On-line quality control method and system based on rheological properties of raw polymer

Abstract

This invention discloses an online quality control method and system based on the rheological properties of raw polymers. Real-time flow information from different sections of the raw polymer melt processing line is acquired and converted into one or more rheological property curves at different times. These rheological property curves at different times are then filtered using feature engineering methods to obtain correlated rheological property curves. These correlated rheological property curves are input into a polymer property calculation device to obtain the predicted product quality value for the current time period. It is then determined whether the predicted product quality value falls within a threshold range. If it does, the current flow information of the raw polymer processing line is considered to meet the requirements; otherwise, it does not meet the requirements and adjustments to the current flow information are necessary. This invention employs online rheological detection technology combined with AI algorithms, resulting in improved efficiency and accuracy of online quality control technology.

Description

Technical Field

[0001] This invention belongs to the field of polymer processing technology, and relates to an online quality control method and system based on the rheological properties of raw material polymers, and a method for determining the quality of polymer materials online. These methods involve rheological technology and artificial intelligence. Background Technology

[0002] In the manufacturing and processing of polymers, it is typically necessary to test polymer quality. This involves measuring polymer properties, such as melt flow, solubility, density, and mechanical properties, according to standard procedures like ASTM, to assess the continuity and reliability of the manufacturing process. The entire sampling and testing process occurs after the production process, taking several hours and making it difficult to obtain samples from each stage of production for testing. If polymer performance indicators do not meet requirements, the entire batch is discarded, and process engineers need to analyze the problem, determine when and at what stage it occurred, and take corrective actions. However, process engineers may need to analyze all data collected throughout the entire production process, making this process expensive, labor-intensive, and slow. Furthermore, the force field strength, flow rate, and channel geometry vary at different stages of the production line, resulting in different shear rates. The polymer melt / solution in the production line is a non-Newtonian fluid, and its rheological properties, such as viscosity, often differ at different shear rates. Therefore, monitoring only samples at the end of the production line is insufficient to pinpoint the specific stage and time of the problem. Since manufacturing plants typically have high productivity, the several-hour lag means that tons of product must be collected before any problems are detected.

[0003] Therefore, there is an unmet need in this field to accelerate the quality control process to reduce costs and provide process engineers with fast, accurate and reliable feedback. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing an online quality control method and system based on the rheological properties of raw material polymers.

[0005] In a first aspect, the present invention provides an online quality control method based on the rheological properties of raw material polymers, the method comprising:

[0006] S1: Real-time flow information of different sections of the raw material polymer melt processing line is acquired through sensor devices; wherein the flow information includes flow velocity and flow velocity gradient distribution, temperature and temperature gradient distribution, stress and stress gradient distribution;

[0007] S2: The collected real-time flow information is converted into one or more rheological property curves at different times. Each rheological property curve includes one or more peaks and / or valleys, and the peaks and valleys characterize the physicochemical properties of the raw material polymer.

[0008] The rheological properties include apparent viscosity, zero-shear viscosity, shear rate, shear stress, relaxation time, thixotropic ring, structural index, non-Newtonian index, critical strain in the linear viscoelastic region, Deborah number, complex modulus, storage modulus, loss modulus, loss factor, and loss tangent.

[0009] S3: The rheological property curves at different times above are filtered using the feature engineering method to obtain the relevant rheological property curves:

[0010] S4: Input the relevant rheological property curve into the polymer property calculation device to obtain the product quality prediction value for the current time period;

[0011] The polymer property calculation device executes one or more trained and validated machine learning algorithms;

[0012] S5: Determine whether the predicted product quality value falls within the threshold range. If so, the current raw material polymer processing line flow information is considered to meet the requirements. Otherwise, it does not meet the requirements and the process parameters of the current raw material polymer processing line need to be adjusted.

[0013] Preferably, the feature engineering method described in step S3 employs one of linear index screening, model screening, or data dimensionality reduction. The linear index screening employs one of Pearson coefficient, R-squared, chi-square test, IV, or WOE; the model screening employs one of decision tree or regression; and the data dimensionality reduction employs one of principal component analysis or variable clustering.

[0014] Preferably, the machine learning model mentioned in step S3 includes one of the following: regression model, classification model, and clustering model.

[0015] The regression model includes one of the following: linear regression, random forest (RF), gradient boosting tree (GBM), and decision tree.

[0016] The classification model includes one of the following: Naive Bayes, Logistic Regression, Neural Network, and Support Vector Machine (SVM).

[0017] The clustering model includes one of the following: K-means clustering, k-modes clustering, and DBSCAN clustering.

[0018] In a first aspect, the present invention provides an online quality control system, comprising:

[0019] The real-time processing data acquisition unit acquires real-time flow information of different sections of the raw material polymer melt processing production line through sensor devices;

[0020] The first calculation unit converts real-time flow information into one or more rheological property curves at different times. Each rheological property curve includes one or more peaks and / or valleys, and the peaks and valleys characterize the physicochemical properties of the raw material polymer.

[0021] The second calculation unit uses feature engineering methods to filter the rheological property curves at different times to obtain relevant rheological property curves.

[0022] The product quality prediction unit inputs the relevant rheological characteristic curves into the polymer property calculation device to obtain the product quality prediction value for the current time period.

[0023] The process parameter control unit adaptively adjusts the process parameters of the current raw material polymer processing line based on the predicted product quality values.

[0024] Thirdly, the present invention provides a computing device, including a memory and a processor, wherein the memory stores executable code, and the processor executes the executable code to implement the method described thereon.

[0025] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the method described thereon.

[0026] This invention employs online rheological detection technology, combined with AI algorithms, which improves the efficiency and accuracy of online quality control technology. Attached Figure Description

[0027] Figure 1 It is a rheological viscosity curve, which shows the polymer's processing properties corresponding to viscosity in different shear rate ranges.

[0028] Figure 2 This is a block diagram illustrating the method of the present invention in the deployment phase.

[0029] Figure 3 This is a block diagram of a polymer property calculation device.

[0030] Figure 4 This is a flowchart illustrating an implementation scheme for carrying out the method of the present invention.

[0031] Figure 5 It is a flowchart showing the implementation scheme for training machine learning algorithms.

[0032] Figure 6 This graph depicts a comparison between melt flow predicted by a machine learning algorithm based on rheological data and actual melt flow measured in the laboratory. Dashed lines indicate a perfect match between the predicted and actual values. Solid lines represent deviations of + / - 20%.

[0033] Figure 7 This graph depicts a comparison between molecular weights predicted by machine learning algorithms based on rheological data and actual molecular weights measured in the laboratory. Dashed lines indicate a perfect match between predicted and actual values. Solid lines represent deviations of + / - 20%.

[0034] Figure 8 The flow channel morphology is shown in the polymer processing production line; (a) is a flow channel with a cylindrical cross-section; (b) is a flow channel with a rectangular cross-section; and (c) is a flow channel with a screw cross-section. Detailed Implementation

[0035] The present invention will be further analyzed below with reference to the accompanying drawings and specific embodiments.

[0036] Rheology is a crucial analytical technique for polymer materials because it describes the flow and deformation of substances. It allows for real-time assessment of the flow properties of polymer melts and solutions during production and provides accurate and well-resolved rheological curves containing both structural and chemical information. Rheological testing relies on the acquisition of physical signals such as stress and torque. Under the influence of an external flow field, polymer molecules undergo physical changes such as stretching and detangling, as well as chemical changes such as chain breakage, leading to changes in rheological properties such as viscosity.

[0037] Rheology is a well-known analytical tool for the characterization, identification, and quantification of polymers. It utilizes signals such as stress and torque during the flow of incoming polymers to obtain information about molecular chain motion and structure. Rheological curves contain information directly or indirectly related to various properties of the polymer sample. Rheological curves are typically displayed as a graph of rheological parameters versus strain rate / frequency, where the strain rate is determined by the flow field velocity and geometry. The strain rate is usually expressed as shear rate in reciprocating seconds (s). -1 Reported in units of 10 s. There are no particular limitations on the range of shear rates for the obtained rheological profiles, but useful ranges include shear rates in the viscosity fingerprint region corresponding to the typical range of extrusion processes, typically around 10 s. -1 approximately 1,000s -1 The typical range of shear rates in the viscosity fingerprint region of blow molding processes is approximately 100 s. -1 approximately 10,000s -1 The shear rate in the viscosity fingerprint region, corresponding to the typical range of the injection molding process, is typically around 10 s. -1 approximately 10,000s -1 The shear rate in the viscosity fingerprint region, corresponding to the typical range of the spinning process, is typically around 1,000 s. -1 approximately 100,000s -1The shear rate in the viscosity fingerprint region, corresponding to a typical range for the coating process, is typically around 10,000 s. -1 From approximately 1,000,000s -1 By fitting the rheological curves, important rheological parameters closely related to the polymer molecular structure can be obtained, including but not limited to zero-shear viscosity, CYA index, and non-Newtonian index.

[0038] Based on considerations of machine resolution and capacity, acquisition time, data analysis time, information density, and other factors understood by those skilled in the art, the frequency interval of the acquired data can be readily determined. Similarly, those skilled in the art can readily determine the average amount of signal used based on the efficiency and limitations of the machine and method.

[0039] Now for reference Figure 2 Instruments used to collect and process online rheological data typically include stress sensors, temperature sensors, torque sensors, probes that communicate with the polymer production process, and machine learning algorithms.

[0040] The online rheological data detection device includes transmitting signals such as stress through a probe and collecting them in a data storage device, and processing the raw mechanical signals into rheological parameters such as viscosity through a computing device, which is then processed by a polymer property computing device, as further described below.

[0041] Signals can be collected from the sensor probe using any convenient means known in the art, such as conventional beam-manipulating optics or fiber optic cables. For online process measurements, data transmission via fiber optics is particularly convenient. A specific advantage of online rheological detection methods is that commonly used sensors can easily acquire real-time flow data such as stress and torque.

[0042] Dispersed stress, torque, and other data are imaged onto the detector in real time. Considering various factors (e.g., resolution, sensitivity to an appropriate shear rate range, and response time), those skilled in the art can easily select a sensor. The sensor response is transmitted to a data subsystem, which generates a set of data points constituting a rheological curve such as viscosity and shear rate (x, y).

[0043] As mentioned above, signals can be transmitted to the data storage device using any convenient means, such as conventional optical elements or fiber optic cables. The sensor may or may not have an immersion element. Sensor probes with immersion elements can be immersed in liquid samples, such as molten polymers in an extruder or polymers dissolved in a solvent.

[0044] Sensors are typically placed at accessible points in the polymer manufacturing process. For example, this accessible point could be near where engineers sample the polymer for offline quality control, such as immediately after the pellet mill and / or after the purge chamber. The sensor probe can be located in a pipe or small container at the accessible point.

[0045] polymer

[0046] The polymer may be a homopolymer, copolymer, or polymer blend. As understood by those skilled in the art, the term "polymer" refers to a polymeric compound prepared by polymerizing monomers of the same or different kinds. Thus, the general term "polymer" includes the terms "homogeneous polymer" and "copolymer," where "homogeneous polymer" refers to a polymer prepared from only one type of monomer, and "copolymer" refers to a polymer prepared from two or more different monomers. As used herein, the term "blend" or "polymer blend" generally refers to a physical mixture of two or more polymers that are not chemically bonded. Such blends may be miscible and may be phase-separated or phase-separated. Polymer blends may contain one or more domain structures resulting from the morphology of the polymer. Domain structures can be determined by X-ray diffraction, transmission electron microscopy, scanning transmission electron microscopy, scanning electron microscopy, and atomic force microscopy, or other methods known in the art.

[0047] The polymer may be a polyolefin. Exemplary polyolefins include, but are not limited to, polyethylene, polypropylene, polyisobutylene, and their homopolymers and copolymers. In some embodiments, the polyolefin is a polypropylene homopolymer or a polypropylene-based copolymer, such as an impact copolymer or a random polymer.

[0048] Polypropylene-based copolymers can consist of linear polymer chains and / or branched polymer chains. Exemplary polypropylene-based copolymers include alternating copolymers, periodic copolymers, block copolymers, random copolymers, or impact copolymers. In some embodiments, the polypropylene-based copolymer is a random copolymer or impact copolymer that optionally contains long-chain branches. As used herein, the term "random copolymer" refers to a copolymer in which different types of monomer units are statistically distributed in the polymer molecule. The polypropylene-based copolymer may be a polypropylene-polyethylene random copolymer, wherein the content of ethylene monomer units is typically up to 7% by weight based on the total weight of the copolymer. The term "impact copolymer" refers to a multiphase polyolefin copolymer in which one polyolefin is a continuous phase (i.e., the matrix) and an elastomeric phase is uniformly dispersed therein. Impact copolymers include, for example, multiphase polypropylene copolymers in which a polypropylene homopolymer is a continuous phase and an elastomeric phase such as ethylene propylene rubber (EPR) is uniformly distributed therein. Impact copolymers are produced by in-vessel processes rather than physical blending.

[0049] In some embodiments, the polyolefin is a polyethylene homopolymer, such as very low density polyethylene, low density polyethylene, linear low density polyethylene, medium density polyethylene, high density polyethylene, and ultra-high molecular weight polyethylene; or a polypropylene-based copolymer, such as ethylene vinyl acetate copolymer.

[0050] The polymer may be a polyester. Exemplary polyesters include, but are not limited to, polyethylene terephthalate (PET), polybutylene terephthalate (PBT), and polyarylates (PAR). In some embodiments, the polyester is a polyester elastomer (TPEE), which is generally polymerized from dimethyl terephthalate, 1,4-butanediol, and polybutanol, and whose chain segments include hard segments and soft segments, and is a thermoplastic elastomer.

[0051] The polymer may be a polyamide. Exemplary polyamides may be at least one of nylon 6, nylon 66, nylon 11, nylon 12, nylon 610, nylon 612, nylon 1010, nylon 46, nylon 7, nylon 9, nylon 13, nylon 6I, and nylon 9T.

[0052] Polymer material processing technology

[0053] The polymer processing technology can be an extrusion process, with a typical range of shear rates in the viscosity fingerprint region of about 10 s⁻¹ to about 1,000 s⁻¹.

[0054] The polymer processing technology can be a blow molding process, with a typical range of shear rates in the viscosity fingerprint region of about 100 s⁻¹ to about 10,000 s⁻¹.

[0055] The polymer processing technology can be an injection molding process, with a typical range of shear rates in the viscosity fingerprint region of about 10 s⁻¹ to about 10,000 s⁻¹.

[0056] The polymer processing technology can be a spinning process, with a typical range of shear rates in the viscosity fingerprint region of about 1,000 s⁻¹ to about 100,000 s⁻¹.

[0057] The polymer processing technology may be a coating process, with a typical range of shear rates in the viscosity fingerprint region typically ranging from about 10,000 s⁻¹ to about 1,000,000 s⁻¹.

[0058] The polymer processing technology can be a unidirectional or bidirectional film stretching process, and the shear rate in the viscosity fingerprint region is typically from about 100 s⁻¹ to about 100,000 s⁻¹.

[0059] The polymer processing technology can be a blown film process, with a typical range of shear rates in the viscosity fingerprint region of about 1,000 s⁻¹ to about 1,000,000 s⁻¹.

[0060] Polymer properties or characteristics

[0061] Polymer properties can be any polymer-related property that can be rheologically analyzed and measured by a person skilled in the art, including molecular weight, molecular weight distribution, degree of branching, melt flow rate, mechanical properties (e.g., tensile or compressive properties), and combinations thereof. As used herein, the term "molecular weight" can refer to index-average molecular weight, weight-average molecular weight, or Z-average molecular weight.

[0062] Mechanical properties can be any polymer-related mechanical properties known to those skilled in the art, including Young's modulus, yield tensile strength, strain hardening modulus, traction yield elongation, and flexural modulus at 1% secant.

[0063] Polymer characteristics may be the amount of one or more additives (e.g., talc, kaolin, glass fiber) in the presence and the amount of one or more comonomers in the presence. Other polymer characteristics known to those skilled in the art can also be determined by the disclosed methods.

[0064] like Figure 3 As shown, a polymer-property computing device includes a processor or central processing unit (CPU), memory, optional configurable hardware logic, and communication systems connected together via a bus device. However, a polymer-property computing device may include other types and numbers of components in other configurations. In this example, the bus is a PCI Express bus, but other bus types and links may also be used.

[0065] The processor within the polymer property computing device can execute one or more computer-executable instructions stored in memory for the methods illustrated and described herein, but the processor can also execute other types and numbers of instructions and perform other types and numbers of operations. The processor may include one or more CPUs or a general-purpose processor having one or more processing cores, such as... Processor, but other types of processors can also be used (e.g., by...). Those processors produced).

[0066] The memory within a polymer-property computing device may include one or more tangible storage media known to those skilled in the art, such as RAM, ROM, flash memory, CD-ROM, floppy disk, hard disk drive, solid-state storage, DVD, or any other memory type or device, including combinations thereof. The memory may store one or more non-transitory computer-readable instructions as illustrated and described herein, which may be executed by a processor. Figure 4 and Figure 5 The exemplary flowcharts shown represent exemplary steps or operations of the technology and may be presented or expressed as one or more non-transitory computer-readable or machine-readable instructions stored in memory, which may be executed by a processor and / or implemented by configuration logic in optional configurable logic.

[0067] Therefore, the memory of the polymer property computing device can store one or more applications that may include computer-executable instructions, which, when executed by the polymer property computing device, cause the polymer property computing device to perform actions, such as sending, receiving, or otherwise processing messages, and cause the polymer property computing device to perform reference operations. Figure 4 and Figure 5 Other actions described and illustrated. The application can be implemented as a module or component of another application, as an operating system extension, module, plugin, etc., and can run in a cloud computing environment and execute in a virtual machine or virtual server managed within that environment. Furthermore, the application (including the polymer computing device itself) can reside in a virtual server running in a cloud computing environment, rather than being tied to one or more specific physical network computing devices. Additionally, the application can run in one or more virtual machines (VMs) running on the polymer computing device. In at least one of the embodiments, the virtual machine running on the polymer computing device can be managed or supervised by a hypervisor.

[0068] Optional configurable hardware logic devices in the polymer property computing device may include dedicated hardware configured to implement one or more steps of the technique as illustrated and described herein. By way of example only, optional configurable logic hardware devices may include one or more of the following: field-programmable gate arrays (“FPGAs”), field-programmable logic devices (“FPLDs”), application-specific integrated circuits (“ASICs”), and / or programmable logic units (“PLUs”).

[0069] The communication system in the polymer property computing device is used for operatively connecting and communicating between the polymer property computing device and the real-time rheological detection device. Both the polymer property computing device and the rheological detection device are connected together via a communication network (e.g., one or more local area networks (LANs) and / or wide area networks (WANs)). However, other types and numbers of communication networks or systems with other types and numbers of connections and configurations with other devices and components may also be used. As examples only, communication networks such as LANs and WANs can use TCP / IP over Ethernet and industry-standard protocols including NFS, CIFS, SOAP, XML, DAP, and SNMP, although other types and numbers of communication networks may also be used.

[0070] Although the polymer property computing device is shown as a single device in this example, in other examples, it may include multiple devices or blades, each with one or more processors, each processor having one or more processing cores that implement one or more steps of the technology. In these examples, one or more devices may have dedicated communication interfaces or memory. One or more devices may utilize the memory, communication interfaces, or other hardware or software components of one or more other communication-connected devices. In other examples, one or more devices included in the polymer property computing device may be standalone devices or integrated with one or more other devices or applications. Furthermore, in these examples, one or more devices of the polymer property computing device may be on the same or different communication networks, for example, including one or more public networks, private networks, or cloud networks.

[0071] Each system can be conveniently implemented using one or more general-purpose computer systems, microprocessors, digital signal processors, and microcontrollers programmed according to the teachings described and shown herein and as will be understood by one of ordinary skill in the art.

[0072] For example, a polymer property computing device can be configured to operate as a virtual instance on the same physical machine. Furthermore, two or more computing systems or devices can replace any single system or device. Therefore, the principles and advantages of distributed processing, such as redundancy and replication, can be implemented as needed to increase the robustness and performance of the device and system. The method can also be implemented on a computer system and extended over any suitable network using any suitable interface mechanism and traffic technology, including, for example, any suitable form of communication traffic (e.g., voice and modem), wireless traffic media, wireless traffic networks, cellular traffic networks, G3 traffic networks, the Public Switched Telephone Network (PSTN), packet data networks (PDN), the Internet, intranets, and combinations thereof.

[0073] The method may also be presented on a non-transitory computer-readable medium storing instructions for one or more aspects of the technology as described and illustrated herein, which, when executed by a processor (or configurable hardware), cause the processor to perform steps required to implement the method as described and illustrated herein.

[0074] Therefore, the present invention also relates to a non-transitory computer-readable medium storing instructions for determining the quality of a polymer based on its rheological properties, the instructions including machine-executable code that, when executed by at least one processor, causes the processor to perform the following steps: (i) acquiring real-time information such as flow rate, temperature, and stress of the polymer from a production line via temperature and stress sensors; (ii) converting the real-time information into a rheological curve using the polymer rheological property calculation device, including plateaus, multiple peaks, and valleys in a chemical fingerprint and structural fingerprint corresponding to one or more polymer properties or characteristics; (iii) calculating the one or more polymer properties or characteristics by comparing the acquired chemical fingerprint and structural fingerprint with stored data of polymer properties and characteristics, wherein the calculation step is performed by executing one or more trained machine learning algorithms; and (iv) determining whether the quality of the polymer sample meets a predetermined quality threshold based on one or more calculated polymer properties or characteristics.

[0075] The present invention also relates to a polymer property calculation device including a processor and a memory, wherein the memory is connected to the processor, the processor being configured to execute program instructions stored in the memory, including: (i) real-time information such as flow rate, temperature, and stress obtained from a polymer production line; (ii) converting the real-time information into rheological curves by the polymer rheological property calculation device, including plateaus, multiple peaks, and valleys in a chemical fingerprint and structural fingerprint corresponding to one or more polymer properties or features; (iii) calculating the one or more polymer properties or features by comparing the obtained chemical fingerprint and structural fingerprint with stored polymer property and feature data, wherein the calculation step is implemented by executing one or more trained machine learning algorithms; and (iv) determining whether the mass of the polymer sample meets a predetermined mass threshold based on one or more calculated polymer properties or features.

[0076] Other aspects, advantages, and features of the invention are set forth in this specification and will become apparent in part to those skilled in the art upon study of the following, or may be learned by practicing the invention. The invention disclosed in this application is not limited to any particular set or combination of aspects, advantages, and features. It is contemplated that various combinations of the aspects, advantages, and features constitute the invention disclosed in this application.

[0077] Example

[0078] A method for determining product quality based on the rheological properties of a raw material polymer (polyethylene) includes the following steps:

[0079] S1: Real-time flow information of different sections of the raw material polymer (polyethylene) melt processing line is acquired through a sensor device; wherein the flow information includes flow velocity and flow velocity gradient distribution, temperature and temperature gradient distribution, stress and stress gradient distribution.

[0080] S2: The collected flow information is converted into shear rate curves, shear stress curves, and viscosity curves at different times. Each curve includes one or more peaks and / or valleys, and the peaks and valleys characterize the physicochemical properties of the raw material polymer (polyethylene).

[0081] If the cross-sectional shape of the flow channel is cylindrical, the shear rate can be obtained according to formulas (1)-(2). and shear stress τ w ;

[0082]

[0083]

[0084] Where L is the flow channel length, R is the flow channel cross-sectional radius, Q is the volumetric flow rate, and ΔP is the pressure drop.

[0085] If the cross-sectional shape of the flow channel is rectangular, the shear rate can be obtained according to formulas (3)-(4). and shear stress τ w ;

[0086]

[0087]

[0088] Where L is the flow channel length, W is the flow channel cross-sectional width, and H is the flow channel cross-sectional height;

[0089] If the flow channel is a screw, the shear rate can be obtained according to formulas (5)-(6). and shear stress τ w ;

[0090]

[0091]

[0092] Where L is the flow channel length, N is the screw speed, D is the barrel diameter, and h is the clearance between the screw and the barrel;

[0093] Table 2: Based on the flow channel geometry in polymer processing lines ( Figure 8 The system calculates rheological data such as shear rate and shear stress by analyzing online acquired signals such as flow velocity Q, stress, and torque.

[0094]

[0095] The apparent viscosity of the melt depends on the shear rate. and shear stress τ w Calculation yields:

[0096]

[0097] in Indicates shear rate The apparent viscosity is below;

[0098] Based on the shear rate, shear stress, and apparent viscosity, the zero-shear viscosity, relaxation time, structural index, and non-Newtonian index are obtained by fitting the Carreau-Yasuda model using equation (8):

[0099]

[0100] According to the Carreau-Yasuda model, the low shear rate region provides information about the polymer's molecular weight, the intermediate shear rate region corresponds to information such as the polymer's molecular weight distribution and degree of branching, while the high shear rate region provides information about the polymer's non-Newtonian index.

[0101] By fitting the zero-shear viscosity, the viscosity in the second Newtonian region, the relaxation time, the structural index CYA, and the non-Newtonian index, rheological curves covering the first Newtonian plateau region, the shear-thinning region, and the second Newtonian plateau region can be obtained. See [link to relevant documentation]. Figure 1 Furthermore, these viscosity data can be correlated with structural parameters. For example, in the first Newtonian plateau region, where the polymer chains have not undergone detangling or orientation changes, the viscosity is called the zero-shear viscosity η0. According to scaling theory (see Michael Rubinstein and Ralph Colby, Polymer Physics, 2003, pp. 366-367, for a discussion of entangled polymer dynamics), the zero-shear viscosity η0 is related to the molecular weight M of the entangled polymer. w The relationship is This allows for online monitoring of polymer molecular weight changes at different stages of the production line. As the shear rate increases, the viscosity curve enters a downward transition region. The rate of viscosity decrease is characterized by parameter a (CYA) in the Carreau-Yasuda model. Empirically, the magnitude of CYA can be correlated with molecular weight distribution and branching degree (see Thomas Mezger, The Rheology Handbook, p. 350, Polymers: Results, and determination of the zero-shear viscosity). Based on this, changes in polymer molecular weight distribution and branching degree at different stages of the production line can be monitored. In the high shear rate region, polymer molecular chains may undergo physical changes such as detangling and orientation, as well as chemical changes such as chain breakage. Therefore, monitoring the viscosity in this region can reveal changes in the polymer supramolecular structure.

[0102] S3: The shear rate curves and shear stress curves at different times above are screened using characteristic engineering methods to obtain the relevant rheological property curves:

[0103] The feature engineering methods include: (1) linear index screening: Pearson coefficient, R-squared, chi-square test, IV, WOE; (2) model screening: decision tree, regression, etc.; (3) data dimensionality reduction: principal component analysis and variable clustering;

[0104] Although rheological properties, exemplified by viscosity profiles, are rich in structural and chemical information, it is challenging to create parametric models that reliably correlate real-time monitoring data from various stages of the production line with the properties / characteristics of the polymer. The main difficulty lies in the sheer volume and dimensionality of the data. Each point collected by the monitoring equipment represents the polymer's flowability under a specific flow field; therefore, each point in a viscosity dataset possesses a unique attribute, and typical online datasets contain 10,000 to 30,000 data points. Data from specific time periods is often manually selected based on expertise in the material's flow behavior, while the remaining information is discarded.

[0105] Therefore, this manual assessment is not suitable for online product characterization. Furthermore, although rheological data can be correlated with polymer properties, this information is not readily apparent simply by looking at viscosity curves and making visual observations.

[0106] In contrast, the method of this invention uses artificial intelligence, and more specifically, machine learning techniques, to develop models capable of predicting relevant polymer properties / characteristics from rheological properties. Furthermore, by employing an online non-destructive fingerprinting method and artificial intelligence, real-time assessment of product specifications can be obtained, thereby reducing the time and cost associated with quality control using traditional laboratory equipment.

[0107] S4: Input the relevant rheological characteristic curve into the polymer property calculation device to obtain the product quality prediction value for the current time period.

[0108] The polymer property calculation device executes one or more trained and validated machine learning algorithms;

[0109] The machine learning models include: regression models, classification models, and clustering models;

[0110] The regression models include: linear regression, random forest (RF) or gradient boosting tree (GBM), and decision tree;

[0111] The classification models include: Naive Bayes, Logistic Regression, Neural Network, and Support Vector Machine (SVM);

[0112] The clustering models include: K-means clustering, k-modes clustering, and DBSCAN clustering;

[0113] In training the machine learning algorithm, rheological curves of known samples are acquired, and polymeric properties / characteristics of the known samples are measured in the laboratory. Polymeric properties / characteristics can be measured using methods known to those skilled in the art. For example, the molecular weight and molecular distribution of the sample can be measured using gel permeation chromatography (GPC). Those skilled in the art can determine an appropriate number of calibration samples based on the model's performance and incremental changes in performance with additional calibration data. The measured polymeric properties / characteristics include those that the model will calculate. The measured polymeric properties / characteristics and the acquired rheological properties are then input into a polymeric property calculation device.

[0114] To obtain the desired polymer properties, a dataset including the measured polymer properties / characteristics and the corresponding rheological data of the samples is entered into a tabular, machine-readable database that can be accessed via an application programming interface (API) or a graphical user interface.

[0115] Since most machine learning methods typically perform poorly if the input variables do not have a zero mean and unit variance, they are usually scaled to obtain a zero mean and a standard deviation equal to 1. In other words, transforming data so that it can be read by machine learning algorithms is a process called feature engineering or variable engineering. Many possible variables are generated during the data collection and feature / variable engineering phases, even if some of them are irrelevant. For example, new variables can be generated by applying basic arithmetic operations between the original features. The feature / variable engineering process may also include: removing instances with missing values ​​or replacing missing values ​​with the mean of a given variable, and identifying and removing outliers. Once the appropriate feature engineering operations are determined, a "data transformation pipeline" can be written to apply the same variable operations to the data presented to the machine learning algorithm. This data can be obtained from known or unknown samples with measured polymer properties / characteristics.

[0116] The next step in the training phase involves feature selection and dimensionality reduction. Specific input features can be highly correlated (containing information not found in any other feature), correlated, weakly correlated (containing some information included in other features), or uncorrelated. During feature selection, a subset of highly correlated features is used in model building. In this step, one or more rheological properties are associated with the measured properties / features to form a subset of the polymer property and feature data. For example, zero-shear viscosity can be used to characterize the molecular weight of a polymer (see Michael Rubinstein and Ralph Colby, Polymer Physics, 2003, pp. 366-367, foradiscussion of entangled polymer dynamics); the CYA index can be used to assess the molecular weight distribution and branching degree of a polymer (see Thomas Mezger, The Rheology Handbook, p. 350, Polymers: Results, and determination of the zero-shear viscosity); and the non-Newtonian index in the shear-thinning region can be used to assess the extrusion swelling effect of the polymer melt (see Thomas Mezger, The Rheology Handbook, p. 350, Polymers: Results, and determination of the zero-shear viscosity).

[0117] Models can be trained using methods known in the art. For example, the collected data is typically divided into two groups: a training dataset (typically about 80% of the collected data) and a detection dataset (typically about 20% of the collected data). The training dataset is used to develop the model, where machine learning algorithms are executed to analyze the training dataset and produce an inference function. Since the selected polymer properties / features used for the training dataset samples are both calculated and measured, the effectiveness of the model can be evaluated by comparing the calculated and measured values. Optimization algorithms are typically used to minimize empirical or structural risk. This is done by tuning the parameters of the inference function (called hyperparameters) to minimize the error between the known output and the model's prediction.

[0118] The trained model (i.e., the machine learning algorithm) is then tasked with calculating polymer properties / features based on the detected dataset. This is done to evaluate the model's inductive ability.

[0119] The trained model can then be applied to production line signals from unknown samples to calculate the desired polymer properties or characteristics.

[0120] Models can be trained to predict one or more polymer properties / characteristics. In some implementations, a separate model is developed for each polymer property / characteristic. In other implementations, multiple polymer properties / characteristics can be determined simultaneously using a model.

[0121] The trained model can be further enhanced through ensemble techniques, where the goal is to combine the predictions of several basic estimators constructed using a given learning algorithm to improve generality / robustness relative to a single estimator. Some ensemble methods include bagging, K-means clustering, K-modes clustering, Gaussian mixture models, DBSCAN clustering, Naive Bayes, decision trees, random forests, neural networks, support vector machines, AdaBoost, gradient boosting, and voting classifiers.

[0122] In some embodiments, the method further includes, prior to the calculation step, classifying the rheological properties in the obtained chemical and structural fingerprints into relevant or irrelevant polymer features using a polymer property calculation device. As an illustrative example, the CYA index can be classified as a relevant polymer feature because its value is related to the degree of branching of the polyolefin and can serve as a basis for polyolefin classification. Another illustrative example: different processing applications correspond to different shear rate regions, and the viscosity of the polymer in the corresponding shear rate region determines its processing performance. Therefore, the viscosity curves of the shear rate regions corresponding to different processing techniques can be used to identify the application scenarios of the polymer.

[0123] In some embodiments, the method further includes reporting one or more calculated polymer properties or characteristics via a polymer property calculation device after the determination step. For example, one or more calculated polymer properties or characteristics may be displayed on the screen of a computer (e.g., a desktop computer, laptop computer, tablet computer, mobile phone, and smartwatch).

[0124] In some implementations, the method further includes using a polymer property calculation device to at least repeatedly perform and adjust steps to refine one or more trained machine learning algorithms by using data from the process. The machine learning algorithms are then continuously refined as product cycles change, thereby improving their predictive capabilities over time.

[0125] The polymer property calculation device can determine whether the quality of a polymer sample meets a predetermined quality threshold based on one or more calculated polymer properties or characteristics, which typically includes acceptable batch-to-batch variation.

[0126] When one or more of the calculated polymer properties or characteristics fall within acceptable ranges, the polymer sample can be considered to meet the quality threshold, and the polymer production process can continue without further adjustments.

[0127] When one or more calculated polymer properties or characteristics fall outside the acceptable range, the polymer sample does not meet the predetermined quality threshold. Therefore, parameters used in the polymer production process can be adjusted using a polymer property calculation device to obtain a polymer with the desired properties or characteristics. These parameters include, but are not limited to, the amount / concentration of reactants (e.g., propylene, ethylene, hydrogen), the amount / concentration of additives, the amount / concentration of polymerization catalysts, temperature, and pressure.

[0128] S5: By comparing the predicted product quality values, if they fall within the threshold range, the current raw material polymer processing line flow information is considered to meet the requirements; otherwise, it does not meet the requirements and adjustments need to be made to the current raw material polymer processing line flow information.

[0129] Tests conducted on polyethylene showed that the properties assessed using rheological techniques and machine learning, as disclosed in this invention, correlated with historical measurements of laboratory data collected using the methods shown in Table 2 below. Figures 6 to 7 The figure provided shows a comparison between the predicted value of a given property determined by a machine learning algorithm based on rheological data and the actual value of that property measured in the laboratory. Specifically, Figure 6 The figure shows a comparison between melt flow predicted by a machine learning algorithm based on real-time rheological data and actual melt flow measured in the laboratory. Figure 7 The graphs depict a comparison between molecular weights predicted by machine learning algorithms based on real-time rheological data and actual molecular weights measured in the laboratory. In each graph, the dashed line in the middle represents a perfect prediction, while the black line represents a deviation of plus or minus 20% from the laboratory value.

[0130] Table 2

[0131]

[0132] from Figures 6 to 7 As can be seen, the machine learning predictions for all the properties shown (including melt flow and molecular weight) fall within the range, indicating that the predicted properties are at most 20% higher or 20% lower than the actual measurements.

[0133] based on Figures 6 to 7 Based on the results shown, those skilled in the art can conclude that the disclosed method using machine learning algorithms based on real-time rheological data is capable of assessing at least the accuracy shown for several key quality control properties of polymers, including molecular properties (melt flow) and chemical properties (molecular weight). Although only some properties are shown, it is expected that the disclosed method using machine learning algorithms based on real-time rheological data will be able to predict other similar polymer properties and can be used for other polyolefin or polymer compositions.

Claims

1. An online quality control method based on the rheological properties of raw material polymers, characterized in that... The method includes: S1: Real-time flow information of different sections of the raw material polymer melt processing line is acquired through sensor devices; wherein the flow information includes flow velocity and flow velocity gradient distribution, temperature and temperature gradient distribution, stress and stress gradient distribution; S2: The collected real-time flow information is converted into one or more rheological property curves at different times. Each rheological property curve includes one or more peaks and / or valleys, and the peaks and valleys characterize the physicochemical properties of the raw material polymer. The rheological properties include apparent viscosity, zero-shear viscosity, shear rate, shear stress, relaxation time, thixotropic ring, structural index, non-Newtonian index, critical strain in the linear viscoelastic region, Deborah number, complex modulus, storage modulus, loss modulus, loss factor, and loss tangent. S3: The rheological characteristic curves at different times are filtered using the feature engineering method to obtain the correlated rheological characteristic curves; the feature engineering method adopts one of linear index filtering, model filtering, and data dimensionality reduction. S4: Input the correlation rheological property curve into the polymer property calculation device to obtain the product quality prediction value within the current time period; wherein the polymer property calculation device executes one or more trained and validated machine learning algorithms; S5: Determine whether the predicted product quality value falls within the threshold range. If so, the current raw material polymer processing line flow information is considered to meet the requirements. Otherwise, it does not meet the requirements and the process parameters of the current raw material polymer processing line need to be adjusted.

2. The method according to claim 1, characterized in that... The raw material polymer is polyethylene.

3. The method according to claim 2, characterized in that... The rheological properties mentioned in step S2 are shear rate, shear stress, and apparent viscosity; If the cross-sectional shape of the flow channel is cylindrical, the shear rate can be obtained according to formulas (1)-(2). and shear stress ; Equation (1) Equation (2) Where L is the channel length, R is the cross-sectional radius of the channel, and Q is the volumetric flow rate. To reduce pressure; If the cross-sectional shape of the flow channel is rectangular, the shear rate can be obtained according to formulas (3)-(4). and shear stress ; Equation (3) Equation (4) Where L is the flow channel length, W is the flow channel cross-sectional width, and H is the flow channel cross-sectional height; If the flow channel is a screw, the shear rate is obtained according to formulas (5)-(6). and shear stress ; Equation (5) Equation (6) Where L is the flow channel length, N is the screw speed, and D is the barrel diameter. The clearance between the screw and the barrel; The apparent viscosity of the melt depends on the shear rate. and shear stress Calculation yields: Equation (7) in Indicates shear rate The apparent viscosity is below; Based on the shear rate, shear stress, and apparent viscosity, the zero-shear viscosity, relaxation time, structural index, and non-Newtonian index are obtained by fitting the Carreau-Yasuda model using equation (8): Equation (8) in Zero tangential viscosity Let be the viscosity in the second Newtonian zone, τ be the relaxation time, a be the structural index CYA, and n be the non-Newtonian index.

4. The method according to claim 1, characterized in that... The linear index selection in step S3 uses one of Pearson coefficient, R-squared, chi-square test, IV, or WOE; the model selection uses one of decision tree or regression; and the data dimensionality reduction uses one of principal component analysis or variable clustering.

5. The method according to claim 1, characterized in that... The machine learning algorithm mentioned in step S4 includes one of the following: regression model, classification model, and clustering model.

6. The method according to claim 5, characterized in that... The regression model mentioned in step S3 includes one of the following: linear regression, random forest (RF), gradient boosting tree (GBM), and decision tree; The classification model includes one of the following: Naive Bayes, Logistic Regression, Neural Network, and Support Vector Machine (SVM). The clustering model includes one of the following: K-means clustering, k-modes clustering, and DBSCAN clustering.

7. An online quality control system for implementing the method of any one of claims 1-6, characterized in that... include: The real-time processing data acquisition unit acquires real-time flow information of different sections of the raw material polymer melt processing production line through sensor devices; The first calculation unit converts real-time flow information into one or more rheological property curves at different times. Each rheological property curve includes one or more peaks and / or valleys, and the peaks and valleys characterize the physicochemical properties of the raw material polymer. The second calculation unit uses feature engineering methods to filter the rheological property curves at different times to obtain relevant rheological property curves: The product quality prediction unit inputs the relevant rheological characteristic curves into the polymer property calculation device to obtain the product quality prediction value for the current time period. The process parameter control unit adaptively adjusts the process parameters of the current raw material polymer processing line based on the predicted product quality values.

8. A computing device comprising a memory and a processor, wherein the memory stores executable code, and the processor, when executing the executable code, implements the method of any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the method of any one of claims 1-6.

Images