Prefabricated concrete component production quality control method, device, equipment and medium

Abstract

The invention relates to a precast concrete component production quality control method, device and equipment and a medium. According to the method, a transfer rule of quality deviation is accurately described by establishing a state space model of inter-process quality deviation transfer, and real-time sensor data is utilized to drive a Kalman filtering algorithm to dynamically estimate a real-time quality state and an accumulative effect of the real-time quality state of each process. Based on the state estimation, a model prediction control algorithm is adopted to generate a technological parameter self-adaptive compensation strategy in a prospective mode, technological parameters of subsequent procedures are adjusted in real time through an execution mechanism to offset preorder deviation, and meanwhile the adjustment effect is fed back to a model for continuous self-updating; therefore, systematic suppression and compensation of the quality deviation of the whole production process of the concrete prefabricated part are achieved, the dimensional precision and strength consistency of the part are remarkably improved, the rejection rate and the rework cost are effectively reduced, and the self-adaptive capacity of a production system to working condition changes is enhanced.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent manufacturing, and particularly relates to a concrete prefabricated component production quality control method, device, equipment and medium. BACKGROUND

[0002] Concrete prefabricated component production involves mold preparation, steel bar binding, concrete pouring, vibration, curing and other processes. The quality deviation of each process (such as mold size error, steel bar position deviation, and improper concrete proportioning) will be transmitted and accumulated step by step, affecting the size accuracy, strength and service life of the final component. The existing quality control method usually detects and adjusts the single process in isolation, ignoring the interaction of the process deviation, resulting in difficulty in accurately predicting and controlling the final quality. The core of this problem lies in the lack of system modeling, real-time identification and dynamic compensation mechanism for the whole process deviation transmission. Therefore, a method integrating mathematical model, real-time monitoring and adaptive adjustment is needed to break the isolation between processes and realize whole-process collaborative control. SUMMARY

[0003] Therefore, it is necessary to provide a concrete prefabricated component production quality control method, device, equipment and medium in view of the above technical problems.

[0004] In a first aspect, the application provides a concrete prefabricated component production quality control method, comprising:

[0005] S1, based on the historical process parameters and historical quality detection data of each production process of the concrete prefabricated component, a mathematical model of the quality deviation transmission between processes is established through multivariate regression analysis to obtain a state space model containing a state transition matrix and a control input matrix;

[0006] S2, real-time acquisition of process parameters and quality data of the current production batch is performed through a sensor network deployed in each production process to obtain a real-time data set;

[0007] S3, the real-time data set is input into the state space model, and the quality state of each process is optimally estimated through a Kalman filtering algorithm to generate a state estimation vector containing a deviation amount and an accumulated influence on subsequent processes;

[0008] S4, based on the state estimation vector, a multi-objective optimization problem is solved through a model predictive control algorithm to generate an adaptive compensation strategy containing process parameter adjustment amounts and adjustment timing of each process;

[0009] S5, the adaptive compensation strategy is converted into a control instruction and issued to the execution mechanism of the corresponding process to drive the execution mechanism to complete the process adjustment according to the specified parameters;

[0010] S6, feeding production data containing mass state before and after adjustment to the state space model, updating the state transition matrix and the control input matrix through an incremental learning algorithm.

[0011] In a second aspect, the application further provides a concrete prefabricated component production quality control device for implementing the method in the first aspect, which comprises:

[0012] a quality transfer modeling module, configured to establish a mathematical model of quality deviation transfer between processes through multivariate regression analysis based on historical process parameters and historical quality detection data of each production process of the concrete prefabricated component, and obtain a state space model containing a state transition matrix and a control input matrix;

[0013] a real-time data sensing module, configured to collect process parameters and quality data of a current production batch in real time through a sensor network deployed in each production process, and obtain a real-time data set;

[0014] a quality state estimation module, configured to input the real-time data set into the state space model, and perform optimal estimation on the quality state of each process in the current production batch through a Kalman filtering algorithm, and generate a state estimation vector containing a deviation amount and a cumulative impact on subsequent processes;

[0015] an adaptive compensation decision module, configured to solve a multi-objective optimization problem through a model predictive control algorithm based on the state estimation vector, and generate an adaptive compensation strategy containing an adjustment amount and an adjustment timing of each process parameter;

[0016] a process execution driving module, configured to convert the adaptive compensation strategy into a control instruction and issue the control instruction to an execution mechanism of a corresponding process, and drive the execution mechanism to complete process adjustment according to the specified parameters;

[0017] a model dynamic optimization module, configured to feed production data containing mass states before and after adjustment to the state space model, and update the state transition matrix and the control input matrix through an incremental learning algorithm.

[0018] In a third aspect, the application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the concrete prefabricated component production quality control method in the first aspect when executing the computer program.

[0019] In a fourth aspect, the application further provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the concrete prefabricated component production quality control method in the first aspect.

[0020] The aforementioned method, device, equipment, and medium for quality control in the production of precast concrete components accurately describe the transmission law of quality deviations by establishing a state-space model for the transmission of quality deviations between processes. It utilizes real-time sensor data to drive a Kalman filter algorithm to dynamically estimate the real-time quality state and its cumulative effect at each process stage. Based on this state estimation, a model predictive control algorithm is used to proactively generate an adaptive compensation strategy for process parameters. The actuators adjust the process parameters of subsequent processes in real time to offset previous deviations, while simultaneously feeding the adjustment effect back to the model for continuous self-updating. This achieves systematic suppression and compensation of quality deviations throughout the entire precast concrete component production process, significantly improving component dimensional accuracy and strength consistency, effectively reducing scrap rates and rework costs, and enhancing the production system's adaptability to changes in operating conditions. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart illustrating a method for quality control in the production of precast concrete components provided by this invention;

[0023] Figure 2 This is a schematic diagram of the process for generating an adaptive compensation strategy in one optional embodiment of the present invention;

[0024] Figure 3 This is a structural schematic diagram of a quality control device for the production of precast concrete components provided by the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0026] refer to Figure 1 The application presents a flowchart illustrating a method for quality control in the production of precast concrete components, which includes the following steps:

[0027] S1. Based on the historical process parameters and historical quality inspection data of each production process of precast concrete components, a mathematical model for the transmission of quality deviations between processes is established through multiple regression analysis, resulting in a state-space model containing a state transition matrix and a control input matrix.

[0028] Specifically, the production process of precast concrete components can encompass core stages such as mold preparation, rebar tying, concrete pouring, vibration, and curing. Corresponding historical process parameters comprehensively cover the key control items for each stage. For example, in the mold preparation stage, this includes mold splicing gaps, mold size calibration values, and mold fixing pressure; in the rebar tying stage, it includes rebar spacing, rebar lap length, and tying point tightening force; in the concrete pouring stage, it includes concrete slump, pouring speed, and pouring temperature; in the vibration stage, it includes vibration frequency, vibration time, and vibrator insertion depth; and in the curing stage, it includes curing temperature, curing humidity, and curing duration. Historical quality inspection data can include intermediate quality inspection results for each stage and the final component quality inspection results. Intermediate inspection results include mold size errors, rebar position offsets, and concrete pouring layer thickness deviations, while final inspection results include component dimensional accuracy, concrete compressive strength, flexural strength, and component appearance defects. After obtaining the historical data, data preprocessing is performed to remove abnormal data caused by sensor failure, manual recording errors, etc. The 3σ criterion can be used to screen data, that is, for a certain parameter's historical data sequence, its mean and standard deviation are calculated, and data that exceed the range of mean plus or minus three times the standard deviation are identified as outliers and removed. At the same time, missing data is filled in using linear interpolation to ensure the integrity and validity of the data.

[0029] Next, we construct the basic relationship of quality deviation transmission between processes through multiple regression analysis. Taking the deviation transmission from the mold preparation process to the rebar tying process as an example, we set the quality deviation of the mold preparation process as the dependent variable and the process parameters of this process as the independent variables. The multiple linear regression equation is shown in the following equation:

[0030]

[0031] In the formula, y represents the quality deviation of the mold preparation process (such as mold size error). This represents the constant term, reflecting the level of basic deviation unaffected by the independent variable; , , These represent the regression coefficients corresponding to each process parameter, used to quantify the degree of influence of different process parameters on quality deviation; , , These represent the process parameters for the mold preparation process (such as mold splicing gap, mold fixing pressure, and number of calibrations). This represents the random error term, which covers the influence of uncontrollable factors such as measurement errors and environmental interference, and its mean is 0.

[0032] The regression coefficients are solved using the least squares method, which involves constructing a residual sum of squares function, taking partial derivatives with respect to each coefficient and setting the partial derivatives to zero to obtain a system of equations. Solving this system determines the regression coefficients, thus revealing the quantitative relationship between process parameters and quality deviation within that process step. Similarly, multiple regression models are established for each individual process step to clarify the influence of process parameters on the quality deviation of each step.

[0033] Based on the multiple regression models for each process, a state-space model is further constructed. The core of the state-space model is the state transition matrix and the control input matrix. The state vector is defined as the quality deviation state of each process at a given time, the control input vector is defined as the process parameter adjustment input for each process at a given time, and the output vector is the quality inspection result vector at a given time. The discrete form of the state-space model is shown in the following equation:

[0034]

[0035]

[0036] In the formula, X(t) represents the state vector at time t, which includes the quality deviations of each process, such as the deviations of the mold preparation process, the rebar binding process, the concrete pouring process, the vibration process, and the curing process; X(t+1) represents the state vector at time t+1; A represents the state transition matrix, which describes the influence relationship between the state vector at time t and the state vector at time t+1. The elements in the matrix reflect the influence coefficients of the deviation of one process at time t on the deviation of another process at time t+1; B represents the control input matrix, which describes the influence of the process parameter adjustment input on the state vector. The elements in the matrix reflect the influence coefficients of the process parameter adjustment amount of one process on its own deviation. U(t) represents the control input vector at time t, which includes the adjustment amount of process parameters for each process step; W(t) represents the process noise vector, which reflects the impact of unpredictable disturbances on the state vector during production, and it follows a Gaussian distribution with a mean of 0; Y(t) represents the output vector at time t, which is the quality inspection result vector; C represents the output matrix, which is used to establish the mapping relationship between the state vector and the output vector; D represents the direct transmission matrix, which can usually be set to 0, since the quality inspection result is mainly determined by the quality deviation state and has no direct transmission relationship with the control input; V(t) represents the measurement noise vector, which reflects the impact of errors on the output vector during the quality inspection process, and it follows a Gaussian distribution with a mean of 0.

[0037] The state transition matrix is ​​constructed based on the quantitative relationship of deviation transmission between processes. The influence coefficient of each process's deviation on the deviation of subsequent processes can be calculated using historical data. For example, the deviation in the mold preparation process will affect the deviation in the rebar tying process; this influence is quantified by the corresponding matrix elements. Simultaneously, the deviation of a particular process itself will also affect the deviation of that process at the next moment due to the accumulation of deviations within the process; this is also quantified by the corresponding matrix elements. The control input matrix is ​​constructed based on the influence coefficient of each process's process parameter adjustment on its own quality deviation. Since the parameter adjustment of a particular process only affects its own deviation, elements in different rows and columns of the matrix are set to 0. By using the correspondence between the deviations of each process and the process parameter adjustments in historical data, the specific element values ​​of the state transition matrix and the control input matrix can be obtained using the least squares method, ultimately yielding a complete state-space model.

[0038] S2. Real-time datasets are obtained by collecting process parameters and quality data of the current production batch in real time through a sensor network deployed in each production process.

[0039] Specifically, based on the technological characteristics and data acquisition requirements of each production process, the type of sensor, sampling frequency, and other parameters are determined, and the sensors are precisely deployed in each process.

[0040] During the mold preparation process, data such as mold dimensional deviation, mold splicing gap, and mold fixing pressure are collected. For mold dimensional deviation, a laser rangefinder sensor is selected and fixed to a support around the mold. Multiple sensors are deployed along the length, width, and height of the mold to ensure comprehensive coverage of the key dimensional detection surfaces and to obtain real-time deviations between the actual and design dimensions of various parts of the mold. For mold splicing gap, a fiber optic displacement sensor is used, embedded on both sides of the mold splice seam to directly detect the gap size. Mold fixing pressure is collected by a pressure sensor installed at the bottom of the mold fixing bolts to monitor the bolt tightening pressure in real time, indirectly reflecting the mold's fixing status.

[0041] The rebar tying process can collect data such as rebar spacing, rebar lap length, and tying point tightening force. Rebar spacing and lap length are collected using vision sensors (industrial cameras) paired with LED strip light sources. The vision sensors are mounted above the rebar tying platform at an appropriate distance. Machine vision algorithms (such as edge detection and Hough transform) process the collected images to identify the rebar outline and position, and calculate the spacing between adjacent rebars and the lap length. Tying point tightening force is collected using tension sensors connected to tying tools (such as electric tying guns). The tension of the tying thread is monitored in real time during the tying process, serving as a quantitative indicator of the tying point tightening force.

[0042] The concrete pouring process can collect data such as concrete slump, pouring speed, and pouring temperature. Concrete slump is measured using an ultrasonic slump meter, which achieves non-contact measurement by utilizing the correlation between the propagation speed of ultrasonic waves in concrete and slump. This instrument is installed at the outlet of the pouring funnel to acquire real-time slump data. Pouring speed is collected using a flow sensor installed on the concrete delivery pipeline to monitor the volumetric flow rate of the concrete in real time, and the pouring speed is calculated in conjunction with the pouring time. Pouring temperature is measured using a thermocouple temperature sensor; the sensor probe is inserted into the concrete to monitor temperature changes in real time during the pouring process.

[0043] The vibration compaction process can collect data such as vibration frequency, vibration time, and vibratory rod insertion depth. Vibration frequency is collected using a vibration sensor fixed to the vibratory rod, which monitors the vibration frequency in real time. Vibration time is collected via a time relay linked to a sensor; the time relay starts timing when the vibratory rod starts and stops timing when the vibratory rod stops. Vibratory rod insertion depth is collected using a wire-type displacement sensor. The sensor's fixed end is mounted on the frame of the vibration equipment, and the wire end is connected to the top of the vibratory rod. As the vibratory rod inserts into the concrete, the wire descends, and the sensor calculates the insertion depth based on the change in the wire's length.

[0044] The curing process can collect data such as curing temperature, curing humidity, and curing time. Curing temperature and humidity are measured using temperature and humidity sensors. Multiple sensors are evenly deployed inside the curing kiln to ensure a comprehensive reflection of the temperature and humidity distribution of the curing environment. Curing time is collected through a timer linked to the door control switch of the curing kiln. The timer starts when the component enters the curing kiln and the door control switch is closed, and ends when the component exits the kiln and the door control switch is opened.

[0045] Sensors at each stage of the process form a sensor network via industrial Ethernet (such as Profinet or EtherNet / IP). The raw data collected by the sensors is transmitted to the edge computing gateway via a network switch. The gateway first preprocesses the data: a moving average filtering method is used to remove random noise; for example, for continuous sampling data from a sensor, an appropriate window size is selected, and the average value of the data within the window is calculated as the filtered data. The Grubbs criterion is used to remove outliers; that is, for the preprocessed dataset, the mean and standard deviation are calculated. If the difference between a data point and the mean is greater than the product of the Grubbs threshold and the standard deviation, it is considered an outlier and removed. Simultaneously, linear interpolation of adjacent data is used to fill in missing values. The preprocessed dataset is integrated according to the format "process number-parameter type-acquisition time-parameter value" to form a real-time dataset, which is then transmitted in real-time to the central control system's database via a communication module, providing data support for model calculations in subsequent steps.

[0046] S3. Input the real-time dataset into the state-space model, and use the Kalman filter algorithm to make the optimal estimate of the quality status of each process, generating a state estimation vector that includes the deviation and the cumulative impact on subsequent processes.

[0047] Specifically, based on the discrete form of the state-space model, both the process noise vector and the measurement noise vector follow a Gaussian distribution with a mean of 0. The covariance matrix of the process noise vector is denoted as Q, and the covariance matrix of the measurement noise vector is denoted as R. The initial state estimate is denoted as... It selects the mean of the quality deviation of each process in the historical data to form the initial state vector; the initial error covariance matrix is ​​denoted as... The method selects a matrix composed of the variance and covariance of the quality deviations of each process in historical data. If the elements in the matrix are the variances of the deviations of the same process, it reflects the uncertainty of the initial deviation estimation of that process; if they are the covariances of the deviations of different processes, it reflects the correlation uncertainty of the initial deviation estimations of different processes. The process noise covariance matrix Q and the measurement noise covariance matrix R are calculated from historical data. For example, for the process noise vector, a sample of the difference between the state vector and the state transition matrix at time t+1, the state vector at time t, the control input matrix, and the control input vector at time t in multiple sets of historical data is selected, and the covariance matrix of this sample is calculated as Q; for the measurement noise vector, a sample of the difference between the output vector and the output matrix at time t, and the state vector at time t in multiple sets of historical data is selected, and the covariance matrix of this sample is calculated as R.

[0048] The Kalman filtering process then proceeds iteratively, consisting of a prediction step and an update step. The purpose of the prediction step is to predict the current state and error covariance based on the state estimate from the previous time step. Assume that the optimal state estimate at time t-1 has already been obtained. With error covariance matrix The predicted state value at time t is shown in the following formula:

[0049]

[0050] In the formula, Represents the predicted state value at time t based on information from time t-1; A is the state transition matrix; t-1 is the optimal state estimate; B is the control input matrix; U(t-1) is the control input vector at time t-1. If t=1, then U(0) takes the initial process parameters, and is a 0 vector when there is no adjustment.

[0051] The prediction error covariance matrix at time t is shown in the following equation:

[0052]

[0053] In the formula, P(t|t-1) represents the prediction error covariance matrix at time t based on the information at time t-1; Let A be the transpose of the state transition matrix A; P(t-1|t-1) is the error covariance matrix at time t-1; Q is the process noise covariance matrix. This formula takes into account the propagation of error and the influence of process noise during the state transition process.

[0054] The purpose of the update step is to refine the predicted state by incorporating the real-time measurement data (i.e., the output vector Y(t) corresponding to the real-time dataset) to obtain the optimal state estimate. First, the Kalman gain is calculated, as shown in the following equation:

[0055]

[0056] In the formula, K(t) represents the Kalman gain at time t; P(t|t-1) is the prediction error covariance matrix at time t; Let C be the transpose of the output matrix; R is the measurement noise covariance matrix. The Kalman gain is used to balance the magnitude of prediction error and measurement error. If the measurement noise is small, the Kalman gain is large, and more attention is paid to the measurement data for correction; if the prediction error is small, the Kalman gain is small, and more attention is paid to the prediction result.

[0057] Then, the optimal state estimate at time t is calculated, as shown in the following equation:

[0058]

[0059] In the formula, This represents the optimal state estimate at time t; Let Y(t) be the predicted state value at time t; K(t) be the Kalman gain at time t; Y(t) be the output vector (real-time measurement data) at time t; and C be the output matrix. In the formula... To measure the residual, which is the difference between the actual measured value and the predicted measured value, the predicted state is corrected by Kalman gain to obtain a more accurate optimal state estimate.

[0060] Finally, update the error covariance matrix as shown in the following equation:

[0061]

[0062] In the formula, P(t|t) represents the optimal error covariance matrix at time t; I is the identity matrix; K(t) is the Kalman gain at time t; C is the output matrix; P(t|t-1) is the prediction error covariance matrix at time t. This formula is used to update the error covariance at the current time and provide parameters for the prediction step at the next time.

[0063] The optimal state estimate is obtained through the iterative calculation of the Kalman filter described above. This is the state estimation vector, which contains the deviation of each process and its cumulative impact on subsequent processes. The deviation of each process is directly reflected in the corresponding element of the state estimation vector, while the cumulative impact on subsequent processes is reflected through the state transition matrix. For example, the impact of the deviation of a certain process on the deviation of the next process is the product of the deviation of that process and the corresponding element of the state transition matrix, and the impact on the deviation of the process after that is the product of that product and another corresponding element of the state transition matrix. In this way, the cumulative impact value of the deviation of a certain process on the deviations of all subsequent processes can be calculated, and this value is used as additional information in the state estimation vector. This comprehensively reflects the current quality status of each process and its impact on subsequent processes, providing a basis for the generation of subsequent adaptive compensation strategies.

[0064] S4. Based on the state estimation vector, solve the multi-objective optimization problem through the model predictive control algorithm to generate an adaptive compensation strategy that includes the adjustment amount and timing of process parameters for each process.

[0065] Specifically, this step aims to solve a multi-objective optimization problem based on the state estimation vector using a model predictive control algorithm, ensuring that the generated compensation strategy can effectively eliminate quality deviations while also taking into account production efficiency and cost.

[0066] First, the objective function for multi-objective optimization is determined. Considering the actual needs of precast concrete component production, the objectives of multi-objective optimization mainly include three aspects: first, minimizing the quality deviation of each process to ensure the final component quality meets standards; second, minimizing the adjustment of process parameters to avoid equipment wear or production interruptions due to significant adjustments; and third, minimizing the impact of the adjustment process on the production cycle to ensure production efficiency. Based on this, the objective function is constructed as follows:

[0067] 1) The objective function for minimizing quality deviation is shown in the following equation:

[0068]

[0069] In the formula, This represents the objective function for minimizing quality deviation; n represents the prediction time domain, i.e., predicting the quality state at n future moments. This represents the state estimation vector at time t+k based on the information at time t; for The transpose of ; This represents the weight matrix, whose elements are set according to the degree of influence of each process deviation on the final quality. The weight matrix satisfies the properties of being non-negative and symmetric, ensuring that the influence weights of different process deviations are reasonably allocated.

[0070] 2) The objective function for minimizing the adjustment amount is shown in the following equation:

[0071]

[0072] In the formula, This represents the objective function that minimizes the adjustment amount; n-1 represents the control time domain, i.e., determining the adjustment amount for the next n-1 time steps. This represents the vector of process parameter adjustments at time t+k, i.e., the change in the control input vector. for The transpose of ; This represents the weight matrix, whose elements are set according to the cost of adjusting the parameters of each process. The weight matrix also satisfies the properties of being non-negative and symmetric. The higher the cost of the process adjustment, the larger the corresponding weight coefficient, in order to reduce large adjustments.

[0073] 3) The objective function for minimizing the impact of the production cycle is shown in the following equation:

[0074]

[0075] In the formula, This represents the objective function for minimizing the impact of the production cycle; This represents the production cycle extension time corresponding to the adjustment amount at time t+k. The larger the adjustment amount, the longer the production cycle extension time usually is. This represents the weighting coefficient, which is set according to the importance of the production cycle to ensure a balanced weighting of each objective in the overall objective function.

[0076] Combining the above three objectives, the multi-objective optimization problem is transformed into a single-objective optimization problem using the weighted summation method. The overall objective function is shown in the following equation:

[0077]

[0078] In the formula, Represent the overall objective function; , , Let represent the weight coefficients of the three objectives, and satisfy . Specific values ​​can be set according to the priorities of quality, cost, and efficiency in actual production. For example, when quality is the priority, When the value is large and efficiency is the priority, The value is relatively large.

[0079] Secondly, the constraints of the optimization problem are determined. These constraints mainly include constraints on the adjustment range of process parameters, constraints on the capability of actuators, and constraints on the timing between processes, as detailed below:

[0080] A) Constraints on the adjustment range of process parameters: The adjustment amount of process parameters must be within the range limited by the equipment performance and process requirements to avoid damage to the equipment due to the adjustment amount exceeding the equipment's load-bearing capacity, or deterioration of component quality due to exceeding the allowable range of the process.

[0081] B) Actuator capability constraints: The actuator speed corresponding to the adjustment amount must be within the maximum speed range of the actuator to ensure that the actuator can complete the adjustment within the specified time and avoid the decrease in adjustment accuracy due to excessive speed or the impact on production progress due to excessively slow speed.

[0082] C) Inter-process timing constraints: The adjustment time of the subsequent process must be later than the sum of the adjustment time and execution time of the previous process, to ensure that the subsequent process can be adjusted after the adjustment of the previous process is completed, so as to avoid the conflict between the adjustments and ensure the orderly progress of the production process.

[0083] After defining the objective function and constraints, the optimization problem is solved through the rolling optimization process of the model predictive control algorithm. First, based on the state estimation vector at the current time t... Using a state-space model, the quality state is predicted for the next n time steps, taking into account the influence of the control input vector on the state during the prediction process. Then, within the control time domain, the optimal control input sequence is solved under constraints with the goal of minimizing the overall objective function. The solution process can employ a genetic algorithm, encoding the control input sequence into chromosomes, with each gene corresponding to an adjustment value. Using the overall objective function as the fitness function, the sequence iteratively evolves through selection, crossover, and mutation operations until the chromosome with the highest fitness is obtained, which is the optimal control input sequence.

[0084] Finally, based on the rolling optimization principle of model predictive control, only the first element of the optimal control input sequence is selected as the adjustment amount of the process parameters at the current moment. Combined with the production sequence of each process, the order of adjustment for each process is determined, generating an adaptive compensation strategy. For example, if the state estimation vector shows a large deviation in a certain process, and this deviation has a significant impact on subsequent processes, then that process is adjusted first. Subsequently, based on the deviation of subsequent processes, adjustments are completed before their production begins, forming a complete compensation strategy that includes the adjustment amount and timing of each process. This ensures that the compensation measures can accurately and promptly eliminate quality deviations without affecting the overall production process.

[0085] S5. Convert the adaptive compensation strategy into control commands and send them to the actuators of the corresponding processes to drive the actuators to complete the process adjustment according to the specified parameters.

[0086] Specifically, firstly, the actuators for different processes are selected based on the process adjustment requirements. The execution mechanism for the mold preparation process is a servo motor, used to adjust the mold size and splicing gap. Its control parameters include motor speed, rotation angle, and running time. The execution mechanism for the rebar tying process is a robotic arm and an electric tying gun. The robotic arm is used to adjust the position and spacing of the rebars. Its control parameters include the angle of each joint of the robotic arm and the movement speed. The electric tying gun is used to adjust the fastening force of the tying points. Its control parameter is the output torque of the tying gun. The execution mechanism for the concrete pouring process is a flow control valve and a temperature regulator. The flow control valve is used to adjust the concrete pouring speed. Its control parameter is the valve opening. The temperature regulator is used to adjust the concrete pouring temperature. Its control parameter is the heating power. The execution mechanism for the vibration process is a vibrating motor and a lifting device. The vibrating motor is used to adjust the vibration frequency. Its control parameter is the motor frequency. The lifting device is used to adjust the insertion depth of the vibrating rod. Its control parameter is the lifting height. The execution mechanism for the curing process is a temperature controller and a humidifier. The temperature controller is used to adjust the curing temperature. Its control parameter is the set temperature. The humidifier is used to adjust the curing humidity. Its control parameter is the humidification rate.

[0087] Next, the adaptive compensation strategy is converted into control commands. The process parameter adjustments in the compensation strategy are converted into control parameter values ​​for the actuators. This conversion process is based on pre-calibrated conversion relationships. For example, the compensation strategy adjustments in the mold preparation process are converted into control parameters for the servo motor. The conversion process is based on a pre-calibrated correspondence between the servo motor rotation angle and the mold adjustment amount, calculating the required rotation angle, speed, and running time of the motor to form control parameters. Similarly, the compensation strategy adjustments in the concrete pouring process are converted into control parameters for the flow control valve. The conversion process is based on a pre-calibrated correspondence between the valve opening and the pouring speed, calculating the required valve opening and adjustment time to form control parameters, ensuring smooth valve switching.

[0088] The converted control parameters are encapsulated into control commands according to the communication protocol format of the actuators. Each actuator receives commands using an industrial communication protocol. Different actuators can select different protocols according to their needs; for example, servo motors use the Modbus-RTU protocol, robotic arms use the EtherCAT protocol, and flow control valves use the Profinet protocol. Taking the Modbus-RTU protocol as an example, the frame structure of the control command includes the slave address, function code, register address, data, and checksum. The slave address is used to identify the corresponding actuator, ensuring that the command is accurately sent to the target actuator; the function code is used to identify the command type, for example, a function code corresponds to a command type of writing a single register; the register address is used to identify the register corresponding to the control parameter, and different control parameters are stored in different registers; the data is the value of the control parameter, which needs to be converted to the format specified by the protocol; the checksum is used to verify whether errors have occurred during command transmission, usually using CRC checksum. Multiple control parameters can be set through multiple command frames, ensuring that the actuators can obtain complete control information.

[0089] After the control commands are generated, they are sent to the corresponding actuators via industrial buses (such as RS485 or EtherCAT). During command issuance, a master-slave communication method is used, with the central control system acting as the master station and each actuator as a slave station. The master station sends control commands to the slave stations sequentially according to the process order. After sending each command frame, it waits for a response signal from the slave station. If no response is received within a specified time, the command is resent, with a maximum of a preset number of retries to ensure the reliability of command transmission and prevent adjustments from being missed due to command loss.

[0090] After receiving control commands, the actuators parse the control parameters within the commands and drive their own actions to complete process adjustments. For example, after the servo motor parses the control parameters, it starts the motor driver and drives the motor to rotate at the set speed. Simultaneously, the motor's built-in encoder provides real-time feedback on the rotation angle. When the feedback angle reaches the set value, the driver stops the motor, completing the mold adjustment. After parsing the control parameters, the robotic arm coordinates the movements of each joint motor through the controller, moving to the target position at the set speed to adjust the spacing and position of the reinforcing bars. During the adjustment process, a vision sensor monitors the position of the reinforcing bars in real time. If the deviation from the target position exceeds the allowable range, the controller fine-tunes the robotic arm's movements to ensure adjustment accuracy.

[0091] During the operation of the actuator, the central control system monitors the adjustment effect in real time through the sensor network. For example, after the mold is adjusted, the laser rangefinder collects the actual size of the mold, compares it with the design size, and calculates the deviation after adjustment. If the deviation is less than the allowable threshold, the adjustment is deemed qualified; if the deviation is greater than the allowable threshold, a secondary adjustment command is generated and reissued to the actuator until the adjustment is qualified, ensuring that the process adjustment meets the requirements of the compensation strategy and providing a guarantee for the quality control of subsequent processes.

[0092] S6. Feed back the production data, including the quality status before and after the adjustment, to the state space model, and update the state transition matrix and control input matrix through incremental learning algorithm.

[0093] Specifically, this step updates the state transition matrix and control input matrix of the state space model through an incremental learning algorithm, thereby achieving dynamic optimization of the model and ensuring that the model can adapt to changes in the production process in real time, thus improving the accuracy of quality control.

[0094] First, the feedback data can include quality status data and process parameter data before and after adjustment. Specifically, it includes: the state estimation vector before adjustment (the predicted state at the current moment), the actual state vector after adjustment (the actual quality deviation data at the current moment collected by sensors), the control input vector before adjustment, and the control input vector after adjustment (i.e., the control input corresponding to the adjustment amount in the compensation strategy). After acquiring this data, preprocessing is performed. First, data alignment ensures that the data before and after adjustment correspond one-to-one in the time dimension, avoiding data mismatch due to time deviations that could affect the accuracy of model updates. Second, data standardization converts the quality status data and control input data into standardized data within the [0,1] interval using the min-max standardization method, as shown in the following formula:

[0095]

[0096] In the formula, This represents the standardized data; Represents the original data; This indicates the historical minimum value of the parameter; This represents the historical maximum value of the parameter. Data standardization can eliminate the impact of differences in the magnitude of different parameters on model updates, ensuring a balanced contribution of each parameter to the model update.

[0097] Next, we selected an incremental learning algorithm and determined its core parameters. Considering the real-time and dynamic nature of precast concrete component production data, we chose an incremental learning algorithm based on gradient descent. This algorithm can gradually update model parameters using single samples or small batches of samples, avoiding the drawback of traditional batch learning requiring model retraining. The core parameters of the algorithm include the learning rate and the forgetting factor. The learning rate controls the step size of each parameter update and needs to be moderate. If the value is too large, it will cause parameter oscillations and fail to converge to the optimal value. If the value is too small, the update speed will be too slow and unable to adapt to changes in production in a timely manner. The forgetting factor is used to balance the weights of new data and historical data. The value should be close to 1 to ensure that historical data has a larger weight, the model update is more stable, and noise in the new data does not have an excessive impact on the model.

[0098] Then, the update formulas for the state transition matrix and the control input matrix are derived based on the incremental learning algorithm. According to the discrete form of the state-space model, ignoring the process noise vector (since the influence of process noise is already reflected in the covariance matrix), the simplified relationship shown in the following equation can be obtained:

[0099]

[0100] The above equation can be rewritten in matrix form as shown below:

[0101]

[0102] In the formula, Y represents the output vector, which is composed of state vectors at multiple time points; This represents the regression matrix, with each row containing the state vector and control input vector at the corresponding time point; The parameter vector to be updated is formed by concatenating the state transition matrix and the control input matrix column by column. This represents the error vector, which encompasses the effects of model simplification and data noise.

[0103] The goal of incremental learning is to minimize the sum of squared errors. Taking the partial derivatives of the parameter vector and setting them to zero, combined with gradient descent and the forgetting factor, we obtain the following formula for updating the parameter vector:

[0104]

[0105] In the formula, This represents the updated parameter vector; This represents the parameter vector before the update; m represents the number of samples in the mini-batch. Indicates the learning rate; Representing the regression matrix The transpose of ; Y represents the output vector; Indicates the forgetting factor; Represents the identity matrix.

[0106] The updated parameter vector is split into a state transition matrix and a control input matrix, and the update formula for the state transition matrix is ​​shown below:

[0107]

[0108] In the formula, A(t+m) represents the updated state transition matrix; A(t) represents the state transition matrix before the update; X(t+k) represents the state vector at time t+k; X(t+k+1) represents the state vector at time t+k+1; B(t) represents the control input matrix before the update; and U(t+k) represents the control input vector at time t+k.

[0109] The update formula for the control input matrix is ​​shown below:

[0110]

[0111] In the formula, B(t+m) represents the updated control input matrix; B(t) represents the original control input matrix.

[0112] In the specific calculation process, a single element of the state transition matrix or control input matrix can be updated. By substituting the state vector, control input vector, and other data at the corresponding time, the updated value of that element can be calculated. In this way, all elements of the entire state transition matrix and control input matrix can be updated.

[0113] After the model is updated, the updated state-space model is validated to ensure that its accuracy meets the requirements. The validation method is as follows: select multiple sets of feedback data from the updated model, input the control input vector into the model to obtain the predicted state vector, and calculate the root mean square error between the predicted value and the actual state vector, as shown in the following formula:

[0114]

[0115] In the formula, RMSE represents the root mean square error; N represents the number of validation data sets; This represents the actual state vector of the i-th group; This represents the predicted state vector of the updated model for the i-th set of data. If the root mean square error (RMSE) is less than the RMSE before the update and less than the preset error threshold, the model update is considered successful. If the RMSE is greater than the error threshold, the learning rate and forgetting factor are adjusted, and the model is updated again until the RMSE meets the requirements.

[0116] Through the incremental learning process described above, the state transition matrix and control input matrix of the state space model can be integrated with new production data in real time, dynamically reflecting the changing patterns of quality deviation transmission between processes and the impact of process parameter adjustments. This ensures that the model remains consistent with the actual production process, providing more accurate mathematical model support for quality control of subsequent batches and achieving continuous optimization of the production quality of precast concrete components.

[0117] The aforementioned method for quality control in the production of precast concrete components accurately describes the transmission law of quality deviations by establishing a state-space model for the transmission of quality deviations between processes. It uses real-time sensor data to drive a Kalman filter algorithm to dynamically estimate the real-time quality state and its cumulative effect of each process. Based on this state estimation, a model predictive control algorithm is used to proactively generate an adaptive compensation strategy for process parameters. The process parameters of subsequent processes are adjusted in real time by the actuator to offset the deviations of the preceding processes. At the same time, the adjustment effect is fed back to the model for continuous self-updating. This achieves systematic suppression and compensation of quality deviations throughout the entire production process of precast concrete components, significantly improving the dimensional accuracy and strength consistency of components, effectively reducing scrap rate and rework costs, and enhancing the adaptability of the production system to changes in operating conditions.

[0118] In one optional embodiment, based on historical process parameters and historical quality inspection data of each production process of precast concrete components, a mathematical model for the transmission of quality deviations between processes is established through multiple regression analysis, resulting in a state-space model containing a state transition matrix and a control input matrix, including the following steps:

[0119] S11. Define a state vector containing the key quality characteristics of each process and a control vector containing the adjustable process parameters of each process; wherein, the dimension of the state vector is determined by the number of key quality characteristics, and the dimension of the control vector is determined by the number of adjustable process parameters.

[0120] Specifically, based on the technological characteristics of the entire precast concrete component production process, key quality characteristics that meet the criteria of being "quantifiable, directly related to deviation transmission, and detectable," as well as adjustable process parameters that are "adjustable through actuators and whose adjustment effects can be fed back," are selected. State vector. The deviation state used to characterize the key quality characteristics of the entire process at a certain moment is defined as follows: In the formula, n represents the total number of key quality characteristics, which determines the dimension of the state vector ( (dimensional column vector); This represents the deviation value of the i-th critical quality characteristic (such as mold size deviation, rebar position deviation). Control vector. Used to characterize the adjustment state of all adjustable process parameters at a given moment, defined as follows: In the formula, m represents the total number of adjustable process parameters, which determines the dimension of the control vector. (dimensional column vector); This represents the adjustment amount for the j-th adjustable process parameter (such as mold calibration displacement or binding gun torque adjustment). The vector dimension is dynamically adjusted based on the process differences of component type (beam, slab, column), always adhering to the principle of "characteristics / number of parameters = vector dimension".

[0121] S12. Extract state sequences and control sequences from historical process parameters and historical quality inspection data; construct a data sample set for model training based on the state sequences and control sequences.

[0122] Specifically, data is filtered in ascending order by collection timestamp, using production batches as the unit. The state sequence is a state vector composed of key quality characteristic values ​​at the same timestamp. Formation, that is (T represents the total number of condition monitoring cycles); the control sequence and the state sequence are time-aligned, and the control vector is formed by combining the adjustable process parameter adjustment records before condition monitoring. Formation (the control input at time t-1 affects the quality state at time t), without adjustment records. All elements are 0. After data extraction, preprocessing is performed. Time series alignment is achieved through timestamp matching, outliers are removed using the Dixon criterion, and Z-score standardization is used to eliminate parameter magnitude differences. The standardization formula is: In the formula, z represents the standardized data. This is the original data. This is the historical average of this parameter. This represents the historical standard deviation of the parameter. Finally, a triplet sample set of "previous time step state - current control input - current time step state" is constructed. The number of samples must satisfy "sample number ≥ 3 × (n + m)" to ensure the effectiveness of parameter solution.

[0123] S13. Based on the data sample set, solve the parameter matrix of the state transition equation using the least squares method to obtain the state transition matrix that quantifies the influence weight of the deviation of the preceding process on the current process.

[0124] Specifically, we first construct a simplified state transition equation, focusing on the influence of the preceding state on the current state. The formula is as follows:

[0125]

[0126] In the formula, for dimensional state transition matrix, elements This represents the weight of the influence of the j-th critical quality characteristic deviation at time t-1 on the i-th characteristic deviation at time t; This is the residual vector (covering measurement error and random disturbance).

[0127] Based on the sample set constructed using S12 (taking N=T-1 valid samples), the solution obtained using the least squares method is as follows, with the objective of minimizing the sum of squared residuals:

[0128]

[0129] In the formula, The current state observation matrix ( dimension), The preceding state matrix ( (Dimension). Verification after solving. Values ​​in the range [-1, 1] and not directly related to the process Approximately 0 to ensure a reasonable physical meaning.

[0130] S14. Based on the data sample set, the parameter matrix of the control input equation is solved by the least squares method to obtain the control input matrix that quantifies the effect of the control parameters on the quality state.

[0131] Specifically, we first construct a simplified equation for the control input, focusing on the impact of the control input on the current state. The formula is as follows:

[0132]

[0133] In the formula, for dimensional control input matrix, elements This indicates the effect of the adjustment amount of the j-th adjustable process parameter at time t-1 on the deviation of the ith critical quality characteristic at time t. This represents the residual vector (covering adjusted hysteresis and environmental disturbances). Based on the sample set of S12 (the same N samples as S13), the least squares method is used to minimize the sum of squared residuals.

[0134]

[0135] In the formula, For the preceding control matrix ( (Dimension). Verification after solving. The symbols conform to the process logic (e.g., an increase in mold calibration displacement corresponds to a decrease in dimensional deviation). (negative), and unrelated Approximately 0 ensures accurate quantification of the adjustment's effectiveness.

[0136] S15. Combining the state transition matrix and the control input matrix, construct the state-space model expression containing the process noise term to obtain the state-space model.

[0137] Specifically, this step integrates the aforementioned matrices and introduces process noise to construct a complete state-space model. Process noise vector. It follows a vector with zero mean and a covariance matrix of multivariate Gaussian distribution ( ), Solving based on sample composite residuals:

[0138]

[0139] In the formula, This is the composite residual vector.

[0140] Based on this, the core state equation is constructed: .

[0141] Simultaneously, supplementary output equations are provided to correlate the model state with actual detection data: In the formula, For the observation vector ( dimension, (Number of detected features) For the observation matrix ( Dimension, directly related ), The observed noise vector (following the rules) , (For the observation noise covariance matrix). The final result contains... , , , A complete state-space model enables an accurate description of the dynamic evolution of quality states.

[0142] In this embodiment, by setting state vectors (key quality characteristics) and control vectors (adjustable process parameters), constructing a standardized data sample set, and using the least squares method to accurately solve the state transition matrix (quantifying the weight of inter-process deviation transmission) and control input matrix (quantifying the effect of control parameters on quality), a state-space model that fits actual production is finally constructed. This provides an accurate and reliable mathematical basis for subsequent real-time quality state estimation and compensation strategy generation, and solves the problem that traditional models are difficult to quantify process correlation and control effects.

[0143] In one optional embodiment, a real-time dataset is input into a state-space model, and the quality status of each current process is optimally estimated using a Kalman filter algorithm to generate a state estimation vector containing deviation amounts and cumulative impacts on subsequent processes. This includes the following steps:

[0144] S21. Based on the state-space model and the mass state estimate of the previous time step, make predictions through the state transition equation to generate the prior state estimate vector and the prior estimate error covariance matrix at the current time step.

[0145] Specifically, this step relies on the state transition logic in the state-space model. Based on the posterior quality state estimate determined at the previous moment, it calculates the prior state estimate vector at the current moment through the state transition equation and simultaneously updates the prior estimate error covariance matrix. Specifically, the formula for calculating the prior state estimate vector is:

[0146]

[0147] In the formula, This is the prior state estimation vector at the current time (time t), reflecting the preliminary prediction of the current quality state based on information from the previous time. This is the posterior state estimation vector of the previous time step (t-1), which is the optimal quality state obtained from the previous round of Kalman filtering; This is the state transition matrix, which quantifies the impact of the previous process state on the current process. To control the input matrix, the process parameter adjustments are correlated with changes in quality status; This is the control input vector from the previous moment, i.e., the adjustment amount of the process parameters in the previous round.

[0148] The prior estimation error covariance matrix, calculated simultaneously, is used to describe the uncertainty of the prior state estimation, and is formulated as follows:

[0149]

[0150] In the formula, Let be the prior estimation error covariance matrix at time t, where the matrix elements reflect the degree of correlation between the estimation errors of each quality characteristic; Let be the posterior estimation error covariance matrix at time t-1; State transition matrix Transpose of; This is the process noise covariance matrix, used to quantify the impact of random disturbances on quality status during production.

[0151] S22. Collect physical measurement data of the current process through a sensor network, and convert the physical measurement data into an observation vector with the same dimension as the prior state estimation vector based on the observation model.

[0152] Specifically, this step converts the physical measurement data collected by sensors into observation vectors that match the dimension of the prior state estimation vector, ensuring that the measurement data can be directly compared with the model prediction results. First, physical measurement data for the current process is collected through a sensor network deployed at each process stage. This data covers the actual measured values ​​of various key quality characteristics, such as measured values ​​of mold dimensions, rebar positions, and concrete strength. Then, based on the observation model in the state-space model, the physical measurement data is mapped to observation vectors. The core relationship of the observation model is:

[0153]

[0154] In the formula, Let be the observation vector at the current moment, whose dimension is the same as the prior state estimation vector. Consistency is ensured to guarantee compatibility in subsequent calculations; The observation matrix is ​​used to establish the mapping relationship between the mass state vector and the physical measurement data. If a certain physical measurement data directly corresponds to a certain mass characteristic deviation, the corresponding element in the observation matrix is ​​1; otherwise, it is 0. This represents the current true quality state vector. To observe the noise covariance matrix, the impact of sensor measurement errors on the observation results is quantified. Through this mapping process, scattered physical measurement data are integrated into an observation vector of a unified dimension, laying the foundation for subsequent residual calculation and state correction.

[0155] S23. Based on the prior estimation error covariance matrix and the observation noise covariance matrix, calculate the Kalman gain matrix using a recursive algorithm.

[0156] Specifically, this step calculates the Kalman gain matrix using a recursive algorithm. This matrix is ​​used to balance the uncertainty of the prior state estimate with the reliability of the observation data, determining the contribution of the observation residuals to the state correction. The formula for calculating the Kalman gain matrix is:

[0157]

[0158] In the formula, Let be the Kalman gain matrix at time t; Let be the prior estimation error covariance matrix at time t. The larger the value, the higher the uncertainty of the prior state estimation, and the more the Kalman gain tends to increase to rely more on the observation data. Observation matrix Transpose of; For the observation noise covariance matrix, a larger value indicates lower reliability of the observation data, and the Kalman gain tends to decrease to rely more on prior estimates. The recursive algorithm here is characterized by the fact that the calculation of the Kalman gain only depends on the prior covariance matrix, observation matrix, and observation noise covariance matrix at the current time step. Furthermore, the prior covariance matrix is ​​recursively derived from the posterior covariance matrix at the previous time step, eliminating the need to store all historical data and ensuring the algorithm's real-time performance.

[0159] S24. Calculate the difference between the observation vector and the theoretical observation value after the prior state estimation vector is transformed by the observation model to obtain the observation residual vector.

[0160] Specifically, firstly, based on the prior state estimation vector and the observation matrix, the theoretical observation value, i.e., the ideal measurement result corresponding to the prior state estimation, is calculated using the following formula:

[0161]

[0162] In the formula, The theoretical observation value at time t; The observation matrix; Let be the prior state estimation vector at time t.

[0163] Subsequently, the observation residual vector, i.e., the difference between the actual observed value and the theoretical observed value, is calculated using the following formula:

[0164]

[0165] In the formula, Let be the observation residual vector at time t; The actual observation vector at time t (obtained by S22); Let t be the theoretical observation value at time t. The magnitude of the observation residual vector directly reflects the accuracy of the prior state estimation. The smaller the residual, the closer the prior estimation is to the true state, and the smaller the subsequent correction. Conversely, a larger residual requires a larger correction.

[0166] S25. Multiply the Kalman gain matrix by the observation residual vector to generate the state correction term, and add the state correction term to the prior state estimation vector to generate the posterior state estimation vector at the current time.

[0167] Specifically, the state correction term is first generated by multiplying the Kalman gain matrix by the observation residual vector, as shown in the formula:

[0168]

[0169] In the formula, This is the state correction term at time t; Let be the Kalman gain matrix at time t; Let be the observation residual vector at time t. The magnitude of the state correction term is determined by both the Kalman gain and the residuals. If the observation data is reliable ( Small) and high prior estimation uncertainty ( If the value is large, the correction term will be larger and will adjust the prior estimate more significantly.

[0170] Subsequently, the state correction term is added to the prior state estimation vector to obtain the posterior state estimation vector, as shown in the formula:

[0171]

[0172] In the formula, Let be the posterior state estimation vector at time t, representing the optimal estimation result of the current quality state of each process. This vector contains the quality deviation of each process (such as mold size deviation, rebar position deviation, etc.), and implicitly implies the cumulative impact of the current deviation on subsequent processes through the inherent correlation of the state transition matrix (such as the current mold deviation through...). (The impact of the deviation in the rebar tying process at the next moment) ultimately forms a state estimation vector that meets the requirements.

[0173] In this embodiment, relying on the Kalman filter algorithm, a priori estimates are first generated through the state transition equation. Then, the a priori correction is completed by combining the observation vector converted from the sensor measured data, the Kalman gain matrix, and the observation residual. This effectively filters out noise interference in the real-time data, achieving the optimal estimation of the quality deviation of the current process and the cumulative impact of subsequent processes. This provides an accurate basis for the quality status of subsequent compensation strategies and avoids estimation bias caused by data noise.

[0174] refer to Figure 2 In one optional embodiment, based on the state estimation vector, a multi-objective optimization problem is solved using a model predictive control algorithm to generate an adaptive compensation strategy that includes the adjustment amount and timing of process parameters for each process step, including the following steps:

[0175] S31. With the goal of optimizing the final component quality state to be close to the design target value, establish a multi-objective cost function that includes state error term and control cost term.

[0176] Specifically, this step constructs a multi-objective cost function around the goal of "final component quality approaching the design target," balancing quality achievement with adjustment costs by integrating state errors and control costs. The first step is to design the objective vector. This vector is composed of the design standard values ​​of key quality characteristics of each process (such as a design value of 0 for mold size deviation, a design value of 0 for rebar position deviation, etc.), and is used to quantify the target that the quality state needs to approximate. Based on this, the multi-objective cost function includes two parts: one is the state error term, which is used to measure the deviation between the quality state and the design target in the prediction time domain; the other is the control cost term, which is used to limit equipment damage or production interruption caused by excessive adjustment of process parameters.

[0177] The cost function expression can be:

[0178]

[0179] In the formula, For prediction in the time domain (i.e., predicting the future) (quality state at each moment) The state prediction value at time t+k based on time t; The state error weight matrix (a non-negative symmetric matrix) assigns greater weight to key characteristics that affect the final quality (such as concrete strength deviation); To control the time domain (i.e., to determine the future) Adjustment amount at each moment). This is the vector of process parameter adjustments at time t+k; To control the cost weight matrix (a non-negative symmetric matrix), parameters with high adjustment costs (such as rebar position adjustment) are given greater weight to avoid over-adjustment.

[0180] S32. Using the current state estimation vector as the initial condition, solve the problem of minimizing the multi-objective cost function in the prediction time domain to generate the optimal control sequence for multiple future processes.

[0181] Specifically, this step solves the cost function minimization problem in the prediction time domain to generate the optimal control sequence for future processes, with the initial condition being the state estimation vector at the current time. First, based on the established state-space model, ... Starting with predicting the future The quality state at time 1 The prediction process incorporates the influence of control inputs on the state, i.e. The predicted state is then substituted into the cost function, and the optimal control sequence is solved using a quadratic programming algorithm with the objective of minimizing J. This sequence contains future The optimal process parameter adjustment for each process step ensures that the quality status gradually approaches the design target within the prediction time domain, while minimizing the control cost.

[0182] S33. Extract the optimal control quantity of the current process from the optimal control sequence and generate process parameter adjustment instructions.

[0183] Specifically, this step extracts the adjustment amount for the current process from the optimal control sequence, generates specific process parameter adjustment instructions, and follows the core logic of "rolling optimization" in model predictive control—only the control amount at the current moment is executed, and the control amounts at subsequent moments need to be re-solved based on new state estimates. Specifically, the first element is selected from the optimal control sequence. This element represents the optimal control quantity to be applied in the current process, corresponding to the adjustment values ​​of various process parameters (such as mold calibration displacement adjustment, vibration frequency adjustment, etc.). This control quantity is then converted into executable adjustment instructions in the format of "process-parameter-adjustment value," for example, "mold preparation process - length direction calibration displacement - +0.03mm" or "vibration process - vibration frequency - +5Hz," ensuring that the instructions can be directly associated with the control parameters of subsequent actuators.

[0184] S34. Verify whether the process parameter adjustment instructions meet the physical constraints of the equipment and the process specification constraints. For instructions that do not meet the constraints, make feasibility corrections to obtain the corrected instructions.

[0185] Specifically, this step verifies the feasibility of the adjustment instructions. Through dual verification using both equipment physical constraints and process specification constraints, infeasible instructions are corrected. Equipment physical constraints are determined by the performance limits of the actuator, including upper and lower limits for process parameter adjustments (e.g., the maximum calibration displacement of a servo motor is ±0.1mm; exceeding this limit will prevent the equipment from achieving the desired result) and upper limits for adjustment speed (e.g., the maximum moving speed of a robotic arm is 5mm / s to avoid positioning deviations due to excessive speed). Process specification constraints are determined by the production standards of precast concrete components, including the allowable fluctuation range of key parameters (e.g., the adjustment range for concrete slump is ±10mm; exceeding this limit will affect casting performance) and time limits for adjustments between processes (e.g., curing temperature adjustments must be initiated within one hour of casting completion).

[0186] For instructions that do not meet the constraints, a "truncation correction" or "timing offset" strategy is adopted: if the adjustment amount exceeds the upper or lower limit, it is truncated to the nearest constraint boundary value (e.g., if the adjustment amount of +0.12mm exceeds the upper limit of +0.1mm, it is corrected to +0.1mm); if the adjustment timing violates the specifications, the start time is offset to a time period that meets the requirements (e.g., the curing temperature adjustment is advanced to 1 hour after pouring) to ensure that the corrected instructions can be executed within the range of equipment capacity and process standards.

[0187] S35. Integrate the corrected instructions with the corresponding timing information to generate an adaptive compensation strategy.

[0188] Specifically, timing information is determined in conjunction with the production progress of the process, including the start time and duration of adjustment instructions. The start time matches the process execution node; for example, mold adjustment needs to be completed 30 minutes before the rebar tying process begins to avoid affecting subsequent processes. The duration is determined by the action time of the actuator; for example, adjusting the vibration frequency from 50Hz to 55Hz requires 2 seconds to ensure parameter stability.

[0189] Each revised instruction is associated with its corresponding "start time - duration" and sorted according to the execution sequence of the process (e.g., mold preparation → rebar binding → concrete pouring → vibration → curing). This results in an adaptive compensation strategy that includes "process name - process parameters - adjustment amount - start time - duration". For example, "mold preparation process - length calibration displacement - +0.1mm - 30 minutes before rebar binding - 5 seconds", and "curing process - temperature - +2℃ - 1 hour after pouring - 60 minutes". This strategy is used for subsequent control instruction conversion and actuator driving.

[0190] In this embodiment, a multi-objective cost function (taking into account both quality state error and control cost) is used as the core. The optimal control sequence is solved in the prediction time domain. After extracting the control quantity of the current process, the physical constraints of the equipment and the process specification constraints are verified and corrected. Finally, the time sequence is integrated to generate a compensation strategy, which not only ensures that the quality of the final component approaches the design target, but also avoids equipment damage or production interruption caused by excessive adjustment, thus achieving a balance between quality control and production feasibility.

[0191] In one optional embodiment, using the current quality state estimation vector as initial conditions, a multi-objective cost function minimization problem is solved in the prediction time domain to generate the optimal control sequence for multiple future processes, including the following steps:

[0192] S41. Initialize the state control sequence and set the initial state of the state control sequence based on the quality state estimation vector; wherein, the state control sequence includes a state vector composed of key quality characteristics of each process at all times in the prediction time domain and a control vector composed of adjustable process parameters of each process.

[0193] Specifically, the state control sequence covers all moments within the prediction time domain. Each moment corresponds to a set of "state vector-control vector" pairs, meaning the sequence as a whole includes the key quality characteristic states and adjustable process parameter adjustment states at each moment within the prediction time domain. During initialization, the length of the prediction time domain is first determined (based on the continuity of the production process and the adjustment response speed, e.g., covering 3-5 subsequent process nodes, denoted as N at the end moment). Then, the quality state estimation vector at the current moment is directly used as the initial state of the sequence (i.e., the state vector at moment k). (where k is the current decision moment). This initial state reflects the actual quality deviation of each process and is the starting point for subsequent prediction and optimization. The initial value of the control vector is set based on historical adjustment experience or the default state of the equipment, and can usually be initialized to a zero vector (i.e., ...). This indicates that no process parameter adjustments will be applied under the initial assumptions, and predictions will be made based on the current state.

[0194] S42. Construct a multi-objective cost function based on the initial state, with the following expression:

[0195]

[0196] in, Indicates the total cost. This represents the state vector at the end of the prediction time domain, Represents the target state vector. Represents the terminal state weight matrix. This represents the state vector at time j. Represents the state error weight matrix. This represents the control vector at time j. Let represent the control cost weight matrix, and k represent the current decision time.

[0197] Specifically, this step constructs a multi-objective cost function based on the initial state, quantifying the combined requirements of "quality state approximation target" and "control cost minimization". In the function expression, J is the total cost, which comprehensively reflects the superposition effect of quality deviation and control cost. The state vector at the end of the time domain is the predicted state vector, representing the predicted final quality state; The target state vector is determined by the component design requirements (such as setting the target values ​​of key characteristics like mold size deviation and rebar position deviation to 0 to ensure that the final component meets the design standards). This is the terminal state weight matrix, used to emphasize the importance of the quality state at the end moment (if the terminal state is directly related to the final component quality, its diagonal elements can have a greater value than the weights at intermediate moments, and the off-diagonal elements are 0, since terminal errors with different characteristics are not directly related). Let j be the state vector at time j (k≤j≤N-1), representing the quality state at each intermediate time point during the prediction process; This is the state error weight matrix. The diagonal elements are set according to the importance of key quality characteristics (such as the concrete density deviation that affects the strength of the component, the corresponding element has a value greater than that of the secondary characteristic), and the off-diagonal elements are 0. Let be the control vector at time j, representing the adjustment amount of the process parameters at that time; To control the cost weight matrix, the diagonal elements are adjusted according to the process parameters (for parameters that require shutdown for adjustment, the corresponding element value is greater than the online adjustment parameter), and the off-diagonal elements are 0; k is the current decision time, which is the starting time node for prediction and optimization.

[0198] S43. Use a sequential quadratic programming algorithm to solve the multi-objective cost function, transforming the nonlinear optimization problem into a series of quadratic programming subproblems; for each quadratic programming subproblem, solve for the optimal control increment under the condition of satisfying state constraints and control constraints.

[0199] Specifically, firstly, because the relationship between quality state and control input may be nonlinear (e.g., the nonlinear correlation between concrete strength deviation and curing temperature adjustment), the original nonlinear cost function is Taylor-expanded at the current iteration point, approximating it as a quadratic function. Simultaneously, the nonlinear state constraints (e.g., key quality characteristic deviations not exceeding the process allowable upper limit) and control constraints (e.g., process parameter adjustments not exceeding the equipment's maximum adjustment range) are linearized, forming the standard form of a quadratic programming subproblem: minimizing the quadratic approximate cost function under the "linearized constraint condition." For each quadratic programming subproblem, the core of the solution is to obtain the optimal control increment (i.e.,... ,in This is the optimal solution to the subproblem. (This is the control vector value for the current iteration). This increment satisfies the dual requirements of "reducing state error and keeping control costs controllable", and strictly complies with the physical limits of the equipment and process specifications.

[0200] S44. Determine the optimal step size through linear search and update the state control sequence of the current iteration point; terminate the calculation when the cost function value difference of M consecutive iterations is less than the preset tolerance, and output the optimal control sequence; where M is the preset value.

[0201] Specifically, a linear search method is first used to determine the optimal step size, using the control increment obtained from S43. For the search direction, find a step size α in the interval [0,1] such that the new state control sequence after substituting the step size ( , The cost function J corresponding to the state transition relationship is minimized to avoid iteration divergence due to excessively large step sizes or computation time extension due to excessively small step sizes. The state control sequence at the current iteration point is then updated, and the cost function value corresponding to the new sequence is calculated. The iteration termination condition is set as "the difference in cost function values ​​after M consecutive iterations is less than a preset tolerance," where M is the preset number of iterations to stabilize (usually 3-5 iterations to ensure stable results), and the tolerance is a preset accuracy threshold (e.g., ...). (This can be adjusted according to the accuracy requirements of quality control). When this condition is met, the iteration stops, and the control sequence at this time is output. This sequence is the set of optimal control vectors at each moment in the prediction time domain, providing a basis for extracting the adjustment instructions for the current process.

[0202] In this embodiment, by initializing the state control sequence and constructing a multi-objective cost function, the nonlinear optimization problem is decomposed into a quadratic programming subproblem that can be solved efficiently using a sequential quadratic programming algorithm. The optimal step size update sequence is determined by combining linear search, and termination is controlled by the iterative cost difference threshold. This ensures the efficiency, accuracy, and stability of solving the optimal control sequence, provides reliable optimal control parameters to support the generation of compensation strategies, and avoids the problems of slow solution and low accuracy of traditional optimization methods.

[0203] The aforementioned method for quality control in the production of precast concrete components accurately describes the transmission law of quality deviations by establishing a state-space model for the transmission of quality deviations between processes. It uses real-time sensor data to drive a Kalman filter algorithm to dynamically estimate the real-time quality state and its cumulative effect of each process. Based on this state estimation, a model predictive control algorithm is used to proactively generate an adaptive compensation strategy for process parameters. The process parameters of subsequent processes are adjusted in real time by the actuator to offset the deviations of the preceding processes. At the same time, the adjustment effect is fed back to the model for continuous self-updating. This achieves systematic suppression and compensation of quality deviations throughout the entire production process of precast concrete components, significantly improving the dimensional accuracy and strength consistency of components, effectively reducing scrap rate and rework costs, and enhancing the adaptability of the production system to changes in operating conditions.

[0204] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0205] Based on the same inventive concept, this application also provides an apparatus for implementing the above-mentioned method for quality control of precast concrete component production. The solution provided by this apparatus is similar to the solution described in the above method; therefore, the specific limitations in one or more embodiments of the precast concrete component production quality control apparatus provided below can be found in the limitations of the precast concrete component production quality control method described above, and will not be repeated here.

[0206] In one exemplary embodiment, such as Figure 3 As shown, a quality control device 30 for precast concrete component production is provided to implement the methods in the above-described method embodiments. The device includes:

[0207] The quality transfer modeling module 31 is used to establish a mathematical model for the transfer of quality deviations between processes based on historical process parameters and historical quality inspection data of each production process of precast concrete components, and to obtain a state-space model that includes a state transition matrix and a control input matrix.

[0208] The real-time data sensing module 32 is used to collect process parameters and quality data of the current production batch in real time through a sensor network deployed in each production process to obtain a real-time dataset.

[0209] The quality status estimation module 33 is used to input the real-time dataset into the state space model, and use the Kalman filter algorithm to make the optimal estimation of the quality status of each process, generating a state estimation vector that includes the deviation and the cumulative impact on subsequent processes.

[0210] The adaptive compensation decision module 34 is used to solve multi-objective optimization problems based on the state estimation vector and through model predictive control algorithm, and generate an adaptive compensation strategy that includes the adjustment amount and adjustment time of process parameters for each process.

[0211] The process execution drive module 35 is used to convert the adaptive compensation strategy into control commands and send them to the corresponding process actuators to drive the actuators to complete the process adjustment according to the specified parameters.

[0212] The model dynamic optimization module 36 is used to feed back production data containing the quality state before and after adjustment to the state space model, and update the state transition matrix and control input matrix through incremental learning algorithm.

[0213] Embodiments of this application also provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the aforementioned method embodiments.

[0214] Embodiments of this application also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described method embodiments.

[0215] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0216] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for controlling the production quality of a concrete precast component, characterized by, The method comprises: S1, based on the historical process parameters and historical quality detection data of each production process of the concrete prefabricated component, a mathematical model of quality deviation transmission between processes is established through multiple regression analysis to obtain a state space model containing a state transition matrix and a control input matrix; S2, real-time acquisition of process parameters and quality data of the current production batch is performed through a sensor network deployed in each production process to obtain a real-time data set; S3, the real-time data set is input into the state space model, and the quality state of the current process is optimally estimated through a Kalman filtering algorithm to generate a state estimation vector containing a deviation amount and a cumulative influence on subsequent processes; S4, based on the state estimation vector, a multi-objective optimization problem is solved through a model predictive control algorithm to generate an adaptive compensation strategy containing process parameter adjustment amounts and adjustment timing of each process; S5, the adaptive compensation strategy is converted into control instructions and issued to the execution mechanism of the corresponding process to drive the execution mechanism to complete process adjustment according to the specified parameters; S6, production data containing the quality state before and after adjustment are fed back to the state space model, and the state transition matrix and the control input matrix are updated through an incremental learning algorithm.

2. The method of claim 1, wherein, The method comprises: S11, a state vector containing key quality characteristics of each process and a control vector containing adjustable process parameters of each process are defined; wherein the dimension of the state vector is determined by the number of the key quality characteristics, and the dimension of the control vector is determined by the number of the adjustable process parameters; S12, a state sequence and a control sequence are extracted from the historical process parameters and the historical quality detection data; a data sample set for model training is constructed according to the state sequence and the control sequence; S13, based on the data sample set, a parameter matrix of a state transition equation is solved through a least squares method to obtain a state transition matrix quantifying the influence weight of the previous process deviation on the current process; S14, based on the data sample set, a parameter matrix of a control input equation is solved through a least squares method to obtain a control input matrix quantifying the influence efficiency of the control parameters on the quality state; S15, the state transition matrix and the control input matrix are combined to construct a state space model expression containing a process noise term to obtain the state space model.

3. The method of claim 1, wherein, The method comprises: S21, according to the state space model and the quality state estimation value at the last time, a prior state estimation vector and a prior estimation error covariance matrix at the current time are generated through a state transition equation. S22, collecting physical measurement data of a current process through the sensor network, and converting the physical measurement data into an observation vector with the same dimension as the prior state estimation vector based on an observation model; S23, calculating a Kalman gain matrix through a recursive algorithm based on the prior estimation error covariance matrix and an observation noise covariance matrix; S24, calculating a difference between the observation vector and a theoretical observation value of the prior state estimation vector after being transformed by the observation model, to obtain an observation residual vector; S25, multiplying the Kalman gain matrix and the observation residual vector to generate a state correction term, and adding the state correction term to the prior state estimation vector to generate a posterior state estimation vector at a current time.

4. The method according to any one of claims 1 to 3, characterized in that, The adaptive compensation strategy includes process parameter adjustment amounts and adjustment time sequences of each process, and is generated by solving a multi-objective optimization problem through a model predictive control algorithm based on the state estimation vector, and includes: S31, establishing a multi-objective cost function including a state error term and a control cost term, with the final component quality state approaching a design target value as an optimization objective; S32, solving a minimization problem of the multi-objective cost function in a prediction time domain with the state estimation vector at the current time as an initial condition, to generate an optimal control sequence of future multiple processes; S33, extracting an optimal control amount of a current process from the optimal control sequence to generate a process parameter adjustment instruction; S34, verifying whether the process parameter adjustment instruction meets device physical constraints and process specification constraints, and performing feasibility correction on the instruction that does not meet the constraints to obtain a corrected instruction; S35, integrating the corrected instruction and corresponding time sequence information to generate the adaptive compensation strategy.

5. The method of claim 4, wherein, Solving the minimization problem of the multi-objective cost function in the prediction time domain with the quality state estimation vector at the current time as an initial condition to generate an optimal control sequence of future multiple processes includes: S41, initializing a state control sequence, and setting an initial state of the state control sequence based on the quality state estimation vector; wherein the state control sequence includes state vectors composed of key quality characteristics of each process at all times in the prediction time domain and control vectors composed of adjustable process parameters of each process; S42, constructing the multi-objective cost function based on the initial state, and the expression is: wherein, denotes the total cost, denotes the state vector at the end of the prediction horizon, denotes the target state vector, denotes the terminal state weight matrix, denotes the state vector at time j, denotes the state error weight matrix, denotes the control vector at time j, denotes the control cost weight matrix, k denotes the current decision time; S43, solving the multi-objective cost function by using a sequential quadratic programming algorithm, converting a nonlinear optimization problem into a series of quadratic programming sub-problems, and solving an optimal control increment for each quadratic programming sub-problem under the condition of meeting state constraints and control constraints; S44, determining an optimal step length through linear search to update the state control sequence of a current iteration point, and terminating the calculation when a difference between cost function values of consecutive M iterations is less than a preset tolerance, and outputting the optimal control sequence; wherein M is a preset value.

6. A device for controlling the quality of production of concrete precast elements, for implementing the method according to any one of claims 1 to 5, characterized in that, The device includes: The quality transfer modeling module is configured to establish a mathematical model of quality deviation transfer between processes by multivariate regression analysis based on historical process parameters and historical quality detection data of each production process of the concrete prefabricated component, and obtain a state space model including a state transition matrix and a control input matrix; The real-time data sensing module is configured to collect process parameters and quality data of a current production batch in real time through a sensor network deployed in each production process, and obtain a real-time data set; The quality state estimation module is configured to input the real-time data set into the state space model, and perform optimal estimation on quality states of the current processes by a Kalman filtering algorithm, and generate a state estimation vector including a deviation amount and a cumulative influence on subsequent processes; The adaptive compensation decision module is configured to solve a multi-objective optimization problem by a model predictive control algorithm based on the state estimation vector, and generate an adaptive compensation strategy including process parameter adjustment amounts and adjustment time sequences of the processes; The process execution driving module is configured to convert the adaptive compensation strategy into control instructions and issue the control instructions to an execution mechanism of a corresponding process, and drive the execution mechanism to complete process adjustment according to specified parameters; The model dynamic optimization module is configured to feed production data including quality states before and after adjustment to the state space model, and update the state transition matrix and the control input matrix by an incremental learning algorithm. 7.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-6 when the computer program is executed by the processor. The processor executes the computer program to implement the method in any one of claims 1 to 5.

8. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the method in any one of claims 1 to 5.

Images