Rapid material changing and cleaning method and system for full-biodegradable multiphase composite material

Abstract

The invention relates to the technical field of high polymer material processing and modification, and discloses a rapid material changing and cleaning method and system for a full-biodegradable multi-phase composite material, and the method comprises the following steps: a data interface module obtains front and back formulas and equipment state data containing a screw wear coefficient; the cleaning decision engine performs reasoning on data characteristics by using a machine learning model to generate a three-section type strategy including physical stripping, chemical dissolution and wetting and precise flushing; the formula execution module controls the feeding and granulating unit to execute the strategy according to the time sequence; and the online monitoring and feedback module judges whether the cleaning is qualified or the circulation is triggered based on the melt spectrum matching degree. A self-adaptive cleaning process is generated by combining formula differences and equipment abrasion states, three-section type time sequence operation is utilized to cooperate with physical shearing and chemical softening effects, and online spectrum closed-loop monitoring is matched, so that the problems that full-biodegradable materials are not thoroughly replaced and cleaned and the materials are wasted are solved, and efficient and automatic intelligent cleaning is achieved.

Description

Technical Field

[0001] This invention relates to the field of polymer material processing and modification technology, specifically to a rapid material replacement and cleaning method and system for fully biodegradable multiphase composite materials. Background Technology

[0002] With increasingly stringent environmental regulations, the application of fully biodegradable materials such as polylactic acid (PLA) and polybutylene adipate / terephthalate (PET) is rapidly expanding in packaging, agriculture, and disposable products. To meet the demands of different applications for material strength, heat resistance, and degradation cycles, plant fibers, inorganic mineral powders, or reactive chain extenders are often added to the matrix resin during production to form multiphase composite materials. The complexity of this material system presents challenges for the frequent switching between multiple product types in extrusion granulation production lines.

[0003] Fully biodegradable materials are generally heat-sensitive, and are prone to degradation, cross-linking, or carbonization under high temperatures or strong shear, resulting in stubborn deposits on screw edges, barrel walls, and dead corners of the flow channel. Current production processes typically involve directly rinsing the substrate resin of the next product during material changeover. Because the melt flow is laminar and pure resin has limited mechanical stripping ability against surface deposits, this method of washing material with material often leads to long rinsing times, large amounts of transition material, and difficulty in completely removing residues containing plant fibers or highly filled powders, affecting the appearance and performance of the next batch of product.

[0004] Although dedicated screw cleaning agents exist on the market, existing cleaning processes mostly employ fixed operating procedures, namely setting constant temperature and speed for discharge. This rigid process fails to consider the specific differences in composition, viscosity, and color between batches of the formula, and also ignores the impact of extruder screw wear on shearing efficiency after long-term use. When the formula variation is small or the equipment is new, a fixed process may result in excessive waste of cleaning agent; while when the formula difference is significant or the screw wear is severe, leading to increased clearance, conventional cleaning processes cannot provide sufficient shearing force to remove contaminants, resulting in feed change failure.

[0005] Furthermore, current technologies primarily rely on operators visually inspecting color changes in the extrudate to determine whether cleaning is satisfactory. This manual judgment method is not only highly subjective and suffers from slow response times, but also struggles to detect trace residues invisible to the naked eye. In the precision processing of fully biodegradable materials, even trace amounts of crosslinking compounds or foreign resin residues can become stress concentration points or degradation initiation points in subsequent processing, thus posing potential quality risks. Therefore, the industry needs an intelligent cleaning solution that can adaptively adjust the process based on formulation differences and equipment status, and objectively determine the cleaning endpoint in real time. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a rapid material replacement and cleaning method and system for fully biodegradable multiphase composite materials. This solves the problems in the production of fully biodegradable materials, such as incomplete material replacement and cleaning due to complex formulation components, the inability of fixed cleaning processes to adapt to varying formulations, and material waste and low production efficiency caused by relying on manual experience to determine the cleaning endpoint.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a rapid material replacement and cleaning method and system for fully biodegradable multiphase composite materials, comprising: The first aspect of this invention provides a rapid material replacement and cleaning method for fully biodegradable multiphase composite materials. This method is applied to a central intelligent control unit and controls the material replacement and cleaning process through data-driven strategy generation and closed-loop feedback mechanism.

[0008] In this method, the data interface module responds to the material change command and reads the first formula data, the second formula data to be produced, and the equipment status data of the current production line. The equipment status data includes the screw wear coefficient, which is calculated by weighting the cumulative running time of the screw and the historical integral value of the main motor torque, and is used to quantitatively characterize the gap state between the screw and the barrel.

[0009] The cleaning decision engine performs inference calculations on the above data based on its built-in machine learning model, and outputs a cleaning strategy. Before the inference calculation, the system converts the multidimensional data into a numerical feature vector, which consists of three parts: first, the formula component difference characteristics, calculating the absolute value of the difference in plant fiber content and the difference in total inorganic powder filler content between the first and second formulas; second, the physicochemical property difference characteristics, calculating the Euclidean distance between the two formulas in the CIELAB color space and the difference in melt flow rate of the matrix resin; and third, the aforementioned equipment operating status characteristics. Based on this feature vector, the cleaning decision engine outputs a cleaning strategy that includes the cleaning material sequence, material usage at each stage, and screw speed setting curve.

[0010] The formula execution module analyzes the cleaning strategy and sends instructions to the multi-channel precision feeding unit and the composite granulation unit, executing the following three stages sequentially: The first stage is the physical stripping stage. The system controls the feeding of a highly filled cleaning masterbatch, which uses polypropylene carbonate or polylactic acid as a carrier and contains 40% to 60% inorganic mineral powder by mass. The melt flow rate at 190℃ and a load of 2.16 kg is less than 3 g / 10 min. In terms of process, the screw speed is increased to 70% to 85% of the rated speed, and the barrel temperature is set 10℃ to 20℃ lower than the normal processing temperature. This setting utilizes the high shear stress generated by the high-viscosity material under low-temperature, high-speed conditions to physically strip away the strong adhesions on the inner wall of the barrel and the surface of the screw.

[0011] The second stage is the chemical dissolution and wetting stage. The system switches to the chemical cleaning material containing reactive additives. In the matrix of this cleaning material, there are maleic anhydride grafted polymers dispersed at 8% to 15% by mass percentage and 1% to 3% of citrate-based bio-based high-boiling solvents. Technologically, the screw speed is reduced to 30% to 50% of the rated speed, and the barrel temperature is further reduced by 5°C to 15°C based on the physical peeling stage. This setting aims to extend the residence time of the material, and utilize the reaction activity of polar groups and the penetration of solvents to cause the residues to swell and soften.

[0012] The third stage is the precision rinsing stage. The system controls the feeding of the pure matrix resin specified by the second formulation data, and gradually restores the barrel temperature and screw speed to the standard production settings of the second formulation data. At the same time, the back pressure of the screen changer is adjusted, and the target resin is used to displace and discharge the cleaning material and softened pollutants remaining in the previous stage.

[0013] During the execution process, the online monitoring and feedback module analyzes the melt spectral data collected by the near-infrared spectrometer in the online quality detection unit. The analysis process includes performing standard normal variate transformation and second derivative processing on the original spectrum, and calculating the mathematical correlation between the real-time spectrum and the pre-stored standard fingerprint spectrum of the pure matrix resin of the second formulation using the principal component analysis method. When the matching degree reaches the preset threshold and maintains the preset time length, the cleaning is determined to be qualified. If the execution time of the precision rinsing stage exceeds the preset maximum rinsing time and the matching degree does not meet the standard, the system determines it as a stubborn pollution state and triggers the formulation execution module to automatically jump back to the chemical dissolution and wetting stage to re-execute the cleaning cycle.

[0014] The second aspect of the present invention provides a rapid material change cleaning system for a fully biodegradable multiphase composite material, including a data interface module, a cleaning decision engine, a formulation execution module, and an online monitoring and feedback module.

[0015] The data interface module is connected to the production execution system through an industrial fieldbus or Ethernet, and is used to obtain the first formulation data, the second formulation data, and the equipment status data.

[0016] The cleaning decision engine is connected to the data interface module and has a built-in machine learning model, which is used to calculate and generate a cleaning strategy including three-stage process parameters based on the input data.

[0017] The formulation execution module is connected to the cleaning decision engine and is used to analyze the cleaning strategy and control the multi-channel precise feeding unit and the composite granulation unit to perform physical peeling, chemical dissolution and wetting, and precision rinsing operations.

[0018] The online monitoring and feedback module is connected to the near-infrared spectrometer and is used to process the melt spectral data in real time, calculate the spectral matching degree to determine the cleaning end point, and feedback the determination result to the formulation execution module.

[0019] This invention provides a rapid material replacement and cleaning method and system for fully biodegradable multiphase composite materials. It offers the following advantages: 1. This invention employs a sequential combination of physical stripping, chemical dissolution and wetting, and precision rinsing processes. By constructing a low-temperature, high-shear stress field during the physical stripping stage, the frictional action of the highly filled masterbatch is used to strip away the hard coke deposits on the barrel wall. In conjunction with the reactive additives in the chemical dissolution stage, the residues in the dead corners of the screw are softened. This phased treatment mechanism targets the characteristics of different types of contaminants in multiphase composite materials. Compared with the traditional single resin rinsing method, it improves the cleaning efficiency for materials containing plant fibers and highly filled powders, and shortens the material changeover downtime.

[0020] 2. This invention introduces a machine learning-based cleaning decision engine. The system converts the differences in components, physicochemical properties, and wear coefficients of the equipment screws before and after the formulation into numerical feature vectors, and infers and generates customized cleaning material sequences and process parameters. This overcomes the limitations of traditional fixed cleaning programs and can dynamically match the cleaning intensity according to the formulation range and the actual aging state of the equipment. While ensuring thorough cleaning, it effectively reduces the consumption of expensive biodegradable cleaning materials.

[0021] 3. This invention establishes an online closed-loop feedback mechanism based on melt spectral analysis. It monitors the melt in real time using a near-infrared spectrometer and calculates its mathematical correlation with the standard fingerprint spectrum of the target formulation. It uses quantitative data to replace manual experience in determining the cleaning endpoint and automatically triggers process rollback when stubborn contamination is detected. This eliminates the subjective error and lag of manual visual judgment, ensures the purity of the first batch of products after production conversion, and realizes fully automated closed-loop control of the material change process. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the rapid material change and cleaning system architecture according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating the rapid material change and cleaning method according to an embodiment of the present invention.

[0023] Among them, 100 is a multi-channel precision feeding unit; 200 is a compound granulation unit; 300 is an online quality detection unit; 310 is a near-infrared spectrometer; 400 is a central intelligent control unit; 410 is a data interface module; 420 is a cleaning decision engine; 430 is a formula execution module; and 440 is an online monitoring and feedback module. Detailed Implementation

[0024] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] See attached document Figure 1 This invention provides a rapid material change and cleaning system for fully biodegradable multiphase composite materials. The system is integrated into or communicates with a central intelligent control unit 400 to coordinate and control multiple precision feeding units 100, composite granulation units 200, and online quality detection units 300, so as to achieve rapid switching of production formulas and equipment cleaning.

[0026] The rapid material change cleaning system includes a data interface module 410, a cleaning decision engine 420, a formula execution module 430, and an online monitoring and feedback module 440.

[0027] The data interface module 410 is configured as the system's input terminal, connecting to the production execution system and the underlying controller of the equipment via an industrial fieldbus or Ethernet. Industrial fieldbuses include, but are not limited to, Profibus-DP, CANopen, DeviceNet, or EtherCAT high-speed real-time Ethernet protocols; Ethernet connections support the OPC UA unified architecture or the MQTT message queue telemetry transmission protocol to ensure high-frequency data interaction and interoperability between heterogeneous devices. The data interface module 410 receives material change commands and acquires in real-time the first formulation data currently in production, the second formulation data to be produced, and equipment status data. The first and second formulation data include resin component ratios, plant fiber content, inorganic filler types and contents, and masterbatch information. Equipment status data includes screw configuration information, screw wear coefficient, die type, and filter usage time.

[0028] The cleaning decision engine 420 communicates with the data interface module 410 and has a built-in pre-trained machine learning model. The cleaning decision engine 420 performs calculations based on the multi-dimensional data transmitted from the data interface module 410 and outputs a structured cleaning strategy. The cleaning strategy includes the cleaning material sequence, material usage at each cleaning stage, screw speed setting curve, and temperature setting curves for each temperature zone of the barrel.

[0029] The formulation execution module 430 is connected to the cleaning decision engine 420 and the multi-channel precision feeding unit 100. The formulation execution module 430 analyzes the cleaning strategy output by the cleaning decision engine 420, generates control commands, and sends them to the loss-in-weight weighing scales of the multi-channel precision feeding unit 100 and the main controller of the composite granulation unit 200. The formulation execution module 430 controls the start-up and shutdown sequence and feeding rate of the cleaning masterbatch hopper, chemical cleaning hopper, and resin hopper, and synchronously adjusts the extruder screw speed and barrel temperature to execute a phased gradient cleaning process.

[0030] The online monitoring and feedback module 440 is communicatively connected to the near-infrared spectrometer 310 in the online quality inspection unit 300. The online monitoring and feedback module 440 receives melt spectral data collected by the near-infrared spectrometer 310 and performs feature extraction and matching degree calculation on the spectral data. The online monitoring and feedback module 440 feeds the calculation results back to the formula execution module 430. When the calculation results meet the preset termination conditions, the system is triggered to end the cleaning process and switch to normal production mode.

[0031] See attached document Figure 2 This process is executed automatically by the aforementioned system and covers the entire process from receiving instructions to confirming the completion of cleaning.

[0032] The system first responds to externally triggered material change requests through the data interface module 410. The data interface module 410 reads the first formula parameters currently in operation on the production line and loads the second formula parameters to be put into production. At the same time, the data interface module 410 reads the current cumulative screw running time and historical torque records from the equipment controller to calculate the current equipment wear status.

[0033] Subsequently, the cleaning decision engine 420 receives the aforementioned data and transforms the component differences between the first and second formulations and the equipment status into feature vectors. The cleaning decision engine 420 uses a built-in model to reason about the feature vectors, determining the required combination of cleaning materials, the estimated quality of the cleaning materials at each stage, and the trajectory of process parameters during the cleaning process, thereby generating a cleaning strategy data package.

[0034] Based on the cleaning strategy data package, the formulation execution module 430 sequentially controls the multi-channel precision feeding unit 100 and the composite granulation unit 200 to perform a three-stage cleaning process. In the first stage, the formulation execution module 430 controls the feeding of highly filled cleaning masterbatch and sets a high screw speed and a low processing temperature for physical stripping. In the second stage, the formulation execution module 430 controls the switching to a chemical cleaning material containing reactive additives, reducing the screw speed and further lowering the temperature for chemical dissolution and wetting. In the third stage, the formulation execution module 430 controls the feeding of the matrix resin of the second formulation and restores the process parameters to the standard production settings of the second formulation for precision rinsing.

[0035] During the third phase of execution, the online monitoring and feedback module 440 continuously analyzes the real-time data transmitted back by the near-infrared spectrometer 310. The online monitoring and feedback module 440 compares the real-time spectrum with the pre-stored standard spectrum of the pure matrix resin of the second formulation. When the matching degree between the two reaches a preset threshold and remains at that threshold for a preset time, the online monitoring and feedback module 440 determines that the equipment cleaning is qualified and sends a cleaning completion signal to the formulation execution module 430. The formulation execution module 430 then adjusts the proportions of the multi-channel precision feeding unit 100, feeding all components according to the second formulation, completing the production switchover.

[0036] If the spectral matching degree still fails to reach the preset threshold after the third stage has been performed for more than the preset maximum rinsing time (e.g., 20 minutes), the online monitoring and feedback module 440 will determine it as a stubborn contamination state. At this time, the system will automatically issue a manual intervention alarm to the operator, or automatically jump back to the second stage according to the preset degradation strategy, and re-execute a short-cycle chemical cleaning cycle to avoid the system falling into an endless ineffective rinsing state.

[0037] In this embodiment, before performing strategy calculations, the cleaning decision engine 420 first preprocesses the unstructured or heterogeneous data obtained from the data interface module 410, transforming it into a numerical feature vector of a unified dimension. This numerical feature vector serves as the basic input for the machine learning model's inference, covering differences in formulation components, differences in physicochemical properties, and equipment operating status. For the differences in formulation components, the cleaning decision engine 420 calculates the absolute value of the difference in the mass percentage of key components between the first and second formulations.

[0038] Specifically, the system extracts the content data of plant fibers and inorganic powder fillers separately. For plant fibers, the system calculates the absolute value of the difference between the plant fiber content in the first formulation and the plant fiber content in the second formulation. This characteristic reflects the degree of risk of physical adhesion or carbonization accumulation deep in the screw threads and on the inner wall of the barrel. For inorganic powder fillers, the system calculates the total difference between the two in terms of inorganic components such as hydroxyapatite, talc, or calcium carbonate. This characteristic is related to the shear strength required during the cleaning process and the amount of physical stripping medium used.

[0039] The differences in physicochemical properties primarily include color difference and melt rheological difference. The cleaning decision engine 420 acquires standard color swatch data for the first and second formulations. This data is based on the CIELAB color space and includes lightness (L), red-green axis component (a), and yellow-blue axis component (b). The system calculates the Euclidean distance between the two formulations in the CIELAB space to obtain the color difference characteristic value. This value directly determines the resin volume and time required for the third-stage precision rinsing; the greater the color difference, the greater the required displacement rinsing volume. The melt rheological difference is obtained by calculating the difference in melt flow rate of the two formulation matrix resins under standard test conditions, used to characterize the hindering effect of abrupt changes in interfacial viscosity on the flow of the cleaning material.

[0040] To address resin matrix compatibility, the cleaning decision engine 420 establishes a matrix change index. The system compares the main matrix resin types of the first and second formulations. When the main matrix resin changes from poly(adipic acid) or butylene terephthalate to different categories such as polylactic acid or polyglycolic acid, the system generates a Boolean value characteristic for the matrix change. Furthermore, the system incorporates a Hansen solubility parameter database for common biodegradable resins, calculating the solubility parameter distance between the first and second formulation resin matrices as a polarity difference characteristic. This polarity difference characteristic indicates the degree of polarity matching of the solubilizer or solvent required in the second-stage chemical cleaning.

[0041] To address the characteristics of equipment operation, the system incorporates screw wear coefficient and flow channel resistance coefficient. The screw wear coefficient is not a single time variable but a composite index calculated based on historical equipment operating data. The system retrieves the cumulative operating time of the extruder since the last screw replacement or maintenance and performs a weighted calculation combining this time with the average load rate or historical torque integral value of the main motor. A higher screw wear coefficient indicates a larger gap between the screw threads and the inner wall of the barrel, resulting in lower shearing efficiency during cleaning. The model accordingly increases the set amount of cleaning material or adjusts the speed compensation. The flow channel resistance coefficient is calculated in real-time based on the current filter mesh size and the melt pressure difference across the screen changer, used to correct the back pressure setting during the cleaning process.

[0042] Furthermore, the core algorithm architecture of the cleaning decision engine 420 is built upon ensemble learning theory, preferentially employing gradient boosting decision trees or random forest algorithms. This type of algorithm can effectively handle high-dimensional heterogeneous data and capture the complex nonlinear mapping relationship between input features and cleaning process parameters. Compared to traditional single logistic regression or rule engines, the ensemble tree model combines multiple weak learners to form a strong learner, exhibiting higher robustness and prediction accuracy when processing small-sample, imbalanced industrial production data.

[0043] The construction of the cleaning decision model begins with the preparation of the training dataset. The system collects historical material changeover logs from experimental production lines and actual industrial sites as raw samples. Each sample contains two parts of data: material changeover scenario description data as the input feature vector and validated optimal cleaning scheme data as the target label. The input feature vector covers the aforementioned differences in formulation components, physicochemical properties, and equipment status characteristics. The target label is a set of cleaning process parameters that have been verified and confirmed to be qualified in actual production, specifically including the type identification code of the cleaning material, the actual consumption mass of the cleaning material at each stage, the set value sequence of screw speed, and the temperature adjustment value sequence of each temperature zone of the barrel.

[0044] During the training phase of the cleaning decision model, the algorithm employs a supervised learning paradigm. Taking gradient boosting decision trees as an example, the model is initialized with a constant prediction value, and then new decision trees are added iteratively round by round. The goal of generating each new tree is to fit the residual between the model's prediction value from the previous round and the true target label. When constructing each decision tree, the algorithm traverses all input features and all possible split points under that feature, calculates the gain after splitting, and selects the features that can minimize the loss function and the split threshold as nodes. Through multiple iterations, the cleaning decision model gradually reduces the prediction error, ultimately forming an additive model that can accurately predict the cleaning strategy.

[0045] To address the diversity of cleaning strategy outputs, the cleaning decision model of this invention is designed as a multi-objective output structure or a model group composed of multiple sub-models. Specifically, for the selection of cleaning material sequences, the cleaning decision model is trained using a multi-classification loss function, outputting the probability distribution of the selection of various cleaning masterbatches and chemical cleaning materials, and selecting the combination with the highest probability as the recommended sequence. For continuous numerical parameters such as cleaning material dosage, screw speed, and temperature, the cleaning decision model is trained using a regression loss function, outputting specific numerical prediction results. This multi-task parallel processing mechanism ensures that the generated cleaning strategy achieves optimal accuracy in both material selection and process parameter precision.

[0046] Furthermore, the cleaning decision model possesses the capability to assess feature importance. After training, the model calculates the importance score of each input feature based on the average gain brought by each feature during decision tree splitting. For example, the model identifies that the difference in plant fiber content has a high weight for predicting the amount of material used in the first-stage physical cleaning, while the color difference feature dominates the prediction of the time for the third-stage precision rinsing. This weighting information is used for model pruning and optimization, and also provides quantitative data support for subsequent process analysis. After passing the evaluation on the training and validation sets, the cleaning decision model is encapsulated and deployed in the cleaning decision engine 420 for inference responses to real-time material change requests.

[0047] In addition, the system has an online incremental learning function. After each cleaning process, the data interface module 410 automatically collects the actual cleaning time, material consumption, and final spectral matching curve, and combines them with the feature vector before cleaning to form new sample data. The cleaning decision engine 420 uses this new sample to fine-tune and update the built-in model. By adjusting the weights of the leaf nodes of the decision tree, the model can adapt to the performance drift caused by equipment aging, and achieve continuous iterative optimization of the cleaning strategy.

[0048] The dynamic cleaning strategy generated by the cleaning decision engine 420 is represented as a structured control data package containing temporal logic and multi-dimensional process parameters. This data package, after encoding, can be directly parsed by the formulation execution module 430 into a sequence of machine instructions for the underlying controller. The strategy data package first includes the feeding sequence and quantitative parameters of the cleaning medium. The data package explicitly specifies the specific material identification codes to be used in each stage of the three-stage cleaning process, corresponding to high-filling physical cleaning masterbatch, reactive chemical cleaning material, and the target formulation matrix resin as the final replacement medium. Corresponding to the material type, the data package provides precise mass setpoints for the cleaning material at each stage. These setpoints are specific values ​​derived by the model based on the current formulation variation and equipment contamination level, directly mapped to the target total amount of material fed by each loss-in-weight scale in the multi-channel precision feeding unit 100, ensuring that the consumption of cleaning material is minimized while meeting the cleaning cleanliness requirements.

[0049] The dynamic cleaning strategy further incorporates a time-varying control curve for the screw speed. This curve defines the dynamic trajectory of the screw speed as the cleaning process progresses, rather than a single constant speed setting. Structurally, this is represented as an ordered sequence of coordinates containing timestamps and speed setpoints. The strategy typically involves maintaining a high speed plateau during the physical stripping stage to remove strong adhesions using high shear rates, linearly decreasing to a low speed plateau during the chemical cleaning stage using a set slope to significantly extend material residence time, and smoothly increasing the speed back to the standard production speed for the second formulation during the precision rinsing stage. This curve data also includes acceleration / deceleration slope limit parameters to prevent torque overshoot or mechanical shock to the equipment caused by drastic speed fluctuations.

[0050] Furthermore, the strategy data package defines in detail the time-varying control curve for barrel temperature. This curve sets temperature adjustment targets and timing for each independent temperature zone distributed along the material conveying direction of the extruder. The strategy data specifies lowering the temperature setpoint for specific temperature zones during the physical cleaning phase, aiming to enhance the physical scraping effect by increasing melt viscosity; it also specifies a temperature holding window for the chemical cleaning phase, matching the optimal reactivity temperature range of the chemical additives to prevent additive decomposition due to excessively high temperatures or poor plasticization due to excessively low temperatures. The temperature control curve also incorporates a lead parameter based on the equipment's thermal inertia to ensure that the actual barrel temperature responds closely to process requirements. Finally, the data package integrates process safety thresholds, including the maximum permissible melt pressure and torque percentage upper limit during the cleaning process, providing safe operating boundaries for the execution module.

[0051] Specifically, the formula execution module 430 incorporates a graded alarm and melt bypass mechanism. When the real-time monitored melt pressure at the die head exceeds the first safety threshold (e.g., 25 MPa), the system prioritizes triggering the automatic draining action of the screen changer or reducing the screw speed; if the pressure continues to exceed the second safety threshold (e.g., 35 MPa), the system immediately triggers a shutdown interlock and opens the pressure relief valve to prevent gas expansion caused by the degradation of the high-temperature cleaning material from damaging the equipment.

[0052] The first stage executed by the formulation execution module 430 is the high-shear physical stripping stage. At the initial stage of the cleaning process, the formulation execution module 430, based on the instructions generated by the cleaning decision engine 420, controls the multi-channel precision feeding unit 100 to stop the supply of the first formulation raw materials and opens the feeding channel of the dedicated physical cleaning masterbatch hopper. The core process objective of this stage is to utilize the principles of fluid dynamics to construct a high-intensity shear stress field in the screw thread, screw groove root, and inner wall of the barrel, thereby forcibly stripping away the plant fiber carbon deposits, degradation gel points, and highly adhesive inorganic filler agglomerates that have adhered firmly due to prolonged heating and oxidation through mechanical force.

[0053] To achieve the aforementioned physical stripping effect, the physical cleaning masterbatch selected in this invention possesses specific rheological characteristics and composition. This masterbatch uses polypropylene carbonate or polylactic acid as the carrier resin. These two resins exhibit good interfacial compatibility with fully biodegradable material systems, avoiding phase separation contamination that might occur with the introduction of dissimilar plastics such as polyethylene or polypropylene. Using this carrier resin as a matrix, the physical cleaning masterbatch is filled with a high proportion of inorganic mineral powder, preferably calcium carbonate or wollastonite, at a mass percentage of 40%–60%. These inorganic powders, after particle size screening, can form a dense suspended phase of solid particles during melt flow, improving the overall rigidity and coefficient of friction of the melt. Simultaneously, this cleaning masterbatch is designed to have an extremely low melt flow rate, below 3 g / 10 min at 190°C and a load of 2.16 kg. This high viscosity characteristic ensures that laminar slippage is unlikely to occur during extrusion, thereby efficiently converting the mechanical energy transmitted by the screw into a tangential drag force against the wall adhesions.

[0054] In terms of specific process parameter execution, the formula execution module 430 controls the temperature control system of the composite granulation unit 200, adjusting the set temperature of each temperature zone of the barrel to a range 10℃~20℃ lower than the normal processing temperature. The physical mechanism of lowering the temperature lies in utilizing the characteristic that polymer viscosity increases exponentially with decreasing temperature, artificially increasing the apparent viscosity of the melt in the barrel, thereby increasing the shear stress on the wall. At the same time, the formula execution module 430 controls the main motor to rapidly increase the screw speed to 70%~85% of the rated speed. The high-speed operation causes the material in the screw channel to undergo high-frequency shearing and tensile deformation. Combined with the friction of high viscosity and high filler material, a strong scouring and scraping effect is formed on the dead corner area of ​​the screw surface, peeling off large pieces of adhering material and accumulated old material layers from the metal surface and carrying them out of the die head.

[0055] This scraping effect, rheologically speaking, manifests as the disruption of the laminar sublayer of the boundary layer by utilizing the yield stress characteristics of the highly filled masterbatch. Conventional resin melts have near-zero flow velocity at the wall surface, easily forming a stagnant layer; however, the physical cleaning masterbatch in this embodiment exhibits Bingham fluid characteristics, flowing only when the applied shear stress exceeds its yield value. This characteristic forces the melt to slide forward as a whole in an approximately plug flow form on the screw advance face, thereby maximally removing the static adsorbed layer on the inner wall of the barrel.

[0056] After the first stage is completed, the formulation execution module 430 executes the second stage of low-shear chemical dissolution and wetting process. Although most of the macroscopic deposits have been removed, microscopic polymer film layers or stubborn adhesions due to degradation may still remain in the dead corners of the die head flow channel, the edges of the filter plate, and the deep grooves of the screw. The second stage process aims to remove these residues that are difficult to reach by physical shear through chemical thermodynamic action, that is, by using solubility parameter matching and interfacial chemical reactions. The formulation execution module 430 switches the feed channel according to the instruction and feeds the chemical cleaning material containing reactive functional groups and penetrating solvents into the extrusion system.

[0057] The chemical cleaning agent used in this stage is a specially designed compound system. Its matrix resin is poly(butylene adipate / terephthalate) or polypropylene carbonate, which has certain compatibility with both the first and second formulations. Within this matrix, 8%–15% by weight of a maleic anhydride-grafted polymer, such as maleic anhydride-grafted poly(butylene adipate / terephthalate), is uniformly dispersed. The maleic anhydride groups possess extremely high polarity and reactivity, enabling them to chemically interact with or form hydrogen bonds with polar degradation products (such as hydroxyl or carboxyl-terminated oligomers) remaining on the metal surface, thereby reducing the interfacial bonding energy between the residue and the metal surface. In addition, 1%–3% by weight of a bio-based high-boiling-point solvent or plasticizer is added to the cleaning agent, preferably a citrate compound such as triethyl acetylcitrate. These small molecule additives, with their small molecular volume, can quickly penetrate into the interior of the residual polymer hardened layer. Through solvation, they disrupt the van der Waals forces between polymer chains, causing the residual layer to swell and soften, transforming it from a dense solid state into a loose gel state.

[0058] To ensure the thorough chemical dissolution and wetting processes described above, the process control strategy in this stage differs significantly from that in the first stage. The formulation execution module 430 controls the main motor to significantly reduce its speed, setting the screw speed to 30%–50% of its rated speed. This low-speed operation reduces the axial conveying velocity of the material within the barrel, significantly extending the residence time of the chemically cleaned material in the extrusion system, providing ample time for solvent molecule diffusion and interfacial reactions of active groups. Simultaneously, the formulation execution module 430 controls the temperature control system to further reduce the barrel temperature by 5°C–15°C compared to the first stage. This low-temperature setting prevents the low-molecular-weight solvent from evaporating too quickly or undergoing thermal decomposition at high temperatures, ensuring an effective concentration gradient in the melt. Furthermore, the lower temperature increases the melt strength of the base resin, helping to completely encapsulate the swollen and softened residue during screw propulsion and smoothly eject it via laminar flow displacement, preventing the formation of new wall-attached layers due to excessively low viscosity.

[0059] After the second stage of chemical cleaning material completes the predetermined residence time and cycle, the formulation execution module 430 automatically enters the third stage, namely the precision displacement and online confirmation stage. The process objective of this stage is to use the pure matrix resin in the target formulation that will be put into production to completely remove the chemical cleaning material containing dissolved contaminants remaining in the extruder, establish a melt flow environment that meets the quality standards of the new product, and achieve a smooth transition from cleaning conditions to normal production conditions.

[0060] The formulation execution module 430 first performs a material switching action, cutting off the supply channel of the chemical cleaning material and activating the loss-in-weight weighing scale of the matrix resin hopper in the second formulation. The system preferably selects the lightest-colored and most fluid resin component in the second formulation as the displacement medium, such as pure polylactic acid or poly(butylene adipate / terephthalate) chips. Choosing pure resin instead of a full-component mixture as the displacement medium is, on the one hand, to avoid interference from inorganic fillers or pigments in the full formulation with the subsequent near-infrared spectroscopy baseline calibration; on the other hand, it utilizes the lower interfacial tension of pure resin to form a near-piston flow melt advance front, thereby efficiently pushing the softened residues adhering to the screw surface and barrel inner wall in the previous stage layer by layer to the die outlet.

[0061] Simultaneously with material replacement, the formulation execution module 430 initiates a dynamic restoration program for process parameters. For temperature control, the system controls the heating elements in each temperature zone of the barrel to operate at a preset heating rate, gradually raising the barrel temperature from the low-temperature melting state of the second stage to the standard processing temperature of the second formulation. This heating process must be precisely synchronized with the material conveying progress to ensure that the newly introduced resin receives sufficient plasticizing energy, while avoiding cavitation caused by the rapid expansion of volatiles in the residual chemical cleaning material due to excessively rapid heating. For screw speed control, the system no longer maintains the ultra-low speed state of the second stage, but instead linearly increases the screw speed to 80%~90% of the rated production speed of the second formulation. Increasing the speed increases the melt volumetric flow rate, shortens the material residence time and distribution tailing in the barrel, and helps to quickly replace stale material in dead zones.

[0062] At this stage, the system introduces rheology-based pressure balance logic. The recipe execution module 430 monitors the melt pressure sensor feedback at the die head in real time and maintains a pressure level within the barrel sufficient to suppress melt wall slippage by adjusting the back pressure of the screen changer or melt pump. The stable back pressure environment forces the melt to fill the screw groove space, eliminating voids that could lead to cleaning dead zones.

[0063] To quantify the cleaning endpoint and avoid material waste caused by over-cleaning, the system utilizes an online monitoring and feedback module 440 in conjunction with a near-infrared spectrometer 310 installed at the extruder die head to construct a real-time quality closed-loop control loop. The core task of this control loop is to convert the microscopic chemical composition changes of the melt flow into calculable digital indicators, thereby replacing traditional manual visual observation.

[0064] In terms of hardware deployment and signal acquisition, the optical probe of the near-infrared spectrometer 310 adopts a high-temperature and high-pressure immersion design, directly screwed into the die head flow channel of the composite granulation unit 200 or the adapter flange hole after the screen changer. The probe end face is kept flush with the melt flow direction to ensure the acquisition of the most representative central flow layer data and avoid the generation of flow dead zones. The near-infrared spectrometer 310 transmits optical signals through optical fiber, uses a halogen tungsten lamp light source to irradiate the melt, and collects diffuse reflected light or transmitted light after scattering and absorption inside the melt. The grating beam splitting system inside the spectrometer decomposes the composite light into monochromatic light of different wavelengths, and the indium gallium arsenide detector array converts the light intensity signal into an electrical signal. The online monitoring and feedback module 440 continuously scans at a frequency of milliseconds, acquiring wavenumbers typically covering 4000~12500 cm⁻¹. -1 The original absorbance spectral data carries the overtone and combination frequency information of chemical bonds such as CH bonds, OH bonds, and NH bonds in the melt.

[0065] In terms of data processing and feature extraction, the online monitoring and feedback module 440 incorporates a spectral preprocessing algorithm unit. This unit first performs standard normal transformation and second-derivative processing on the original spectrum to eliminate background noise interference caused by melt pressure fluctuations, bubble scattering, or baseline drift, highlighting the waveform details of characteristic absorption peaks. To further overcome the problems of light source attenuation or probe contamination caused by long-term operation, the system introduces a dynamic dark current correction and background spectrum update mechanism. Before each cleaning process begins, in an empty state or at the moment of injection of pure transparent resin, the system automatically acquires the current background spectrum as a new benchmark to subtract environmental thermal radiation and systematic errors in the optical path system, ensuring the absolute accuracy of absorbance calculations.

[0066] Subsequently, the online monitoring and feedback module 440 calls upon a pre-established standard fingerprint spectrum library of the second-formulation pure matrix resin. This standard fingerprint spectrum is a spectral data model of a specific resin entered under ideal purity conditions. The online monitoring and feedback module 440 employs chemometric algorithms, such as principal component analysis or partial least squares, to calculate the mathematical correlation between the currently acquired dynamic spectrum and the standard fingerprint spectrum in real time. The system focuses on absorbance residuals at characteristic wavelengths; for example, for residues containing plant fibers, it monitors the change in absorption intensity in the cellulose hydroxyl peak region; for carbon black or inorganic filler residues, it monitors the baseline rise across the entire wavelength range.

[0067] In terms of determining the cleaning endpoint and controlling execution, the system defines a strict composite criterion logic. The online monitoring and feedback module 440 calculates a dimensionless matching degree index in real time, which characterizes the similarity between the current melt composition and the target pure resin, with a value range of 0% to 100%. The system sets the judgment threshold to 99.5%. However, a single point of instantaneous high matching degree is insufficient to trigger the end of cleaning because melt flow is unstable. Therefore, this invention introduces a time window stabilization mechanism, requiring the matching degree index to remain above 99.5% continuously within a preset time length, such as 60s to 120s, and the standard deviation of the spectral data within this time window to be lower than a preset fluctuation threshold. Only when the above stability conditions are met simultaneously can the online monitoring and feedback module 440 confirm that the cleaning is qualified, generate a cleaning completion status code, and send it to the cleaning decision engine 420 and the formula execution module 430. Upon receiving the signal, the system immediately terminates the third stage of pure resin rinsing and instructs the multi-channel precision feeding unit 100 to feed production materials according to the complete process sheet of the second formula, achieving a smooth switch.

Claims

1. A rapid material replacement and cleaning method for fully biodegradable multiphase composite materials, characterized in that, The application to the central intelligent control unit (400) includes the following steps: The data interface module (410) receives the material change command and reads the first formula data currently being produced, the second formula data to be produced, and the equipment status data including the screw running time and torque history data; The cleaning decision engine (420) uses a built-in machine learning model to reason about the first formula data, the second formula data and the equipment status data, and outputs a cleaning strategy that includes the cleaning material sequence, the material usage at each stage and the screw speed setting curve. The formulation execution module (430) analyzes the cleaning strategy and sends control commands to the multi-channel precision feeding unit (100) and the composite granulation unit (200) to execute the physical peeling stage using physical cleaning masterbatch, the chemical dissolution and wetting stage using chemical cleaning material, and the precision rinsing stage using the second formulation matrix resin in sequence according to the time sequence. The online monitoring and feedback module (440) analyzes the melt spectral data collected by the near-infrared spectrometer (310) in the online quality detection unit (300). When the matching degree between the real-time spectrum and the standard spectrum reaches the preset threshold and is maintained for a preset time, the cleaning is deemed qualified.

2. The rapid material replacement and cleaning method for fully biodegradable multiphase composite materials according to claim 1, characterized in that, Before inference, the cleaning decision engine (420) converts the first formula data, the second formula data, and the equipment status data into a numerical feature vector, the numerical feature vector including: The differences in formulation components are calculated from the absolute value of the difference in plant fiber content between the first formulation data and the second formulation data, and the difference in the total amount of inorganic powder fillers. The differences in physicochemical properties are calculated from the Euclidean distance between the first formulation data and the second formulation data in the CIELAB color space, and the difference in the melt flow rate of the matrix resin. The equipment operating status characteristics are composed of the screw wear coefficient, which is calculated by weighting the cumulative operating time of the screw with the historical integral value of the main motor torque.

3. The rapid material replacement and cleaning method for fully biodegradable multiphase composite materials according to claim 1, characterized in that, During the physical stripping stage, the formula execution module (430) controls the multi-channel precision feeding unit (100) to feed the physical cleaning masterbatch, which is a high-filling cleaning masterbatch. The composite granulation unit (200) controls the screw speed to increase to 70% to 85% of the rated speed and adjusts the barrel temperature to a range of 10°C to 20°C lower than the normal processing temperature.

4. The rapid material replacement and cleaning method for fully biodegradable multiphase composite materials according to claim 3, characterized in that, The high-filling cleaning masterbatch uses polypropylene carbonate or polylactic acid as the carrier resin and is filled with 40% to 60% inorganic mineral powder by mass. The melt flow rate at 190°C and 2.16 kg load is less than 3 g / 10 min.

5. The rapid material replacement and cleaning method for fully biodegradable multiphase composite materials according to claim 3, characterized in that, During the chemical dissolution and wetting stage, the formulation execution module (430) controls the multi-channel precision feeding unit (100) to switch to chemical cleaning material containing reactive additives, controls the composite granulation unit (200) to reduce the screw speed to 30% to 50% of the rated speed, and further reduces the barrel temperature by 5°C to 15°C based on the physical peeling stage.

6. The rapid material replacement and cleaning method for fully biodegradable multiphase composite materials according to claim 5, characterized in that, The chemical cleaning material contains 8% to 15% by mass of maleic anhydride graft polymer and 1% to 3% by mass of citrate-based bio-based high-boiling solvent in its matrix resin.

7. The rapid material replacement and cleaning method for fully biodegradable multiphase composite materials according to claim 1, characterized in that, During the precision rinsing stage, the formula execution module (430) controls the multi-channel precision feeding unit (100) to feed the pure matrix resin specified in the second formula data, controls the composite granulation unit (200) to gradually restore the barrel temperature and screw speed to the standard production settings of the second formula data, and adjusts the back pressure of the screen changer in the composite granulation unit (200).

8. The rapid material replacement and cleaning method for fully biodegradable multiphase composite materials according to claim 1, characterized in that, The online monitoring and feedback module (440) performs standard normal variable transformation and second derivative processing on the melt spectral data, and uses principal component analysis to calculate the mathematical correlation between the real-time spectrum and the pre-stored standard fingerprint spectrum of the second formulation pure matrix resin as the matching degree.

9. The rapid material replacement and cleaning method for fully biodegradable multiphase composite materials according to claim 1, characterized in that, If the execution time of the precision rinsing stage exceeds the preset maximum rinsing time and the matching degree does not reach the preset threshold, the online monitoring and feedback module (440) determines that it is a stubborn contamination state and triggers the formula execution module (430) to automatically jump back to the chemical dissolution and wetting stage to re-execute the cleaning cycle.

10. A rapid material replacement and cleaning system for fully biodegradable multiphase composite materials, characterized in that, The rapid material replacement and cleaning method for fully biodegradable multiphase composite materials according to any one of claims 1 to 9 includes: The data interface module (410) is connected to the production execution system via an industrial fieldbus or Ethernet to acquire formula data and equipment status data; The cleaning decision engine (420) is connected to the data interface module (410) and has a built-in machine learning model, which is used to calculate and generate a cleaning strategy containing three-stage cleaning process parameters based on the data of the data interface module (410). The formula execution module (430) is connected to the cleaning decision engine (420) and is used to analyze the cleaning strategy and control the multi-channel precision feeding unit (100) and the composite granulation unit (200) to perform physical stripping, chemical dissolution and wetting and precision rinsing operations. The online monitoring and feedback module (440) is connected to the near-infrared spectrometer (310) in the online quality detection unit (300) and is used to determine the cleaning endpoint based on the melt spectrum matching degree and feed it back to the formula execution module (430).

Images