Method for intelligently regulating and controlling moisture content during production of nylon reinforced material

Abstract

The invention belongs to the technical field of material production, and provides a method for intelligently regulating and controlling the moisture content during production of a nylon reinforced material, which comprises the following steps of: obtaining the predicted moisture content and the predicted confidence coefficient after pelletizing on the basis of various parameters which are acquired in real time, come from a production line and are associated with the moisture content; on the basis of the difference value between the predicted moisture content after grain cutting and the preset target moisture content, a feed-forward regulation and control instruction is calculated and generated through a preset first control algorithm; on the basis of the difference value between the actual water content, obtained through real-time measurement, after grain cutting and the preset target water content, a feedback regulation and control instruction is generated through calculation of a preset second control algorithm; dynamically calculating and generating a feedforward weight and a feedback weight based on the prediction confidence; based on the feed-forward weight and the feedback weight, the feed-forward regulation and control instruction and the feedback regulation and control instruction are subjected to weighted fusion, a comprehensive control instruction is obtained and sent to a drying system, and accurate regulation and control of the moisture content in the nylon reinforced material production process are achieved.

Description

Technical Field

[0001] This invention relates to the field of materials production technology, and in particular to a method for intelligent control of moisture content during the production of nylon reinforced materials. Background Technology

[0002] Nylon-reinforced materials are widely used in the automotive, electronics, and aerospace industries due to their excellent mechanical properties and processing adaptability. The moisture content after pelletizing is a core indicator determining the consistency of product performance—excessive moisture leads to decreased mechanical properties, increased surface defects, and even melt processing cracking; insufficient moisture, on the other hand, can cause embrittlement and shorten product lifespan. Therefore, accurately controlling the moisture content after pelletizing is crucial for ensuring product quality stability.

[0003] Traditional production processes typically include core steps such as raw material premixing, melt extrusion, water cooling of the strips, pelletizing, homogenization, and drying. Among these, the drying step, as a crucial dehumidification process, faces significant technical bottlenecks: high energy consumption, long drying time, and low control precision limit production efficiency; fixed heating modes are prone to localized overheating or uneven drying, further exacerbating product quality fluctuations.

[0004] Existing control strategies often rely on empirical adjustment of a single parameter (such as ambient humidity or cooling water temperature) or on PID feedback control with a fixed threshold. These methods are ill-suited to complex operating conditions such as fluctuations in the initial moisture content of raw materials and changes in melt pressure. They result in large deviations in moisture content control, significant fluctuations in product yield, and an inability to meet the demands of high-precision production.

[0005] Traditional control modes lack dynamic response capabilities. They fail to effectively utilize the confidence information of predictive models, employing aggressive adjustments even with low prediction confidence, which can easily lead to system oscillations. Conversely, if a conservative mode is used with high confidence, it struggles to respond quickly to changes in operating conditions, resulting in control lag. This response mode cannot adapt to the dynamic fluctuations of production processes.

[0006] Therefore, it is necessary to provide a method for intelligent control of moisture content during the production of nylon reinforced materials to solve the above-mentioned technical problems. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides an intelligent method for controlling the moisture content during the production of nylon reinforced materials. Through multi-parameter fusion prediction, dual-control algorithm collaboration, dynamic response mode switching, and closed-loop optimization mechanism, the method achieves precise control of the moisture content during the production of nylon reinforced materials.

[0008] This invention provides a method for intelligently controlling the moisture content during the production of nylon reinforced materials, the method comprising the following steps: Based on real-time collection of multiple parameters from the production line that are related to moisture content, the predicted moisture content after pelleting and the corresponding prediction confidence level are obtained. Based on the difference between the predicted moisture content after pelleting and the preset target moisture content, a feedforward control command is generated by calculating using a predetermined first control algorithm. Based on the difference between the actual moisture content after pelleting obtained by real-time measurement and the preset target moisture content, a feedback control command is generated by a predetermined second control algorithm. Weighting coefficients are dynamically calculated based on the predicted confidence level, wherein the weighting coefficients include feedforward weights and feedback weights. Based on the feedforward weight and feedback weight, the feedforward control command and the feedback control command are weighted and fused to obtain the final comprehensive control command; The integrated control command is sent to the drying system, and the response mode of the drying system is adaptively adjusted according to the comparison result of the predicted confidence level and the dynamic switching threshold, wherein the dynamic switching threshold is generated based on the distribution characteristics of the historical predicted confidence level.

[0009] Preferably, the step of obtaining the predicted moisture content and corresponding prediction confidence level after pelletizing based on multiple parameters collected in real time from the production line and related to moisture content includes: Real-time data collection of raw material initial moisture content, ambient humidity, extruder melt pressure, melt temperature, and cooling water temperature in the production line; The initial moisture content of the raw materials, ambient humidity, extruder melt pressure, melt temperature and cooling water temperature are collected as input parameters and input into the pre-trained moisture value chain prediction model. The input parameters are mapped and processed by the moisture value chain prediction model to output the predicted moisture content after pelleting and the corresponding prediction confidence level.

[0010] Preferably, the step of calculating the difference between the predicted moisture content after pelleting and the preset target moisture content using a predetermined first control algorithm to generate a feedforward control command includes: The difference between the predicted moisture content after pelleting and the target moisture content is calculated as the prediction control error; The predictive control error is input into the predetermined first control algorithm; The first control algorithm processes the predictive control error and outputs the feedforward control command.

[0011] Preferably, the difference between the actual moisture content after pelleting obtained from real-time measurement and the preset target moisture content is calculated using a predetermined second control algorithm to generate feedback control commands, including: The difference between the actual moisture content after pelleting and the target moisture content is calculated as the actual control error; The actual control error is input into the predetermined second control algorithm; The second control algorithm processes the actual control error and outputs the feedback control command.

[0012] Preferably, both the first control algorithm and the second control algorithm are PID control algorithms.

[0013] Preferably, the weight coefficients are dynamically calculated based on the predicted confidence level, wherein the weight coefficients include feedforward weights and feedback weights, including: Based on a preset nonlinear mapping relationship, feedforward weights are calculated from the predicted confidence level, wherein the nonlinear mapping relationship is configured such that when the predicted confidence level is lower than a predefined first threshold, a first preset ratio is applied. Reduce the feedforward weights; when the prediction confidence is higher than a predefined second threshold, adjust according to a second preset ratio. Increase the feedforward weight, wherein the second threshold is greater than the first threshold. , ; The feedback weights are generated based on the feedforward weights.

[0014] Preferably, the generation of the dynamic switching threshold includes: Collect multiple historical prediction confidence data within a preset historical time period to form a historical prediction confidence dataset; Statistical analysis was performed on the historical prediction confidence dataset to determine the distribution characteristic parameters; The dynamic switching threshold is calculated and generated based on the distribution characteristic parameters.

[0015] Preferably, the step of adaptively adjusting the response mode of the drying system based on the comparison result of the predicted confidence level and the dynamic switching threshold includes: The predicted confidence level is numerically compared with the dynamic switching threshold to generate a comparison result; The response mode type is determined based on the comparison results, wherein the response mode type includes a conservative response mode and an aggressive response mode. When the prediction confidence is greater than or equal to the dynamic switching threshold, the conservative response mode is selected, otherwise the aggressive response mode is selected. Configure the control parameters of the drying system according to the response mode type, and apply the configured control parameters to the drying system.

[0016] Compared with related technologies, the intelligent control method for moisture content during the production of nylon reinforced materials provided by this invention has the following beneficial effects: This invention integrates five key parameters, including initial moisture content of raw materials, ambient humidity, extruder melt pressure, melt temperature, and cooling water temperature, and outputs the predicted moisture content and confidence level after pelletizing through a pre-trained moisture value chain prediction model, thereby improving prediction accuracy. A weighted fusion strategy of feedforward PID control (based on prediction error) and feedback PID control (based on actual error) is adopted, and the feedforward / feedback weights are dynamically calculated by combining the prediction confidence to achieve accurate synthesis of control commands; A dynamic switching threshold is generated based on the distribution characteristics of historical prediction confidence. When the prediction confidence is higher than the threshold, an aggressive response mode (such as large step adjustment) is adopted, and when it is lower than the threshold, a conservative response mode (such as small step adjustment) is switched to adaptively to production fluctuations. Through a closed-loop control process of "prediction-feedforward-feedback-weight fusion-mode switching", the deviation between the actual moisture content after pelleting and the target value is reduced, the product qualification rate is improved, and energy consumption is reduced. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a method for intelligently controlling the moisture content during the production of nylon reinforced materials, as provided by the present invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the drawings, not all structures. Moreover, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0019] It should also be noted that, for ease of description, the accompanying drawings show only the parts relevant to the invention and not all of them. Before discussing exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the operations (or steps) as sequential processes, many of the operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but it may also have additional steps not included in the drawings. The process may correspond to a method, function, procedure, subroutine, subroutine, etc.

[0020] It should be noted that the intelligent moisture content control method for nylon reinforced material production described in this invention relies on an integrated intelligent control system. The physical structure of this system mainly includes a data acquisition layer, a computation and control core layer, and an actuator layer. These layers interact and transmit commands through an industrial network, collectively forming a closed-loop control system.

[0021] In terms of physical structure, the data acquisition layer is responsible for acquiring various parameters related to moisture content in real time. It consists of a series of sensors, including a raw material moisture detection unit located before the raw material premixing section, environmental parameter sensors distributed in the production workshop and key workstations, process parameter sensors integrated on the twin-screw extruder, cooling water temperature sensors installed in the cooling water pipeline, and an online moisture meter located at the pelletizer outlet. The computing and control core layer consists of an industrial computer, including a data integration and communication module, a moisture value chain prediction model, an intelligent decision-making and control module, and an adaptive mode scheduler. The actuator layer mainly consists of the drying system, including the actuators of equipment such as dehumidifying dryers and hot air circulation systems.

[0022] The sensors in the data acquisition layer upload real-time data to the data integration and communication module of the computing and control core layer via fieldbus or industrial Ethernet. The output of the moisture value chain prediction model inside the core layer is transmitted to the intelligent decision and control module. This module also receives feedback data from the online moisture meter. After completing the calculation, it sends the comprehensive control command and mode selection signal to the drying system in the actuator layer, thus forming a closed-loop control loop of "measurement-prediction-decision-execution-remeasurement".

[0023] The system works by comprehensively sensing the status information of the entire production process through the data acquisition layer, using a predictive model to proactively estimate moisture change trends through the core computing and control layer, and combining this with real-time feedback data to generate optimal control commands through a confidence-driven fusion method. Finally, the actuator layer dynamically adjusts the drying process based on the commands and adaptive modes, thereby achieving high-precision, low-energy-consumption intelligent control of the moisture content of nylon reinforced materials. For details, please refer to Figure 1 As shown, the control method includes the following steps: S1: Based on real-time data collected from the production line and related to moisture content, the predicted moisture content after pelletizing and the corresponding prediction confidence level are obtained.

[0024] Specifically, step S1 includes the following steps: S11: Real-time data collection of raw material initial moisture content, ambient humidity, extruder melt pressure, melt temperature, and cooling water temperature in the production line.

[0025] In this embodiment, before starting the production line, the moisture content of the nylon base material in the raw material silo is first detected. An online near-infrared moisture meter (measuring range 0-5%) is used to scan the raw material on the conveyor belt in real time, acquiring the initial moisture content data of the raw material every 30 seconds. Under normal production conditions, this value is typically between 0.8% and 1.5%. Ambient humidity is collected by digital temperature and humidity sensors installed near the premixing zone and the extruder inlet, with a monitoring range covering 10%-90%RH, and a sampling frequency set to once per second.

[0026] The extruder melt pressure (measurement range 0-40MPa) and melt temperature (measurement range 150-300℃) are collected by a melt pressure sensor and a thermocouple, respectively. The sensors are installed from the fifth zone of the extruder to the die head, and the data is uploaded in real time with a period of 100ms. The cooling water temperature is monitored using a PT100 resistance temperature detector (measurement range 5-50℃), installed in the return pipe of the cooling water tank, and records data twice per second.

[0027] S12: The collected initial moisture content of raw materials, ambient humidity, extruder melt pressure, melt temperature and cooling water temperature are used as input parameters and input into the pre-trained moisture value chain prediction model.

[0028] In this embodiment, the data preprocessing module first performs timestamp alignment and outlier filtering on the received multi-source heterogeneous data. Then, it combines five parameters—initial raw material moisture content, ambient humidity, melt pressure, melt temperature, and cooling water temperature—into a standardized input vector. This vector is pushed to a moisture value chain prediction model deployed on an industrial server via the OPCUA protocol. This prediction model employs a multilayer perceptron (MLP) architecture based on an attention mechanism, and its network structure specifically includes the following components: Input layer: It has 5 neuron nodes, corresponding to the five input parameters mentioned above. Each input feature is Z-score standardized before entering the network, so that its mean is 0 and its standard deviation is 1, which effectively improves the stability of model training.

[0029] Feature fusion layer: A three-layer fully connected neural network structure is adopted (with 64, 32, and 16 neurons respectively), with each layer followed by a ReLU activation function and a BatchNormalization layer. In particular, an attention mechanism module is embedded between the second and third layers. This module automatically calculates the importance score of each feature dimension through trainable weight parameters (with melt pressure feature weight accounting for approximately 35% and initial moisture content of raw materials feature weight accounting for approximately 30%), realizing dynamic focusing on key parameters.

[0030] Output Layer: Employs a dual-branch structure. The first branch outputs the predicted moisture content after pelletizing via a linear activation function, typically ranging from 0.05% to 0.3%. The second branch outputs the prediction confidence score (dimensionless, ranging from 0 to 1) via a sigmoid activation function, obtained by nonlinearly transforming the prediction variance calculated internally by the network. The model is trained on historical production data, using the Adam optimizer (learning rate 0.001) to minimize the mean squared error between the predicted and actual values, with 500 training iterations. When the input parameters are within the typical operating conditions covered by the model training data (e.g., melt pressure 20-25 MPa, raw material moisture content 1.0%-1.2%), the confidence score can reach above 0.85; when abnormal parameter combinations occur, the confidence score automatically drops below 0.6. The final output data is encapsulated in JSON format for use by downstream control modules.

[0031] S13: The input parameters are mapped by the moisture value chain prediction model to output the predicted moisture content after pelleting and the corresponding prediction confidence level.

[0032] In this embodiment, the importance of features is calculated in parallel using multiple decision trees within the model. Melt pressure (with a weight of approximately 35%) and initial moisture content of raw materials (with a weight of approximately 30%) are identified as key influencing factors.

[0033] The model aggregates the prediction results of all trees through an ensemble learning mechanism and outputs two key values: one is the predicted moisture content after pelleting (unit: %), which is usually between 0.05% and 0.3%; the other is the prediction confidence (dimensionless, value range 0-1), which is obtained by calculating the dispersion of the prediction results of each subtree.

[0034] When the input parameters are within the typical operating conditions covered by the model training data (e.g., melt pressure 20-25 MPa, raw material moisture content 1.0%-1.2%), the confidence level can reach above 0.85; when abnormal parameter combinations occur (e.g., cooling water temperature suddenly rises by more than 5°C), the confidence level will automatically drop below 0.6. The final output data is encapsulated in JSON format, containing information such as timestamp, predicted value, and confidence level, for use by downstream control modules.

[0035] S2: Based on the difference between the predicted moisture content after pelleting and the preset target moisture content, a feedforward control command is generated by calculating using a predetermined first control algorithm.

[0036] Specifically, step S2 includes the following steps: S21: Calculate the difference between the predicted moisture content after pelleting and the target moisture content, as the prediction control error.

[0037] In this embodiment, after obtaining the predicted moisture content after pelletizing from the moisture value chain prediction model, the predicted value is compared with a preset target moisture content. The target moisture content is preset according to product grade requirements, typically set to 0.15% for PA66-GF30 injection molding grade materials (allowable fluctuation range ±0.05%). The error calculation module uses... The formula is used for calculation, where To predict moisture content, The target moisture content is set at 10 times per second. When the predicted value deviates from the target value by more than 0.1%, it is automatically marked as a significant deviation. The error value is displayed in real-time on the control interface and simultaneously transmitted to a historical database for archiving and subsequent statistical analysis. Under normal production conditions, this error value typically remains within ±0.08%.

[0038] S22: Input the predictive control error into the predetermined first control algorithm.

[0039] In this embodiment, the calculated predictive control error is transmitted to the PID control algorithm processing unit via a real-time data bus. The algorithm employs a positional digital PID architecture, and its discrete control formula is:

[0040] In the formula, Representing the The output value of the controller at each sampling time, that is, the original value of the feedforward control command; This is a discrete time series index, representing the current sampling time. Representing the The input error value at each sampling time point, in this scheme specifically refers to the predictive control error. That is, the difference between the moisture content predicted by the model and the target moisture content; This is the proportional gain coefficient (value 2.8), used to amplify the current error signal and determine the system's response strength to instantaneous deviations; The sampling period of the controller (value is 100ms), which is the time interval for calculating and updating control commands; The integral time constant (value 200s) is used to adjust the speed at which the controller eliminates historical accumulated errors. Represents the time from the start time to the current time. The sum of all errors, i.e. the integral term, reflects the historical cumulative effect of the errors; The differential time constant (valued at 25s) is used to adjust the ability to suppress the trend of error change; This represents the difference between the error at the current time and the error at the previous time, i.e., the differential term. It reflects the rate of change of the error and is used to predictively suppress overshoot and improve system stability.

[0041] The overall control logic of this algorithm consists of the superposition of proportional, integral, and derivative actions: the proportional action (P) reacts immediately and proportionally to the magnitude of the current predicted deviation; the integral action (I) clears historical accumulated deviations, specifically eliminating small but persistent steady-state errors; and the derivative action (D) performs predictive control based on the deviation change trend, suppressing rapid deviation growth and allowing the system to smoothly approach the target value. In this application, the parameter-tuned PID algorithm can intelligently generate forward-looking feedforward control commands based on the future moisture deviation output by the prediction model, thereby correcting the deviation before it actually occurs.

[0042] S23: The first control algorithm performs calculations on the predictive control error and outputs the feedforward control command.

[0043] In this embodiment, the PID algorithm performs adaptive calculations based on error characteristics: when When using P control mode; When using PI control mode; The full PID control mode is activated. The algorithm output undergoes output limiting processing, constraining the command value to a standard unit range of -4.0 to +4.0, corresponding to a temperature adjustment range of -12℃ to +12℃ for the drying system (conversion factor of 3℃ / unit). The final generated feedforward control command contains three core pieces of information: basic adjustment amount, adjustment direction (heating / cooling), and duration (default 2 sampling periods, 200ms). This command is encapsulated using the Modbus TCP protocol, accompanied by a CRC checksum and time synchronization identifier, and sent to the command buffer queue for subsequent weight fusion processing. Key parameters during command generation, including real-time error values, PID output values, and limiting status, are recorded simultaneously for control effect evaluation and online algorithm optimization.

[0044] S3: Based on the difference between the actual moisture content after pelleting obtained by real-time measurement and the preset target moisture content, a feedback control command is generated by calculating the difference through a predetermined second control algorithm.

[0045] Specifically, step S3 includes the following steps: S31: Calculate the difference between the actual moisture content after pelleting and the target moisture content, and use it as the actual control error.

[0046] In this embodiment, the actual moisture content of the pellets after pelleting is acquired in real time using an online near-infrared moisture meter installed at the pelletizer outlet, with a sampling frequency set to 5 times per second. The control unit performs moving average filtering on 10 consecutive sample values ​​and compares them with a preset target moisture content (e.g., 0.15%). The formula for calculating the actual control error is as follows: ,in This is the actual moisture content measurement value after filtering. This is the target value for the process. The system is equipped with an error validity verification mechanism. When an abnormality is detected in the moisture analyzer signal (such as three consecutive sampling values ​​exceeding the reasonable range of 0.05%-0.4%), it automatically switches to the backup calibration mode. Under normal operating conditions, the fluctuation range of this error value is usually controlled within ±0.03%, and all error data are stored in the real-time database with quality labels.

[0047] S32: Input the actual control error into the predetermined second control algorithm.

[0048] In this embodiment, the verified actual control error is transmitted to the second PID control algorithm module via a real-time data interface. This feedback control algorithm adopts the same PID control formula architecture as the first control algorithm.

[0049] However, different parameter tuning was used to address the characteristics of feedback control: proportional coefficient The integral time constant is 1.8 (more moderate than the 2.8 of feedforward control, to avoid overreacting to measurement noise). The derivative time constant is 300s (longer than the 200s of feedforward control, providing a slower integral action to prevent saturation). =15s (shorter than the 25s of feedforward control, enhancing the ability to suppress measurement noise). This algorithm also uses an incremental digital implementation, with the control cycle set to 200ms in sync with the moisture meter sampling cycle. The algorithm input has an adaptive dead-zone adjustment function, when... When the error is less than 0.02%, it automatically filters out minor fluctuations. It is also equipped with a rate of change constraint module to enable second-order filtering when the error changes by more than 0.03% in a single period.

[0050] S33: The second control algorithm performs calculations on the actual control error and outputs the feedback control command.

[0051] In this embodiment, the feedback PID algorithm adopts a piecewise strategy based on the dynamic characteristics of the error: when When using PI control mode; PD control mode is enabled when; Full PID control is executed. The algorithm output is processed by a rate limiting module to limit the command variation within a single cycle to no more than ±1.5 units, with the final output value constrained within the range of -3.0 to +3.0 standard units (corresponding to a drying temperature adjustment range of -9℃ to +9℃). The generated feedback control command includes three elements: reference correction amount (unit: standard units), direction of action (increase / decrease), and duration (default 3 cycles, 600ms). The command data is sent to the command fusion queue via TCP / IP protocol packets with sequence numbers and synchronized with the feedforward command. The system simultaneously records the entire process information, including the original moisture meter reading, filtered data, and intermediate PID calculation values, for control effect evaluation and fault diagnosis analysis.

[0052] S4: Dynamically calculate and generate weight coefficients based on the predicted confidence level, wherein the weight coefficients include feedforward weights and feedback weights.

[0053] Specifically, step S4 includes the following steps: S41: Based on a preset nonlinear mapping relationship, feedforward weights are calculated from the predicted confidence level, wherein the nonlinear mapping relationship is configured such that when the predicted confidence level is lower than a predefined first threshold, feedforward weights are calculated according to a first preset ratio. Reduce the feedforward weights; when the prediction confidence is higher than a predefined second threshold, adjust according to a second preset ratio. Increase the feedforward weight, wherein the second threshold is greater than the first threshold. , .

[0054] In this embodiment, the prediction confidence level C (range 0-1) from the water value chain prediction model is received. First, the confidence level validity is verified. When the confidence level fluctuation exceeds 0.2 for three consecutive sampling periods, data smoothing processing is initiated.

[0055] The implementation of nonlinear mapping relationships is based on the principle of piecewise linear interpolation: for example, setting the first threshold to 0.7 and the second threshold to 0.9, when... At that time, feedforward weights ;when hour, ;when hour, The weight calculation module updates every 200ms and performs amplitude limiting on the output value to ensure... It always stays within the range of 0-1.

[0056] S42: Generate the feedback weights based on the feedforward weights.

[0057] In this embodiment, feedback weight Complementary calculation relationship The results are obtained directly. The calculation process uses double-precision floating-point arithmetic to ensure that the absolute error of the weight sum does not exceed 0.001. The final generated weight coefficient set includes a timestamp and quality identifier.

[0058] S5: Based on the feedforward weight and feedback weight, the feedforward control command and the feedback control command are weighted and fused to obtain the final comprehensive control command.

[0059] In this embodiment, feedforward control instructions and feedback control instructions are retrieved from the instruction buffer queue, respectively. Each instruction data packet contains a timestamp, instruction value, and quality identifier. First, a time synchronization check is performed to ensure that the timestamp deviation between the two instructions does not exceed 50ms. If this deviation exceeds 50ms, the nearest neighbor interpolation algorithm is used for time alignment. Then, the instruction values ​​are checked for range. Feedforward instruction values ​​must be within the range of -4.0 to +4.0 standard units, and feedback instruction values ​​must be within the range of -3.0 to +3.0 standard units. When an outlier is detected, valid data from the previous cycle is automatically used for replacement. Simultaneously, the validity of the weighting coefficients is checked, confirming that both the feedforward and feedback weights are between 0 and 1, and that the deviation of their sum does not exceed ±0.005.

[0060] A weighted average algorithm is used for instruction fusion. Before fusion, the temperature adjustment amount of the feedforward instruction and the standard unit of the feedback instruction are converted into the temperature adjustment amount that the drying system can directly execute. The conversion coefficient is uniformly set at 3℃ / standard unit.

[0061] The integrated control command after fusion is output-limited to restrict the final command value to the actual adjustment range of -10℃ to +10℃. The limiting method adopts a combination of soft and hard limiting strategies: when the command value exceeds ±8℃, soft limiting is activated, and an S-curve is used for smooth transition; when it exceeds ±10℃, hard limiting is activated, and the value is directly truncated to the boundary value.

[0062] S6: Send the integrated control command to the drying system, and adaptively adjust the response mode of the drying system according to the comparison result of the predicted confidence level and the dynamic switching threshold, wherein the dynamic switching threshold is generated according to the distribution characteristics of the historical predicted confidence level.

[0063] Specifically, step S6 includes the following steps: S61: Compare the predicted confidence level with the dynamic switching threshold to generate a comparison result.

[0064] The generation of the dynamic switching threshold includes: First, collect multiple historical prediction confidence data within a preset historical time period to form a historical prediction confidence dataset.

[0065] In this embodiment, historical prediction confidence data for the most recent seven consecutive production days were collected, with a sampling frequency of one minute, resulting in approximately 10,080 valid data points. The data collection scope covered different production batches, raw material specifications, and operating conditions to ensure the representativeness of the dataset. During the collection process, obvious outliers (such as data points with more than 10 consecutive identical confidence values ​​or instantaneous jumps exceeding 0.5) were automatically removed, and missing data was imputed using the moving average of the previous and next five sampling points. The preprocessed data was stored in a dedicated database in time series format, with each data point accompanied by a timestamp, production batch number, and quality identifier, forming a standardized historical prediction confidence dataset.

[0066] Secondly, statistical analysis is performed on the historical prediction confidence dataset to determine the distribution characteristic parameters.

[0067] In this embodiment, the kernel density estimation method is used to fit the probability distribution of the historical dataset and calculate the skewness, kurtosis, and confidence interval of the distribution curve. First, the data is divided into equal-width bins with an interval of 0.05, and the frequency distribution of each interval is statistically analyzed. Then, the central tendency index (mean maintained between 0.75 and 0.85) and the dispersion index (standard deviation typically between 0.12 and 0.18) of the distribution are calculated.

[0068] Specifically, the system identifies high-density distribution areas (regions with a probability density exceeding 0.6), and the boundaries of these areas serve as an important basis for threshold setting. The analysis process employs a sliding window mechanism, with each analysis including the most recent 2000 data points to ensure the timeliness of the statistical results.

[0069] Finally, the dynamic switching threshold is calculated and generated based on the distribution characteristic parameters.

[0070] In this embodiment, the dynamic switching threshold is calculated using the percentile method based on the distribution characteristic analysis results. Specifically, the 85th percentile of the historical data distribution is selected as the threshold benchmark value to ensure that a conservative response mode is preferred in most cases (85% of operating conditions).

[0071] Weighting factors are introduced during the calculation process: recent data (last 24 hours) has a weight of 0.6, intermediate data (24-72 hours) has a weight of 0.3, and early data (72 hours ago) has a weight of 0.1, allowing the threshold to dynamically adapt to changes in production conditions. The final dynamically generated switching threshold is limited to the range of 0.65-0.95 and is automatically updated every 24 hours. A smooth transition strategy is used during the update process to avoid threshold jumps exceeding 0.1.

[0072] After generating the dynamic switching threshold, it needs to be compared with the predicted confidence level. The process is as follows: The real-time prediction confidence level C is compared with the dynamic switching threshold T. Output a high confidence flag when the confidence level is high, and output a low confidence flag when the confidence level is low.

[0073] S62: Determine the response mode type based on the comparison result, wherein the response mode type includes a conservative response mode and an aggressive response mode. When the prediction confidence is greater than or equal to the dynamic switching threshold, the conservative response mode is selected; otherwise, the aggressive response mode is selected.

[0074] In this embodiment, based on the comparison results, one of two predefined response modes is selected. When (At high confidence) Select conservative response mode. In this mode, the control parameters of the drying system are set as follows: proportional coefficient. The integral time is 1.2. The derivative time is 400s. The control cycle was extended from 20 seconds to 500 milliseconds. When the confidence level is low, the aggressive response mode is enabled, and the corresponding parameters are adjusted as follows: It is 2.5. It is 150 seconds. The control cycle has been shortened to 100ms, instead of 10 seconds. Mode switching adopts a gradual transition method, gradually adjusting to the target parameters within 30 seconds to avoid abrupt changes in control commands.

[0075] S63: Configure the control parameters of the drying system according to the response mode type, and apply the configured control parameters to the drying system.

[0076] In this embodiment, the selected response mode parameters are sent to the drying system controller in real time via a configuration file. The controller employs a dual-buffering mechanism to ensure the stability of parameter switching: current operating parameters are stored in the active memory area, and new parameters are pre-loaded into the backup memory area. Upon receiving a switching command, a seamless switch is completed via a hardware interrupt. During execution, key indicators of the drying system are continuously monitored, including heater power, wind speed, and material temperature change rate. In conservative response mode, the temperature change rate is limited to no more than 2°C / min; in aggressive response mode, the temperature change rate is allowed to reach 5°C / min, but an overshoot protection mechanism is set up, automatically activating damping adjustment when the actual temperature deviates from the target temperature by more than 3°C.

[0077] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0078] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0079] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

Claims

1. A method for intelligently controlling the moisture content during the production of nylon reinforced materials, characterized in that, The control method includes the following steps: Based on real-time collection of multiple parameters from the production line that are related to moisture content, the predicted moisture content after pelleting and the corresponding prediction confidence level are obtained. Based on the difference between the predicted moisture content after pelleting and the preset target moisture content, a feedforward control command is generated by calculating using a predetermined first control algorithm. Based on the difference between the actual moisture content after pelleting obtained by real-time measurement and the preset target moisture content, a feedback control command is generated by a predetermined second control algorithm. Weighting coefficients are dynamically calculated based on the predicted confidence level, wherein the weighting coefficients include feedforward weights and feedback weights. Based on the feedforward weight and feedback weight, the feedforward control command and the feedback control command are weighted and fused to obtain the final comprehensive control command; The integrated control command is sent to the drying system, and the response mode of the drying system is adaptively adjusted according to the comparison result of the predicted confidence level and the dynamic switching threshold, wherein the dynamic switching threshold is generated based on the distribution characteristics of the historical predicted confidence level.

2. The intelligent moisture content control method for the production of nylon reinforced materials according to claim 1, characterized in that, The method for obtaining the predicted moisture content and corresponding prediction confidence level after pelletizing based on multiple parameters collected in real time from the production line and associated with moisture content includes: Real-time data collection of raw material initial moisture content, ambient humidity, extruder melt pressure, melt temperature, and cooling water temperature in the production line; The initial moisture content of the raw materials, ambient humidity, extruder melt pressure, melt temperature and cooling water temperature are collected as input parameters and input into the pre-trained moisture value chain prediction model. The input parameters are mapped and processed by the moisture value chain prediction model to output the predicted moisture content after pelleting and the corresponding prediction confidence level.

3. The intelligent moisture content control method for the production of nylon reinforced materials according to claim 2, characterized in that, The difference between the predicted moisture content after pelleting and the preset target moisture content is calculated using a predetermined first control algorithm to generate a feedforward control command, including: The difference between the predicted moisture content after pelleting and the target moisture content is calculated as the prediction control error; The predictive control error is input into the predetermined first control algorithm; The first control algorithm processes the predictive control error and outputs the feedforward control command.

4. The intelligent moisture content control method for the production of nylon reinforced materials according to claim 3, characterized in that, The difference between the actual moisture content after pelleting, obtained from real-time measurement, and the preset target moisture content is calculated using a predetermined second control algorithm to generate feedback control commands, including: The difference between the actual moisture content after pelleting and the target moisture content is calculated as the actual control error; The actual control error is input into the predetermined second control algorithm; The second control algorithm processes the actual control error and outputs the feedback control command.

5. The intelligent moisture content control method for the production of nylon reinforced materials according to claim 4, characterized in that, Both the first control algorithm and the second control algorithm are PID control algorithms.

6. The intelligent moisture content control method for the production of nylon reinforced materials according to claim 5, characterized in that, The weight coefficients are dynamically calculated based on the predicted confidence level, wherein the weight coefficients include feedforward weights and feedback weights, including: Based on a preset nonlinear mapping relationship, feedforward weights are calculated from the predicted confidence level, wherein the nonlinear mapping relationship is configured such that when the predicted confidence level is lower than a predefined first threshold, a first preset ratio is applied. Reduce the feedforward weights; when the prediction confidence is higher than a predefined second threshold, adjust according to a second preset ratio. Increase the feedforward weight, wherein the second threshold is greater than the first threshold. , ; The feedback weights are generated based on the feedforward weights.

7. The intelligent moisture content control method for the production of nylon reinforced materials according to claim 6, characterized in that, The generation of the dynamic switching threshold includes: Collect multiple historical prediction confidence data within a preset historical time period to form a historical prediction confidence dataset; Statistical analysis was performed on the historical prediction confidence dataset to determine the distribution characteristic parameters; The dynamic switching threshold is calculated and generated based on the distribution characteristic parameters.

8. The intelligent moisture content control method for the production of nylon reinforced materials according to claim 7, characterized in that, The adaptive adjustment of the drying system's response mode based on the comparison result between the predicted confidence level and the dynamic switching threshold includes: The predicted confidence level is numerically compared with the dynamic switching threshold to generate a comparison result; The response mode type is determined based on the comparison results, wherein the response mode type includes a conservative response mode and an aggressive response mode. When the prediction confidence is greater than or equal to the dynamic switching threshold, the conservative response mode is selected, otherwise the aggressive response mode is selected. Configure the control parameters of the drying system according to the response mode type, and apply the configured control parameters to the drying system.

Images