Powder flow precision control and real-time adjustment method for powder quantitative packaging

Abstract

The present application provides a powder flow precision control and real-time adjustment method for powder quantitative packaging, comprising: through periodic injection of lightweight excitation signal and accurate collection of input and output time series data, using recursive extended least squares method to identify the dynamic model parameters of the actuator, combining the preset nonlinear rule mapping relationship to automatically convert into optimal PID control parameters, and using exponential weighted average to realize parameter smooth transition and avoid system disturbance; according to the closed-loop performance index, the data acquisition and identification cycle are adjusted adaptively, and continuous optimization in operation is realized, the present application can match the characteristic change of the actuator in real time, improve the powder flow regulation precision and system robustness.

Description

Technical Field

[0001] This invention relates to the field of industrial automation and process control technology, and in particular to a method for precise control and real-time adjustment of powder flow rate for quantitative powder packaging. Background Technology

[0002] In the field of existing powder quantitative packaging systems, precise flow control and real-time adjustment are core aspects affecting product metering accuracy and industrial automation levels. Currently, mainstream technical solutions mostly employ PID controllers tuned with fixed parameters or empirical methods. The proportional, integral, and derivative parameters are manually set during equipment commissioning or initial operating condition setup, and control commands are sent to the powder flow actuators, such as motors or valves, at a constant output frequency. This type of technology is widely used in scenarios such as general powder packaging, food additives, and automatic metering of pharmaceutical and chemical raw materials, and has significant advantages in control algorithm stability and ease of operation. However, with the increasing complexity of industrial environments, the variety of actual actuators and the dynamic drift in response speed due to manufacturing differences, wear, temperature changes, and load fluctuations lead to problems such as difficulty in maintaining control accuracy over long periods and insufficient parameter generalization ability in closed-loop systems. Some existing solutions attempt to introduce intelligent adaptive control, fuzzy algorithms, neural network prediction mechanisms, or online parameter tuning strategies to enhance system robustness; however, these methods typically have complex system architectures and high demands on data processing capabilities and computing resources, making them difficult to promote on a large scale in resource-constrained industrial automation scenarios. Most currently available technologies focus on improving the accuracy of control algorithms, while neglecting the significant time-scale difference between the output frequency of the control algorithm and the actual response speed of the actuator. During closed-loop regulation, if the controller's command refresh frequency exceeds the physical response rate of the actuator, the actuator cannot provide timely feedback, easily leading to a series of phenomena such as control lag, command backlog, and system oscillation. This problem is particularly prominent when multiple actuator models are used alternately or operating conditions change frequently, directly resulting in increased powder flow error, response overshoot, decreased control stability, and excessive packaging errors. Current industry standards generally lack a dynamic matching mechanism between control algorithms and actuator response characteristics, and lack an integrated technical approach for real-time perception, identification, and adjustment of actuator dynamic parameters. Existing PID self-tuning or neural network prediction models based on empirical parameters, while possessing a certain degree of intelligence, either rely on large amounts of historical data for training, limiting generalization, or have complex structures, making real-time performance difficult to guarantee, and also fail to effectively solve the problems of control lag and cross-platform parameter matching. Summary of the Invention

[0003] In order to solve the above-mentioned technical problems, the present invention provides a method for precise control and real-time adjustment of powder flow rate for quantitative powder packaging.

[0004] The technical solution of this invention is implemented as follows: a method for precise control and real-time adjustment of powder flow rate for quantitative powder packaging, comprising: S1: When the powder quantitative packaging system is started or reset, the control system outputs a set of lightweight excitation signals, which are small-amplitude step signals or pseudo-random sequence signals, to stimulate the actuator to produce identifiable dynamic response behavior. S2: Synchronously acquire the excitation signal as the input control quantity and the actual action response signal of the actuator as the output response quantity to form an input-output time series data pair. The response signal includes the time trajectory of the change in motor speed or the change in valve opening. S3: Perform online system identification processing on the input-output time series data pair, estimate the dynamic characteristic parameters of the actuator using the recursive augmented least squares method, and obtain its equivalent first-order inertial plus pure time delay form of the transfer function model, which includes three core parameters: time constant, pure time delay and system gain. S4: Based on the dynamic characteristic parameters of the equivalent transfer function model, the model parameters are converted into initial PID controller parameters through a preset rule mapping relationship. The rule mapping relationship is generated by offline simulation of a large number of typical actuator models and optimization combined with Ziegler-Nichols-type tuning criteria, and outputs three basic control parameters: proportional gain, integral time and derivative time. S5: Load the initial PID controller parameters into the real-time control loop, and periodically trigger the excitation signal injection and data acquisition process during normal system operation to obtain the latest input-output time series data pair under the current working condition, which is used to detect the dynamic characteristic drift of the actuator caused by wear, temperature rise or load change; S6: Based on the latest collected data, repeat the online system identification process, update the equivalent transfer function model at the current moment, and recalculate the corresponding PID controller parameters according to the updated model parameters to generate a new set of target control parameters; S7: Determine whether the difference between the parameters of the old and new PID controllers exceeds the preset threshold. If it does, start the parameter smoothing transition mechanism and use an exponential weighted average method to gradually update the proportional gain, integral time and derivative time item by item to avoid sudden changes in control output that could cause system disturbances. S8: Continuously monitor the deviation between the powder flow feedback signal and the set value, perform closed-loop regulation in combination with the parameters of the currently applied PID controller, and record the control performance indicators. If the deviation convergence speed decreases or the overshoot increases within several consecutive regulation cycles, the trigger interval for the next system identification is shortened to achieve enhanced dynamic adaptability.

[0005] The method for precise control and real-time adjustment of powder flow rate for quantitative powder packaging provided by this invention has the following beneficial effects: (1) This invention introduces a dynamic characteristic sensing mechanism based on lightweight excitation signals and online system identification, which realizes the automatic extraction of key response characteristics such as actuator time constant, delay time and damping ratio. Then, based on the established model-parameter mapping relationship, it autonomously generates the optimal PID gain parameters, effectively overcoming the defects of control performance degradation caused by differences in actuator models, aging and wear or fluctuations in operating conditions. It significantly improves the response consistency and adjustment accuracy of the flow control loop under different equipment configurations and operating stages. Especially in the face of industrial scenarios with frequent start-stop and large changes in material flowability, it can quickly complete parameter matching, greatly shorten the system debugging cycle and reduce overshoot, thereby ensuring the metering accuracy and process stability in the powder packaging process. (2) This invention designs a periodic re-identification and parameter self-update mechanism, and combines it with an exponential weighted average smooth transition strategy to achieve progressive optimization and adjustment of PID parameters while ensuring control continuity, thus avoiding process disturbances that may be caused by sudden switching. This closed-loop adaptive architecture breaks through the limitations of traditional control modules that require "one-time tuning and long-term use", enabling the control system to continuously track the health status of the actuator and changes in the working environment, significantly enhancing the robustness and reliability of the overall system. At the same time, the entire parameter tuning process is based on structured modeling and rule mapping, without using black-box models such as neural networks and fuzzy inference, and has clear engineering interpretability and real-time computation feasibility. It can be efficiently deployed on conventional industrial controllers without the need for additional high-performance computing units or a large amount of training data, solving the common problems of uncontrollable models, high resource consumption, and high implementation costs in existing intelligent tuning methods. (3) This method constructs a complete technical chain of "excitation injection - data acquisition - system identification - parameter mapping - dynamic update", forming a full-link automated closed loop from the perception of the characteristics of the underlying actuator to the optimization of the upper-level control parameters. It truly realizes the active cognition and dynamic adaptation of the control algorithm to the hardware characteristics, and promotes the upgrade of the flow control module from static configuration to intelligent evolution. This architecture is not only suitable for high-precision control of a single type of actuator, but also supports mixed use scenarios of multiple brands and specifications of motors or valves, which greatly improves the compatibility and scalability of the powder packaging system. It is particularly suitable for practical application environments with frequent production line transformation and equipment replacement. By dynamically and deeply coupling the control strategy with the execution end, the system can complete accurate modeling and parameter tuning using only conventional input and output signals without adding complex sensing devices. This significantly reduces the dependence on the professional skills of operators and improves the level of automation and operation and maintenance efficiency. Attached Figure Description

[0006] Figure 1This is a flowchart of the method for precise control and real-time adjustment of powder flow rate for quantitative powder packaging according to the present invention. Figure 2 This is a sub-flowchart of the method for precise control and real-time adjustment of powder flow rate for quantitative powder packaging according to the present invention. Figure 3 This is another sub-flowchart of the method for precise control and real-time adjustment of powder flow rate for quantitative packaging of powder according to the present invention. Detailed Implementation

[0007] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0008] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0009] like Figure 1 As shown, this invention provides a method for precise control and real-time adjustment of powder flow rate for quantitative powder packaging, specifically including: S1: When the powder quantitative packaging system is started or reset, the control system outputs a set of lightweight excitation signals, which are small-amplitude step signals or pseudo-random sequence signals, to stimulate the actuator to produce identifiable dynamic response behavior. S2: Synchronously acquire the excitation signal as the input control quantity and the actual action response signal of the actuator as the output response quantity to form an input-output time series data pair. The response signal includes the time trajectory of the change in motor speed or the change in valve opening. S3: Perform online system identification processing on the input-output time series data pair, estimate the dynamic characteristic parameters of the actuator using the recursive augmented least squares method, and obtain its equivalent first-order inertial plus pure time delay form of the transfer function model, which includes three core parameters: time constant, pure time delay and system gain. S4: Based on the dynamic characteristic parameters of the equivalent transfer function model, the model parameters are converted into initial PID controller parameters through a preset rule mapping relationship. The rule mapping relationship is generated by offline simulation of a large number of typical actuator models and optimization combined with Ziegler-Nichols-type tuning criteria, and outputs three basic control parameters: proportional gain, integral time and derivative time. S5: Load the initial PID controller parameters into the real-time control loop, and periodically trigger the excitation signal injection and data acquisition process during normal system operation to obtain the latest input-output time series data pair under the current working condition, which is used to detect the dynamic characteristic drift of the actuator caused by wear, temperature rise or load change; S6: Based on the latest collected data, repeat the online system identification process, update the equivalent transfer function model at the current moment, and recalculate the corresponding PID controller parameters according to the updated model parameters to generate a new set of target control parameters; S7: Determine whether the difference between the parameters of the old and new PID controllers exceeds the preset threshold. If it does, start the parameter smoothing transition mechanism and use an exponential weighted average method to gradually update the proportional gain, integral time and derivative time item by item to avoid sudden changes in control output that could cause system disturbances. S8: Continuously monitor the deviation between the powder flow feedback signal and the set value, perform closed-loop regulation in combination with the parameters of the currently applied PID controller, and record the control performance indicators. If the deviation convergence speed decreases or the overshoot increases within several consecutive regulation cycles, the trigger interval for the next system identification is shortened to achieve enhanced dynamic adaptability.

[0010] Step S1: When the powder quantitative packaging system is started or reset, the control system outputs a set of lightweight excitation signals. These excitation signals are small-amplitude step signals or pseudo-random sequence signals, used to stimulate the actuators to produce identifiable dynamic response behavior. Specifically, this includes: S1.1: Based on the triggering conditions of system startup or reset events, a control command activation signal is generated as the starting mark of the excitation signal injection process to ensure that the excitation process is only executed during the system initialization phase or after the state is reset, so as to avoid interfering with the stability of closed-loop control during normal adjustment. Based on the system start or reset event flag signal output by the operation status detection module of the powder quantitative packaging system, an event triggering identification method (parameters: event type code, state machine identifier) ​​is adopted to achieve accurate identification of the system initialization phase. The state machine decision algorithm (parameters: main control loop state register value, actuator enable bit) is used to make logical judgments on the current operating state of the control system and obtain Boolean judgment results that meet the start or reset conditions. Furthermore, a control command generation algorithm (parameters: trigger source type, signal encoding format) is adopted to convert the judgment result into a control command activation signal and obtain a start flag data packet for driving the subsequent excitation signal injection process; Furthermore, through the safety interlock logic judgment module (parameters: closed-loop control enable flag, actuator stop signal), the safety screening of the control command activation signal is realized, and the activation signal output that has passed the interlock verification is generated to ensure that the excitation process is only executed during the initialization phase or after the state is reset; Furthermore, a signal buffering and timing loading method (parameters: buffer duration, main control loop sampling period) is adopted to load the verified activation signal into the control buffer according to the time base of the main control loop, and generate a start trigger pulse sequence synchronized with the main loop; By using an event-triggered recognition algorithm and a signal interlocking processing method, the judgment result of the previous step is transformed into a highly reliable start flag for the excitation signal injection process, achieving the expected technical effect of not interfering with the stability of normal closed-loop control throughout the excitation process. For example, in a powder quantitative packaging system, the system startup detection module uses the running status word of the main control PLC as input. When bits 0 and 3 in the running status word are both set, it is determined to be a startup event; a state machine register value of 5 indicates a reset state. Either condition being met triggers the event trigger identification module, generating a Boolean judgment result with a logic value of 1. The control instruction generation module encodes this judgment result into an 8-bit binary control signal "00010010", which is used as an activation signal data packet and transmitted to the safety interlock logic processing unit. The interlock module reads that the closed-loop control enable flag register value is 1 and the actuator stop signal is true, determining that the safety condition is valid and directly transmitting the activation signal. The buffer loading module sets the buffer duration to 2 sampling cycles, with a sampling cycle of 10ms. It loads the activation signal into the buffer area and outputs a trigger pulse sequence after aligning it with the sampling step size of the main loop. The pulse sequence is read by the subsequent excitation signal injection module, which initiates the generation process of a lightweight step or pseudo-random sequence signal. Throughout the execution process, the closed-loop control stability test results show that there is no increase in overshoot and the system delay time remains unchanged, achieving complete isolation between the excitation process and the normal adjustment process. S1.2: Based on the preset excitation signal template library, select a lightweight excitation signal form suitable for the current actuator type. The template library contains two types of standard signals: small amplitude step signals and pseudo-random binary sequences (PRBS). The amplitude range is limited to 5% to 10% of the rated control amount to ensure that the excitation provides a sufficient dynamic response without causing significant fluctuations in powder flow. After receiving the control command activation signal output by S1.1 as the starting condition, the actuator type matching algorithm (parameters: actuator model identifier, drive mode, rated control quantity) is used to automatically identify and classify the type of the currently connected actuator. Furthermore, by using a template library retrieval method (parameters: actuator type index, signal type classification), the standard signal form of the corresponding actuator can be selected from the preset excitation signal template library, and its amplitude setting range can be extracted as candidate configuration data at the same time. Furthermore, an amplitude constraint calculation method (parameters: rated control quantity, amplitude percentage boundary) is adopted to achieve quantization setting of the candidate signal amplitude, and the actual amplitude is calculated according to the following formula:

[0011] in, This represents the actual amplitude of the signal. This is the rated control quantity. The selected amplitude percentage (between 5 and 10). This calculation ensures that the excitation signal elicits a sufficient dynamic response without causing significant fluctuations in powder flow rate; Furthermore, a signal integrity verification method (parameters: amplitude setting value, actuator safe operating range) is adopted to realize the simulated loading verification of the selected signal form and amplitude, output the signal excitation simulation curve and its spectrum analysis results, confirm that it covers the main dynamic modes of the actuator and avoids entering the saturation region or nonlinear operating region; By using signal selection and amplitude constraint algorithms, a continuous link output from the template library to a lightweight excitation signal adapted to the current actuator is achieved, ensuring that the original signal data that meets the system identification requirements is obtained in the subsequent discretization process. For example, on an electric valve actuator with a rated control quantity of 200 units of opening, the system type identification module outputs a type code of VALVE-SRV-200, which is matched to the PRBS signal template in the template library. The amplitude percentage is set to 7, and the actual amplitude is calculated using the formula: The output is 14 units of opening. The signal integrity verification module simulates the PRBS signal, and the results show that the spectrum has concentrated energy and no high-frequency spikes in the 0.5Hz to 3Hz range, ensuring coverage of the valve's main inertial modes and avoiding flow disturbances. Finally, the selected PRBS signal and actual amplitude parameters are sent to the subsequent time series discretization processing module. Field test results show that the actuator produces a stable and identifiable response trajectory when receiving the signal, meeting the system's identification accuracy requirements and significantly improving the stability of model parameter estimation. S1.3: The selected excitation signal is discretized into a time series and converted into a digital control sequence with a fixed sampling period using the high-precision timing module of the industrial controller. The sampling period is synchronized with the main control loop to ensure that the excitation signal is strictly aligned with the subsequently acquired response data on the time axis, thereby obtaining an equidistant input-output data pair that can be used for system identification. S1.4: The discretized excitation signal is loaded as a virtual control output into the actuator drive channel. Without affecting the actual material supply, it is applied to the motor or valve actuator through a low-power drive method to produce small but observable motion changes, forming a dynamic excitation with sufficient spectral richness to cover the main dynamic modes of the actuator. S1.5: Monitor whether the actuator enters the linear operating range during the application of the excitation signal, and determine whether there is saturation, dead zone or nonlinear distortion based on the real-time feedback signal; if non-ideal response characteristics are detected, automatically adjust the amplitude of the excitation signal or switch to another type of signal form, and re-execute the injection process to ensure that the acquired input-output data meets the basic assumptions of the subsequent recursive augmented least squares method for the linearity and time invariance of the system.

[0012] Step S2: Synchronously acquire the excitation signal as the input control quantity and the actual action response signal of the actuator as the output response quantity to form an input-output time series data pair. The response signal includes the time trajectory of changes in motor speed or valve opening. Specifically, it includes: S2.1: Obtain a lightweight excitation signal generated by the control system during the startup or reset phase as an input control quantity. The excitation signal is a small-amplitude step signal or a pseudo-random sequence signal. Perform signal synchronization marking processing based on the time-domain characteristics of the excitation signal to generate a timestamped input signal sequence and obtain a time reference for matching the output response. The lightweight excitation signal generated by the control system during the startup or reset phase is acquired as the input control quantity. The excitation signal is directly read based on the industrial controller signal buffer module to ensure that the signal amplitude and time characteristics remain original and intact during acquisition. A time-domain feature extraction algorithm (parameter: sampling period synchronized with the main control loop) is used to realize the time-domain characteristic analysis of the excitation signal waveform data, including amplitude vector, duration, and step or jump position index within the signal sequence, thus laying the foundation for subsequent accurate labeling; Furthermore, by using a timestamp generation method (parameters: based on the high-precision clock of the industrial controller, with a resolution of not less than 1ms), point-by-point time positioning of the excitation signal sequence is achieved. An absolute timestamp is added to each sampling point to form a time stamp matrix, ensuring that the time reference of the input control quantity corresponds completely with the acquisition time of the actuator response quantity. Furthermore, through the signal synchronization tagging processing algorithm (parameter: adopting event triggering mode, the event source is the amplitude change threshold of the excitation signal), the synchronization code embedding of key points of the input signal is realized, that is, the synchronization flag bit is inserted into the corresponding abrupt change point inside the excitation signal to form an explicit alignment anchor point for matching the output response; By using time characteristic verification processing, the consistency between the time stamp matrix and the synchronization identifier sequence is checked, erroneous identifiers caused by sampling jitter are eliminated, and the input control quantity sequence with accurate timestamp and synchronization code is output, so as to achieve accurate matching with the subsequent output response quantity; For example, in a powder packaging system with a rated frequency of 50Hz, the controller outputs a small step signal with an amplitude of 7% of the rated control value during the startup phase. The signal sampling period is set to 20ms, and the high-precision clock resolution of the industrial controller is configured to 1ms. Using a time-domain feature extraction algorithm, an amplitude change event is detected at the 100th sampling point of the excitation signal, triggering the embedding of a synchronization identifier. A timestamp generation method is used to form a time stamp matrix containing the absolute time of each sampling point; for example, the timestamp corresponding to the 100th sampling point is... Seconds. After time characteristic verification, the erroneous marker of the 250th sampling point was removed, resulting in a 500-point input control sequence with synchronization anchors. In actual verification, when this sequence was time-aligned with the output response, the data pairing error was less than [missing value]. The system achieves high-precision synchronization between the input signal and the actuator response signal during the system identification process, providing a stable and reliable time reference for the adaptive tuning of PID parameters. S2.2: Collect the actual action response signal generated by the actuator under the drive of the excitation signal as the output response quantity. The response signal includes the speed change trajectory output by the motor speed sensor or the opening change time sequence fed back by the valve positioner. Use a high-precision sampling module to capture data at a sampling frequency of not less than 100Hz and extract the original observation data of the dynamic response of the actuator. S2.3: Perform time synchronization and alignment processing on the acquired input control quantity and output response quantity. Based on the timestamp reference generated in S2.1, use a linear interpolation algorithm to compensate for the time deviation caused by the asynchronous sampling of sensors, generate a set of strictly corresponding time-paired data points, and form an equal-time interval input-output time series data pair that can be used for system identification. S2.4: The aligned input-output time series data pairs are preprocessed, including removing DC bias, filtering high-frequency noise interference, and normalization. The output response signal is smoothed using a second-order Butterworth low-pass filter to reduce the impact of measurement noise on the system's identification accuracy, resulting in a clean dynamic response data sequence. S2.5: Encapsulate the preprocessed input-output time series data pairs into a structured data format to generate a standard dataset that meets the input requirements of the recursive augmented least squares method. This dataset serves as the direct input for the next stage of online system identification, ensuring that the dynamic characteristic parameter estimation process has a high-quality data foundation.

[0013] like Figure 2 As shown, step S3 involves performing online system identification processing on the input-output time series data pair, estimating the dynamic characteristic parameters of the actuator using the recursive augmented least squares method, and obtaining its equivalent first-order inertial plus pure time delay transfer function model. This model includes three core parameters: time constant, pure time delay, and system gain. Specifically, it includes: S3.1: Based on the time series data pairs consisting of the input control quantity (excitation signal) and the output response quantity (actuator action time trajectory) generated in the previous steps, an initial identification dataset is constructed; specifically, the original input is subjected to data structure encapsulation processing to form a standardized input format adapted to the recursive augmented least squares method, and the output is an initial identification dataset that can be used for algorithm initialization. S3.2: The initial identification dataset is preprocessed by using a sliding window filtering method to remove high-frequency noise interference and eliminating DC bias components through differential operation. The denoised data sequence is processed to construct a regression vector. An extended observation matrix is ​​generated using the lag input and output terms to enhance the excitation sufficiency and numerical stability in the parameter estimation process. The reconstructed input-output regression structure is output for recursive calculation. S3.3: Based on the reconstructed input-output regression structure, the recursive augmented least squares method is applied to estimate the dynamic parameters of the actuator online. Each time a new data point is received, a parameter update iteration is performed. The discrete model coefficients mapped by the proportional gain, time constant and pure time delay are successively corrected to continuously optimize the parameter convergence performance and output a dynamic characteristic parameter estimation sequence that evolves over time and tends to stabilize. Based on the reconstructed input-output regression structure, the recursive augmented least squares method (parameters: the initial covariance matrix is ​​set as a diagonal matrix and each diagonal element takes a large positive value to reflect the initial uncertainty; the forgetting factor λ is set between 0.95 and 0.99 to balance the parameter tracking speed and noise suppression capability) is used to achieve real-time estimation of the coefficients of the discretized model of the actuator. Furthermore, the iterative formula is updated through parameters.

[0014] The system performs an iterative update of the model coefficients upon receiving each new sample data, and obtains the coefficient vector at the latest time step. ,in The gain vector obtained from augmented least squares calculation. Let be the parameter estimation vector at time k-1. Let k be the output response at time k. This is the transpose of the regression vector at time k; Furthermore, the covariance update formula is used.

[0015] To achieve stepwise convergence control of the variance of parameter estimation, where Let be the covariance matrix at time k. It is the covariance matrix at time k-1, and the ability to track time-varying characteristics is enhanced under the influence of the forgetting factor λ; Furthermore, the discretized model coefficients are analyzed according to the mapping relationship of three types of physical parameters: proportional gain, time constant, and pure time delay, to obtain the estimated physical quantity values ​​for each iteration and form a dynamic characteristic parameter estimation sequence. Furthermore, by setting a convergence criterion, the parameter sequence is monitored in real time, and the rate of change detection method is used to ensure that the relative changes in proportional gain, time constant and pure time delay are all below the preset stability threshold, so as to confirm that the parameters tend to stabilize. By combining the recursive augmented least squares algorithm with parameter convergence determination, the regression structure of the previous step is transformed into a high-precision parameter time series that can reflect the dynamic characteristics of the actuator in real time, thereby achieving stable output of the dynamic model under online identification conditions. For example, in a powder quantitative packaging system, for a servo motor actuator with a rated speed of 3000 rpm, the forgetting factor λ is set to 0.98, the diagonal elements of the initial covariance matrix P(0) are 100, and the excitation signal amplitude is taken as 7% of the rated control value. During operation, input-output data are collected at a sampling frequency of 200 Hz and a regression vector is constructed. The recursive augmented least squares update formula was applied to calculate the result. For example, the parameter vector at the 500th sampling point =(0.85, -0.30, 0.12), and after mapping analysis, the system gain K=0.85, time constant T=0.35s, and pure time delay L=0.12s are obtained. Subsequently, the model is iterated until the 1000th sampling point. When the rate of change of each parameter is below the 1% stability threshold, the model is considered converged, and a stable dynamic characteristic parameter sequence is output. This achieves high-speed and accurate dynamic modeling of the motor under the current operating conditions, significantly improving the response consistency of powder flow control. S3.4: Based on the converged dynamic characteristic parameter estimation sequence, the final stable parameter set is extracted and the model is restored to convert it into a first-order inertial plus pure time delay transfer function in the continuous domain. Among them, the system gain, time constant and pure time delay are used as core modeling parameters. After inverse Z-transform and matching with the model structure, the standardized equivalent transfer function model is output as a mathematical representation of the dynamic behavior of the actuator. Based on the dynamic characteristic parameter estimation sequence output by the recursive augmented least squares method, a stability determination method (parameters: convergence speed threshold and fluctuation amplitude threshold) is adopted to identify the stable state of the parameter sequence. Furthermore, by using a parameter set filtering method (parameters: variance threshold 0.001, mean drift threshold 0.0005), the final parameter set within the stable interval is extracted, and a ternary parameter set of recording time constant, pure time delay, and system gain is obtained. Furthermore, a model reduction processing method (parameters: discretization coefficients and sampling period) is adopted to map the discrete domain model coefficients to the first-order inertial plus pure time delay model basis in the continuous domain, and to generate the mathematical expression of each core parameter in the continuous time domain. Furthermore, the Z-transform inverse algorithm (parameters: sampling period h and discrete transfer function coefficient set) is used to realize the continuous domain transformation of discrete domain data, and the proportional relationship between time constant and lag time is corrected by combining the model structure matching method, thereby generating transfer function parameters that meet the standard. Furthermore, the equivalent transfer function model is expressed through formula calculation, where the transfer function takes the form:

[0016] in, The system gain represents the steady-state amplitude of the output with a unit input. is a time constant, representing the rate at which the system response reaches steady state; Pure time delay represents the time interval between the input action and the output response. For the Laplace operator; The model constructed using the above formula serves as a standardized mathematical representation of the dynamic behavior of the actuator, forming a data structure that can be directly called by the subsequent PID parameter mapping module. By using Z-transform inverse reasoning and structure matching, the converged dynamic parameter sequence is transformed into a standardized continuous domain transfer function model, achieving a full restoration of the dynamic characteristics of the actuator in the mathematical domain. For example, in a powder quantitative packaging system, the sampling period is set to 0.01 seconds. The estimated time constant output by the recursive augmented least squares method is stable within a range of 2.35 seconds with fluctuations less than ±0.002 seconds. The pure lag time series is stable within a range of 0.48 seconds with fluctuations less than ±0.001 seconds. The system gain sequence has a mean of 4.2 and a fluctuation range of ±0.003. The stability interval is identified using a stability threshold variance of 0.001 and a mean drift of 0.0005, and the aforementioned parameter set is extracted. The discrete model coefficients are converted to a continuous domain form through Z-transform to obtain the time constant. Pure delay time System gain The standardized model. According to the formula... A transfer function model was constructed, which had a mean square error of 0.0009 in the comparison between simulation and actual output, meeting the accuracy requirements of subsequent PID parameter mapping and achieving high consistency between control commands and actuator response characteristics. S3.5: Perform validity verification on the equivalent transfer function model. Calculate the mean square error index based on the dynamic trajectory comparison between the model simulation output and the actual acquired response. If the error exceeds a preset threshold, trigger the data resampling instruction and return to the initial data construction process. Otherwise, encapsulate the confirmed valid dynamic characteristic parameter set into a reliable model output for subsequent PID parameter mapping module calls, ensuring that the control strategy adjustment is based on high-precision modeling. Based on the equivalent first-order inertial plus pure time delay form transfer function model and its core parameters output from the preceding step S3.4, this model is selected as the input object for simulation verification.

[0017] like Figure 3 As shown, step S4 involves converting the model parameters into initial PID controller parameters based on the dynamic characteristic parameters of the equivalent transfer function model through a preset rule mapping relationship. This rule mapping relationship is generated by offline simulation of numerous typical actuator models and optimization using Ziegler-Nichols-type tuning criteria, outputting three basic control parameters: proportional gain, integral time, and derivative time. Specifically, this includes: S4.1: Based on the equivalent first-order inertia plus pure time delay form of the transfer function model output from the previous steps, obtain the three core parameters it contains: time constant, pure time delay, and system gain. These parameters are used as input conditions for rule mapping to establish a quantitative correlation with PID control parameters. S4.2: Based on the typical actuator dynamic model library built by offline simulation, and combined with the Ziegler-Nichols type tuning criterion, perform closed-loop response optimization simulation for each type of model to generate a dataset of corresponding optimal PID parameter combinations. Use this dataset to train the parameter mapping rule table to form the nonlinear mapping function relationship between time constant and proportional gain, pure time delay and integral time, and system gain and derivative time. The typical actuator dynamic model library built based on offline simulation uses model library data as input objects, covering multiple sets of first-order inertial plus pure time delay model parameters with different time constants, pure time delay and system gain, to achieve full coverage simulation of the response characteristics of common actuators in powder control system. Using Ziegler-Nichols-type tuning criteria (parameters: open-loop critical gain, critical oscillation period), closed-loop response simulations were performed for each type of dynamic model, and the proportional element was adjusted. Integral Time With differential time This achieves optimization of overshoot and rise time in the step response, resulting in the optimal combination of PID control parameters under the current model parameters. Furthermore, a multidimensional dataset is generated using the optimal parameter combination output from the closed-loop simulation, where sample attributes include time constants. Pure delay time System gain The three input parameters and the corresponding three output parameters, namely proportional gain, integral time, and derivative time, realize a structured mapping of input-output pairs; Furthermore, a nonlinear regression modeling method (parameters: polynomial regression order, regularization coefficient) is used to jointly fit the input and output of the dataset, and construct a nonlinear mapping function between time constant and proportional gain, pure time delay and integral time, and system gain and derivative time. The above nonlinear mapping function solidifies the quantization relationship obtained from offline simulation into a rule mapping table, realizing a repeatable conversion mechanism from the actuator dynamic characteristic parameter space to the PID control parameter space. For example, in an offline simulation environment for a powder packaging system, a dynamic model library containing 50 typical actuators is established, where the time constant range is... Instant seconds, pure time delay range is Instant seconds, system gain range is to For each set of parameters in the model library, closed-loop simulations were performed using Ziegler-Nichols-type tuning criteria, and the critical gain was determined. With critical oscillation period Using the formula , , Calculate the corresponding PID parameter set. Combine all model parameters with the optimal PID parameters to form a sample matrix, and perform third-order polynomial regression (with regularization coefficients of 1). Fit them respectively and , and , and The relationship is used to obtain a continuously differentiable nonlinear mapping function. The results of this step show that the fitted function curve can smoothly cover the response characteristics of different actuators, significantly improving the accuracy and stability of parameter transformation in the actual mapping process, and ensuring that the initial PID parameters matching the dynamic characteristics of the current actuator can be quickly generated during the online self-tuning stage; S4.3: Substitute the currently acquired time constant, pure time delay, and system gain into the trained nonlinear mapping function to calculate preliminary candidate values ​​for proportional gain, integral time, and derivative time, so as to achieve the initial conversion from actuator dynamic characteristics to PID parameter space; S4.4: Boundary constraint processing is performed on the calculated candidate PID parameters. Upper and lower limit thresholds are set based on the safe operating range of the industrial control site. Parameters exceeding the allowable range are truncated and corrected. A proportional-integral coordinated adjustment coefficient is introduced to verify the consistency of the parameter combination and generate an initial PID controller parameter set that meets the stability conditions. Based on the candidate PID controller parameter set output by S4.3, the input object includes proportional gain. Integral Time and differential time Three numerical values ​​and their corresponding calculation sources. A boundary constraint processing method (parameters: on-site operation safety specifications, rated control range of the actuator) is adopted to realize the legality verification and correction of candidate PID parameters; Furthermore, by setting upper and lower limit thresholds (parameters: upper and lower limit of proportional gain, shortest and longest integral time, shortest and longest derivative time), the values ​​of each parameter are compared between intervals, and truncation correction is performed on parameters that exceed the allowable range, so as to obtain a set of parameter values ​​that meet the conditions for safe industrial operation. Furthermore, a proportional-integral coordinated adjustment coefficient calculation method (parameters: proportional gain value, integral time value, coordinated adjustment objective function) is adopted to quantitatively evaluate the mutual matching degree between the proportional and integral components within the parameter group, and to fine-tune the parameter values ​​based on the coordinated adjustment results, so as to ensure the consistency of the proportional and integral components in dynamic response characteristics. Furthermore, by using a consistency check algorithm (parameters: adjusted proportional gain, integral time, derivative time, and closed-loop stability margin index), a global stability check of the combination of the three parameters is performed, generating a modified initial PID controller parameter set that meets the preset stability conditions. By using boundary constraints and coordinated adjustment, the candidate PID parameters from the previous step are transformed into initial PID parameter data that meet the safety and stability requirements of industrial sites and can be directly applied, thereby achieving the steady-state operation target of closed-loop control during the startup phase. For example, in a powder quantitative packaging system, candidate PID parameters = , = Second, = Seconds. The on-site safety operation specifications set the proportional gain range to be [value missing]. The integration time range is seconds, the differential time range is Seconds. Applying interval comparison and truncation correction processing, ... Cut off to , and The adjustment remains unchanged. During the coordinated adjustment phase, the adjustment coefficient is used. = right Make fine adjustments and correct the formula as follows: ,get The corrected parameter set was confirmed through closed-loop stability margin evaluation. The system meets the conditions for stable operation. In actual operation tests, the controller response time with this parameter group is shortened, the flow regulation overshoot is significantly reduced, and the system achieves precise and stable control of powder flow. S4.5: Encapsulate the verified initial PID controller parameters into a loadable format and output them to the control loop initialization module as the default control parameter configuration during the startup phase of the powder flow closed-loop control system, ensuring that the control algorithm has the adjustment capability that matches the dynamic characteristics of the actuator when it first runs.

[0018] Step S5: Load the initial PID controller parameters into the real-time control loop, and periodically trigger the excitation signal injection and data acquisition process during normal system operation to obtain the latest input-output time series data pairs under the current operating conditions, used to detect dynamic characteristic drift of the actuator caused by wear, temperature rise, or load changes. Specifically, this includes: S5.1: Based on the initial PID controller parameters output in the previous step S4, write them into the parameter register of the real-time control loop to initialize the three control parameters of proportional gain, integral time and derivative time, so as to ensure that the closed-loop control system has the adjustment capability to match the current actuator in the start-up phase and generate the initial control response characteristics. Based on the verified initial PID controller parameter set output from the preceding step S4, the three independent control parameters, proportional gain, integral time, and derivative time, are mapped to the parameter register address space of the real-time control loop. The hardware-level initialization loading of the controller parameters is achieved by using a register writing method (parameters: data width is 32-bit floating point type, writing mode is blocking confirmation response). Furthermore, a parameter consistency verification algorithm (parameters: verification method is CRC32, and the verification length includes a continuous storage unit containing all three parameters) is used to verify the integrity of the data after loading, and the verification pass flag is used as the confirmation basis for successful writing. Furthermore, by utilizing the parameter application instruction set of the control loop initialization module (parameters: including command codes for setting the proportional loop gain, setting the integral loop time constant, and setting the derivative loop time constant), the initial parameters in the registers are mapped to the coefficient register group of the PID calculation core, and a closed-loop control law matching the dynamic characteristics of the actuator is generated. Furthermore, by using a rapid detection method for system response characteristics (parameters: excitation signal amplitude limited to 3% of the rated control quantity, sampling period of 1ms), the output performance of the control loop under the action of initialization parameters is evaluated, and dynamic indicators such as initial overshoot, rise time, and settling time are generated. By combining the above-mentioned register writing, data verification, control law loading and initial performance detection, the initial PID parameters obtained in the previous step are transformed into real-time usable closed-loop control operation coefficients, so as to achieve the expected technical effect of matching the dynamic characteristics of the current actuator in the start-up phase of the closed-loop system. For example, in a powder quantitative packaging system, the actuator is a servo motor with a rated speed of 3000 rpm, and the initial PID parameters obtained by the calculation in step S4 are the proportional gain. Integral Time Second, differential time Seconds. In S5.1, the controller's parameter register starts at address 0x2000, the data bus width is 32 bits, and writing uses blocking mode to ensure that the next processing stage is only entered after the parameter update is complete. After the parameter is written, a CRC32 check is performed, and the obtained check value is completely consistent with the expected value, confirming that the data is correct. The proportional gain is adjusted using a dedicated command from the initialization module. Load the integral time into the PID proportional coefficient register. The seconds are loaded into the integral time constant register, and the derivative time is... The load is added to the differential time constant register in seconds. After loading is complete, the system outputs a small-amplitude step excitation signal, corresponding to a powder flow rate change of no more than 3% of the rated value. The sampling frequency is 1kHz, and the initial dynamic response of the closed-loop system is detected. The recorded rise time is 0.42 seconds, the overshoot is within the acceptable range, and the settling time is 0.95 seconds, verifying the high degree of matching between the PID parameters and the current dynamic characteristics of the actuator. S5.2: Set a periodic triggering mechanism to generate triggering instructions based on a preset time interval or the changing trend of control performance indicators, which is used to start a new round of lightweight excitation signal injection process. This triggering mechanism serves as the driving source for dynamic characteristic monitoring and outputs timed or adaptive triggering signals. After the real-time control loop is started and the initial control response characteristics are generated, the input conditions include the preset periodic monitoring time interval parameter, the performance index change trend data output by the closed-loop control performance evaluation module, and the current timestamp reference signal of the control system clock source or scheduling task management module. A periodic timed task scheduling method is adopted (parameter: fixed time interval). (Setting the detection cycle as required by the process), realizes the generation of time-driven excitation signal injection trigger conditions; Furthermore, by using a performance change trend analysis algorithm (parameters: inputs are the deviation convergence rate, overshoot, and steady-state error for several consecutive periods), the degradation state of control performance is identified, and supplementary parameters for adaptive triggering conditions are obtained. ; Furthermore, the time interval adjustment formula is used:

[0019] in For a new cycle, For a fixed period, This is a performance degradation adjustment coefficient; when the performance degradation trend is significant, Value less than To shorten the interval, the trigger cycle is adaptively shortened, and a new timed trigger schedule is generated; Furthermore, an event-driven and time-driven hybrid scheduling mechanism is adopted (parameter: simultaneously receiving trigger requests from the timed task table and performance monitoring events) to achieve unified queue management of trigger instruction generation, and to arrange the generated trigger instructions into the execution priority queue, ensuring that the control system can start the lightweight excitation signal injection process at the appropriate time; Furthermore, by using a trigger signal encoding and marking method (parameters including trigger type, generation time, and applicable actuator identifier), the trigger signal is encapsulated in a structured message format and transmitted to the execution strategy module, thereby achieving accurate identification and routing of trigger information within the control system. By combining periodic timed task scheduling with performance adaptive adjustment algorithms, the initial control parameter loading state of the previous step is transformed into time-varying adaptive trigger signal data, thereby achieving stable output and accurate triggering effect of the dynamic characteristic monitoring drive source. For example, in a set of rated control cycles of In a second-level powder quantitative packaging system, Set as The input to the performance change trend analysis algorithm is the most recent second. Deviation convergence speed per cycle Second, Second, Second, Seconds and Seconds, overshoot remains The steady-state error increased to The performance degradation judgment module outputs an adjustment coefficient. = Substitute into the formula A new cycle is derived Seconds. The scheduling module updates the scheduled task table, changing the trigger time from the original every [time period]. seconds changed to per The excitation signal injection task is executed once per second. The trigger instruction encapsulation includes the type marker "adaptive adjustment" and the system clock corresponding to the generation time. The solution, applicable to valve actuator A, achieved faster capture of actuator characteristic drift during the verification phase, maintaining constant overshoot and reducing deviation convergence speed. Second; S5.3: Based on the trigger command, the control system outputs a set of lightweight excitation signals, which are small-amplitude step signals or pseudo-random sequence signals, and acts on the input end of the actuator to stimulate it to produce identifiable dynamic response behavior, while avoiding significant disturbance to the powder flow rate, and generating a low-interference excitation signal sequence. After the periodic trigger signal has been generated by the previous step S5.2, the input objects are the trigger command output by the scheduling module and the drive interface parameter configuration of the actuator, including the current actuator type, electrical drive characteristics and mechanical response range; The excitation signal template matching method (parameters: actuator type, rated control range) is adopted to accurately select the template applicable to the current actuator and determine the signal form as a small amplitude step or pseudo-random binary sequence signal with an amplitude limited to 5%~10% of the rated control range; Furthermore, by using a time series discretization algorithm (parameters: sampling period and main control loop synchronization signal), the selected excitation signal is converted into a digital control sequence that is strictly consistent with the main control loop period, and time synchronization index data is generated to ensure the consistency of the time axis of subsequent response signal acquisition. Furthermore, a low-power drive loading method is adopted (parameter: drive power accounts for no more than 10% of rated power) to apply the discrete excitation digital sequence to the actuator drive channel, avoiding significant disturbance to the material flow and covering the main dynamic mode frequency bands of the actuator; Furthermore, by using a linear working interval monitoring algorithm (parameters: saturation detection threshold, dead zone length determination), the response characteristics of the actuator are detected in real time during the application of the excitation signal. If the response curve shows saturation or dead zone phenomena, the amplitude of the excitation signal is dynamically adjusted or the signal type is switched, and the discretization and loading process is re-executed to ensure that the obtained dynamic response meets the linear and time-invariant assumptions of the system identification. By using the above-mentioned signal template matching, discretization, low-power loading and linear interval monitoring processing methods, the trigger command is transformed into a low-interference signal sequence that can be directly used for dynamic excitation of the actuator, so as to achieve the expected technical effect of both exciting an identifiable dynamic response and not affecting the accuracy of powder flow measurement. For example, in a servo motor actuator used for high-precision powder packaging, the excitation form selected by the signal template matching method is a pseudo-random binary sequence (PRBS), with an amplitude of 0.08 times the rated control quantity and a sampling period of 0.01s, consistent with the frequency of the main control loop. A time-series discretization algorithm converts the PRBS signal into a 500-point digital sequence and adds a timestamp index. A low-power drive loading method applies this signal to the servo drive channel, with a drive power of 0.07 times the rated power. The measured initial response curve covers the main inertial modes and damping characteristics of the motor. A linear operating range monitoring algorithm detects that the response curve has no saturation or dead zone, determining the signal to be valid. The generated low-interference excitation signal sequence is recorded by the acquisition module. The signal energy distribution is uniform and the spectrum covers the target frequency band, significantly reducing the fitting residual identified by the subsequent system and greatly reducing the dynamic matching error of the control strategy. S5.4: Synchronously acquire the excitation signal as the input control quantity and the actual action response signal of the actuator as the output response quantity to form a new input-output time series data pair. The response signal includes the time trajectory of the change in motor speed or the change in valve opening, and obtain a measured dataset reflecting the dynamic characteristics of the actuator under the current working condition. S5.5: Preprocess the acquired input-output time series data pairs, including denoising, time alignment, and normalization, to eliminate the effects of sensor noise and communication delay, and generate high-quality dynamic response data that can be used for online system identification, serving as the data basis for detecting actuator characteristic drift.

[0020] Step S6: Based on the latest collected data, repeatedly perform the online system identification process, update the equivalent transfer function model at the current moment, and recalculate the corresponding PID controller parameters according to the updated model parameters to generate a new set of target control parameters. Specifically, this includes: S6.1: Obtain the latest input-output time series data pair after activation by the periodic triggering mechanism, where the input is a lightweight excitation signal and the output is the actual action response signal of the actuator; based on this data pair, perform parameter estimation using the recursive augmented least squares method to dynamically correct the influence of noise terms on system modeling and obtain a regression coefficient vector with time-varying adaptability. S6.2: Based on the regression coefficient vector obtained in the previous step, construct a discretized dynamic model of the actuator, and convert it into an equivalent first-order inertial plus pure time delay transfer function model in the continuous domain through Z-transform; the model contains three core parameters: time constant, pure time delay and system gain, which serve as quantitative indicators characterizing the dynamic characteristics of the actuator under the current working conditions. Based on the regression coefficient vector output by S6.1, a discrete-time system modeling method (parameters: including lag order, sampling period, and regression structure type) is used to map the coefficient vector to the coefficient set of discrete-domain difference equations. Furthermore, the discrete domain transfer function representation is generated analytically through the difference equation, and the controller sampling period is used as the time reference for the transformation from the Z domain to the continuous domain to obtain the discretized system function data containing zeros, poles and delay terms. Furthermore, the inverse Z-transform method (parameter: sampling period) is employed to realize the structural transformation from the discrete domain model to the equivalent continuous domain model, and low-order approximations corresponding to the main dynamic modes are selected. During this process, the equivalent delay time is calculated using the delay mapping formula for the pure time lag characteristic, where is the phase delay and is the angular frequency, ensuring that the lag time quantization accuracy meets the subsequent control parameter mapping requirements. Furthermore, by using model order reduction and parameter fitting methods (parameter: first-order inertial pure time delay model structure), the continuous domain model is fitted to a standard form, and the system gain K, time constant T, and pure time delay L are extracted from the fitting process. Furthermore, for the mapping from discretized coefficients to continuous domain parameters, variable substitution is performed according to the bilinear transformation formula to ensure the physical consistency and engineering usability of the time constant and gain value after the transformation; By constructing a discretized model, performing inverse Z-transformation, and processing parameter fitting, the regression coefficient vector results from the previous step are transformed into the continuous domain equivalent first-order inertia plus pure time delay transfer function parameter data of the actuator under the current operating conditions, thereby achieving a quantitative characterization of dynamic characteristics. For example, in a powder quantitative packaging system, the sampling period T is configured to be 0.05s. The regression coefficient vector obtained in S6.1 is [a1=-0.67, a2=0.12, b0=0.05, b1=0.04]. The discrete domain transfer function is obtained by modeling with difference equations. By inversely deducing through Z-transform and combining with the delay mapping formula to extract the pure time delay L=0.15s, the model is fitted with order reduction to obtain the continuous domain equivalent transfer function, where the system gain K=1.1 and the time constant T=0.35s. This parameter set is directly used in subsequent mapping steps to generate PID controller parameters that match the dynamic characteristics of the current actuator. Verification results show that the overshoot of the closed-loop system is significantly reduced, the settling time is shortened, and the response consistency remains stable over a long period of time. S6.3: Take the time constant, pure time delay, and system gain in the equivalent transfer function model generated in the previous step as input conditions and substitute them into the preset rule mapping relationship; this rule mapping relationship is generated based on a large number of typical actuator dynamic models in offline simulation and combined with Ziegler-Nichols-type tuning criteria, and is used to establish a nonlinear mapping channel from the model parameter space to the PID parameter space. S6.4: Based on the above rule mapping relationship, perform parameter transformation processing to calculate the corresponding proportional gain, integral time, and derivative time, three basic PID control parameters; this set of parameters constitutes a new set of target PID controller parameters, which are designed to match the actual dynamic response capability of the current actuator; Based on the mapped regular function relationship and the dynamic characteristic parameters of the equivalent transfer function model, a parameter mapping calculation method (input parameters: time constant T, pure time delay L, system gain K) is adopted to realize the quantitative conversion from the actuator model parameter space to the PID control parameter space. Furthermore, by inputting the nonlinear relationship between the system gain and the time constant into the calculation formula through the proportional gain calculation function, the proportional gain Kp is derived:

[0021] in, For system gain, It is a time constant. This is the pure time delay; By substituting the time constant and pure time delay into the formula using the PID integral time calculation function, the integral time Ti is obtained:

[0022] Among them, integration time Used to eliminate steady-state errors and accelerate system convergence; The time constant and pure time delay are converted into differential time Td according to a rule-based mapping relationship using the PID differential time calculation function:

[0023] Among them, differential time Used to improve the dynamic response performance of the system; Furthermore, by using a parameter consistency verification method (checking the boundary constraints of proportional gain, integral time, and derivative time within the safe operating range of industrial control), abnormal parameters are truncated and corrected, while maintaining the stability matching of the three control parameter combinations. Furthermore, through parameter result encapsulation and processing, the proportional gain, integral time, and derivative time are output in a control register adaptation format to generate a new set of target PID parameters to match the dynamic response capability of the current actuator; Through the above parameter mapping and calculation, the model parameters of the previous step are transformed into PID parameters that can be directly applied to the control loop, thereby achieving the optimization effect of the control strategy that adapts to the response characteristics of the actuator. For example, in a powder quantitative packaging system, the updated actuator equivalent transfer function model parameters are: time constant T = 0.35s, pure time delay L = 0.12s, and system gain K = 3.5. The proportional gain calculation formula is used:

[0024] The calculated proportional gain is approximately 2.94. The integral time calculation formula is used as follows:

[0025] The integration time is 0.59 seconds. The differential time is calculated using the following formula:

[0026] The calculation result was 0.06 seconds. Boundary constraint testing confirmed that all parameters were within the set safe range, and these were encapsulated into a new set of target PID controller parameters: Kp=2.94, Ti=0.59s, Td=0.06s. After applying these parameters, the system exhibited significantly reduced overshoot, shorter settling time, and decreased steady-state error in powder flow regulation tests, achieving rapid matching with changes in actuator response characteristics and improved flow control performance. S6.5: Temporarily store the newly generated target PID controller parameters in the control buffer area as input objects for the subsequent parameter smooth transition mechanism; at the same time, mark the group of parameters as pending activation, and wait for step S7 to determine whether to start the update process based on the difference judgment result, so as to ensure the continuity and stability of the control loop operation.

[0027] Step S7: Determine whether the difference between the parameters of the old and new PID controllers exceeds a preset threshold. If it does, activate the parameter smoothing transition mechanism, using an exponentially weighted average method to progressively update the proportional gain, integral time, and derivative time item by item, to avoid system disturbances caused by sudden changes in control output. Specifically, this includes: S7.1: Obtain the PID controller parameters applied in the previous cycle as a historical control parameter set. The historical control parameter set includes three basic parameters: proportional gain, integral time, and derivative time. Based on this parameter set as a reference input, it is used to compare the parameters with the newly calculated target control parameters item by item to identify the change in control parameters caused by the drift of the actuator dynamic characteristics. S7.2: Calculate the difference between the parameters of the old and new PID controllers. Specifically, calculate the absolute deviation values ​​of the proportional gain, integral time and derivative time, and compare each deviation value with the corresponding preset parameter change threshold. If the deviation of any parameter exceeds its corresponding threshold, it is determined that the control parameter has changed significantly, triggering the subsequent smooth update logic to prevent the control output from jumping drastically and affecting the stability of the powder flow rate. S7.3: Generate control decision signals based on the judgment results. If the difference does not exceed the preset threshold, the current PID parameters remain unchanged and the current adjustment process ends. If the threshold is exceeded, a start command is output to the parameter smoothing transition module to activate the exponential weighted average mechanism and prepare to perform progressive fusion processing on the new and old PID parameters to generate a dynamic control parameter sequence in the intermediate transition state. Based on the difference magnitude calculation results in step S7.2, a conditional judgment algorithm is adopted (input parameters: absolute deviation values ​​of new and old proportional gain, integral time, and derivative time, and corresponding preset thresholds) to use the difference state of each parameter as logical judgment input to realize the branch selection function of the control strategy. Furthermore, by using the logical AND gate operation method (parameter: three Boolean values ​​for difference judgment), the aggregated evaluation of significant change conditions is realized, and unified change state identification data is generated for use as the triggering basis for subsequent control decisions; Furthermore, for the changed state identifier data, a state machine-driven mechanism (parameters: state transition table and trigger condition set) is adopted to realize the switching between two operating modes: "keeping parameters unchanged" and "smooth transition of starting parameters", and obtain the mode selection signal output; Furthermore, when the system determines that it has entered the "maintain parameter unchanged" mode, the parameter holding processing unit (parameter: current PID parameter register value) is used to realize the function of discarding new parameters and continuing to run the current control parameters, and generates an adjustment process end signal; Furthermore, when determining to enter the "startup parameter smooth transition" mode, the decision signal encoding module (parameters: mode selection signal, update instruction code) realizes the standardized encoding and output of the startup instruction, and sends the instruction to the parameter smooth transition module, preparing to perform an exponential weighted average operation on the proportional gain, integral time and derivative time in S7.4 to generate a dynamic control parameter sequence for the intermediate transition state; Through the above condition judgment and mode switching algorithm, the results of the difference magnitude analysis are transformed into precise control mode selection data, so as to achieve reasonable control of the parameter update time node and avoid control performance fluctuations caused by over-adjustment. For example, in a certain powder packaging production line, the historical proportion gain is... The newly calculated proportional gain is The set threshold for proportional gain change is Historical points time is seconds, the newly calculated integration time is Seconds, set the change threshold to Seconds; historical differential time is seconds, the newly calculated differential time is Seconds, set the change threshold to Seconds. Using the absolute deviation formula. Calculate the differences of the three terms separately to obtain the proportional gain difference as follows: The integral time difference is The differential time difference is All three parameters exceeded their respective thresholds. The AND gate returned a true result, the state machine transitioned to the "smooth transition of startup parameters" mode, and the encoding module generated update instruction codes. The data is then sent to the S7.4 module, where the proportional gain, integral time, and derivative time will be adjusted based on the exponential weighted average strategy. On-site monitoring shows that the flow fluctuation amplitude has been significantly reduced, the actuator response curve exhibits continuous and smooth transition characteristics, and the stability of the control system operation has been significantly improved. S7.4: Perform incremental parameter updates under the exponential weighted average method, that is, recursively calculate the three parameters of proportional gain, integral time and derivative time respectively, and use a weighted strategy to gradually approach the target parameter value. The smoothing coefficient is preset according to the system response inertia to ensure that the control command output rhythm matches the response capability of the actuator. Based on the determination result of parameter difference exceeding the threshold in step S7, the proportional gain between the historical control parameter set and the target control parameter set is obtained. Integral Time and differential time Three values ​​serve as inputs for the exponentially weighted average smooth transition; The parametric recursive calculation method is used (parameter: smoothing coefficient). The proportional gain is gradually updated using historical parameter values ​​and target parameter values. The calculation formula is as follows:

[0028] in, For the updated proportional gain parameters, The proportional gain parameter of the previous control cycle. For the newly calculated target proportional gain parameter, Used to control the update amplitude, the value is based on the inertial response analysis results of the actuator; Furthermore, the smoothing adjustment values ​​for the integral time and the derivative time are calculated using the same recursive formula, as follows:

[0029]

[0030] An iterative update mechanism is adopted, in which the calculated intermediate parameter values ​​are gradually replaced with the old values ​​in the real-time control loop at a fixed iteration period, and the change of the response curve is monitored based on the flow feedback signal after each iteration to verify whether the update step size is adapted to the dynamic response characteristics of the current actuator. Furthermore, by adaptively adjusting the calculation method of the smoothing coefficient α, its value is dynamically changed based on the statistical analysis of instantaneous overshoot and steady-state time, so as to achieve a balance between update speed and system stability, and enable the parameter approximation process to adapt quickly while avoiding the risk of oscillation in the control loop. By using the above exponential weighted average recursive calculation, the target control parameters from the previous step are progressively integrated with the historical control parameters to output an intermediate PID parameter set, thereby achieving precise matching between the control command output rhythm and the actual dynamic capability of the actuator. For example, in the control system of a powder quantitative packaging machine, the historical control parameters are as follows: , Second, seconds, target parameters are respectively , Second, seconds, smoothing coefficient = Substituting into the formula, the intermediate proportional gain of the first cycle is calculated as follows: The integration time is seconds, the differential time is Seconds. After loading the above intermediate parameters, the real-time control loop shows that the powder flow response curve is smooth and without oscillation, and the actuator action is highly consistent with the controller output rhythm, achieving significantly improved control stability; S7.5: Outputs smoothed intermediate PID parameters to the real-time control loop and continuously monitors the dynamic response behavior of the powder flow feedback signal during parameter switching. Records control performance indicators such as overshoot, settling time, and steady-state error as a basis for evaluating parameter transition effects. If abnormal fluctuations are detected, the update is interrupted and the system reverts to the original parameter group to ensure system operation safety and control robustness.

[0031] Step S8: Continuously monitor the deviation between the powder flow feedback signal and the set value, perform closed-loop adjustment in conjunction with the parameters of the currently applied PID controller, and record the control performance indicators. If the deviation convergence speed decreases or the overshoot increases within several consecutive adjustment cycles, shorten the trigger interval for the next system identification to achieve enhanced dynamic adaptability. Specifically, this includes: S8.1: Obtain the real-time flow feedback signal collected by the powder flow sensor and the preset target flow setpoint, calculate the instantaneous deviation between the two, and use it as the input condition for closed-loop control adjustment; use the deviation to perform PID adjustment calculation, output the corresponding control command to the actuator to drive it to adjust the action amplitude, and generate a precise flow regulation response. S8.2: Based on the deviation sequence data within multiple consecutive control cycles, calculate the control performance indicators, including three core parameters: deviation convergence speed, overshoot, and settling time; use the above indicators as a quantitative basis for evaluating the adaptability of the current PID controller parameters to form a discriminable performance degradation feature set; S8.3: Determine whether the recorded control performance indicators show a continuous deterioration trend. Specifically, if the deviation convergence speed decreases by more than the preset ratio or the overshoot increases by more than the threshold within three or more consecutive adjustment cycles, it is determined that the current control performance has significantly deteriorated, triggering the activation conditions of the dynamic adaptive enhancement mechanism. S8.4: Based on the performance degradation judgment result, dynamically shorten the trigger interval of the next online system identification process; by multiplying the original period by an adjustment coefficient less than 1 (such as 0.7) to generate a new sampling interval, increase the actuator characteristic detection frequency to accelerate the ability to capture drift states. S8.5: Write the updated system identification trigger interval into the timed task scheduling module, which will serve as the time reference for the periodic trigger excitation signal injection and data acquisition process in S5, thereby realizing a closed-loop optimization link from control performance monitoring to adaptive adjustment of system identification rhythm.

[0032] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0033] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and rules of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for precise control and real-time adjustment of powder flow rate for quantitative powder packaging, characterized in that, Includes the following steps: S1: When the powder quantitative packaging system is started or reset, the control system outputs an excitation signal; S2: Synchronously acquire the excitation signal as the input control quantity and the actual action response signal of the actuator as the output response quantity to form an input and output time series data pair; S3: Perform online system identification processing on the input and output time series data pairs, estimate the dynamic characteristic parameters of the actuator using the recursive augmented least squares method, and obtain its equivalent first-order inertial plus pure time delay form of the transfer function model. S4: Based on the dynamic characteristic parameters of the equivalent transfer function model, the model parameters are converted into initial PID controller parameters through a preset rule mapping relationship, and the basic control parameters are output, specifically including: Based on the transfer function model output in step S3, its core parameters are obtained and used as input conditions for rule mapping. Based on the typical actuator dynamic model library built by offline simulation, closed-loop response optimization simulation is performed on each type of model in combination with Ziegler-Nichols tuning criteria to generate a dataset of corresponding optimal PID parameter combinations. The dataset is then used to train a parameter mapping rule table to form a nonlinear mapping function relationship. Substitute the core parameters of the currently acquired transfer function model into the trained nonlinear mapping function to calculate the preliminary candidate PID parameters. The candidate PID parameters are subjected to boundary constraint processing. The upper and lower limit thresholds are set based on the safe operating range of the industrial control site. Parameters that exceed the allowable range are truncated and corrected. The proportional-integral coordinated adjustment coefficient is introduced to verify the consistency of the parameter combination and generate the verified PID controller parameter group. Specifically, the nonlinear mapping function relationship table is formed by using a nonlinear regression modeling method to jointly fit the input and output of the dataset and constructing nonlinear mapping functions between time constant and proportional gain, pure time delay and integral time, and system gain and differential time. S5: Load the initial PID controller parameters into the real-time control loop, and periodically trigger the excitation signal injection and data acquisition process during normal system operation; S6: Based on the latest collected data, repeat the online system identification process, update the equivalent transfer function model at the current moment, and recalculate the corresponding PID controller parameters according to the updated model parameters; S7: When the difference between the old and new PID controller parameters exceeds a preset threshold, the parameter smooth transition mechanism is activated to progressively update the basic control parameters item by item.

2. The method for precise control and real-time adjustment of powder flow rate for quantitative powder packaging according to claim 1, characterized in that, Following step S7, the following is also included: S8: Continuously monitor the deviation between the powder flow feedback signal and the set value, perform closed-loop regulation in combination with the parameters of the currently applied PID controller, and record the control performance indicators. If the deviation convergence speed decreases or the overshoot increases within several consecutive regulation cycles, the trigger interval for the next system identification will be shortened.

3. The method for precise control and real-time adjustment of powder flow rate for quantitative powder packaging according to claim 1, characterized in that, The excitation signal is a small-amplitude step signal or a pseudo-random sequence signal.

4. The method for precise control and real-time adjustment of powder flow rate for quantitative powder packaging according to claim 1, characterized in that, The actual action response signal includes the time trajectory of changes in motor speed or valve opening.

5. The method for precise control and real-time adjustment of powder flow rate for quantitative powder packaging according to claim 1, characterized in that, Step S3 specifically includes: Based on the input and output time series data pairs generated in step S2, an initial identification dataset is constructed. The initial identification dataset is preprocessed to output the reconstructed input and output regression structure; Based on the reconstructed input and output regression structure, the dynamic parameters of the actuator are estimated online. Each time a new data point is received, a parameter update iteration is performed. The proportional gain, time constant, and the discretized model coefficients mapped by the pure time delay are successively corrected, and the dynamic characteristic parameter estimation sequence is output. Based on the dynamic characteristic parameter estimation sequence, the final stable parameter set is extracted and model reduction processing is performed to convert it into a first-order inertial plus pure time delay transfer function in the continuous domain, and a standardized equivalent transfer function model is output. The standardized equivalent transfer function model is validated. Based on the dynamic trajectory comparison between the model simulation output and the actual acquisition response, the mean square error index is calculated. If the mean square error index exceeds a preset threshold, a data resampling instruction is triggered and the initial data construction process is returned. Otherwise, the set of valid dynamic characteristic parameters is encapsulated as a reliable model output.

6. The method for precise control and real-time adjustment of powder flow rate for quantitative powder packaging according to claim 5, characterized in that, The transfer function model includes three core parameters: time constant, pure time delay, and system gain.

7. The method for precise control and real-time adjustment of powder flow rate for quantitative powder packaging according to claim 1, characterized in that, Step S5 specifically includes: Based on the initial PID controller parameters output in step S4, write them into the parameter register of the real-time control loop, initialize the three control parameters of proportional gain, integral time and derivative time, and generate the initial control response characteristics. Set up a periodic triggering mechanism to generate trigger commands based on preset time intervals or changes in control performance indicators; Based on the trigger command, the control system outputs an excitation signal, which acts on the input end of the actuator to stimulate it to produce a identifiable dynamic response behavior and generate a low-interference excitation signal sequence. The excitation signal is simultaneously acquired as the input control quantity, and the actual action response signal of the actuator is acquired as the output response quantity, forming a new input-output time series data pair; The new input-output time series data pair is preprocessed to generate high-quality dynamic response data.

8. The method for precise control and real-time adjustment of powder flow rate for quantitative powder packaging according to claim 7, characterized in that, The preprocessing includes denoising, time alignment, and normalization.

Images