Concrete mix proportion engineering reliability online evaluation and closed-loop control method and system

Abstract

This invention discloses an online reliability assessment and closed-loop control method and system for concrete mix design, relating to the field of intelligent concrete production control technology. The method includes: correcting the original mix design output from a data-driven model based on engineering constraints to obtain a compliant mix design; calculating the correction offset between two mix designs; calculating risk characteristics based on the correction offset within a rolling statistical window, and obtaining a comprehensive engineering risk value through normalization, weighted fusion, and monotonic mapping; inputting the comprehensive engineering risk value into a three-state machine with hysteresis, switching between normal, degraded, and circuit-breaker states based on preset thresholds and hysteresis conditions, and generating control decisions; issuing corresponding hard control commands; and recording and storing audit traceability information for each batch. This invention achieves reliability quantification and closed-loop hard control in scenarios without real-value feedback, improving production robustness and traceability.

Description

Technical Field

[0001] This invention belongs to the field of intelligent concrete production control technology. Specifically, it relates to a method and system for online evaluation and closed-loop control of concrete mix proportion engineering reliability based on correction offset information for intelligent concrete mixing plants. It is particularly suitable for online production scenarios where it is impossible to obtain real-time test values ​​of concrete mixture performance or hardening performance, and for quantitative evaluation and safety control of mix proportions output by data-driven models. Background Technology

[0002] In concrete mix design applications, key parameters such as unit water consumption, water-cement ratio, and sand ratio collectively determine the workability, strength development, and durability of the mixture. With the development of data-driven methods, multi-objective regression models or deep learning models can output these multiple engineering parameters given material properties and engineering conditions, providing technical support for intelligent manufacturing.

[0003] However, during project implementation, especially under online production conditions, it is often impossible to obtain real-time test results of the performance indicators or hardening performance indicators of the batch of concrete mix, such as the slump, spread, and compressive strength after hardening. This core constraint causes traditional evaluation methods based on the "error between predicted and actual values" to fail in this scenario, thus lacking direct and immediate feedback on the quality of AI model output.

[0004] In existing technologies, online assurance of model output quality has the following main shortcomings:

[0005] (1) Monitoring methods based solely on input drift or confidence: These methods focus on monitoring changes in the model input distribution or output confidence. However, in scenarios lacking real-time performance test values, their indicators are difficult to directly correspond to whether the mix proportion output violates key engineering constraints, and cannot provide an actionable risk assessment of the engineering usability of the output results.

[0006] (2) Using only constraint correction or projection methods: This type of method can correct the model output to the feasible region of the constraints to obtain a compliant mix ratio, but usually only outputs the corrected result and does not quantify the risk information reflected by the "degree of conflict between the model output and the engineering constraints". Therefore, it cannot provide an early warning signal of reliability degradation to the control system.

[0007] (3) Alarm or manual intervention as the main methods: Existing solutions mostly remain at the level of prompting, alarm or manual review, lacking a hard isolation mechanism that links risk signals with industrial control links (such as PLC / controller), such as freezing write, switching write sources, rolling back to stable versions, etc., making it difficult to block unreliable mix ratios from entering the production execution link in a timely manner.

[0008] Therefore, there is an urgent need for an online evaluation and control mechanism that can quantify the "conflict / offset degree" generated by the model output after engineering constraint correction into a risk signal in the absence of real-time performance test values, and trigger the linkage actions of industrial control links such as state switching, write freeze / write source switching and stable version rollback based on the risk signal. Summary of the Invention

[0009] To address the aforementioned deficiencies in existing technologies, the present invention aims to provide a method and system for online evaluation and closed-loop control of concrete mix design reliability. This system addresses the challenge of quantitatively evaluating the engineering reliability of mix designs output by data-driven models under online production conditions where real-time detection values ​​of concrete mixture performance or hardening performance cannot be obtained. Furthermore, it addresses the need to automatically execute industrial-grade hard control actions (including write freeze, write source switching, and stable version rollback) based on the evaluation results. This achieves real-time blocking of unreliable mix designs and closed-loop assurance of production safety.

[0010] This invention solves the above-mentioned technical problems through the following technical solution: a method for online evaluation and closed-loop control of the reliability of concrete mix design, comprising:

[0011] Obtain the original mix proportions output by the data-driven model for the current production batch;

[0012] Based on engineering constraints, the original mix proportion is modified to meet compliance requirements, resulting in a compliant mix proportion that satisfies the engineering constraints.

[0013] Calculate the correction offset between the compliant mix ratio and the original mix ratio;

[0014] Within a rolling statistical window that includes multiple recent production batches, risk characteristics, including at least the correction trigger rate and the offset intensity, are calculated based on the corrected offset. The risk characteristics are then normalized and weighted and fused before being monotonically mapped to obtain a comprehensive engineering risk value.

[0015] The comprehensive engineering risk value is input into a three-state machine with hysteresis mechanism. The three-state machine switches between normal state, degraded state and circuit breaker state and generates control decisions based on preset thresholds and hysteresis conditions.

[0016] Issuing hard control instructions corresponding to the control decision, including: when in the normal state, allowing the compliant mix ratio to the production execution unit; when in the degraded state, blocking the original mix ratio and switching the writing source to the backup mix ratio; when in the circuit breaker state, blocking all mix ratio writing, so that the production execution unit maintains the currently executed mix ratio until the hysteresis condition is met and the blocking is lifted.

[0017] Record and link audit traceability information for each production batch to support quality traceability and version rollback.

[0018] To address the problem that traditional assessment methods based on the error between predicted and actual values ​​fail due to the inability to obtain real-time test results of concrete mix performance or hardening properties under online production conditions, this invention introduces a compliance correction step. This step corrects the original mix proportions output by the model to within the feasible region that satisfies engineering constraints, and calculates the correction offset as an observable surrogate quantity characterizing the consistency between the model output and engineering constraints. Thus, even without access to actual test values, it still achieves a quantitative characterization of the consistency between the model output and engineering constraints. This technique transforms the "degree of constraint conflict" into a calculable statistic online, providing a reliable data foundation for subsequent risk assessment.

[0019] Existing methods, which only monitor input drift or model confidence, struggle to directly correlate whether mix proportion outputs violate critical engineering constraints. This invention, within a rolling statistical window, calculates risk characteristics based on corrected offsets, including at least the correction trigger rate and offset intensity. These characteristics are then normalized, weighted, and fused before being monotonically mapped to obtain a comprehensive engineering risk value. This technique transforms raw offset information into risk indicators with clear engineering significance, comprehensively reflecting the frequency, magnitude, and evolution trend of model output deviations from engineering constraints, providing a quantitative basis for hierarchical control.

[0020] Existing solutions often rely on alarms or manual intervention, lacking a hard isolation mechanism for integration with the industrial control chain. This invention introduces a three-state machine with hysteresis, using a comprehensive engineering risk value as input. Based on preset thresholds, the machine switches between normal, degraded, and circuit-breaker states, generating control decisions. The hysteresis mechanism effectively prevents frequent state switching due to fluctuations in risk values, ensuring the stability of the production process. The hierarchical control architecture allows the system to take differentiated measures based on risk levels, achieving progressive safety protection from early warning to circuit breaker.

[0021] To address the limitations of existing technologies in implementing hard control actions such as write freeze, write source switching, and stable version rollback, this invention addresses this issue by issuing hard control commands corresponding to control decisions. These commands enable specific control actions in three states: in the normal state, compliant mix ratios are allowed; in the degraded state, the original mix ratio is blocked and switched to a backup mix ratio; and in the circuit breaker state, all mix ratio writes are blocked while maintaining the currently executing mix ratio. This technique directly translates risk assessment results into executable commands for the production controller or programmable logic controller, achieving real-time blocking of unreliable mix ratios and closed-loop assurance of production safety, completely changing the passive situation of relying on manual intervention.

[0022] Existing technologies lack systematic recording of control decisions, making it difficult to support quality accountability and version rollback. This invention establishes a complete digital archive for each production batch by recording and linking audit traceability information. This information can not only be used for post-production quality traceability, but more importantly, it provides data support for automatic rollback to the previous stable version. When the system recovers from a circuit breaker state, it can accurately identify and switch back to a reliable mix ratio version based on historical records, forming a complete closed-loop governance system.

[0023] Furthermore, the engineering constraints include at least one of physical equality constraints and engineering inequality constraints; wherein, the physical equality constraints include at least one of volume conservation constraints and mass-volume conversion constraints based on material density, and the engineering inequality constraints include at least one of water-cement ratio range constraints, sand ratio range constraints, total cementitious material range constraints, and air content constraints.

[0024] This invention, by specifying engineering constraints, clarifies the physical equation constraints (such as volume conservation and mass-volume conversion) and engineering inequality constraints (such as water-cement ratio range, sand ratio range, total cementitious material range, and air content constraints) upon which compliance correction is based. This ensures that the corrected mix proportion not only meets basic physical laws but also conforms to the engineering requirements of actual production processes and material properties, thereby improving the engineering feasibility and safety of the correction results and providing a reliable constraint benchmark for subsequent risk quantification.

[0025] Furthermore, the corrected trigger rate and offset intensity are calculated as follows:

[0026] ;

[0027] ;

[0028] in, This represents the correction trigger rate of the j-th output dimension; n represents the number of samples within the scrolling statistics window. This represents the correction offset of the i-th sample in the j-th output dimension; This represents the offset threshold for the j-th output dimension; This indicates an indicator function; the value is 1 if the condition within the parentheses is true, and 0 otherwise. This represents the offset strength of the j-th output dimension; This represents the scaling factor of the j-th output dimension;

[0029] The offset threshold is determined based on a multiple of the weighing system resolution or a proportion of the allowable engineering deviation; the scale factor is the standard deviation of the corrected offset under historical stable conditions or the width of the allowable engineering range.

[0030] This invention provides specific calculation formulas for the correction trigger rate and offset intensity, and specifies engineering setting rules for the offset threshold and scale factor (such as based on the weighing system resolution or engineering allowable deviation). This makes the quantification of risk characteristics have clear physical meaning and operability, accurately reflects the frequency and magnitude of the correction offset on each output dimension, provides standardized input for the calculation of comprehensive risk value, and ensures the flexibility and consistency of parameter settings under different production scenarios.

[0031] Furthermore, the risk characteristics also include at least one of the following: a corrected continuity index, a coupled corrected strength index, and a corrected offset dispersion index; wherein the corrected continuity index is calculated in the following manner:

[0032] ;

[0033] in, This represents the corrected continuity index for the j-th output dimension; n represents the number of samples within the rolling statistics window. , These represent the correction offsets of the i-th and i+1-th samples in the j-th output dimension, respectively. This represents the offset threshold for the j-th output dimension; Indicates an indicator function; This represents the logical AND operation;

[0034] The coupling correction strength index is calculated as follows:

[0035] ;

[0036] in, This represents the coupling correction strength index between the j-th output dimension and the k-th output dimension; This represents the correction offset of the i-th sample in the k-th output dimension; This represents the offset threshold for the k-th output dimension; , These represent the scaling factors of the j-th and k-th output dimensions, respectively.

[0037] The corrected offset dispersion index is calculated as follows:

[0038] If all corrected offset vectors within the rolling statistics window are zero, then the corrected offset dispersion index is 0;

[0039] otherwise:

[0040] ;

[0041] CDI stands for Corrected Offset Dispersion Index; Let represent the cosine similarity between the p-th sample and the q-th sample after correction. The cosine similarity is zero when the norm of any correction offset vector is zero.

[0042] This invention introduces derived risk features such as corrected continuity, coupled corrected strength, and corrected offset dispersion, and characterizes them through explicit mathematical formulas. These features can capture complex risk patterns that traditional single-parameter monitoring cannot identify: the corrected continuity index reflects the trend change of offset, which helps to warn of systemic deviations; the coupled corrected strength index reveals the synergistic risks when multiple parameters exceed thresholds simultaneously, and can identify chain reactions caused by raw material fluctuations or model mismatch; the corrected offset dispersion index quantifies the degree of dispersion of offset directions within a window, providing a basis for assessing the consistency of risk evolution. The addition of these multi-dimensional features significantly improves the ability of the comprehensive risk value to characterize the engineering risk status.

[0043] Furthermore, the comprehensive engineering risk value is calculated as follows:

[0044] ;

[0045] Where R represents the comprehensive engineering risk value; L represents the number of risk characteristics; The weight coefficient represents the l-th risk characteristic; This represents the normalized value of the l-th risk feature;

[0046] When the risk characteristic is a coupling correction strength index, the corresponding weighting coefficient The value is 1, and the normalized value of the coupling correction strength index is: , This represents the normalized value of the coupling strength index between the j-th and k-th output dimensions. This represents the corresponding weighting coefficient;

[0047] The normalization is achieved through monotonic mapping, specifically: for the original statistics of risk characteristics The normalized value is calculated based on the low and high quantiles of its historical distribution. Here, clip() restricts the calculation result to the interval [0,1], Q L and Q H These represent the lower quantile and the upper quantile, respectively.

[0048] This invention specifies a concrete method for calculating the comprehensive engineering risk value, including fusing weighted normalized risk characteristics using an exponential monotonic mapping and a normalization method based on historical quantiles. The exponential mapping ensures that the risk value responds sensitively in the low-risk range and naturally saturates in the high-risk range, conforming to the nonlinear characteristics of engineering risk perception. Normalization based on low and high quantiles eliminates the influence of different feature dimensions and utilizes historical distributions to achieve robust scaling, making the risk value comparable across different production stages. Simultaneously, a two-layer weighted structure is adopted for the coupling correction strength index, preserving the independence of each coupling pair while ensuring the rationality of the overall risk contribution.

[0049] Furthermore, the preset thresholds include a first threshold, a second threshold, and a recovery threshold, and satisfy the condition that the first threshold < the second threshold < the recovery threshold; the switching rule of the three-state machine is as follows:

[0050] When the overall engineering risk value is less than the first threshold, it is in or remains in a normal state;

[0051] When the comprehensive engineering risk value is greater than or equal to the first threshold and less than the second threshold, it is in or remains in a downgraded state.

[0052] When the comprehensive engineering risk value is greater than or equal to the second threshold, the circuit breaker is activated.

[0053] The hysteresis condition includes: when the three-state machine is in a circuit breaker state or a degraded state, recovery to the normal state is only allowed after the comprehensive engineering risk value of multiple consecutive rolling statistical windows is lower than the recovery threshold.

[0054] This invention refines the switching rules and hysteresis conditions of the three-state machine, clarifies the progressive relationship between the first threshold, the second threshold, and the recovery threshold (recovery threshold < first threshold < second threshold), and stipulates that recovery from a circuit breaker state or a degraded state to a normal state requires satisfying a hysteresis condition where the risk value for multiple consecutive windows is lower than the recovery threshold. This mechanism effectively prevents frequent state switching caused by short-term fluctuations in risk values, ensuring the stability of production control. Simultaneously, the setting of tiered thresholds enables risk-level response, allowing the system to adopt differentiated control strategies at different risk levels, balancing production continuity and safety.

[0055] Furthermore, the backup mix ratio includes a preset conservative template mix ratio or a stable version mix ratio determined by audit traceability information; the stable version mix ratio is determined by selecting a mix ratio version where multiple consecutive windows are in a normal state and the risk value is below a first threshold.

[0056] This invention clarifies the specific sources of backup mix proportions, including preset conservative template mix proportions and stable version mix proportions determined by audit traceability information, and provides a method for identifying stable versions (mix proportion versions whose most recent consecutive multi-window operation was in a normal state and whose risk value was below a first threshold). This limitation ensures that the backup mix proportions used have reliable historical performance when switching write sources in a degraded state, enabling rapid restoration of safe production; at the same time, the stable version identification method based on audit records provides an objective basis for automatic rollback, avoiding the subjectivity and lag of manual selection.

[0057] Furthermore, the audit traceability information includes at least: the original mix ratio, compliant mix ratio, corrected offset, comprehensive engineering risk value, model version, threshold version, control status, control action, and work order number for each production batch.

[0058] This invention specifies the concrete information included in audit traceability information. The complete recording of this information provides comprehensive data support for post-event quality traceability, responsibility identification, and system optimization. Simultaneously, the recording of model versions and threshold versions enables accurate identification and restoration of the corresponding model and parameter configurations when the system needs to revert to a previous stable version, achieving a closed loop in version governance.

[0059] Furthermore, the hard control commands are mapped via an industrial protocol to one of the following operations on a production controller or programmable logic controller: writing to a recipe register, setting a write protection flag, closing a write channel, or switching a recipe version number.

[0060] This invention maps hard control commands to industrial operations on production controllers or programmable logic controllers. These operations directly correspond to the underlying execution units of the industrial control system, enabling control decisions to be implemented quickly using standardized industrial protocols and achieving precise intervention in the writing of mix proportions. At the same time, the enumeration of multiple operation modes provides a flexible adaptation solution for different industrial control architectures, ensuring the wide applicability of the method of this invention in various mixing plant control systems.

[0061] Based on the same concept, the present invention also provides an online assessment and closed-loop control system for the reliability of concrete mix design, including a module configured to perform the online assessment and closed-loop control method for the reliability of concrete mix design as described above.

[0062] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0063] By introducing "corrected offset" as an observable surrogate quantity, the degree of conflict between model output and engineering constraints is transformed into a statistical feature that can be calculated online. Without relying on the actual detection value, the engineering reliability of the AI ​​model output quality is quantitatively evaluated, solving the problem that traditional methods based on prediction error fail in this scenario.

[0064] By using a three-state machine (normal state / degraded state / fuse state) to classify and respond to risk levels, and by directly executing hard control actions such as write release, write source switching, and write freeze on the PLC or weighing control unit through the industrial control interface, the risk assessment results are transformed into production intervention in real time. This achieves automatic blocking and safety fuse for unreliable mixing ratios, completely changing the passive mode that relies on manual alarms or interventions.

[0065] The introduction of the hysteresis mechanism (recovery from a high-risk state requires multiple consecutive windows of risk value to be below the recovery threshold) effectively prevents frequent state switching caused by risk value fluctuations, ensuring the stability of the production process; the fusion of multi-dimensional risk characteristics (correction continuity, coupling correction strength, correction offset dispersion, etc.) can identify complex risk patterns that cannot be captured by single-parameter monitoring (such as systematic deviations and parameter coupling risks), improving the accuracy of risk quantification.

[0066] Complete audit traceability records (including original mix ratio, compliant mix ratio, corrected offset, risk value, model version, threshold version, control status, etc.) not only support post-event quality accountability, but also provide a data foundation for automatic rollback to the previous stable version, realizing closed-loop governance from risk identification, control execution to version recovery, and significantly improving the maintainability and compliance of intelligent production. Attached Figure Description

[0067] 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.

[0068] Figure 1 This is a flowchart of the online evaluation and closed-loop control method for the reliability of concrete mix design in this invention.

[0069] Figure 2 This is a schematic diagram illustrating the principle of online risk feature aggregation and calculation based on a scrolling window in this embodiment of the invention.

[0070] Figure 3 This is a schematic diagram illustrating the dynamic evolution of the comprehensive engineering risk value R with the production batch sequence in an embodiment of the present invention; Detailed Implementation

[0071] 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.

[0072] 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.

[0073] Example 1

[0074] To address the technical challenges of scenarios lacking online feedback of true values, this invention provides an online reliability assessment and closed-loop control method for concrete mix design. It proposes using the correction offset between the original mix design output by the data-driven model and the compliant mix design after engineering constraints as an observable surrogate quantity for the consistency between the model output and engineering constraints. By quantifying the statistical risk characteristics of this observable surrogate quantity online, it replaces traditional true value error feedback, thereby driving a three-state machine. This state machine directly performs hard control actions such as writing, freezing, switching, and rolling back on the PLC or weighing control unit through an industrial control interface. Simultaneously, it generates audit traceability records containing versioned information, ultimately forming a pluggable "quality and safety gate" module independent of the data-driven model. This solves the engineering challenges of "reliability" and "safety" of AI prediction results in actual production.

[0075] Reference Figure 1 The method for online evaluation and closed-loop control of the reliability of concrete mix design includes the following steps:

[0076] Step S1: Obtain the original mix proportions output by the data-driven model for the current production batch.

[0077] The task is issued as a production batch based on the mix proportion of each truckload or batch of concrete, denoted as sample i. For each newly arrived production batch, the system first obtains the original mix proportion vector output by the data-driven model for that production batch through the parameter acquisition module.

[0078] Data-driven models can be pre-trained multi-objective regression models, deep learning models, or other machine learning models. Their inputs typically include the characteristics of raw materials in the current production batch (such as cement grade, sand and gravel moisture content, admixture type, etc.), engineering design requirements (such as strength grade, slump target value), and environmental conditions (such as temperature and humidity). The output is the key parameters of the mix proportion that meet the given conditions.

[0079] The original mix ratio vector is denoted as:

[0080] (1)

[0081] in, Represents the original combination ratio vector for the i-th sample; This represents the original mix proportion parameters for the i-th sample in the j-th output dimension; m represents the number of mix proportion parameters. In typical applications, the original mix proportion parameters include, but are not limited to, the following engineering parameters: unit water consumption, water-cement ratio, sand ratio, total amount of cementitious materials, admixture dosage, and air content. The specific meaning and units of each parameter are determined according to the concrete mix design specifications; for example, the unit water consumption is kg / m³. 3 The water-cement ratio is a dimensionless ratio, and the sand ratio is a percentage, etc.

[0082] The parameter acquisition module can obtain the data-driven model output in one of the following ways:

[0083] Direct Invocation: If the data-driven model is deployed as an API service, the parameter acquisition module requests the model service in real time via HTTP or RPC interface, passing in the input feature data of the current production batch and receiving the returned original mix ratio vector.

[0084] Local computation: If the data-driven model is embedded within the production control system (such as an industrial computer or edge computing node), the parameter acquisition module directly calls the local model inference engine, inputs feature data, and obtains the original mix ratio vector output by the model.

[0085] To ensure data consistency and traceability, while obtaining the original mix proportions, the parameter acquisition module also records information such as the model version number, input feature snapshots, and timestamps, and passes this information to subsequent steps and the audit traceability module.

[0086] The execution timing of step S1 is synchronized with the production cycle: whenever a new production batch task is generated, a model call and parameter acquisition operation are triggered. In continuous production scenarios, the interval between adjacent samples may range from tens of seconds to several minutes, and step S1 needs to be completed within milliseconds to meet real-time requirements.

[0087] In some implementations, the original mix proportions output by the model may not directly include all the required parameters; for example, only some core parameters may be output, while other parameters are supplemented by fixed formulas or rules.

[0088] Step S2: Based on engineering constraints, the original mix proportion is modified to meet compliance requirements, resulting in a compliant mix proportion that satisfies the engineering constraints.

[0089] To obtain the original mix proportions for the current production batch Subsequently, the system corrects the mix proportions through the compliance correction module to ensure that the final mix proportions used in production meet all predefined engineering constraints. The core objective of step S2 is to retain as much information as possible about the original mix proportions while ensuring engineering feasibility, thereby providing a reliable correction offset for subsequent risk quantification.

[0090] Engineering constraints include at least one physical equality constraint and at least one engineering inequality constraint, the specific contents of which are determined by concrete mix design specifications, production process requirements and raw material characteristics.

[0091] In specific implementations, physical equality constraints include, but are not limited to:

[0092] Volume conservation constraint: The sum of the volumes of all components (cement, water, sand, stone, admixtures, mineral admixtures, etc.) must be equal to the total volume of 1 cubic meter of concrete; Mass-volume conversion constraint: The conversion relationship between specific parameters established based on material density, such as the relationship between the total amount of cementitious materials and the amount of each cementitious component.

[0093] Engineering inequality constraints include, but are not limited to: water-cement ratio range constraints, sand ratio range constraints, total cementitious material range constraints, unit water consumption range constraints, admixture dosage range constraints, and air content range constraints. The specific values ​​of these engineering inequality constraints are pre-configured according to engineering design requirements and can be dynamically adjusted as raw material batches, environmental conditions, or production targets change.

[0094] The compliance correction module takes the original mix design as input and obtains the compliant mix design by solving the following constrained optimization problem:

[0095] (2)

[0096] (3)

[0097] in, Let represent the compliant mix ratio vector to be solved, with dimension and same; The weighted Euclidean norm is represented by the weight matrix W, which reflects the relative importance of the magnitude of the correction of different parameters (for example, a small change in the water-cement ratio may have a large impact on the strength and can be given a higher weight). Represents the constraint function for the k-th project inequality; M represents the constraint function of the l-th physical equation; C N represents the number of engineering inequality constraints. C This indicates the number of physical equality constraints.

[0098] In practice, the above constraints can usually be expressed in linear or quadratic form, so the optimization problem can be transformed into a quadratic programming (QP) problem, which is convenient for efficient solution.

[0099] To meet the real-time requirements of online production, the system employs a numerical optimization algorithm to solve the aforementioned optimization problem within a limited time. Based on the mathematical properties of the constraints, one of the following methods can be selected:

[0100] Quadratic programming solver: When all constraints are linear and the objective function is quadratic, it uses the interior point method or the effective set method to solve the problem, with a typical solution time in the millisecond range.

[0101] Sequential Quadratic Programming (SQP): When the constraints are nonlinear, the SQP algorithm can be used to iteratively approximate the optimal solution.

[0102] Projection gradient method: For large-scale problems, the projection gradient method can be used to project the iteration points into the feasible region.

[0103] The system sets the maximum solution time threshold T. max (e.g., 500 milliseconds), if the solver is in T max If no feasible solution is returned, or if the problem is determined to be infeasible, the solution failure handling process is triggered.

[0104] If the solution fails, it indicates that the compliance correction has failed, and the system will immediately execute a rollback strategy to ensure uninterrupted production and manageable risks. Specific rollback methods include the following two:

[0105] One approach uses a pre-set conservative template mix proportion: This template is pre-defined by domain experts based on historical production experience, meets engineering constraints under all operating conditions, and can be directly used as a compliant mix proportion. Another approach uses a stable version mix proportion determined by audit traceability information: The system searches the audit database for the mix proportion version where multiple consecutive windows were in a normal state, the risk value was below the first threshold, and the data was successfully written, and uses this version as the compliant mix proportion for the current batch.

[0106] Regardless of the rollback method used, the system marks the failure event in the audit traceability record and records the reason for the failure (such as timeout or infeasibility) for subsequent analysis and optimization.

[0107] After successful solution, the optimal solution to the optimization problem is the compliant mix proportion, which satisfies all engineering constraints and has the smallest weighted deviation from the original mix proportion. This compliant mix proportion will be used as input for subsequent steps:

[0108] Used to calculate the correction offset; directly written to the production execution unit in normal state; and used as a reference for the spare mix ratio in degraded or circuit-breaker state.

[0109] For example, the unit water consumption in a certain batch of original mix proportions is 175 kg / m³. 3 However, engineering constraints require that the unit water consumption must not be less than 180 kg / m³. 3 (Due to the high water absorption rate of the raw materials). The compliance correction module constructs an optimization problem, aiming to minimize the weighted distance from the original mix proportion, while simultaneously satisfying the lower limit constraint on unit water consumption and other parameter constraints. The solver obtains the compliant mix proportion after 50 iterations, where the unit water consumption is adjusted to 180 kg / m³. 3 Meanwhile, parameters such as the water-cement ratio and sand ratio were also fine-tuned accordingly to ensure volume conservation. The solution process took 120 milliseconds, which was within the time limit, and the correction was successful.

[0110] Step S3: Calculate the correction offset between the compliant mix proportion and the original mix proportion.

[0111] After obtaining the compliant mix proportion for the current production batch, the system executes step S3 through the offset calculation module to calculate the corrected offset between the compliant mix proportion and the original mix proportion. This corrected offset is the core observable surrogate quantity of this invention, used to quantify the degree of consistency between the model output and engineering constraints when real-time detection values ​​of concrete performance cannot be obtained, and to provide basic data for subsequent risk feature extraction.

[0112] The correction offset is defined as the difference between the compliant mix ratio vector and the original mix ratio vector, denoted as:

[0113] (4)

[0114] in, The original mix ratio vector output by the data-driven model; This is the compliant mix ratio vector that meets the engineering constraints, obtained after the compliance correction in step S2. To correct the offset.

[0115] Correct offset It is an m-dimensional vector, whose components This represents the adjustment of the compliant mix ratio relative to the original mix ratio in the j-th output dimension. This component can be positive or negative: a positive value indicates that the original value is adjusted upwards during the compliance correction process (e.g., the unit water consumption is increased to meet the lower limit constraint), while a negative value indicates that it is adjusted downwards.

[0116] In practical engineering applications, the offset calculation module usually also calculates the following auxiliary quantities simultaneously to simplify the subsequent statistical process of risk characteristics:

[0117] Offset magnitude: This is the norm of the correction offset, used to measure the overall strength of the correction for this production batch. It can be the Euclidean norm or the weighted norm.

[0118] Single-dimensional trigger event: For each dimension j, define a trigger indicator variable. :

[0119] (5)

[0120] in, This represents the offset threshold for the j-th output dimension (determined by engineering experience or system resolution). This indicates an indicator function; the value is 1 if the condition within the parentheses is true, and 0 otherwise. This variable... Used to indicate whether the dimension has exceeded the threshold correction.

[0121] The aforementioned auxiliary quantities are not essential for step S3, but they provide direct input for calculating the trigger rate and offset strength in subsequent step S4, avoiding repeated traversal of window data and thus improving computational efficiency. In practice, whether to calculate them together can be determined based on system resources and design preferences.

[0122] The correction offset directly reflects the degree of conflict between the model output and engineering constraints: if the correction offset is a zero vector, it means that the original mix proportion itself already meets all engineering constraints, requiring no correction, and the model output is completely consistent with the engineering requirements. If the correction offset is a non-zero vector, it means that the original mix proportion violates at least one engineering constraint, and compliance correction must adjust it. In this case, The magnitude and direction of the values ​​contain risk information: a large absolute value means a serious deviation from engineering constraints, a continuous offset in the same direction may indicate a systematic bias, and offsets in multiple dimensions at the same time may reflect raw material fluctuations or model mismatch.

[0123] In online production scenarios lacking real-time feedback of actual performance test values, the correction offset, as the only observational data that can be stably obtained within the same industrial control link, assumes the surrogate role of replacing the traditional "prediction-actual error". It is based on this surrogate quantity that subsequent steps can extract multi-dimensional risk characteristics through statistical analysis, thereby driving state machine decisions and hard control execution.

[0124] After completing the offset calculation, the offset calculation module will pass the corrected offset along with the batch's identification information (such as work order number and timestamp) to the following modules:

[0125] Risk Feature Calculation and Assessment Module: Used for extracting risk features within a scrolling statistics window; Audit Traceability Module: Persistently stored as part of audit information, supporting subsequent quality traceability and version rollback.

[0126] Meanwhile, if a rollback strategy is adopted in step S2 (i.e., the compliant mix ratio is not obtained through optimization), the correction offset will be calculated based on the compliant mix ratio after rollback and the original mix ratio. This offset also has engineering significance, reflecting the deviation between the original output and the safety benchmark.

[0127] For example, the unit water consumption in the original mix proportion of a certain batch After compliance correction, it was obtained Then correct the offset. Assuming this batch involves a total of 6 mixing ratio parameters, and the remaining parameters remain unchanged before and after correction (such as sand ratio, water-cement ratio, etc.), then the complete correction offset vector is: .

[0128] Simultaneously, the offset amplitude was calculated to be 5 (absolute value), and the triggering event for the unit water consumption dimension was marked. (Assuming) Other dimensions trigger events are 0. These data will be included in the rolling statistics window as contributions to this sample.

[0129] Step S4: Within a rolling statistical window that includes multiple recent production batches, calculate risk characteristics based on the corrected offset, including at least the corrected trigger rate and offset intensity. After normalizing and weighting the risk characteristics, obtain the comprehensive engineering risk value through monotonic mapping.

[0130] After calculating the correction offset for each production batch, the system executes step S4 through the risk feature calculation and evaluation module. The core of step S4 is to perform statistical analysis on the correction offset of all samples within a dynamically updated rolling statistical window, extract multi-dimensional risk features that can characterize the consistency between the model output and engineering constraints, and synthesize them into a quantitative engineering risk value through normalization, weighted fusion, and nonlinear mapping, providing input for subsequent state machine decision-making.

[0131] The system maintains a rolling statistical window of length n to store the corrected offset vectors of the most recent n production batches. The window length n is a configurable parameter, typically ranging from 10 to 200. The specific value needs to be determined based on the production cycle time, raw material fluctuation frequency, and computing resources. For example, in a continuous production scenario, n=50 can be used.

[0132] The update step size 'b' of the scroll statistics window can also be configured:

[0133] When b=1, the window slides forward once for each new production batch, that is, the oldest sample is removed and the newest sample is added; when b>1, the window slides once for every b batches added, which is suitable for scenarios where you want to reduce the calculation frequency.

[0134] When restarting the system after initialization or production interruption, if the actual number of samples in the window is less than the preset minimum number of samples (e.g., 10), risk calculation will not be performed temporarily, or all risk features will be set to zero to avoid random errors caused by small sample statistics.

[0135] For each output dimension j (j=1,2,…,m) within the rolling statistical window, the system calculates the following basic risk characteristics based on the corrected offset of that dimension across all samples:

[0136] Correction trigger rate: This characterizes the frequency with which the correction offset of the j-th output dimension exceeds the engineering constraint threshold within the rolling statistics window. Its calculation formula is as follows:

[0137] (6)

[0138] in, This represents the correction trigger rate for the j-th output dimension; n represents the number of samples in the current scrolling statistics window. This represents the correction offset of the i-th sample in the j-th output dimension; This represents the offset threshold for the j-th output dimension; This indicates an indicator function; the value is 1 if the condition within the parentheses is true, and 0 otherwise. The correction trigger rate ranges from [0,1]. The larger the value, the more frequently the threshold correction occurs in this dimension, and the higher the risk.

[0139] Offset Intensity: The offset intensity, based on the triggering conditions, further considers the magnitude of the over-threshold correction. Its calculation formula is as follows:

[0140] (7)

[0141] in, This represents the offset strength of the j-th output dimension; This represents the scaling factor of the j-th output dimension, used to normalize the offset magnitude so that the offset intensity of different dimensions is comparable.

[0142] The scale factor can be determined in one of the following two ways:

[0143] Historical standard deviation method: Take the sample standard deviation of the corrected offset for this dimension under normal production conditions (i.e., when the system is in a normal state and the risk value is stable); Engineering range method: Directly take 1 / 4 to 1 / 6 of the allowable engineering range width for this dimension. For example, if the allowable range for unit water consumption is 170~190 kg / m³ 3 Therefore, the scale factor can be taken as 5 kg / m. 3 .

[0144] The offset intensity comprehensively reflects the average magnitude of the over-threshold correction. A larger value indicates not only frequent triggering but also severe offset. The offset threshold is the benchmark for determining whether an "over-threshold correction" has occurred, and its setting should reflect the project's tolerance for parameter fluctuations. In practical implementation, one of the following rules can be used to determine it:

[0145] Based on the weighing system resolution: Take P times the weighing system resolution, where P∈[1,3]. For example, if the weighing resolution per unit water consumption is 1kg, then the offset threshold can be taken as 2.

[0146] Based on the allowable deviation ratio: take Q times the allowable deviation, where Q∈[0.05,0.20]. For example, if the allowable deviation of the sand ratio is ±2%, then the offset threshold can be taken as 0.4% (corresponding to Q of 0.2).

[0147] Absolute threshold: For proportional parameters (such as sand ratio), absolute values ​​can also be used directly, such as an offset threshold of 1%.

[0148] To more comprehensively characterize risk patterns, one or more derived risk features can be calculated in addition to the basic risk features. It should be noted that even without calculating derived features, an effective risk quantification scheme can be constituted solely based on the modified trigger rate and offset strength; however, introducing derived features can enhance the ability to identify complex risks.

[0149] The correction continuity index is used to characterize the trend of adjacent samples on the same dimension maintaining a consistent correction offset direction and continuously exceeding the threshold. The specific calculation formula is as follows:

[0150] (8)

[0151] in, This represents the modified continuity index for the j-th output dimension; , These represent the correction offsets of the i-th and i+1-th samples in the j-th output dimension, respectively. Represents the logical AND operation.

[0152] When the correction continuity index is close to 1, it indicates the existence of a persistent, systematic bias in the same direction.

[0153] The coupling correction strength index is used to characterize the coupling risk of different dimensions simultaneously exceeding the threshold correction in the same sample. The specific calculation formula is as follows:

[0154] (9)

[0155] in, This represents the coupling correction strength index between the j-th output dimension and the k-th output dimension; This represents the correction offset of the i-th sample in the k-th output dimension.

[0156] The corrected offset dispersion index is used to reflect the degree of dispersion in the direction and magnitude distribution of the corrected offset vectors among the samples within the rolling statistical window:

[0157] If all corrected offset vectors within the rolling statistics window are zero vectors, then the corrected offset dispersion index is 0; otherwise:

[0158] (10)

[0159] CDI stands for Corrected Offset Dispersion Index; Let represent the cosine similarity between the p-th sample and the q-th sample after correction. The cosine similarity is zero when the norm of any correction offset vector is zero.

[0160] Since different risk characteristics have different dimensions and numerical ranges, they must be normalized before weighted fusion to map them to a uniform [0,1] interval. Step S4 uses a quantile normalization method based on historical distribution, as follows:

[0161] The raw statistics for a certain risk characteristic (For example, adjusting trigger rate, offset strength, etc.), let the low quantile and high quantile of its historical distribution be Q, respectively. L and Q H (For example, if we take 0.1 and 0.9 respectively), then the normalized value is:

[0162] (11)

[0163] The `clip()` function restricts the calculation result to the range [0,1], assigning 0 to values ​​less than 0 and 1 to values ​​greater than 1. The lower and higher quantiles can be dynamically updated based on a historical audit database (e.g., recalculated daily or weekly) or preset to fixed values ​​based on engineering experience.

[0164] Through this normalization, each risk characteristic is transformed into a dimensionless relative risk index, with the value closer to 1 indicating a higher risk level and the value closer to 0 indicating a lower risk level.

[0165] After obtaining the normalized values ​​of all risk characteristics, the system performs a linear weighted summation, and then obtains the final comprehensive engineering risk value through an exponential monotonic mapping. The specific calculation formula is as follows:

[0166] (12)

[0167] Where R represents the comprehensive engineering risk value; L represents the number of risk characteristics; This represents the weighting coefficient of the l-th risk characteristic. ; This represents the normalized value of the l-th risk feature.

[0168] Weighting coefficient This reflects the importance of each risk characteristic in the overall risk assessment and can be determined in the following ways:

[0169] Expert experience method: Concrete process experts directly assign values ​​based on the impact of each risk characteristic on project safety; Historical calibration method: Through regression analysis of historical high-risk events, the weights are optimized so that R can best distinguish between normal and abnormal states; Adaptive adjustment: During system operation, the weights are dynamically fine-tuned based on the actual control effect.

[0170] For the coupling correction strength index, since it already contains a weighted sum of multiple coupling pairs, it is usually treated as a composite feature: let there be C pairs of coupling dimensions, then it corresponds to a composite feature. ,in This is the normalized coupling correction strength index. These are the internal weighting coefficients. In the external weighted fusion, the weights corresponding to this composite feature are... It is usually set to 1 to avoid excessive weighting.

[0171] The comprehensive engineering risk value has the following excellent properties:

[0172] R∈[0,1), when all normalized risk features are 0, R=0, indicating no risk; R increases monotonically with any risk feature; when risk features accumulate continuously, R approaches 1, but will never reach 1, which conforms to the saturation characteristic of risk perception; in the low-risk range (when R is small), the exponential mapping is approximately linear and can respond sensitively to changes in risk; in the high-risk range, the exponential mapping naturally saturates, avoiding excessive amplification of noise.

[0173] For example, suppose a scrolling window has n=50 elements and dimension m=6. Statistical analysis shows the correction trigger rate for the unit water consumption dimension (j=1). Offset intensity Simultaneously, the corrected continuity CCI1 = 0.7 and the coupling correction strength CCS for unit water consumption and admixture dosage were calculated. 12 =0.4. Assuming the historical quantiles of each risk characteristic are known, after normalization, we get:

[0174] RF1= =0.9, RF2= =0.7, RF3=CCI1'=0.8, RF4= =0.5 (after normalization);

[0175] With corresponding weight coefficients set to 0.3, 0.3, 0.2, and 1.0 (note that the outer weight of the coupling feature is 1), the weighted sum is 1.14. The comprehensive engineering risk value R = 1 - exp(-1.14) ≈ 0.68. This value R = 0.68 will be input into the state machine in step S5 to determine the production control state of the current window.

[0176] Figure 2 The principle of online risk feature aggregation and calculation based on a scrolling window is demonstrated. Figure 2 The following core modules are presented in a hierarchical flowchart format, from left to right:

[0177] Real-time offset correction data stream: The corrected offset generated by each production batch continuously flows into the rolling statistics window, serving as the raw data source for all subsequent risk feature calculations.

[0178] Risk Feature Extraction:

[0179] Trigger Frequency and Intensity Statistics: Statistics are performed on each output dimension within the window to calculate the corrected trigger rate and offset intensity, reflecting the frequency and magnitude of model output deviations from engineering constraints. Correction Continuity: A correction continuity index is calculated by jointly determining the offset direction of adjacent samples and the state exceeding the threshold, characterizing the trend of systematic deviations. Coupling Correction Strength: Coupling risk verification is performed on events that simultaneously exceed the threshold in different dimensions, calculating the coupling correction strength index to identify the linkage risk between parameters. Offset Dispersion: A corrected offset dispersion index is calculated through cosine similarity analysis between sample offset vectors within the window, reflecting the degree of concentration or dispersion of risk distribution.

[0180] Normalization mapping: The various risk features (including basic features and derived features) calculated above are normalized based on the low and high quantiles of the historical distribution to eliminate the influence of dimensions and map each feature to a unified [0,1] interval.

[0181] Engineering weighting and fusion: The normalized features are linearly weighted and summed according to the preset weight coefficients, and then the comprehensive engineering risk value is obtained through exponential monotonic mapping. This value is used as the final output of the current window to characterize the overall risk level of the model output relative to the engineering constraints.

[0182] Sample Behavior Consistency Analysis: The comprehensive engineering risk value reflects the overall consistency of the correction behavior of all samples within the window, providing a quantitative basis for subsequent three-state machine decision-making.

[0183] Step S5: Input the comprehensive engineering risk value into a three-state machine with hysteresis mechanism. The three-state machine switches between normal state, degraded state and circuit breaker state according to the preset threshold and hysteresis condition and generates control decisions.

[0184] After obtaining the comprehensive engineering risk value R of the current rolling statistical window, the system executes this step S5 through the production control state machine module. The core of step S5 is to input the quantified risk value R into a three-state state machine with a hysteresis mechanism. The state machine switches between three predefined production control states according to the preset thresholds (risk thresholds) and hysteresis conditions, and outputs corresponding control decisions, providing an instruction basis for subsequent industrial hard control execution.

[0185] The three-state state machine includes three mutually exclusive production control states, corresponding to different risk levels and control strategies respectively:

[0186] Normal state (NORMAL): It indicates that the engineering risk of the current production process is at a low level, the consistency between the model output and engineering constraints is good, and the production can be carried out normally using the compliance-adjusted mix ratio (i.e., the compliant mix ratio). In the normal state, the system trusts the model output and does not intervene.

[0187] Degraded state (DEGRADED): It indicates that the risk is at a medium level, and the model output has shown a certain degree of unreliability (such as frequent or large-scale correction offsets), but it has not reached the severity that must be immediately fused. In the degraded state, the system takes preventive control measures, prohibits the use of the original mix ratio output by the model, and instead switches to the backup mix ratio (such as a conservative template or a historically stable version) to prevent potential risks from expanding.

[0188] Fuse state (FUSE): It indicates that the risk has reached a dangerous level, and the model output seriously deviates from the engineering constraints. Continuing to use it may cause quality accidents. In the fuse state, the system executes an emergency fusing operation, blocks all mix ratio writing channels, and makes the production execution unit maintain the currently written mix ratio until the risk is lifted.

[0189] There are one-way or two-way migration paths between the three states, but not any two states can be directly switched. Instead, they strictly follow the rules based on the risk value R and the preset thresholds.

[0190] To achieve state switching, the system pre-sets three key risk thresholds:

[0191] The first threshold T1: Used to distinguish the boundary between the normal state and the degraded state. When R < T1, it is determined as low risk and should be in the normal state; when R ≥ T1, the risk enters the medium and above regions. The second threshold T2: Used to distinguish the boundary between the degraded state and the fuse state. When R ≥ T2, it is determined as high risk and should enter the fuse state. The recovery threshold Tr: Used to control the condition for recovering from a high-risk state (degraded state or fuse state) to a low-risk state. Only when R continuously remains below Tr is the state machine allowed to recover upward.

[0192] The above thresholds must satisfy a strict magnitude relationship: Tr < T1 < T2. This relationship ensures the effectiveness of the hysteresis mechanism: the recovery threshold being lower than the first threshold means that recovering from a high-risk state requires a lower risk level than entering, thus avoiding frequent state switches due to minor fluctuations in the risk value.

[0193] The specific values of the thresholds can be determined based on engineering experience and historical data. A typical setting method is as follows:

[0194] Collect statistical data on the comprehensive engineering risk value R during historical normal production; set the first threshold T1 as the 85th percentile of R (i.e., 85% of the normal sample risk values are lower than this value); set the second threshold T2 as the 98th percentile of R; set the recovery threshold Tr as the 70th percentile of R. In actual deployment, the thresholds can also be directly specified by domain experts according to the risk tolerance and adjusted dynamically according to feedback during system operation.

[0195] In the case of no hysteresis mechanism, the state machine determines the next state according to the comparison result of the comprehensive engineering risk value R of the current window with the threshold, following the following rules:

[0196] Enter or remain in the normal state: If R < T1, the state machine enters or remains in the normal state; enter or remain in the degraded state: If T1 ≤ R < T2, the state machine enters or remains in the degraded state; enter the fuse state: If R ≥ T2, the state machine enters the fuse state (note: it cannot directly enter the degraded state from the fuse state and must go through the recovery process).

[0197] However, if executed exactly according to the above rules, when the risk value R fluctuates near the threshold boundary, the state machine may frequently switch back and forth between the normal state and the degraded state or between the degraded state and the fuse state, resulting in unstable production control (i.e., the "jitter" phenomenon). Therefore, the present invention introduces a hysteresis mechanism. The core of the hysteresis mechanism is that when recovering from a high-risk state (degraded state or fuse state) to a low-risk state, it cannot switch immediately just because R of a single window is lower than the threshold, but must meet a more stringent condition - R of multiple consecutive rolling statistical windows is lower than the recovery threshold Tr.

[0198] Specifically, the hysteresis recovery condition is defined as follows:

[0199] When the state machine is in the fuse state, if the comprehensive engineering risk value R of consecutive Kr rolling statistical windows all satisfies R < Tr, the state machine exits the fuse state and enters the normal state; when the state machine is in the degraded state, if the comprehensive engineering risk value R of consecutive Kr rolling statistical windows all satisfies R < Tr, the state machine also exits the fuse state and enters the normal state. Among them, Kr is a preset positive integer, called the hysteresis window number, and the typical value range is 3 to 10 (for example, Kr = 5). The larger Kr is, the higher the recovery threshold, the more stringent the system's requirement for continuous risk mitigation, and the stronger the anti-jitter ability.

[0200] It should be emphasized that the switch from the low-risk state to the high-risk state (i.e., the risk increase process) is not restricted by the hysteresis mechanism and still executes immediately according to the basic rules. This is because risk increase requires timely response and should not be delayed.

[0201] Combining the basic switching rules and the hysteresis recovery conditions, the complete state transition logic of the three-state state machine can be summarized as follows:

[0202] The current state is the normal state: if R < T1, the normal state is maintained; if T1 ≤ R < T2, it is immediately switched to the degraded state (risk increase); if R ≥ T2, it is immediately switched to the fuse state (risk sharply increases).

[0203] The current state is the degraded state: if R ≥ T2, it is immediately switched to the fuse state (risk further increases); if T1 ≤ R < T2, the degraded state is maintained; if R < T1, it does not immediately recover, but starts to count the number of windows where R < Tr is continuously satisfied (note: the recovery threshold Tr < T1, so R < T1 does not guarantee R < Tr). Only when R of consecutive Kr windows all satisfy R < Tr, is it allowed to recover to the normal state; otherwise, as long as R ≥ Tr in one window, the counter is cleared and it continues to stay in the degraded state.

[0204] The current state is the fuse state: if R ≥ T2, the fuse state is maintained; if R < T2, it cannot immediately recover and must wait until the condition that R < Tr of consecutive Kr windows is satisfied before it is allowed to exit the fuse state and enter the normal state.

[0205] Through the above hysteresis mechanism, the state machine can effectively filter short-term fluctuations of the risk value, ensure that production resumes to the normal mode only after the risk is truly mitigated, thereby improving the stability and robustness of the system.

[0206] Each time the state machine switches states or remains in a certain state, it generates a clear control decision, which will be passed to the subsequent industrial control interface module (step S6). The control decision at least includes the following information: current state identifier: normal state, degraded state, or fuse state; recommended action type: such as "release", "switch write source", "freeze write"; additional parameters (optional): such as when it is necessary to switch the write source, specify which specific standby mix ratio (conservative template or stable version) to use.

[0207] The state machine itself does not directly execute industrial control operations, but outputs in the form of decision signals, and a dedicated industrial control interface module is responsible for converting them into specific PLC instructions. This separated design enhances the modularity and scalability of the system.

[0208] Exemplarily, assume that the risk thresholds set for a certain concrete mixing plant are: Tr = 0.4, T1 = 0.6, T2 = 0.9, and the number of hysteresis windows Kr = 5. The current state is the normal state.

[0209] Scenario 1 (risk increase): The risk values of several consecutive windows rise to R = 0.7, satisfying T1 ≤ R < T2. The state machine immediately switches to the degraded state and generates a decision of "switch write source".

[0210] Scenario 2 (risk fluctuation): In the degraded state, the R value of a certain window drops to 0.55 (less than T1 but greater than Tr). Since R < Tr is not satisfied, the hysteresis counter does not start, and the state remains in the degraded state. The R value of the next window rises back to 0.8, still in the degraded state. This kind of fluctuation will not cause the state to recover, avoiding frequent switching.

[0211] Scenario 3 (continuous risk mitigation): In the degraded state, the R values of 6 consecutive windows are: 0.38, 0.35, 0.30, 0.28, 0.25, 0.22. The first 5 windows all satisfy R < Tr = 0.4, so the counter reaches Kr at the end of the 5th window. The state machine resumes to the normal state at the beginning of the 6th window. At the same time, a decision of "resume release" is generated.

[0212] Scenario 4 (fuse and recovery): The R value of a certain window suddenly rises to 0.95 ≥ T2. The state machine immediately enters the fuse state and generates a decision of "freeze write". Subsequently, the risk gradually decreases, but the R values of the first 4 windows are lower than T2 but still higher than Tr (such as 0.5, 0.45, 0.42, 0.41), not meeting the recovery conditions. Subsequently, the R values of 5 consecutive windows are all lower than Tr (such as 0.38, 0.35, 0.32, 0.30, 0.28). At this time, the hysteresis recovery conditions are met, the state machine exits the fuse state, and directly resumes to the normal state according to the current R value, generating a decision of "unfreeze".

[0213] Figure 3 It shows the dynamic evolution process of the comprehensive engineering risk value R with the production batch sequence, as well as the state transitions of the three-state state machine with a hysteresis mechanism under different risk thresholds. Figure 3 The abscissa of is the production batch sequence i (arranged in chronological order), and the ordinate is the comprehensive engineering risk value R (the value range is [0, 1)). Figure 3 It intuitively shows the three-state state machine and the hysteresis mechanism described in the present invention:

[0214] Risk increase stage: In the early stage of the batch sequence, R gradually rises and exceeds T1, and the state machine switches from the normal state to the degraded state; then R continues to rise and exceeds T2, and the state machine immediately enters the fusing state. This process reflects the immediate response characteristic of risk increase.

[0215] Hysteresis recovery stage: In the later stage of the batch sequence, R begins to decline. First, it drops below T1, but at this time the state machine does not immediately recover. Instead, it continues to remain in the fusing state (the area marked "in hysteresis" in the figure). It is not until R is continuously lower than the recovery threshold Tr for multiple windows that the state machine exits the fusing state and returns to the normal state. This process reflects the filtering effect of the hysteresis mechanism on short-term risk fluctuations, effectively preventing frequent state switches.

[0216] Figure 3 It clearly shows the setting relationship of Tr < T1 < T2 and the core role of the recovery threshold in the hysteresis mechanism.

[0217] Step S6: Issue a hard control instruction corresponding to the control decision.

[0218] In step S5, the three-state state machine with a hysteresis mechanism generates corresponding control decisions (including the current state and the recommended action type) according to the comparison result of the current comprehensive engineering risk value R and the preset threshold. Step S6 is executed by the industrial control interface module. Its core task is to map the abstract control decision into an executable instruction for the production controller or programmable logic controller (PLC), and issue it to the underlying execution unit through the industrial communication protocol, so as to directly intervene in the mixing ratio writing process.

[0219] The industrial control interface module serves as a bridge connecting the upper-level decision-making system and the lower-level industrial control equipment. Its input receives control decisions from the state machine, and its output exchanges data with the production controller or PLC via industrial communication networks (such as Modbus, Profibus, OPCUA, EtherNet / IP, etc.). Depending on the control decision, this module can issue at least one of the following types of instructions: Mixing ratio setpoint / formula task sheet write instruction: writes the parameter values ​​of the compliant or backup mixing ratio to the specified storage location; Write freeze / write protection instruction: blocks all mixing ratio write operations, ensuring the production execution unit maintains the currently executing mixing ratio; Formula version switching instruction: switches the write source to a preset conservative template mixing ratio or a stable version mixing ratio determined by audit traceability information; Quality inspection strategy adjustment instruction (optional): adjusts the online detection frequency or sampling rules according to the risk level.

[0220] The above instructions can be written to at least one of the following: the recipe register of the batching and weighing control unit, the recipe table ID, or a task sheet field. Freezing can be achieved by setting the write protection flag, closing the write channel, or locking the write value to the last safe value. Switching can be achieved by issuing at least one of the following: a specified recipe version number, a template ID, or a stable version ID.

[0221] Based on the three states of the tri-state machine, the industrial control interface module executes the corresponding hardware control instructions. The specific mapping relationship is as follows:

[0222] Normal State: When the state machine is in the normal state, it indicates that the current risk is low and the model output is in good agreement with the engineering constraints. At this time, the industrial control interface module executes the write release command:

[0223] Write Enable: Ensure the write channel is open; Write Source Selection: Point the write source to the compliant mix ratio output by this system; Parameter Write: Write each component of the compliant mix ratio vector to the corresponding recipe register, recipe table ID, or task sheet field of the PLC; Execution Confirmation: Read back the written value or read the status word returned by the PLC to confirm successful writing.

[0224] Degraded State: When the state machine is in a degraded state, it indicates that the model output has a moderate risk and the original model output should not be used directly. At this time, the industrial control interface module executes the write source switching instruction:

[0225] Write Source Selection: Switches the write source to the backup mix design. Backup mix designs include: a preset conservative template mix design: a safe mix design pre-configured by concrete process experts that meets engineering constraints under all working conditions; and a stable version mix design determined by audit traceability information: a mix design version identified by the system from the audit database where multiple consecutive windows were in a normal state, the risk value was below the first threshold, and the write operation was successful. Parameter Write: Writes the selected backup mix design parameters to the corresponding storage location on the PLC. Block Confirmation: Ensures the original mix design is no longer written (this can be achieved by setting the write protection flag or closing the original channel).

[0226] Fuse State: When the state machine is in the fuse state, it indicates that the risk has reached a dangerous level, and all mix ratio writing operations must be immediately interrupted. At this time, the industrial control interface module executes the write freeze command:

[0227] Write Blocking: By setting the write protection flag, disabling the write channel, or clearing the write enable bit, the PLC rejects any new mix ratio write requests. Current Value Maintenance: Ensures the PLC maintains the last successfully written mix ratio (which can be a compliant mix ratio, a conservative template, or a stable version) before the circuit breaker is triggered as the currently executed mix ratio. Status Indication: Optionally displays the "Fuse Broken" status on the HMI. Waiting for Recovery: Continuously monitors the state machine status; once the state machine exits the circuit breaker state and generates a release decision, the release operation is executed.

[0228] In one specific implementation, the aforementioned write, freeze, and switch controls can be mapped to a set of control variables in a production control system, interacting with the PLC via an industrial protocol. This mapping method includes:

[0229] Write enable bit ben: Used to control the opening and closing of the write channel. When ben=1, write operations are allowed; when ben=0, any write operation is prohibited, achieving write freeze.

[0230] Write source selection bit bsrc: Used to select the source of the current mix ratio. When bsrc=1, the write source is the compliant mix ratio output by this system; when bsrc=0, the write source is switched to the backup mix ratio (such as the preset conservative template mix ratio or the stable version mix ratio determined by audit traceability information).

[0231] Parameter register: Used to store the specific value of the j-th mixing ratio parameter (e.g., unit water consumption, water-cement ratio, sand ratio, etc.). Each parameter corresponds to an independent register or register group. The PLC or weighing control unit directly reads the set value from these registers to execute production.

[0232] The above variables can be mapped to PLC coil or register addresses via industrial protocols (such as Modbus).

[0233] To ensure the reliable execution of hard control commands, industrial control interface modules must have the following mechanisms:

[0234] Post-write verification: After writing critical parameters, the module immediately reads back for verification. If a mismatch is found, it retryes. If the retry fails, an exception is reported and an audit log is triggered. Communication failure handling: If communication with the PLC is interrupted, the module should cache the instruction and continuously attempt to reconnect. Once communication is restored, the instruction will be executed synchronously. Safety rollback: In extreme cases (such as when the PLC is unresponsive), the write enable is forcibly reset to zero via hardwiring.

[0235] Each instruction execution (including success, failure, or retry) must be recorded in the audit traceability information with detailed information, including at least:

[0236] Timestamp: The time the instruction was issued; Work Order Number: The unique identifier of the current production batch; Threshold Version: The version number of the currently used risk threshold configuration; Model Version: The version number of the data-driven model used to generate the original mix ratio; Control Status: The current state of the state machine (normal / degraded / circuit-breaker); Control Action: The specific instruction type (release / switch / freeze) and parameter values ​​executed; Execution Result: Success / failure and reason; PLC Feedback: Readback verification results, communication status, etc.

[0237] Step S7: Record and link the audit traceability information for each production batch to support quality traceability and version rollback.

[0238] After completing the entire process from obtaining the original mix ratio to issuing the hard control command for the current production batch, the system executes step S7 through the audit traceability module. The core task of step S7 is to persistently store the key data of each production batch in a structured and correlated manner, forming complete audit traceability information, and providing reliable data support for subsequent quality traceability, responsibility identification, system optimization, and automatic rollback to the previous stable version.

[0239] The audit traceability module operates independently of other functional modules, spanning the entire online assessment and closed-loop control process. It doesn't just start working in step S7; instead, it continuously collects data at each stage from steps S1 to S6. Once all operations in the current batch are completed, the collected information is integrated into a complete audit record and written to the audit database.

[0240] The design of the audit traceability module follows these principles:

[0241] Completeness: Records cover the entire data chain from model output to control execution, ensuring traceability at any stage. Relevance: All information within the same batch is linked through a unique identifier (such as a work order number), supporting cross-table queries and joint analysis. Immutability: Once audit records are written, they can only be appended to, ensuring the legal validity of the data. Readability: The data storage format should facilitate manual review and program analysis, typically using structured databases (such as SQLite, MySQL) or time-series databases.

[0242] Audit traceability information should include at least the following:

[0243] Work Order Number: The work order number is a unique business identifier for each production batch, generated by the production management system (such as MES or ERP). This number serves as the primary key for all traceability information, used to associate all data for that batch.

[0244] Original mix proportions: The original mix proportion vector output by the data-driven model for the current batch. It should include the specific values ​​and units for all dimensions when recorded.

[0245] Compliant Mix Ratio: The mix ratio vector that meets engineering constraints after compliance correction in step S2. If the correction fails and a rollback strategy is adopted, the actual conservative template or stable version mix ratio used is recorded here, along with a correction failure flag.

[0246] Correction offset: The vector of difference between the compliant mix ratio and the original mix ratio.

[0247] Comprehensive engineering risk value R: The comprehensive engineering risk value of the current rolling window calculated in step S4.

[0248] Model version: A version identifier for the data-driven model that generated the original matching ratios. The model version number should uniquely correspond to a set of model files (including weights, architecture, training data range, etc.) so that the model output at that time can be reproduced when needed.

[0249] Threshold Version: The version identifier of the currently used risk threshold configuration. The threshold version number should uniquely correspond to a set of parameters, including but not limited to: offset threshold, scaling factor, risk thresholds T1 / T2 / Tr, hysteresis window number Kr, and weighting coefficients when calculating the R value.

[0250] Control State: The final output state of the three-state machine in step S5, i.e., normal state, degraded state, or circuit breaker state. It is recorded as an enumeration value or corresponding numerical code (e.g., 0-normal, 1-degraded, 2-circuit breaker).

[0251] Control actions: The actual operations performed by the industrial control interface module in step S6 include: releasing compliance ratio in normal state, switching write source in degraded state, and freezing write in circuit breaker state.

[0252] To enhance traceability, the following information can be added: timestamp (batch start time, completion time of each step), raw material batch (batch number of cement, sand, gravel, and admixtures used in the previous batch), environmental conditions (temperature, humidity, etc. during production), input feature snapshot (feature vectors input to the data-driven model (such as raw material moisture content, target slump, etc.) to facilitate the reproduction of model reasoning scenarios), and execution results (whether the PLC writing was successful, whether the readback verification passed, etc.).

[0253] The audit traceability module organizes the above information in a structured form, with each record corresponding to a production batch.

[0254] The second core use of audit traceability information is version rollback, which means that when the system recovers from a degraded or circuit breaker-triggered state, it can automatically identify and switch to a reliable and stable version. The implementation mechanism is as follows:

[0255] When it is necessary to switch to the "stable version mix ratio determined by audit traceability information" in the backup mix ratio, the system executes the following query logic: filter records from the audit database within the most recent period (e.g., the past 24 hours); filter out batches where multiple consecutive windows (e.g., 5) are in a normal state and the corresponding comprehensive engineering risk value R is less than the first threshold T1; among these batches, select the latest consecutive batches in time, and take the compliant mix ratio of the last batch as the stable version mix ratio; record the model version and threshold version of this stable version to ensure context consistency during subsequent switches.

[0256] When the state machine is in a degraded state and decides to switch to a stable version, or when it needs to be released again after recovering from the circuit breaker state, the system: reads the value of the stable version mix ratio from the audit database; at the same time, records the model version and threshold version of the stable version used this time (although the model may have been updated at this time, the version information at that time is still used for traceability during rollback); and writes the stable version mix ratio into the PLC through the industrial control interface.

[0257] Each version rollback event must also be recorded in the audit traceability information, including: the reason for the rollback (downgrade state switching / circuit break recovery), the target version to be rolled back (such as "stable version v3" and its corresponding batch number), and the changes in control status before and after the rollback.

[0258] Example 2

[0259] Taking a certain intelligent concrete mixing plant as an example, the mix proportion vector includes six parameters (unit water consumption, water-cement ratio, sand ratio, total amount of cementitious materials, admixture dosage, and air content), and the rolling statistical window length is set to 50 production batches. At a certain moment, due to fluctuations in cement batches, the unit water consumption predicted by the artificial intelligence model is consistently lower than expected. The specific process is as follows:

[0260] Acquisition and Correction: For the current batch, the unit water consumption in the original mix proportion output by the model is 175 kg / m³. 3 After optimization by the compliance correction module based on engineering constraints (such as volume conservation and water-cement ratio range), the unit water consumption in the compliant mix design is adjusted to 180 kg / m³. 3 This results in a corrected offset of +5 kg / m. 3 .

[0261] Risk quantification: Set the offset threshold for unit water consumption as 1 kg / m³ 3 The scale factor is 5 kg / m 3 Within a window of 50 samples, the unit water consumption correction offset for all samples exceeded the offset threshold, thus the correction trigger rate was 1; and the correction offset for each triggered sample was 5 kg / m³. 3 Therefore, the offset intensity is 1. Within the rolling statistical window, the number of adjacent pairs of samples that are in the same direction and both exceed the offset threshold is 49, so the corrected continuity index is 1. Considering the coupling risk of unit water consumption and admixture dosage, there are 10 samples within the window that simultaneously exceed the offset threshold for both, and the average coupling intensity is 0.5, so the coupling correction intensity index is 0.1. At the same time, the cosine similarity of the offset vectors within the window yields a corrected offset dispersion of 0.2. After normalizing the above features (using the original values ​​directly in this embodiment), the weight coefficients are taken as 0.20, 0.20, 0.15, 0.25, and 0.50, respectively, and substituted into the comprehensive engineering risk value formula:

[0262] R=1−exp(−(0.20×1+0.20×1+0.15×1+0.25×0.2+0.50×0.1))=1−exp(−0.65)≈0.48

[0263] State transition and control: Preset risk thresholds T1=0.6, T2=0.9, recovery threshold Tr=0.5, and hysteresis window number Kr=1. When the comprehensive engineering risk value rises to 0.7 (greater than T1) due to increased coupling and dispersion in subsequent windows, the three-state machine with hysteresis mechanism switches from the normal state to the degraded state. The industrial control interface module then switches the compliant mix ratio output from the model to the preset conservative template mix ratio via the PLC, thereby preventing high-risk mix ratios from entering the production execution chain.

[0264] Audit Log: The system generates an audit traceability log for this batch, which includes at least the work order number, timestamp, original mix ratio, compliant mix ratio, corrected offset, comprehensive engineering risk value, model version, threshold version, control status (degraded state), and executed control actions (switching write source) for post-event analysis and quality traceability.

[0265] Hysteresis Recovery: After manual intervention and replacement of raw materials, the model output gradually stabilized. The system monitored that the comprehensive engineering risk value for 10 consecutive windows was below the recovery threshold Tr=0.5, meeting the hysteresis recovery condition. The state machine automatically recovered from the degraded state to the normal state, and the original mix ratio output based on the data-driven model was reactivated, achieving safe hysteresis.

[0266] Example 3

[0267] This invention also provides an online assessment and closed-loop control system for the reliability of concrete mix design. The system 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 online assessment and closed-loop control method for the reliability of concrete mix design in this invention.

[0268] Although not shown, the system 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 the RAM. The processor, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0269] 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.

[0270] In another specific embodiment, a concrete mix design reliability online assessment and closed-loop control system includes multiple functional modules, which are configured to execute the concrete mix design reliability online assessment and closed-loop control method in the embodiments of the present invention.

[0271] 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 method for online evaluation and closed-loop control of engineering reliability of concrete mix proportion, characterized in that, The method includes: Obtain the original mix proportions output by the data-driven model for the current production batch; Based on engineering constraints, the original mix proportion is modified to meet compliance requirements, resulting in a compliant mix proportion that satisfies the engineering constraints. Calculate the correction offset between the compliant mix ratio and the original mix ratio; Within a rolling statistical window that includes multiple recent production batches, risk characteristics, including at least the correction trigger rate and the offset intensity, are calculated based on the corrected offset. The risk characteristics are then normalized and weighted and fused before being monotonically mapped to obtain a comprehensive engineering risk value. The comprehensive engineering risk value is input into a three-state machine with hysteresis mechanism. The three-state machine switches between normal state, degraded state and circuit breaker state and generates control decisions based on preset thresholds and hysteresis conditions. Issuing hard control instructions corresponding to the control decision, including: when in the normal state, allowing the compliant mix ratio to the production execution unit; when in the degraded state, blocking the original mix ratio and switching the writing source to the backup mix ratio; when in the circuit breaker state, blocking all mix ratio writing, so that the production execution unit maintains the currently executed mix ratio until the hysteresis condition is met and the blocking is lifted. Record and link audit traceability information for each production batch to support quality traceability and version rollback.

2. The method according to claim 1, wherein, The engineering constraints include at least one of physical equality constraints and engineering inequality constraints; wherein, the physical equality constraints include at least one of volume conservation constraints and mass-volume conversion constraints based on material density, and the engineering inequality constraints include at least one of water-cement ratio range constraints, sand ratio range constraints, total cementitious material range constraints, and air content constraints.

3. The method according to claim 1, wherein, The corrected trigger rate and offset intensity are calculated as follows: ； ； wherein, represents the modified trigger rate of the jth output dimension; n represents the number of samples within the rolling statistics window; represents the modified offset of the ith sample on the jth output dimension; represents the offset threshold of the jth output dimension; represents an indicator function, which takes the value 1 if the condition in the parentheses is true, and 0 otherwise; represents the offset strength of the jth output dimension; represents the scale factor of the jth output dimension; The offset threshold is determined based on a multiple of the weighing system resolution or a proportion of the allowable engineering deviation; the scale factor is the standard deviation of the corrected offset under historical stable conditions or the width of the allowable engineering range.

4. The method according to claim 1, wherein, The risk characteristics also include at least one of the following: a corrected continuity index, a coupled corrected strength index, and a corrected offset dispersion index; wherein the corrected continuity index is calculated in the following manner: ； wherein, represents the modified continuity index of the jth output dimension; n represents the number of samples within the rolling statistical window; , respectively represent the modified offset of the ith, i+1th sample on the jth output dimension; represents the offset threshold of the jth output dimension; represents the indicator function; represents the logical and operation; The coupling correction strength index is calculated as follows: ； wherein, denotes a coupling correction strength indicator between the jth output dimension and the kth output dimension; denotes a correction offset of the ith sample on the kth output dimension; denotes an offset threshold of the kth output dimension; , denote scale factors of the jth and kth output dimensions, respectively; The corrected offset dispersion index is calculated as follows: If all corrected offset vectors within the rolling statistics window are zero, then the corrected offset dispersion index is 0; otherwise: ； wherein CDI represents a corrected offset dispersion index; represents the cosine similarity between the corrected offset vector of the pth sample and the qth sample, and the cosine similarity is zero when the norm of any corrected offset vector is zero.

5. The method of claim 1, wherein the method further comprises: The comprehensive engineering risk value is calculated as follows: ； Wherein, R represents the comprehensive engineering risk value; L represents the number of risk characteristics; represents the weight coefficient of the lth risk characteristic; represents the normalized value of the lth risk characteristic; When the risk characteristic is a coupling correction strength index, the corresponding weighting coefficient The value is 1, and the normalized value of the coupling correction strength index is: , This represents the normalized value of the coupling strength index between the j-th and k-th output dimensions. This represents the corresponding weighting coefficient; The normalization is achieved through monotonic mapping, specifically: for the original statistics of risk characteristics The normalized value is calculated based on the low and high quantiles of its historical distribution. Where clip() restricts the calculation result to the interval [0,1], Q L and Q H These represent the lower quantile and the upper quantile, respectively.

6. The method for online evaluation and closed-loop control of concrete mix design reliability according to claim 1, characterized in that, The preset thresholds include a first threshold, a second threshold, and a recovery threshold, and satisfy the condition that the recovery threshold < the first threshold < the second threshold; the switching rule of the three-state machine is as follows: When the overall engineering risk value is less than the first threshold, it is in or remains in a normal state; When the comprehensive engineering risk value is greater than or equal to the first threshold and less than the second threshold, it is in or remains in a downgraded state. When the comprehensive engineering risk value is greater than or equal to the second threshold, the circuit breaker is activated. The hysteresis condition includes: when the three-state machine is in a circuit breaker state or a degraded state, recovery to the normal state is only allowed after the comprehensive engineering risk value of multiple consecutive rolling statistical windows is lower than the recovery threshold.

7. The method for online evaluation and closed-loop control of concrete mix design reliability according to claim 1, characterized in that, The backup mix ratio includes a preset conservative template mix ratio or a stable version mix ratio determined by audit traceability information; the stable version mix ratio is determined by selecting a mix ratio version where multiple consecutive windows are in a normal state and the risk value is lower than a first threshold.

8. The method for online evaluation and closed-loop control of concrete mix design reliability according to claim 1, characterized in that, The audit traceability information includes at least: the original mix ratio, compliant mix ratio, corrected offset, comprehensive engineering risk value, model version, threshold version, control status, control action, and work order number for each production batch.

9. The method for online evaluation and closed-loop control of concrete mix design reliability according to claim 1, characterized in that, The hard control commands are mapped via an industrial protocol to one of the following operations on a production controller or programmable logic controller: write to the recipe register, set the write protection flag, close the write channel, or switch the recipe version number.

10. A reliable online assessment and closed-loop control system for concrete mix design, characterized in that, The system includes a module configured to perform the online assessment and closed-loop control method for the reliability of concrete mix design as described in any one of claims 1-9.

Images