Self-adaptive adjusting system for quality stability of reprocessed plastic blow molding product

Abstract

The invention belongs to the technical field of solid waste treatment, and particularly relates to a reprocessed plastic blow molding product quality stability self-adaptive adjusting system. And the blow molding process parameters are optimized by using neural network integration, firefly algorithm optimized fuzzy-PID control, deep reinforcement learning, Bayesian optimization and a chaos optimization algorithm. According to the algorithms, for enterprises with different scales and requirements, the product quality is effectively improved, the production efficiency is improved, the defective rate and the energy consumption are reduced, and diversified quality optimization strategies are provided for the regenerated plastic blow molding industry.

Description

Technical Field

[0001] The present invention belongs to the field of manufacturing of recycled plastic blow molding, and specifically relates to an adaptive adjustment system for the quality stability of recycled plastic blow molding products. Background Art

[0002] Recycled plastic blow molding, as an important plastic product processing method, is widely used in many industries such as packaging, automotive, and medical. During the blow molding process, the product quality is affected by various factors, among which the changes in environmental temperature and humidity have a significant impact on the quality stability of the product. For example, an increase in environmental temperature may cause a decrease in the viscosity of the plastic melt, resulting in a thinner wall thickness of the product; an increase in humidity may affect the drying degree of the plastic raw material, thereby affecting the strength of the product.

[0003] In the traditional recycled plastic blow molding process, operators mainly manually adjust the process parameters of the blow molding machine based on experience and regular product quality inspection results. This method has a certain lag, cannot respond in a timely manner to the rapid changes in environmental factors, easily leads to product quality fluctuations, generates a large number of defective products, and increases production costs. Therefore, it is of great practical significance to develop an adaptive adjustment mechanism that can real-time monitor environmental parameters and product quality and automatically adjust process parameters.

[0004] Fuzzy logic control, as an intelligent control method, can handle complex non-linear relationships and uncertain information, and is particularly suitable for systems where it is difficult to establish an accurate mathematical model. In the field of recycled plastic blow molding, using the fuzzy logic control algorithm to adjust product quality can effectively cope with the challenges brought by environmental factor changes and improve the quality stability of products. Summary of the Invention

[0005] The present invention provides an adaptive adjustment system for the quality stability of recycled plastic blow molding products, including:

[0006] Product quality data acquisition module: In the product quality inspection link of the blow molding production line, there is high-precision inspection equipment for real-time collecting product quality data. The product quality data includes wall thickness distribution data obtained by multi-point measurement at different positions of the product through a wall thickness measuring instrument to evaluate wall thickness uniformity, the fracture strength recorded by tensile testing the product using a tensile testing machine, and product surface defect data identified by appearance inspection equipment. And the inspection equipment is connected to the production line control system and can transmit quality data to the subsequent fuzzy logic control system in real time;

[0007] Environmental parameter acquisition module: Multiple environmental parameter sensors are arranged around the blow molding machine and in the production workshop to monitor the environmental temperature and humidity in real time. The temperature sensor uses a thermocouple or a thermistor sensor, and the humidity sensor selects a capacitive or resistive humidity sensor. The sensors are distributed at different positions to obtain the overall environmental parameters. The collected environmental temperature and humidity data are transmitted to the fuzzy logic control system in real time;

[0008] Input variable fuzzification unit: The product quality detection data and environmental parameters are used as the input variables of the fuzzy logic controller. Each input variable is divided into multiple fuzzy subsets according to the actual value range, and the corresponding membership function is defined for each subset;

[0009] Output variable fuzzification unit: The process parameters of the blow molding machine are used as the output variables of the fuzzy logic controller. Each output variable is divided into multiple fuzzy subsets according to its adjustment range, and the membership function is defined;

[0010] Fuzzy rule formulation module: Based on the knowledge of the recycled plastic blow molding process and expert experience, fuzzy rules describing the fuzzy relationship between input variables and output variables are formulated, and the fuzzy rule base is continuously improved through experiments and production experience accumulation;

[0011] Fuzzy inference unit: Appropriate fuzzy inference methods such as Mamdani inference method or Sugeno inference method are used to derive the fuzzy value of the output variable according to the fuzzy rules and the fuzzification results of the input variables;

[0012] Defuzzification unit: Defuzzification methods such as the centroid method and the maximum membership degree method are used to convert the output fuzzy value obtained by fuzzy inference into the specific adjustment value of the blow molding machine process parameters;

[0013] Real-time monitoring and feedback module: The fuzzy logic control system receives the product quality detection data and environmental parameter data in real time, compares them with the preset product quality standards and the ideal range of environmental parameters, judges whether the product quality fluctuates and whether the environmental parameters deviate from the ideal state. If product quality fluctuations or environmental parameter changes are detected, the adaptive adjustment mechanism is triggered, and the deviation information is fed back to the fuzzy logic control algorithm;

[0014] Process parameter automatic adjustment module: According to the calculation results of the fuzzy logic control algorithm, the process parameters of the blow molding machine are automatically adjusted, including blowing pressure, temperature, blowing time, etc., to ensure that the product quality is always maintained within a stable range;

[0015] Dynamic optimization and learning module: By analyzing the historical production data and product quality data, the fuzzy rule base and membership function are optimized to adapt to the changes in the production process and improve the accuracy and adaptability of the fuzzy logic control algorithm.

[0016] Further, in the product quality data acquisition module, after the product is formed, the wall thickness measuring instrument measures the wall thickness at different positions of the product near the discharge port at regular intervals. Each measurement takes multiple points. The regular interval is 10 seconds, and the measurement positions include the top, middle, and bottom.

[0017] Further, in the product quality data acquisition module, in the product quality inspection area, a certain number of products are randomly selected from every certain production batch and used in a tensile testing machine for tensile strength testing. The certain production batch is one batch for every 100 products, and the certain number is 5.

[0018] Further, in the environmental parameter acquisition module, the temperature sensor collects temperature data every 1 minute, and the humidity sensor collects humidity data every 2 minutes, and the temperature sensor and the humidity sensor are calibrated once a month.

[0019] Further, in the fuzzification processing module, for the wall thickness uniformity, its value range is divided into [0, 100], and divided into 5 fuzzy subsets of "very uneven", "uneven", "average", "even", and "very even", and a triangular membership function is adopted.

[0020] Further, in the fuzzification processing module, taking the blowing pressure as an example, its adjustment range is [0.5, 1.5] MPa, and it is divided into 5 fuzzy subsets of "very low", "low", "moderate", "high", and "very high", and a triangular membership function is adopted.

[0021] Further, in the fuzzy inference and defuzzification module, the Mamdani inference method is used for fuzzy inference. According to the fuzzification results of the input variables, the membership degrees of the wall thickness uniformity and the temperature in their respective fuzzy subsets are determined, and the minimum value of the two is taken as the activation strength of this rule. Then, the fuzzy output set of the blowing pressure is obtained through fuzzy composition operation.

[0022] Further, in the fuzzy inference and defuzzification module, the centroid method is used for defuzzification. For the fuzzy output set of the blowing pressure, the defuzzified blowing pressure adjustment value is determined by calculating the centroid where y i is an element in the fuzzy output set of the blowing pressure, μ(y i ) is its membership degree, and n is the number of set elements.

[0023] Further, the dynamic optimization and learning module of the adaptive adjustment mechanism optimizes and adjusts the fuzzy rule base and the membership function by establishing a historical production data and product quality database and using data mining techniques such as association rule mining and clustering analysis to analyze the data in the database.

[0024] Beneficial effects

[0025] Construct a neural network ensemble model that includes MLP, CNN, and RNN. After training and integration, the model can accurately predict product quality, design a fuzzy-PID controller, and optimize its parameters using the firefly algorithm. In practical applications, data is collected in real time to adjust process parameters, and the stability of product quality is significantly improved. The PPO algorithm is used to learn the optimal strategy. The trained agent is integrated into the production system, which can automatically adjust process parameters according to the real-time state. A chaotic sequence is generated using the Logistic map and mapped to process parameters for chaotic optimization. Searching starts from multiple initial values and combines production feedback to adjust algorithm parameters to solve the problem that traditional methods are prone to falling into local optima. Description of the Drawings

[0026] Figure (1) Schematic diagram of the data acquisition system;

[0027] Figure (2) Flowchart of the fuzzy logic control system;

[0028] Figure (3) Flowchart of the operation of the adaptive adjustment mechanism. Detailed Implementation Manner

[0029] Example 1

[0030] (I) Construction of the data acquisition system

[0031] Collection of product quality data

[0032] In the product quality inspection link of the blow molding production line, high-precision detection equipment is installed to collect product quality data in real time. For the wall thickness uniformity of the product, a wall thickness measuring instrument is used to measure the wall thickness at multiple points in different positions of the product to obtain the wall thickness distribution data for evaluating the wall thickness uniformity. For the strength of the product, a tensile testing machine is used to conduct a tensile test on the product and record the breaking strength of the product. At the same time, an appearance detection device, such as an industrial camera, is used to identify defects on the product surface, such as bubbles and scratches. These detection devices are connected to the production line control system to ensure that the quality data can be transmitted to the subsequent fuzzy logic control system in real time.

[0033] Collection of environmental parameters Around the blow molding machine and in the production workshop, multiple environmental parameter sensors are arranged to monitor the environmental temperature and humidity in real time. The temperature sensor can use a thermocouple or a thermistor sensor, and the humidity sensor can select a capacitive or resistive humidity sensor. The sensors are distributed at different positions to obtain the overall distribution of temperature and humidity in the production environment and avoid the influence of local environmental differences on data accuracy. The collected environmental temperature and humidity data are also transmitted to the fuzzy logic control system in real time.

[0034] (II) Design of the fuzzy logic control system

[0035] Fuzzification processing

[0036] Fuzzification of input variables: The product quality inspection data (wall thickness uniformity, strength, appearance defect degree) and environmental parameters (temperature, humidity) are used as the input variables of the fuzzy logic controller. For each input variable, it is divided into multiple fuzzy subsets according to its actual value range, and the corresponding membership functions are defined for each subset. For example, for wall thickness uniformity, it can be divided into fuzzy subsets such as "very uneven", "uneven", "average", "even", "very even", etc., and triangular or trapezoidal membership functions are used to describe the degree to which the input value belongs to each fuzzy subset. For temperature, it can be divided into fuzzy subsets such as "very low", "low", "moderate", "high", "very high", etc., and the corresponding membership functions are also defined.

[0037] Fuzzification of output variables: The process parameters of the blow molding machine (blowing pressure, temperature, blowing time) are used as the output variables of the fuzzy logic controller. For each output variable, it is also divided into multiple fuzzy subsets according to its adjustment range, and the membership functions are defined. For example, the blowing pressure can be divided into fuzzy subsets such as "very low", "low", "moderate", "high", "very high", etc., and the membership degree of different blowing pressure values in each fuzzy subset is represented by the membership function.

[0038] Fuzzy rule formulation: Based on the process knowledge and expert experience of recycled plastic blow molding, fuzzy rules are formulated. These rules describe the fuzzy relationship between the input variables (product quality data and environmental parameters) and the output variables (process parameters of the blow molding machine). For example, when the product wall thickness is uneven and the environmental temperature is high, the fuzzy rule may stipulate that the blowing pressure is appropriately increased to improve the wall thickness uniformity. The form of the rule is usually "If (condition 1 and condition 2 and...), then (conclusion)". Through a large number of experiments and the accumulation of actual production experience, the fuzzy rule base is continuously improved to ensure the accuracy and effectiveness of the rules.

[0039] Fuzzy inference and defuzzification

[0040] Fuzzy inference: Appropriate fuzzy inference methods, such as Mamdani inference method or Sugeno inference method, are used to derive the fuzzy value of the output variable according to the fuzzy rules and the fuzzification results of the input variables. Taking the Mamdani inference method as an example, it is based on the "and", "or", "not" operations of fuzzy sets, and through fuzzy implication relations and composition operations, the output fuzzy set is obtained from the input fuzzy sets.

[0041] Defuzzification: The output fuzzy value obtained by fuzzy inference is converted into the actual control quantity, that is, the specific adjustment value of the process parameters of the blow molding machine. Common defuzzification methods include the centroid method, the maximum membership degree method, etc. The centroid method determines the defuzzified output value by calculating the centroid of the fuzzy set, and this method can comprehensively consider the influence of each element in the fuzzy set to obtain a relatively smooth control output.

[0042] (3) Implementation of Adaptive Adjustment Mechanism

[0043] Real-time Monitoring and Feedback

[0044] The fuzzy logic control system receives the product quality inspection data and environmental parameter data in real time. By comparing with the preset product quality standards and the ideal range of environmental parameters, it judges whether there are fluctuations in product quality and whether the environmental parameters deviate from the ideal state. If fluctuations in product quality or changes in environmental parameters are detected, the system will trigger the adaptive adjustment mechanism.

[0045] Automatic Adjustment of Process Parameters

[0046] According to the calculation results of the fuzzy logic control algorithm, the process parameters of the blow molding machine are automatically adjusted. For example, when the fuzzy logic control system determines that the blowing pressure needs to be increased to improve the wall thickness uniformity of the product, the system sends an instruction to the control system of the blow molding machine to accurately adjust the blowing pressure to an appropriate value. At the same time, other process parameters such as temperature and blowing time are also automatically adjusted in a similar manner to ensure that the product quality always remains within a stable range.

[0047] Dynamic Optimization and Learning

[0048] To adapt to various changes in the production process, the adaptive adjustment mechanism has the ability of dynamic optimization and learning. By analyzing historical production data and product quality data, the system can continuously optimize the fuzzy rule base and membership functions, improving the accuracy and adaptability of the fuzzy logic control algorithm. For example, if under a certain specific environmental condition, it is found that the current fuzzy rules lead to unstable product quality, the system can adjust the corresponding fuzzy rules according to the actual situation to better cope with the complex and changeable production environment.

[0049] Near the product discharge port of the blow molding production line, a high-precision wall thickness measuring instrument is installed to ensure that the wall thickness can be measured immediately after the product is formed. The wall thickness of the product is measured at different positions (such as the top, middle, and bottom) every certain period of time (such as 10 seconds), and 3 - 5 points are measured each time. The measurement data is recorded and transmitted to the production line control system. At the same time, in the product quality inspection area, every certain production batch (such as every 100 products as a batch), a certain number (such as 5) of products are randomly selected and the tensile strength test is carried out using a tensile testing machine, and the fracture strength data of the products is recorded. In addition, an industrial camera is installed on the product conveyor belt, and image recognition technology is used to detect surface defects such as bubbles and scratches on the product in real time, and data such as the type and severity of the defects are transmitted to the production line control system.

[0050] Implementation of Environmental Parameter Acquisition

[0051] Temperature sensors and humidity sensors are evenly distributed in the heating area of the blow molding machine, around the mold, and in different corners of the production workshop. The temperature sensors collect temperature data every 1 minute, and the humidity sensors collect humidity data every 2 minutes. The collected temperature and humidity data are transmitted to the fuzzy logic control system in real time through wireless or wired communication methods. To ensure the accuracy of the data, the sensors are calibrated regularly. For example, the temperature sensors and humidity sensors are calibrated once a month to ensure that the measurement error is within the allowable range.

[0052] Implementation of Fuzzification

[0053] Implementation of input variable fuzzification: For wall thickness uniformity, according to actual production experience and product quality standards, its value range is divided into [0, 100], where 0 represents very uneven and 100 represents very uniform. It is divided into 5 fuzzy subsets: "very uneven" (0 - 20), "uneven" (20 - 40), "average" (40 - 60), "uniform" (60 - 80), "very uniform" (80 - 100). A triangular membership function is used. For example, for the "uneven" subset, its membership function is:

[0054]

[0055] For other input variables such as strength, degree of appearance defects, temperature, and humidity, fuzzy subset division and membership function definition are carried out in a similar manner.

[0056] Implementation of output variable fuzzification: Taking the blowing pressure as an example, assuming its adjustment range is [0.5, 1.5] MPa, it is divided into 5 fuzzy subsets: "very low" (0.5 - 0.7), "low" (0.7 - 0.9), "moderate" (0.9 - 1.1), "high" (1.1 - 1.3), "very high" (1.3 - 1.5). A triangular membership function is also used. For example, the membership function of the "high" subset is:

[0057]

[0058] Similar fuzzification processing is carried out for output variables such as temperature and blow molding time.

[0059] Implementation of Fuzzy Rule Formulation

[0060] Experts in the field of recycled plastic blow molding and experienced operators are organized to formulate a fuzzy rule base in combination with a large amount of experimental data and actual production cases. For example, the following fuzzy rules are formulated:

[0061] If (wall thickness uniformity is "uneven" and temperature is "high"), then (blowing pressure is "high");

[0062] If (strength is "low" and humidity is "high"), then (blow molding time is "long");

[0063] If (appearance defect degree is "severe" and temperature is "low"), then (blow molding machine temperature is "high").

[0064] By continuously collecting and analyzing production data, optimize and improve the fuzzy rules to ensure that they can accurately reflect the relationship between product quality, environmental parameters and process parameters.

[0065] Implementation of Fuzzy Inference and Defuzzification

[0066] Use Mamdani inference method for fuzzy inference. Take the rule "If (wall thickness uniformity is 'non-uniform' and temperature is 'high'), then (blowing pressure is 'high')" as an example. First, according to the fuzzy results of the input variables, determine the membership degrees of wall thickness uniformity and temperature in their respective fuzzy subsets. Suppose the membership degree of wall thickness uniformity in the 'non-uniform' subset is μ 不均匀 (x 1 ), and the membership degree of temperature in the 'high' subset is μ 高 (x 2 ). Take the minimum of the two as the activation strength of this rule. Then, according to the results of all activated rules, obtain the fuzzy output set of blowing pressure through fuzzy composition operation. min(μ 不均匀 (x 1 ), μ 高 (x 2 ))

[0067] Use the centroid method for defuzzification. For the fuzzy output set of blowing pressure, calculate its centroid: where y i is an element in the fuzzy output set of blowing pressure, μ(y i ) is its membership degree, and n is the number of set elements. The calculated y is the adjusted value of blowing pressure after defuzzification.

[0068] Implementation of Real-time Monitoring and Feedback

[0069] In the fuzzy logic control system, set the ideal ranges of product quality and environmental parameters. For example, the wall thickness uniformity of the product should be maintained above 80, the strength should not be lower than a certain set value, the environmental temperature should be between 20 - 25 °C, and the humidity should be between 40% - 60%. The system will compare the product quality data and environmental parameters collected in real time with the ideal ranges. If the wall thickness uniformity of the product is lower than 80, or the strength is lower than the set value, or the environmental temperature and humidity exceed the ideal ranges, the system will immediately trigger the adaptive adjustment mechanism and feedback the deviation information to the fuzzy logic control algorithm.

[0070] Implementation of Automatic Adjustment of Process Parameters

[0071] After the adaptive adjustment mechanism is triggered, the fuzzy logic control system sends instructions to the control system of the blow molding machine according to the adjusted values of the process parameters obtained from fuzzy reasoning and defuzzification. The control system of the blow molding machine automatically adjusts process parameters such as blowing pressure, temperature, and blowing time according to the instructions. For example, if the fuzzy logic control system calculates that the blowing pressure needs to be increased by 0.1 MPa, the control system will precisely control the blowing pressure adjustment device to increase the blowing pressure to the corresponding value. During the adjustment process, the system monitors the changes in product quality and environmental parameters in real time to ensure that the adjustment effect meets the expectations.

[0072] Implementation of Dynamic Optimization and Learning

[0073] A historical production data and product quality database is established to record product quality data, environmental parameters, and process parameter adjustment situations during each production process. The data in the database is analyzed regularly, and data mining techniques such as association rule mining and clustering analysis are used to find potential relationships and rules among product quality, environmental parameters, and process parameters. For example, through association rule mining, it is found that within a certain specific range of environmental temperature and humidity, there is a certain specific association between the wall thickness uniformity of the product and the blowing pressure and blowing time. Based on these findings, the fuzzy rule base and membership function are optimized and adjusted so that the fuzzy logic control system can better adapt to the changes in the production process and continuously improve the stability of product quality. Through the above specific implementation methods, the adaptive adjustment mechanism for the quality stability of recycled plastic blow molding products based on fuzzy logic control can effectively cope with the impact of environmental factor changes on product quality during the production process, achieve automatic and stable control of product quality, improve production efficiency and product quality, and reduce production costs.

[0074] Example 2

[0075] Data Collection and Problem Modeling: Collect information such as the location information of each production workshop within the enterprise, material requirements and supply situations, energy consumption data of transportation equipment, and distances and transportation times of different routes. Model the production route optimization problem as a variant of the Traveling Salesman Problem (TSP), where each production workshop is a node and the transportation route is an edge, and the goal is to find a transportation route with the lowest total energy consumption that meets the material transportation requirements of all workshops.

[0076] Application of Ant Colony Algorithm: In the ant colony algorithm, ants move between various nodes and select the next node based on the pheromone concentration and heuristic information on the path. The higher the pheromone concentration on a path, the greater the probability of being selected by ants. The heuristic information can be determined based on factors such as the distance and transportation time of the path. For example, the shorter the distance and the less the transportation time, the higher the heuristic information value. After each ant completes a path search, the pheromone concentration on the path is updated according to the total energy consumption of the path it has traveled. The lower the energy consumption of the path, the more the pheromone concentration increases. Through multiple iterations of search by multiple ants, an approximately optimal production and transportation path is gradually found.

[0077] Formulation and Implementation of Optimization Plan: After a certain number of iterations (such as 500 times), the optimal production and transportation path is determined. For example, adjust the transportation sequence and path of materials from the sorting workshop to the cleaning workshop and then to the melting workshop, etc. During the implementation process, utilize the enterprise's logistics management system to arrange the operation of transportation equipment according to the optimized path, and monitor the transportation energy consumption and time in real time.

[0078] Example 3

[0079] Screening of Key Parameters and Energy Saving Optimization Based on Grey Relational Analysis: Data Collection and Grey Relational Analysis: Collect production energy consumption data and data of various production parameters (such as temperature, pressure, rotational speed, raw material ratio, etc.) over a period of time (such as half a year). Use the grey relational analysis method to calculate the grey relational degree between each production parameter and energy consumption. The grey relational degree reflects the degree of association between the parameter and energy consumption. The greater the relational degree, the more significant the impact of the parameter on energy consumption.

[0080] Screening and Optimization of Key Parameters: According to the magnitude of the grey relational degree, screen out the key parameters with a greater impact on energy consumption (such as the top 5 parameters in terms of relational degree ranking). For these key parameters, through experimental design (such as orthogonal experiments) or other optimization algorithms, find the parameter combination that minimizes energy consumption. For example, for the key parameters such as temperature, pressure, and raw material ratio screened out, design orthogonal experiments at different levels and test the energy consumption under different parameter combinations.

[0081] Implementation of Optimization Plan and Effect Evaluation: Implement the optimized parameter combination and continuously monitor the change of energy consumption. Evaluate the energy saving effect by comparing the energy consumption data before and after optimization. At the same time, regularly conduct grey relational analysis again to cope with the change of the impact of possible parameter changes on energy consumption during the production process.

[0082] Example 4

[0083] Production Energy Consumption Prediction and Optimization Based on Neural Network Ensemble: Large-scale waste plastic recycling enterprises have a large production scale and complex processes. A single prediction model is difficult to accurately predict energy consumption, resulting in the inability to formulate precise energy saving strategies in advance.

[0084] Data collection and model construction: Collect a large amount of data such as production energy consumption data, equipment operation parameters, and environmental factors. Construct multiple neural network models with different structures (such as multi-layer perceptrons, recurrent neural networks, convolutional neural networks, etc.), and each model learns and extracts features from the data from different perspectives. For example, multi-layer perceptrons are suitable for processing the non-linear relationship between input and output, recurrent neural networks have good processing capabilities for data with time series characteristics (such as energy consumption data that changes over time), and convolutional neural networks can be used to mine local features in the data. Neural network integration and optimization: Use methods such as weighted average and voting to integrate the prediction results of multiple neural network models to obtain a more accurate energy consumption prediction value. According to the prediction results, combined with the enterprise's production plan and equipment operation conditions, formulate energy-saving optimization strategies. For example, when it is predicted that the energy consumption will increase significantly during a certain period, adjust the production plan in advance and reasonably arrange the start-stop and operation load of the equipment. System implementation and continuous improvement: Apply the neural network integration model to the enterprise's energy consumption management system to predict energy consumption in real time and provide optimization suggestions. Regularly update and optimize the model, retrain the model according to the newly collected data, and adjust the weights and parameters of the model to adapt to the changes in the production process and improve the accuracy of prediction and optimization effect.

[0085] Example 5

[0086] Product quality optimization based on neural network integration, data collection and preprocessing: Collect product quality data from the past year during the blow molding process, including wall thickness uniformity, strength, appearance defects, etc., as well as the corresponding environmental parameters (temperature, humidity) and process parameters (blowing pressure, temperature, blow molding time). Normalize the data and map all data to the [0,1] interval to eliminate the influence of dimensions. Then divide the data into a training set, a validation set, and a test set according to the ratio of 70%, 15%, and 15%.

[0087] Construction of Neural Network Ensemble Model: Construct multiple neural networks with different structures, such as multi-layer perceptron (MLP), convolutional neural network (CNN), and recurrent neural network (RNN). For MLP, set multiple hidden layers, and the number of neurons in each layer is determined based on experience and experiments to capture the non-linear features of the data. CNN is used for feature extraction of product image data (such as appearance defect detection images), and through the combination of convolutional layers, pooling layers, and fully connected layers, it learns the key features in the images. RNN is used to process data with time series characteristics, such as continuous quality monitoring data in the production process, and uses its recurrent structure to remember information from previous time steps. Independently train each neural network using the training set, and adjust the hyperparameters of each network, such as learning rate, number of layers, number of neurons, etc., through the validation set to avoid overfitting. Finally, integrate the trained multiple neural networks, and use the weighted average method to fuse the prediction results of each network, and the weights are determined according to the performance of each network on the validation set.

[0088] Implementation of Quality Optimization: Input the test set data into the neural network ensemble model. The model predicts the product quality, and based on the difference between the prediction result and the target quality, adjusts the process parameters. For example, if the predicted wall thickness uniformity of the product does not meet the standard, the model analyzes the influence weights of each process parameter on the wall thickness uniformity and gives adjustment suggestions for parameters such as blowing pressure and blowing time. In actual production, gradually implement these adjustments and monitor the feedback of product quality in real time to continuously optimize the prediction accuracy of the model and the parameter adjustment strategy.

[0089] Example 6

[0090] Fuzzy-PID Control Optimized Based on Firefly Algorithm: Design a fuzzy-PID controller, combining the flexibility of fuzzy control and the precision of PID control. According to the knowledge of blow molding process, determine the input variables (such as product wall thickness deviation, deviation change rate) and output variables (the three parameters of the PID controller). Fuzzify the input variables, divide them into multiple fuzzy subsets, such as "negative large", "negative medium", "negative small", "zero", "positive small", "positive medium", "positive large", and define the corresponding membership functions. Develop fuzzy rules based on expert experience. For example, when the wall thickness deviation is "positive large" and the deviation change rate is "positive small", appropriately increase, decrease, and keep unchanged to quickly adjust the process parameters to reduce the deviation.

[0091] Firefly Algorithm Optimization: The firefly algorithm is used to optimize the parameters of the fuzzy-PID controller. In the firefly algorithm, each firefly represents a set of parameters of the fuzzy-PID controller. Fireflies update their positions by moving towards brighter (higher fitness) fireflies. The fitness function is designed based on product quality indicators (such as wall thickness uniformity, strength, etc.) and control performance indicators (such as response speed, overshoot, etc.), such that the higher the fitness, the better the controller performance. Through multiple iterations, the firefly algorithm finds the optimal combination of fuzzy-PID controller parameters.

[0092] Control Implementation and Optimization: The optimized fuzzy-PID controller is applied to the process parameter control of the blow molding machine. During the production process, product quality data is collected in real-time, the wall thickness deviation and the rate of change of deviation are calculated, and input into the fuzzy-PID controller to adjust process parameters such as blowing pressure, temperature, and blow molding time in real-time. At the same time, continuously monitor product quality and control effects, and fine-tune the parameters of the firefly algorithm according to the actual situation, such as adjusting the moving step size of fireflies, the attraction coefficient, etc., to further optimize the performance of the fuzzy-PID controller.

[0093] Example 7

[0094] Multi-objective Process Parameter Optimization Based on Deep Reinforcement Learning: Definition of State, Action, and Reward: Define the state space as the set of product quality data (strength, flexibility, appearance score, etc.), environmental parameters (temperature, humidity), and process parameters (current blowing pressure, temperature, blow molding time). The action space is the adjustable range of process parameters. For example, the blowing pressure can be adjusted in the range of [0.5, 1.5] MPa with a step size of 0.05 MPa, and the temperature can be adjusted in the range of [180, 250] °C with a step size of 5 °C, etc. The reward function is designed by combining multiple quality objectives. For example, for products with high strength requirements, a higher reward is given for strength improvement; for products with strict appearance requirements, a higher reward is given for reducing appearance defects. At the same time, penalty terms are set considering factors such as energy consumption to encourage reducing energy consumption while meeting quality requirements.

[0095] Application of Deep Reinforcement Learning Algorithm: Adopt a deep reinforcement learning algorithm based on policy gradients, such as the Proximal Policy Optimization algorithm (PPO). Build a neural network model, whose input layer is the dimension of the state space, the hidden layer uses multiple fully connected layers to extract complex features, the output layer is the dimension of the action space, and outputs the probability of each action. By continuously interacting with the production environment, the agent selects actions according to the current state, and the environment returns rewards and new states. The agent uses the policy gradient algorithm to learn the optimal policy to maximize the long-term cumulative reward. During the training process, use the experience replay mechanism to store and reuse past experiences to improve data utilization and algorithm stability.

[0096] Optimization Strategy Implementation: As the training progresses, the agent gradually learns the process parameter adjustment strategy for optimizing multiple quality objectives simultaneously. In actual production, the trained agent is integrated into the production control system to automatically adjust process parameters according to the real-time production status. For example, when producing toy products, the agent selects the appropriate combination of blowing pressure, temperature, and blow molding time based on the quality indicators such as the flexibility and appearance of the current product and environmental parameters to achieve the best balance between product quality and energy consumption.

[0097] Example 8

[0098] Fine Tuning of Process Parameters Based on Bayesian Optimization: Select Gaussian process regression as the surrogate model to approximate the complex relationship between product quality and process parameters. Collect a certain amount of historical production data, including process parameters (such as blowing pressure, blow molding temperature, raw material formula ratio, etc.) and corresponding product quality indicators (dimensional accuracy, chemical stability test results, etc.). Use this data to train the Gaussian process regression model, which can predict the mean and variance of product quality based on the input process parameters, providing a basis for subsequent optimization.

[0099] Bayesian Optimization Process: The Bayesian optimization algorithm is used to find the optimal process parameters. Bayesian optimization balances between exploring new process parameter combinations and exploiting existing high-quality parameter combinations through continuous iteration. In each iteration, according to the prediction results of the surrogate model, a process parameter point with the maximum expected improvement (EI) is selected for experimentation. The expected improvement measures the expected improvement in product quality that may be brought about by conducting an experiment at this point. After obtaining new product quality data from the experiment, update the surrogate model, recalculate the expected improvement for each point, and continue the next iteration.

[0100] Fine Tuning Implementation: In actual production, production experiments are carried out according to the process parameter points determined by the Bayesian optimization algorithm. For example, first select points with higher expected improvement near the current process parameters for experimentation and record the product quality data. As the iteration progresses, gradually converge to the optimal combination of process parameters. During the optimization process, combined with the real-time feedback of product quality, dynamically adjust the parameters of Bayesian optimization, such as adjusting the balance parameter between exploration and exploitation, to more quickly find the fine process parameters that meet the quality requirements of high-end medical plastic products.

[0101] Example 9

[0102] Global Search for Blow Molding Parameters Based on Chaotic Optimization Algorithm: Select a chaotic mapping, such as the Logistic mapping, to generate a chaotic sequence. The Logistic mapping formula is, where is the value of the th iteration and is the control parameter. When the value of is in the range (3.5699456, 4], the system is in a chaotic state. By adjusting the initial value and the control parameter, a chaotic sequence with randomness, ergodicity, and regularity is generated.

[0103] Application of Chaotic Optimization: Map the blow molding process parameters (such as blowing pressure, blow molding time, tube drawing speed, etc.) to the chaotic sequence. For example, map the values of the chaotic sequence to the range of the process parameter values through linear transformation to obtain a set of initial process parameter combinations. Define a fitness function that comprehensively considers product quality indicators (such as wall thickness uniformity and compressive strength of the tube) and production efficiency indicators (such as production speed and energy consumption). Evaluate the initial process parameter combinations and calculate their fitness values. Then, adjust the process parameters through a chaotic search strategy. For example, under the drive of the chaotic sequence, make small perturbations to the process parameters to generate new parameter combinations and evaluate their fitness values again. Retain the parameter combinations with better fitness values and continue the chaotic search, iterating continuously until the termination condition is met (such as reaching the maximum number of iterations or the fitness value converges).

[0104] Implementation of Global Optimization: In actual production, start the chaotic optimization from multiple different initial chaotic sequence values to increase the probability of finding the global optimal solution. After each optimization obtains a set of relatively optimal process parameter combinations, conduct experimental verification on the production line. According to the feedback from actual production, adjust the parameters of the chaotic optimization algorithm (such as the control parameter of the chaotic mapping, the perturbation amplitude, etc.) to further optimize the search process. Through multiple iterations and experiments, find the global optimal process parameter combination that can balance product quality and production efficiency.

Claims

1. A self-adaptive adjustment system for the quality stability of recycled plastic blow molding products, characterized in that: include: Product quality data acquisition module: In the product quality inspection link of the blow molding production line, high-precision inspection equipment is provided to collect product quality data in real time. The product quality data includes wall thickness distribution data obtained by measuring at different positions of the product using a wall thickness measuring instrument, tensile test data recorded by a tensile testing machine, and product surface defect data identified by an appearance inspection device. The inspection equipment is connected to the production line control system to transmit quality data to the subsequent fuzzy logic control system in real time. Environmental parameter acquisition module: Environmental parameter sensors are arranged around the blow molding machine and in the production workshop to monitor the ambient temperature and humidity in real time. The temperature sensor uses a thermocouple or thermistor sensor, and the humidity sensor uses a capacitive or resistive humidity sensor. The sensors are distributed in different positions to obtain the overall environmental parameters. The collected ambient temperature and humidity data are transmitted to the fuzzy logic control system in real time; Input variable fuzzification unit: product quality test data and environmental parameters are used as input variables of the fuzzy logic controller. Each input variable is divided into fuzzy subsets according to the actual value range, and the corresponding membership function is defined for each subset; Output variable fuzzification unit: The process parameters of the blow molding machine are used as the output variables of the fuzzy logic controller. Each output variable is divided into fuzzy subsets according to its adjustment range, and the membership function is defined; Fuzzy rule formulation module: Based on the recycled plastic blow molding process knowledge and expert experience, fuzzy rules describing the fuzzy relationship between input variables and output variables are formulated, and the fuzzy rule library is continuously improved through experiments and production experience accumulation; Fuzzy reasoning unit: using the appropriate fuzzy reasoning method such as Mamdani reasoning method or Sugeno reasoning method, according to the fuzzy rules and the fuzzification results of the input variables, the fuzzy value of the output variable is derived; Defuzzification unit: The centroid method and maximum membership method are used to defuzzify the output fuzzy value obtained by fuzzy reasoning, and the output fuzzy value is converted into the specific adjustment value of the process parameters of the blow molding machine; Real-time monitoring and feedback module: The fuzzy logic control system receives product quality test data and environmental parameter data in real time, compares them with the pre-set product quality standards and ideal range of environmental parameters, and determines whether the product quality fluctuates and whether the environmental parameters deviate from the ideal state. If product quality fluctuations or environmental parameter changes are detected, the adaptive adjustment mechanism is triggered and the deviation information is fed back to the fuzzy logic control algorithm; Process parameter automatic adjustment module: According to the calculation results of the fuzzy logic control algorithm, the process parameters of the blow molding machine, including blowing pressure, temperature and blowing time, are automatically adjusted to ensure that the product quality always remains within a stable range; Dynamic optimization and learning module: By analyzing historical production data and product quality data, the fuzzy rule base and membership function are optimized to adapt to changes in the production process.

2. The self-adaptive adjustment system for the quality stability of recycled plastic blow molding products according to claim 1 is characterized in that: In the product quality data acquisition module, the wall thickness measuring instrument measures the wall thickness of the product at different positions near the discharge port at regular intervals after the product is formed, measuring multiple points each time. The regular time is 10 seconds, and the measurement positions include the top, middle, and bottom.

3. The self-adaptive adjustment system for the quality stability of recycled plastic blow molding products according to claim 1 is characterized in that: In the product quality data collection module, in the product quality inspection area, a certain number of products are randomly selected from every certain production batch and subjected to tensile strength testing using a tensile testing machine, the certain production batch being 100 products per batch, and the certain number being 5.

4. The self-adaptive adjustment system for the quality stability of recycled plastic blow molding products according to claim 1 is characterized in that: In the environmental parameter acquisition module, the temperature sensor collects temperature data every 1 minute, the humidity sensor collects humidity data every 2 minutes, and the temperature sensor and the humidity sensor are calibrated once a month.

5. The self-adaptive adjustment system for the quality stability of recycled plastic blow molding products according to claim 1 is characterized in that: In the fuzzy processing module, for the wall thickness uniformity, its value range is divided into [0,100], divided into five fuzzy subsets of "very uneven", "uneven", "normal", "even", and "very even", and a triangular membership function is used.

6. The self-adaptive adjustment system for the quality stability of recycled plastic blow molding products according to claim 1 is characterized in that: In the fuzzy processing module, taking the blowing pressure as an example, its adjustment range is [0.5, 1.5] MPa, which is divided into five fuzzy subsets of "very low", "low", "moderate", "high" and "very high", and a triangular membership function is used.

7. The self-adaptive adjustment system for the quality stability of recycled plastic blow molding products according to claim 1 is characterized in that: In the fuzzy reasoning and defuzzification module, the Mamdani reasoning method is used for fuzzy reasoning, and the membership of wall thickness uniformity and temperature in respective fuzzy subsets is determined according to the fuzzification results of the input variables. The minimum value of the two is taken as the activation strength of the rule, and the fuzzy output set of the blowing pressure is obtained through fuzzy synthesis operation.

8. The self-adaptive adjustment system for quality stability of recycled plastic blow molding products according to claim 1, characterized in that: In the fuzzy reasoning and defuzzification module, the centroid method is used for defuzzification. For the fuzzy output set of the blowing pressure, the centroid is calculated. Determine the defuzzified blowing pressure adjustment value, where y i is an element in the fuzzy output set of blowing pressure, μ(y i ) is its membership degree, and n is the number of elements in the set.

9. The self-adaptive adjustment system for quality stability of recycled plastic blow molding products according to claim 1, characterized in that: The dynamic optimization and learning module of the adaptive adjustment mechanism optimizes and adjusts the fuzzy rule base and the membership function by establishing a historical production data and product quality database, using data mining techniques such as association rule mining and cluster analysis to analyze the data in the database.

Images