behavior feature group into a preset machine learning model to output the compressive strength and setting time of the batch of concrete under standard curing conditions.

[0101] The quality regulation module is configured to compare the machine learning model output result with a preset performance target, and if the machine learning model output result exceeds a preset standard range, the batch is determined as a low-quality production batch, and a control parameter combination that needs to be corrected in the applicable control parameter set is determined based on the machine learning model output result, and an updated applicable control parameter set is regenerated according to a preset correction rule; if the machine learning model output result does not exceed the preset standard range, the batch is determined as a high-quality production batch, and the batch of concrete is allowed to be discharged.

[0102] The above formulas are dimensionless values calculated, and the formulas are obtained by collecting a large amount of data to simulate a formula of the most recent real situation, and the preset parameters in the formula are set by a person skilled in the art according to the actual situation.

[0103] It should be understood that the size of the sequence number of each process described above in various embodiments of the present application does not mean the order of execution, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0104] Those skilled in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0105] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system, device and unit described above can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0106] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for intelligent production and quality monitoring of ready-mixed concrete, characterized in that: The following steps are involved: Collect actual weighing data of multiple batches of raw materials, calculate the difference between each raw material and the target ratio, and combine it with the actual measured concrete strength results of the corresponding batch to construct the initial control parameter set for the current production batch; Obtain the temperature and humidity data of the current production environment, and use the obtained environmental data as an influencing factor to act on the initial control parameter set to build an applicable control parameter set for the current batch; Execute concrete mixing production operations based on the applicable control parameter set for the current batch. Collect multiple preset process behavior data during the production process. Perform feature extraction on all collected process behavior data to obtain a process behavior feature set. This feature set is then input into a preset machine learning model to output the compressive strength and setting time of the batch of concrete under standard curing conditions. The output of the machine learning model is compared with the preset performance target. If it exceeds the preset standard range, the batch is judged to be a low-quality production batch, and the control parameter combination that needs to be corrected in the applicable control parameter set is determined based on the output of the machine learning model. The updated applicable control parameter set is regenerated according to the preset correction rules. If it does not exceed the preset standard range, it is judged to be a high-quality production batch and the concrete of this batch is allowed to be discharged; The construction of the initial control parameter set includes the introduction of an error weight function. This function uses the proportion of each raw material in the ratio as the weight basis, and accumulates the deviation values ​​between the actual weighing data of each batch and the theoretical target ratio to obtain the deviation impact value that represents the fluctuation pattern between batches. The deviation impact value is combined with the actual strength measurement results of the corresponding batch of concrete to determine the correlation between the raw material type and the strength response through a mapping relationship, and the raw material types with a correlation exceeding the preset correlation threshold are included in the initial control parameter set; Building an applicable control parameter set for the current batch includes the following steps: The initial control parameter set is subjected to the first-level clustering process, using the deviation between the actual weighing data of the raw materials and the target ratio as the input variable to form several deviation behavior groups; The temperature and humidity data were used to construct a joint environmental feature vector, which was used as the input variable for the second-level clustering process to obtain several environmental response groups. The deviation behavior group and environmental response group corresponding to each batch number are cross-combined to obtain the corresponding control parameter adjustment rule under the cluster label index, and the applicable control parameter set for the current batch is generated based on the control parameter adjustment rule.

2. A method for intelligent production and quality monitoring of ready-mixed concrete according to claim 1, characterized in that: The first-layer clustering and the second-layer clustering both use K-means clustering or Gaussian mixture model.

3. A method for intelligent production and quality monitoring of ready-mixed concrete according to claim 2, characterized in that: The applicable control parameter set for the current batch includes the following four executable control parameters: feeding time window, feeding sequence priority, stirring rhythm change rate, and liquid material premix ratio.

4. A method for intelligent production and quality monitoring of ready-mixed concrete according to claim 3, characterized in that: The process behavior feature group is extracted using a time series-based behavior analysis strategy. The process behavior feature group includes the stirring load change rate, slurry temperature change rate, humidity response delay time, coefficient of variation of viscosity estimation, stirring current fluctuation amplitude, and vibration frequency stable time period. Among them, the stirring load change rate is the slope value of the load rising trend within the preset time interval after the start of stirring, the slurry temperature change rate is the average speed of temperature increase within the time interval, the humidity response delay time is the time delay corresponding to the first time the humidity change amplitude exceeds the set threshold, the coefficient of variation of the viscosity estimate is the ratio of the standard deviation of the load fluctuation in the stirring stable stage to the mean, the stirring current fluctuation amplitude is the difference between the maximum and minimum values ​​of the stirring current in the specified time period, and the vibration frequency stable time period is the length of time from the start of vibration to the time when the frequency curve change rate is lower than the set threshold.

5. A method for intelligent production and quality monitoring of ready-mixed concrete according to claim 4, characterized in that: The machine learning model is constructed based on a regression output structure and is built using any one of the gradient boosted regression tree model, support vector regression model or deep neural network structure. The machine learning model outputs the compressive strength and setting time of the current batch of concrete under standard curing conditions as dual-channel results.

6. A method for intelligent production and quality monitoring of ready-mixed concrete according to claim 5, characterized in that: In the process of comparing the output results of the machine learning model with the preset performance targets, a dual-threshold judgment mechanism is used for quality classification judgment. This judgment mechanism sets a first performance tolerance range and a second performance warning range. When the compressive strength and coagulation time in the output results fall within the first performance tolerance range, the current batch is judged to be a high-quality production batch; when the compressive strength and coagulation time in the output results exceed the second performance warning range, the current batch is judged to be a low-quality production batch.

7. A method for intelligent production and quality monitoring of ready-mixed concrete according to claim 6, characterized in that: The control parameter combinations that need to be modified in the applicable control parameter set based on the output of the machine learning model are: When the output result of the machine learning model is compared with the preset performance target, the output deviation of the output result is decomposed into two components: a compressive strength deviation value and a setting time deviation value. The compressive strength deviation value is obtained by calculating the difference between the compressive strength in the output result and the median of the compressive standard range corresponding to the first performance tolerance range. The setting time deviation value is obtained by calculating the difference between the setting time in the output result and the median of the setting standard range corresponding to the first performance tolerance range. Based on the Pearson correlation coefficient analysis, the correlation coefficient G1 of each control parameter in the applicable control parameter set with the compressive strength and the correlation coefficient G2 with the setting time are obtained, and then the comprehensive correlation strength S=|G1×KY|+|G2×NJ| is calculated, where KY represents the compressive strength deviation value and NJ represents the setting time deviation value. Then, the parameters are sorted in descending order according to the comprehensive correlation strength S, and the top m control parameters are taken to obtain the control parameter combination that needs to be modified in the applicable control parameter set.

8. A system for intelligent production and quality monitoring of ready-mixed concrete, based on the method for intelligent production and quality monitoring of ready-mixed concrete according to any one of claims 1 to 7, characterized in that: Specifically include: The raw material modeling module is used to collect the actual weighing data of multiple batches of raw materials, calculate the difference between each raw material and the target ratio, and build the initial control parameter set for the current production batch based on the actual measured concrete strength results of the corresponding batch; The environmental adaptation module is used to obtain the temperature and humidity data of the current production environment, and use the obtained environmental data as an influencing factor to act on the initial control parameter set to build an applicable control parameter set for the current batch; The process prediction module is used to execute concrete mixing production operations based on the applicable control parameter set for the current batch. During the production process, multiple preset process behavior data are collected, and features are extracted from all the collected process behavior data to obtain a process behavior feature set. This feature set is then input into a preset machine learning model to output the compressive strength and setting time of the batch of concrete under standard curing conditions. The quality control module is used to compare the output of the machine learning model with the preset performance target. If it exceeds the preset standard range, the batch is judged to be a low-quality production batch. Based on the output of the machine learning model, the control parameter combination that needs to be corrected in the applicable control parameter set is determined, and the updated applicable control parameter set is regenerated according to the preset correction rules. If it does not exceed the preset standard range, it is judged to be a high-quality production batch and the concrete batch is allowed to be discharged; The construction of the initial control parameter set includes the introduction of an error weight function. This function uses the proportion of each raw material in the ratio as the weight basis, and accumulates the deviation values ​​between the actual weighing data of each batch and the theoretical target ratio to obtain the deviation impact value that represents the fluctuation pattern between batches. The deviation impact value is combined with the actual strength measurement results of the corresponding batch of concrete to determine the correlation between the raw material type and the strength response through a mapping relationship, and the raw material types with a correlation exceeding the preset correlation threshold are included in the initial control parameter set; Building an applicable control parameter set for the current batch includes the following steps: The initial control parameter set is subjected to the first-level clustering process, using the deviation between the actual weighing data of the raw materials and the target ratio as the input variable to form several deviation behavior groups; The temperature and humidity data were used to construct a joint environmental feature vector, which was used as the input variable for the second-level clustering process to obtain several environmental response groups. The deviation behavior group and environmental response group corresponding to each batch number are cross-combined to obtain the corresponding control parameter adjustment rule under the cluster label index, and the applicable control parameter set for the current batch is generated based on the control parameter adjustment rule.

Images