A cloud platform-based plastic new material research and development data processing method

Abstract

The application relates to the technical field of data processing, and discloses a plastic new material research and development data processing method based on a cloud platform, which comprises the following steps: collecting multi-source heterogeneous plastic research and development original data through an edge node, and pre-processing the data; classifying the processed data and performing hierarchical storage management and optimization; analyzing a research and development calculation task submitted by a user, and dynamically scheduling and distributing cloud computing resources according to task attributes to execute the task; mining and analyzing the stored data to construct a knowledge graph, train a performance prediction model and generate a formula recommendation; and visually displaying data processing results, an analysis process and a task state, and providing permission-based data sharing and collaborative research and development support. The application realizes intelligent management of the whole process of plastic research and development data, can significantly accelerate the research and development cycle, and reduce innovation cost.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a data processing method for the research and development of new plastic materials based on a cloud platform. Background Technology

[0002] The research and development of new plastic materials is a key link in promoting the upgrading of the polymer industry. This process encompasses multiple complex stages, including molecular design, synthesis process optimization, performance testing, and characterization. This process generates massive amounts of multi-source, heterogeneous data, including molecular structure files, quantum chemical calculation descriptors, spectral data, thermal analysis curves, mechanical property indicators, and synthesis process parameters.

[0003] Currently, the industry's R&D data management methods are generally traditional and fragmented. Experimental data is mostly recorded manually by researchers or stored separately on different devices and computers, lacking a unified platform for effective integration. Computationally intensive tasks are limited by the computing power of local computers, resulting in long computation cycles and severely slowing down the pace of R&D iteration. In addition, the inherent patterns between "components-structure-process-performance" contained in historical data are difficult to effectively uncover, R&D decisions heavily rely on expert experience, and trial-and-error costs are high. Data sharing and collaboration between teams are also fraught with difficulties due to the lack of a secure and controllable environment.

[0004] While existing general-purpose cloud computing platforms offer powerful basic computing and storage capabilities, they lack specific optimization for the data characteristics and workflows in the field of new plastic materials research and development. They cannot achieve one-stop processing of automated data collection, lifecycle management, domain-specific intelligent analysis, and secure collaboration throughout the entire research and development process.

[0005] Therefore, this invention proposes a data processing method for the research and development of new plastic materials based on a cloud platform. Summary of the Invention

[0006] The purpose of this invention is to propose a cloud-based data processing method for the research and development of new plastic materials to address the problems of data silos, computing power bottlenecks, insufficient knowledge mining, and low collaborative efficiency in existing technologies.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a data processing method for the research and development of new plastic materials based on a cloud platform, comprising the following steps:

[0008] Step S1: Collect multi-source heterogeneous raw data of plastic research and development through edge nodes deployed on the R&D equipment side, and preprocess the data;

[0009] Step S2: Based on the real-time access frequency of the data, the project stage, and the business value, classify the processed data and perform hierarchical storage management and optimization.

[0010] Step S3: parse the R&D computing tasks submitted by the user, and dynamically schedule and allocate cloud computing resources to execute the tasks based on the task attributes;

[0011] Step S4: Machine learning algorithms are used to mine and analyze the stored data in order to build a knowledge graph, train a performance prediction model, and generate recipe recommendations.

[0012] Step S5 visualizes the data processing results, analysis process, and task status, and provides permission-based data sharing and collaborative R&D support.

[0013] Furthermore, step S1 also includes the following sub-steps:

[0014] S1-1 acquires molecular structure data and quantum chemical descriptors from molecular simulation sources, spectral data from analytical testing equipment, mechanical property data from performance testing equipment, and synthesis process parameters from process control systems via a data acquisition interface.

[0015] S1-2, The collected raw data is cleaned and formatted, including standardizing and encoding molecular structure data, performing baseline correction and characteristic peak extraction on spectral data, unifying units and validating the mechanical property data, and smoothing process parameters.

[0016] S1-3, add metadata identifiers to the preprocessed data, and transmit it in batches to the cloud data lake for persistent storage through a secure channel with identity authentication and transmission encryption. The metadata identifiers include unique material identifiers, data lineage tags, data quality identifiers, and timestamps.

[0017] Furthermore, step S2 also includes the following sub-steps:

[0018] S2-1, based on the real-time access frequency and business value of data, data is classified into hot data, warm data and cold data. Hot data is newly generated and accessed data in the current project, warm data is historical data of completed projects but accessed regularly, and cold data is archived data that has exceeded the legal retention period.

[0019] S2-2 stores hot data in a storage layer composed of solid-state drives, warm data in a standard storage layer composed of hard disk arrays, and cold data in an archive storage layer composed of tape libraries.

[0020] S2-3, implement dynamic copy maintenance for hot data in the storage layer, establish a correlation model between data access frequency and the number of copies, and dynamically increase or decrease data copies based on the access pressure monitored in real time;

[0021] S2-4 initiates a compression process for cold data in the archive storage layer, identifies and separates structured data from redundant raw data files, applies a lossless compression algorithm to the structured data, and applies a lossy compression algorithm to the redundant raw data files.

[0022] The information entropy-based compression algorithm performs the following steps: traversing the dataset to be compressed in units of fixed-size data blocks, calculating the information entropy value of each data block using the Shannon entropy formula; classifying the data blocks into multiple importance levels according to a preset entropy threshold range; dynamically assigning different compression algorithms and compression levels to data blocks of different levels, with the data block with the lowest importance being assigned the lossy compression algorithm with the highest compression ratio.

[0023] Furthermore, step S3 also includes the following sub-steps:

[0024] S3-1, Receive user task requests through a unified task submission interface. The task request includes task type, input parameters and resource requirement description. The task type includes computationally intensive molecular dynamics simulation, data-intensive structure-activity model training and memory-intensive multi-objective formulation optimization.

[0025] S3-2, parse the dependency descriptors in the task request and construct a directed acyclic graph representing the execution order and dependencies between tasks;

[0026] S3-3, Traverse the directed acyclic graph, identify all nodes without predecessor-dependent tasks, form an initial set of executable tasks, and prioritize their execution;

[0027] S3-4. For task branches in a directed acyclic graph that do not have direct or indirect dependencies, they are identified as independent task sets and split into multiple subtasks that can be executed in parallel on different computing nodes.

[0028] S3-5: Allocate computing resource units to the task based on the resource requirements declared in the task request and the real-time resource status of the system.

[0029] In step S3-5, allocating computing resource units to tasks includes: allocating resource units consisting of multiple CPU cores to computing-intensive tasks, allocating GPU accelerator card resource units to data-intensive tasks, and allocating memory resource units to memory-intensive tasks.

[0030] Furthermore, step S4 also includes the following sub-steps:

[0031] S4-1 performs named entity recognition on the text data of R&D documents and experimental reports stored on the platform to extract material, performance, and process entities;

[0032] S4-2, Extract relations from the extracted entities and identify semantic relationships between entities;

[0033] S4-3, the extracted entity and relation pairs are associated and mapped with the structured experimental data inside the platform through unique material identifiers to construct a structured material knowledge graph;

[0034] S4-4: Parse and calculate the geometric and electronic structure descriptors from the molecular structure file, combine them with the corresponding synthesis process parameters to form a feature vector, and use the material's measured performance indicators as supervised learning labels to train a machine learning regression model to generate a quantitative structure-activity relationship prediction model.

[0035] S4-5, based on the historical formula database, uses an unsupervised clustering algorithm to divide the formulas into different groups according to their components and performance characteristics. For the current R&D goals, baseline formulas are selected from the groups, and a recommendation algorithm is used to generate a recommended list of optimized proportions of the formula components.

[0036] The geometric and electronic structure descriptors include: the highest occupied molecular orbital energy level, the lowest unoccupied molecular orbital energy level, the energy level gap, the molecular dipole moment, the static polarizability, the molecular van der Waals surface area, the molecular oil-water partition coefficient, the topological polar surface area, and the molecular refractive index.

[0037] Furthermore, step S5 also includes the following sub-steps:

[0038] S5-1 collects project task status, computing resource utilization and model performance index data in real time from the platform database and monitoring system, and dynamically renders and displays them on a unified monitoring interface through configurable visualization components.

[0039] S5-2, responding to front-end user interaction commands, queries the corresponding process parameter dataset and material performance dataset from the storage system, calls the back-end drawing engine to generate process-performance relationship curves, and generates a molecular structure-activity relationship scatter plot based on molecular structure feature data and activity data. The molecular structure feature data are geometric and electronic structure descriptors obtained by parsing and calculating from molecular structure files.

[0040] S5-3 creates independent project spaces within the platform and configures role-based access control lists to assign data read, write, execute, and management permissions to project members, enabling version control, sharing, and collaborative operation of experimental data, analysis models, and report documents within the authorized scope.

[0041] In step S5-2, the process-performance relationship curve is a two-dimensional curve showing the correspondence between a single process parameter and a single performance index, or a three-dimensional surface plot or contour plot showing the complex relationship between multiple process parameters and performance indexes; the molecular structure-activity relationship scatter plot is a visualization result of projecting molecular structure feature data as feature vectors onto a two-dimensional plane after dimensionality reduction.

[0042] The beneficial effects of the technical solution provided by this invention include at least the following:

[0043] This invention, by constructing a cloud-edge collaborative data acquisition and processing flow, enables the automatic aggregation and standardization of multi-source heterogeneous data throughout the entire plastics R&D process, fundamentally solving the data silo problem and providing a high-quality data foundation for subsequent analysis.

[0044] This invention, through a tiered storage and optimization strategy based on data lifecycle and access frequency, can significantly reduce the storage cost of massive R&D data while ensuring fast access efficiency for high-value data, thus achieving efficient utilization of storage resources.

[0045] This invention, by parsing task attributes and dynamically scheduling and allocating cloud computing resources, can efficiently process intensive tasks, reducing computation time from several weeks to hours, thus greatly improving R&D efficiency.

[0046] This invention integrates machine learning algorithms for deep data mining and intelligent analysis, enabling it to automatically discover structure-property rules, predict material properties, and intelligently recommend formulations from historical data. This transforms experience-driven to data-driven approaches, effectively reducing trial-and-error costs and accelerating the innovation process.

[0047] This invention provides R&D teams with intuitive decision support and a secure and controllable collaborative R&D environment by offering multi-dimensional visualization and role-based collaborative management functions, thereby improving the team's overall collaboration efficiency and knowledge accumulation capabilities. Attached Figure Description

[0048] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a schematic diagram of the method flow provided in an embodiment of the present invention. Detailed Implementation

[0050] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a cloud-based data processing method for the research and development of new plastic materials proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0051] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0052] The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0053] The following description, in conjunction with the accompanying drawings, details a specific scheme for a cloud-based data processing method for the research and development of new plastic materials provided by this invention.

[0054] Please see Figure 1 The diagram illustrates a flowchart of a cloud-based data processing method for the research and development of new plastic materials, according to an embodiment of the present invention, comprising the following steps:

[0055] Step S1: Collect multi-source heterogeneous raw data of plastic research and development through edge nodes deployed on the R&D equipment side, and preprocess the data;

[0056] Step S1 further includes the following sub-steps:

[0057] S1-1 acquires molecular structure data and quantum chemical descriptors from molecular simulation sources, spectral data from analytical testing equipment, mechanical property data from performance testing equipment, and synthesis process parameters from process control systems via a data acquisition interface.

[0058] S1-2, The collected raw data is cleaned and formatted, including standardizing and encoding molecular structure data, performing baseline correction and characteristic peak extraction on spectral data, unifying units and validating the mechanical property data, and smoothing process parameters.

[0059] S1-3 adds metadata identifiers to the preprocessed data and transmits it in batches to the cloud data lake for persistent storage through a secure channel with authentication and encrypted transmission. The metadata identifiers include unique material identifiers, data lineage tags, data quality identifiers, and timestamps.

[0060] It should be noted that sub-step S1-1 is implemented through an edge computing node deployed on-site in the laboratory. This node is equipped with a variety of data acquisition interfaces and can connect to different types of R&D equipment and software systems.

[0061] Data is collected from molecular simulation sources: Here, "molecular simulation sources" mainly refers to high-performance computing clusters or workstations that run molecular dynamics and first-principles calculation simulation software. Edge nodes collect molecular structure data and quantum chemical descriptors by calling the application programming interfaces provided by these software or parsing their output standard files. Molecular structure data includes the three-dimensional coordinates of atoms and chemical bond information; quantum chemical descriptors are physicochemical parameters that describe the electronic structure of molecules, obtained through calculation.

[0062] Spectral data is collected from analytical testing equipment: "Analytical testing equipment" here includes Fourier transform infrared spectrometers and nuclear magnetic resonance spectrometers; edge nodes collect raw spectral data, i.e., curves showing the change of signal intensity with wavelength or frequency, through dedicated communication protocols or network ports provided by the equipment manufacturers.

[0063] Mechanical property data are collected from performance testing equipment: "performance testing equipment" here includes universal tensile testing machine, impact testing machine, and hardness tester; mechanical property data measured by edge node acquisition equipment includes tensile strength, elongation at break, impact strength, and Rockwell hardness.

[0064] Synthesis process parameters are collected from the process control system: The "process control system" here mainly refers to the programmable logic controller or distributed control system that controls the reactor, twin-screw extruder, synthesis or processing equipment; edge nodes collect synthesis process parameters in real time, including reaction temperature, pressure, stirring speed, and feed rate of each component material.

[0065] Sub-steps S1-2 clean and format the collected raw data to ensure data quality and standardize the format, laying the foundation for subsequent analysis; specific processing methods are used for different types of data.

[0066] Standardized encoding of molecular structure data: Converting complex three-dimensional molecular structure information into a standard, linear string representation. This encoding method facilitates computer storage, comparison, and searching of molecular structures, and is a key step in realizing the automatic processing of molecular information by computers.

[0067] Baseline correction and characteristic peak extraction of spectral data: Raw spectral data often contains instrument noise and baseline drift; firstly, baseline correction is performed to eliminate the influence of these non-target factors; then, the position and intensity information of characteristic peaks representing specific chemical bonds or functional groups in the spectral curve are identified and extracted through algorithms, transforming complex curve data into structured data points.

[0068] Unit unification and validity verification of mechanical performance data: Convert mechanical performance data from different standards or equipment into standard units under the International System of Units (SI), such as unifying pressure to megapascals; at the same time, check whether the data is within a reasonable range, such as the strength value should not be negative, and eliminate obvious abnormal values ​​caused by sensor failure or operational errors.

[0069] Smoothing of process parameters: For continuously collected time-series data of process parameters, a mathematical filtering algorithm is used to remove short-term and meaningless fluctuations, and retain trend information that reflects the true state of the process.

[0070] Sub-steps S1-3 add rich background information, i.e. metadata identifiers, to the preprocessed data and upload it to the cloud in a secure manner.

[0071] Unique Material Identifier: Assign a globally unique identification code to each experiment or each formulation sample, like an ID number, to accurately link all data of the material throughout the platform.

[0072] Data lineage labeling: Records the source of data, including which device, when, and which experimental project generated it, to ensure data traceability.

[0073] Data quality label: Marks the quality level of the data block after preprocessing, such as excellent, questionable, or requires review.

[0074] Timestamp: Precisely records the moment when data was generated.

[0075] A secure communication channel with authentication and encryption is established between edge nodes and the cloud to ensure that data is not stolen or tampered with during transmission. Finally, data with complete metadata is uploaded in batches to a data lake in the cloud for permanent storage. The data lake in the cloud is a centralized repository capable of storing massive amounts of raw data in various formats.

[0076] Step S2: Based on the real-time access frequency of the data, the project stage, and the business value, classify the processed data and perform hierarchical storage management and optimization.

[0077] Step S2 further includes the following sub-steps:

[0078] S2-1, based on the real-time access frequency and business value of data, classifies data into hot data, warm data and cold data. Hot data is newly generated and accessed data in the current project, warm data is historical data from completed projects but accessed regularly, and cold data is archived data that has exceeded the legal retention period.

[0079] S2-2 stores hot data in a storage layer composed of solid-state drives, warm data in a standard storage layer composed of hard disk arrays, and cold data in an archive storage layer composed of tape libraries.

[0080] S2-3, implement dynamic copy maintenance for hot data in the storage layer, establish a correlation model between data access frequency and the number of copies, and dynamically increase or decrease data copies based on the access pressure monitored in real time;

[0081] S2-4 initiates a compression process for cold data in the archive storage layer, identifies and separates structured data from redundant raw data files, applies a lossless compression algorithm to the structured data, and applies a lossy compression algorithm to the redundant raw data files.

[0082] Furthermore, the information entropy-based compression algorithm performs the following steps: traverse the dataset to be compressed in units of fixed-size data blocks, and calculate the information entropy value of each data block using the Shannon entropy formula; classify the data blocks into multiple importance levels according to a preset entropy threshold range; dynamically assign different compression algorithms and compression levels to data blocks of different levels, and assign the lossy compression algorithm with the highest compression ratio to the data blocks with the lowest importance.

[0083] It should be noted that "hot data" refers to newly generated data in ongoing R&D projects, as well as data that is frequently accessed and calculated by R&D personnel; for example, experimental data that has just been tested but for which no conclusion has yet been reached, or intermediate results generated by a running molecular simulation task. This type of data needs to be read quickly.

[0084] Warm data refers to historical data from completed projects. This data is no longer frequently modified, but still needs to be accessed periodically for comparative analysis, report generation, or training of new models.

[0085] Cold data refers to data that has exceeded the retention period required by laws, regulations, or the company's internal quality management system. This type of data is rarely accessed and is mainly used to meet compliance requirements.

[0086] Hot data storage layer: Composed of solid-state drives (SSDs). SSDs have extremely high read and write speeds, which can ensure fast access to hot data and meet the needs of high-concurrency computing and real-time querying, but the unit storage cost is relatively high.

[0087] Warm data storage layer: It consists of a disk array composed of multiple mechanical hard drives. The read and write speed of mechanical hard drives is lower than that of solid-state drives, but the unit storage cost is lower. It is suitable for storing warm data with moderate access frequency and large capacity.

[0088] Cold data storage layer: Composed of tape libraries. Tape libraries have very slow read and write speeds, but extremely low unit storage costs, making them suitable for long-term archiving of massive amounts of data.

[0089] Establish a correlation model: The platform continuously monitors the access frequency of each hot data block. The system maintains a mathematical model that maps the access frequency of data to the recommended number of replicas; the higher the access frequency, the more replicas are recommended.

[0090] Dynamic adjustment: When the monitoring system detects a sudden increase in access requests for a data block, such as the performance data of a popular recipe, which leads to increased read latency, it will automatically increase the number of replicas of that data block in the solid-state drive storage layer based on a mathematical model. In this way, multiple access requests can be distributed to different replicas, thereby quickly reducing latency. When the access peak has passed, the system will automatically reduce the number of replicas to free up storage space.

[0091] Data identification and separation: The system first identifies the key structured data and redundant raw data files in the data to be compressed.

[0092] Key structured data: refers to highly intensive and invaluable information in the database, including tabular data, core formula ratios, and final performance indicators.

[0093] Redundant raw data files: These refer to raw waveform files, high-resolution images, and a large number of log files generated during simulation calculations. These files are usually large in size and contain a lot of repetitive or detailed information that is indistinguishable to the human eye or algorithms.

[0094] Differentiated compression: For critical structured data, lossless compression algorithms are used. These algorithms can compress data to a smaller size and completely and accurately restore the original data, ensuring the absolute accuracy of core information. For redundant original data files, lossy compression algorithms are used. These algorithms can achieve extremely high compression ratios, but will discard some details that are not sensitive to human eyes or subsequent analysis, significantly reducing file size within an acceptable range of accuracy loss.

[0095] Further explanation of the information entropy-based compression algorithm: This is a more advanced intelligent compression strategy for redundant original data files in cold data environments. Its execution steps are as follows:

[0096] 1. Calculating Information Entropy: The system divides large files into fixed-size data blocks and then uses the Shannon entropy formula to calculate the information entropy value of each data block. Information entropy is an indicator in information theory that measures the disorder or information content of data; the higher the entropy value, the more random the data, the more information it contains, and the more important it is; the lower the entropy value, the more regular the data, the higher the redundancy, and the less important it is. The Shannon entropy formula is: Here, H(X) is the information entropy of the random variable X, the unit of which depends on the base b of the logarithm, where b is a commonly used bit, and b=2 in this case. n is the number of all possible values ​​of the random variable X. It is the i-th possible value of the random variable X. It is a value. The probability of occurrence satisfies =1, It is a logarithmic operation with base b. When b=2, the unit of entropy is bits.

[0097] 2. Classification by Entropy Value: The system presets several entropy value threshold ranges and classifies data blocks into different importance levels based on the calculated entropy values.

[0098] 3. Compression strategy allocation: Different compression algorithms and intensities are assigned to data blocks of different importance levels; for data blocks with the lowest entropy and the lowest importance, the lossy compression algorithm with the highest compression ratio is assigned to minimize their storage space; for data blocks with higher entropy and higher importance, lossy compression or even lossless compression with a lower compression ratio is adopted.

[0099] Step S3: parse the R&D computing tasks submitted by the user, and dynamically schedule and allocate cloud computing resources to execute the tasks based on the task attributes;

[0100] Step S3 further includes the following sub-steps:

[0101] S3-1 receives user task requests through a unified task submission interface. The task request includes the task type, input parameters, and resource requirement description. The task types include computationally intensive molecular dynamics simulation, data-intensive structure-activity model training, and memory-intensive multi-objective formulation optimization.

[0102] S3-2, parse the dependency descriptors in the task request and construct a directed acyclic graph representing the execution order and dependencies between tasks;

[0103] S3-3: Traverse the directed acyclic graph, identify all nodes without predecessor dependencies, form an initial set of executable tasks, and prioritize their execution.

[0104] S3-4. For task branches in a directed acyclic graph that do not have direct or indirect dependencies, they are identified as independent task sets and split into multiple subtasks that can be executed in parallel on different computing nodes.

[0105] S3-5: Allocate computing resource units to the task based on the resource requirements declared in the task request and the real-time resource status of the system.

[0106] Furthermore, in steps S3-5, allocating computing resource units to tasks includes: allocating resource units consisting of multiple CPU cores to computing-intensive tasks, allocating GPU accelerator card resource units to data-intensive tasks, and allocating memory resource units to memory-intensive tasks.

[0107] It should be noted that task type: explicitly specify the characteristics of the task. The platform predefines several types, including:

[0108] Computationally intensive molecular dynamics simulations: This task requires a large number of complex physical equation calculations, placing extremely high demands on the processor's computing power.

[0109] Data-intensive structure-effect model training: This task requires reading massive amounts of historical data to train machine learning models, which places high demands on data throughput and parallel computing capabilities, making it suitable for acceleration using graphics processing units (GPUs).

[0110] Memory-intensive multi-objective optimization of recipes: This task requires storing and processing a large amount of candidate recipe data in memory simultaneously, which places high demands on memory capacity and bandwidth.

[0111] Input parameters: The specific data required for task execution, such as molecular structure files, training datasets, and optimization target parameters.

[0112] Resource requirements description: The user's estimate or requirements for the computing resources needed, such as how many processor cores or how much memory is required.

[0113] A directed acyclic graph (DAG) is a graphical model consisting of nodes and arrows. In this invention, each node represents a specific subtask, such as "calculate the energy of molecule A first." Arrows represent dependencies, such as "calculate the energy of molecule B" can only begin after "calculate the energy of molecule A" is completed. The directionality indicates that the dependency is unidirectional, and the acyclic nature ensures that the task flow will not fall into an infinite loop and is executable.

[0114] The platform's task scheduler begins to traverse the aforementioned directed acyclic graph, searching for all nodes without any predecessor dependent tasks, i.e., the starting task with no arrow pointing to it; these nodes represent tasks that can be executed immediately; the scheduler collects these nodes to form an initial set of executable tasks and immediately allocates them to idle computing resources for priority execution, which is the key to starting the entire task flow.

[0115] During the scheduling process, the scheduler identifies task branches in the directed acyclic graph that do not have direct or indirect dependencies. These branches can be executed simultaneously without interfering with each other, and the platform identifies these branches as independent task sets. To further improve efficiency, the platform can also automatically break down a large, parallelizable independent task, such as calculating the properties of 10,000 molecules, into hundreds or even thousands of smaller subtasks, such as each subtask calculating 100 molecules. These subtasks are then distributed to a large number of computing nodes in the cloud for simultaneous execution, thereby achieving high-throughput computing.

[0116] Based on the resource requirements declared in the task request and the real-time status of the cloud virtualization resource pool, such as which servers have idle processors and which servers have idle graphics processors, the resource manager allocates appropriate computing resource units for each task or subtask to be executed; the allocation strategy is as follows:

[0117] For computationally intensive tasks, such as molecular dynamics simulations: resource units consisting of multiple central processing unit cores are allocated to provide powerful serial and parallel computing capabilities.

[0118] For data-intensive tasks, such as structure-effect model training: allocate graphics processing unit (GPU) accelerator cards. GPUs have thousands of computing cores, which are suitable for performing large-scale matrix operations required for machine learning training, and can achieve speedup ratios of tens of times higher than those of central processing units.

[0119] For memory-intensive tasks, such as multi-objective recipe optimization: allocate resource units with large memory capacity to ensure that all computational data can be quickly loaded into memory for processing, avoiding performance degradation due to insufficient memory.

[0120] Step S4: Machine learning algorithms are used to mine and analyze the stored data in order to build a knowledge graph, train a performance prediction model, and generate recipe recommendations.

[0121] Step S4 further includes the following sub-steps:

[0122] S4-1 performs named entity recognition on the text data of R&D documents and experimental reports stored on the platform to extract material, performance, and process entities;

[0123] S4-2, Extract relations from the extracted entities and identify semantic relationships between entities;

[0124] S4-3, the extracted entity and relation pairs are associated and mapped with the structured experimental data inside the platform through unique material identifiers to construct a structured material knowledge graph;

[0125] S4-4: Parse and calculate the geometric and electronic structure descriptors from the molecular structure file, combine them with the corresponding synthesis process parameters to form a feature vector, and use the material's measured performance indicators as supervised learning labels to train a machine learning regression model to generate a quantitative structure-activity relationship prediction model.

[0126] S4-5, based on the historical formula database, uses an unsupervised clustering algorithm to divide the formulas into different groups according to their components and performance characteristics. For the current R&D goals, baseline formulas are selected from the groups, and a recommendation algorithm is used to generate a recommended list of optimized proportions of the formula components.

[0127] Furthermore, the geometric and electronic structure descriptors include: the highest occupied molecular orbital energy level, the lowest unoccupied molecular orbital energy level, the energy level gap, the molecular dipole moment, the static polarizability, the molecular van der Waals surface area, the molecular oil-water partition coefficient, the topological polar surface area, and the molecular refractive index.

[0128] It should be noted that the platform uses a pre-trained natural language processing model to automatically analyze R&D documents and experimental reports stored on the platform. The model performs named entity recognition, which involves automatically identifying and extracting entity words representing specific meanings from unstructured text. These entities include:

[0129] Material entities: such as specific polymer names, like polycarbonate, additive names, like titanium dioxide, and solvent names.

[0130] Performance properties: such as tensile strength, heat distortion temperature, and light transmittance.

[0131] Process entities: such as injection molding, twin-screw extrusion, and reaction temperature.

[0132] Based on the identification of entities, the model further analyzes the sentence structure and extracts relationships, that is, it determines the semantic relationships between different entities. For example, it identifies a relationship of "improvement" between "toughening agent A" and "impact strength", or a relationship of "influence" between "process parameter B" and "molecular weight".

[0133] The entities and relationships extracted from the text are then fused with structured data obtained through experiments in the platform's database, such as specific performance values ​​and process parameter values. The key to this fusion is the mapping of associations using unique material identifiers. For example, the performance description of "formula X" mentioned in the text is linked to the actual test results of a vast number of formulas "X" in the database through the unique material identifier "X". Ultimately, all these interconnected entities and relationships together constitute a massive, structured materials knowledge graph that can intuitively display the complex network relationships between materials, processes, and performance, supporting intelligent question answering and knowledge discovery.

[0134] Feature engineering: Starting from the three-dimensional structure file of the molecule, a series of geometric and electronic structure descriptors are analyzed and calculated using computational chemistry methods. These descriptors are numerical and used to quantitatively describe the physicochemical properties of the molecule. These molecular descriptors are combined with the corresponding synthesis process parameters, such as temperature and pressure, to form a feature vector representing each material sample.

[0135] Model training: The actual performance indicators of the material, such as the measured tensile strength value, are used as the target that the model needs to predict, i.e., the supervised learning label. The regression algorithm in machine learning is used to train the model with the feature vector as input and the performance label as output. After training, a quantitative structure-property relationship prediction model is obtained. After that, researchers only need to input the structure of the new molecule and the preset process, and the model can quickly predict its possible performance, greatly reducing the blindness of experiments.

[0136] Formula Clustering: Based on a database of historical successful formulas, the platform uses an unsupervised clustering algorithm to automatically divide thousands of formulas into several groups with similar characteristics according to the component characteristics and final performance of the formulas. Unsupervised clustering is an algorithm that does not require prior knowledge of the answer and automatically discovers the inherent grouping patterns of the data.

[0137] Baseline Screening and Optimization Recommendation: When researchers set new R&D goals, such as requiring a high-toughness, low-cost formulation, the system will automatically select several formulations with the best overall performance from the performance family that best matches the goal as baseline formulations. Then, the system uses a recommendation algorithm to generate a list of optimized proportions of formulation components based on the composition of the baseline formulation and taking into account the synergistic or antagonistic effects between the components, providing researchers with a starting point and direction for innovation.

[0138] Further explanation regarding geometric and electronic structure descriptors: These descriptors are key physicochemical parameters derived from molecular structure calculations and form the basis for the accuracy of prediction models.

[0139] The highest occupied molecular orbital energy level and the lowest unoccupied molecular orbital energy level: These measure the ability of a molecule to donate and accept electrons, respectively, and are closely related to the stability and chemical reactivity of materials.

[0140] Energy level gap: The difference between the two energy levels mentioned above is a key parameter for judging the electrical conductivity and optical properties of a material.

[0141] Molecular dipole moment: Characterizes the strength of molecular polarity, affects intermolecular forces, and is related to the solubility and melting point of materials.

[0142] Static polarizability: describes the ease with which the electron cloud of a molecule deforms under the influence of an external electric field.

[0143] Molecular van der Waals surface area and topological polar surface area: describe the size of molecules and the distribution of surface polar regions, and are related to biological activity and adsorption.

[0144] Molecular oil-water partition coefficient: measures the tendency of molecules to partition in oil and water phases, and is an important parameter for predicting the biocompatibility and environmental behavior of materials.

[0145] Molecular refractive index: related to the electron density and polarizability of molecules, reflecting the optical properties of materials.

[0146] Step S5 visualizes the data processing results, analysis process, and task status, and provides permission-based data sharing and collaborative R&D support.

[0147] Step S5 further includes the following sub-steps:

[0148] S5-1 collects project task status, computing resource utilization and model performance index data in real time from the platform database and monitoring system, and dynamically renders and displays them on a unified monitoring interface through configurable visualization components.

[0149] S5-2 responds to front-end user interaction commands, queries the corresponding process parameter dataset and material property dataset from the storage system, calls the back-end drawing engine to generate process-performance relationship curves, and generates a molecular structure-activity relationship scatter plot based on molecular structure feature data and activity data. The molecular structure feature data are geometric and electronic structure descriptors obtained by parsing and calculating from the molecular structure file.

[0150] S5-3 creates independent project spaces within the platform and configures role-based access control lists to assign data read, write, execute, and management permissions to project members, enabling version control, sharing, and collaborative operation of experimental data, analysis models, and report documents within the authorized scope.

[0151] Furthermore, in step S5-2, the process-performance relationship curve is a two-dimensional curve showing the correspondence between a single process parameter and a single performance index, or a three-dimensional surface plot or contour plot showing the complex relationship between multiple process parameters and performance indexes; the molecular structure-activity relationship scatter plot is a visualization result of projecting molecular structure feature data as feature vectors onto a two-dimensional plane after dimensionality reduction.

[0152] It should be noted that the platform obtains key data reflecting the R&D status in real time from the backend database and system monitoring module, including:

[0153] Project task status: such as whether each computing task is queued, running, completed, or failed.

[0154] Computing resource utilization: such as the CPU utilization, memory usage, and GPU load of cloud servers.

[0155] Model performance metrics: such as the accuracy and error magnitude of the quantitative structure-activity relationship prediction model on the test set.

[0156] This data is transmitted to a unified monitoring interface at the front end, which consists of a variety of configurable visualization components, such as charts, dashboards, and progress bars. The system dynamically populates the data into these components and renders them according to predefined rules, thereby forming real-time updated visualization charts, enabling project managers to have a clear understanding of the overall R&D progress and resource consumption.

[0157] User interaction: Researchers can use drop-down menus and checkboxes on the interface to independently select process parameters, material performance indicators, or molecular structure characteristics data of interest.

[0158] Backend processing and plotting: The interactive commands from the frontend will trigger the backend service. The backend service first quickly retrieves the corresponding process parameter dataset and material performance dataset from the storage system based on the user's selection; then, it calls a dedicated plotting engine to generate the corresponding visualization charts.

[0159] Process-performance relationship curve: used to visually demonstrate how process conditions affect material properties.

[0160] Molecular structure-activity relationship scatter plot: used to study the trend relationship between molecular structure features and a specific activity of a material. In order to display high-dimensional molecular features on a two-dimensional plane, a dimensionality reduction algorithm is first used to process them, and then the result is projected into a scatter plot. Dimensionality reduction algorithm is a technique that simplifies complex high-dimensional data into a low-dimensional representation.

[0161] Advanced visualization: The system supports displaying more complex multivariate relationships; for example, it can generate 3D surface plots to simultaneously show the combined impact of two process parameters on a performance index; or use contour plots to clearly express gradient changes in such complex relationships.

[0162] Project Space Creation: Create an independent, logically isolated project space for each R&D project. This project space has its own independent storage area and computing environment.

[0163] Permission configuration: Within the project space, role-based access control is implemented; the system predefines roles such as "Project Leader", "Experimenter", "Data Analyst" and "Visitor", and configures fine-grained permission combinations for each role, such as data reading, writing, executing calculation models, and managing members.

[0164] Collaborative operations: The project leader assigns roles to team members; thereafter, members can only perform actions within their authorized scope, for example:

[0165] Version control: Every modification to experimental data, analysis models, and report documents is recorded by the system and becomes a historical version, which can be viewed or rolled back at any time to avoid data loss due to accidental operation.

[0166] Sharing and Collaboration: Members can securely share data and models, and collaboratively edit documents online, adding annotations and comments, thereby achieving efficient asynchronous or synchronous collaboration.

[0167] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A data processing method for the research and development of new plastic materials based on a cloud platform, characterized in that, Includes the following steps: Step S1: Collect raw data of multi-source heterogeneous plastic research and development through edge nodes deployed on the R&D equipment side, and preprocess the data; Step S2: Based on the real-time access frequency of the data, the project stage, and the business value, classify the processed data and perform hierarchical storage management and optimization. Step S3: parse the R&D computing tasks submitted by the user, and dynamically schedule and allocate cloud computing resources to execute the tasks based on the task attributes; Step S4: Machine learning algorithms are used to mine and analyze the stored data in order to build a knowledge graph, train a performance prediction model, and generate recipe recommendations. Step S5 visualizes the data processing results, analysis process, and task status, and provides permission-based data sharing and collaborative R&D support. Step S1 further includes the following sub-steps: S1-1 acquires molecular structure data and quantum chemical descriptors from molecular simulation sources, spectral data from analytical testing equipment, mechanical property data from performance testing equipment, and synthesis process parameters from process control systems via a data acquisition interface. S1-2, The collected raw data is cleaned and formatted, including standardizing and encoding molecular structure data, performing baseline correction and characteristic peak extraction on spectral data, unifying units and validating the mechanical property data, and smoothing process parameters. S1-3, add metadata identifiers to the preprocessed data, and transmit it in batches to the cloud data lake for persistent storage through a secure channel with identity authentication and transmission encryption. The metadata identifiers include unique material identifiers, data lineage tags, data quality identifiers, and timestamps. Step S3 further includes the following sub-steps: S3-1, Receive user task requests through a unified task submission interface. The task request includes task type, input parameters and resource requirement description. The task type includes computationally intensive molecular dynamics simulation, data-intensive structure-activity model training and memory-intensive multi-objective formulation optimization. S3-2, parse the dependency descriptors in the task request and construct a directed acyclic graph representing the execution order and dependencies between tasks; S3-3, Traverse the directed acyclic graph, identify all nodes without predecessor-dependent tasks, form an initial set of executable tasks, and prioritize their execution; S3-4. For task branches in a directed acyclic graph that do not have direct or indirect dependencies, they are identified as independent task sets and split into multiple subtasks that can be executed in parallel on different computing nodes. S3-5: Allocate computing resource units to the task based on the resource requirements declared in the task request and the real-time resource status of the system; Step S4 further includes the following sub-steps: S4-1 performs named entity recognition on the text data of R&D documents and experimental reports stored on the platform to extract material, performance, and process entities; S4-2, Extract relations from the extracted entities and identify semantic relationships between entities; S4-3, the extracted entity and relation pairs are associated and mapped with the structured experimental data inside the platform through unique material identifiers to construct a structured material knowledge graph; S4-4: Parse and calculate the geometric and electronic structure descriptors from the molecular structure file, combine them with the corresponding synthesis process parameters to form a feature vector, and use the material's measured performance indicators as supervised learning labels to train a machine learning regression model to generate a quantitative structure-activity relationship prediction model. S4-5, based on the historical formula database, uses an unsupervised clustering algorithm to divide the formulas into different groups according to their components and performance characteristics. For the current R&D goals, baseline formulas are selected from the groups, and a recommendation algorithm is used to generate a recommended list of optimized proportions of the formula components.

2. The data processing method for the research and development of new plastic materials based on a cloud platform according to claim 1, characterized in that: Step S2 further includes the following sub-steps: S2-1, based on the real-time access frequency and business value of data, data is classified into hot data, warm data and cold data. Hot data is newly generated and accessed data in the current project, warm data is historical data of completed projects but accessed regularly, and cold data is archived data that has exceeded the legal retention period. S2-2 stores hot data in a storage layer composed of solid-state drives, warm data in a standard storage layer composed of hard disk arrays, and cold data in an archive storage layer composed of tape libraries. S2-3, implement dynamic copy maintenance for hot data in the storage layer, establish a correlation model between data access frequency and the number of copies, and dynamically increase or decrease data copies based on the access pressure monitored in real time. S2-4 initiates a compression process for cold data in the archive storage layer, identifies and separates structured data from redundant raw data files, applies a lossless compression algorithm to the structured data, and applies a lossy compression algorithm to the redundant raw data files.

3. The data processing method for the research and development of new plastic materials based on a cloud platform according to claim 2, characterized in that: The compression process employs an information entropy-based compression algorithm, which performs the following steps: traversing the dataset to be compressed in units of fixed-size data blocks, and calculating the information entropy value of each data block using the Shannon entropy formula; Based on a preset entropy threshold range, data blocks are classified into multiple importance levels; Different compression algorithms and compression levels are dynamically assigned to data blocks of different importance, with the lowest-importance data blocks assigned the lossy compression algorithm with the highest compression ratio.

4. The data processing method for the research and development of new plastic materials based on a cloud platform according to claim 1, characterized in that: In steps S3-5, allocating computing resource units to tasks includes: allocating resource units consisting of multiple CPU cores to computing-intensive tasks, allocating GPU accelerator card resource units to data-intensive tasks, and allocating memory resource units to memory-intensive tasks.

5. The data processing method for the research and development of new plastic materials based on a cloud platform according to claim 1, characterized in that: The geometric and electronic structure descriptors include: highest occupied molecular orbital energy level, lowest unoccupied molecular orbital energy level, energy level gap, molecular dipole moment, static polarizability, molecular van der Waals surface area, molecular oil-water partition coefficient, topological polar surface area, and molecular refractive index.

6. The data processing method for the research and development of new plastic materials based on a cloud platform according to claim 1, characterized in that: Step S5 further includes the following sub-steps: S5-1 collects project task status, computing resource utilization and model performance index data in real time from the platform database and monitoring system, and dynamically renders and displays them on a unified monitoring interface through configurable visualization components. S5-2, responding to front-end user interaction commands, queries the corresponding process parameter dataset and material performance dataset from the storage system, calls the back-end drawing engine to generate process-performance relationship curves, and generates a molecular structure-activity relationship scatter plot based on molecular structure feature data and activity data. The molecular structure feature data are geometric and electronic structure descriptors obtained by parsing and calculating from molecular structure files. S5-3 creates independent project spaces within the platform and configures role-based access control lists to assign data read, write, execute, and management permissions to project members, enabling version control, sharing, and collaborative operation of experimental data, analysis models, and report documents within the authorized scope.

7. The cloud-based data processing method for the research and development of new plastic materials according to claim 6, characterized in that: In step S5-2, the process-performance relationship curve is a two-dimensional curve showing the correspondence between a single process parameter and a single performance index, or a three-dimensional surface plot or contour plot showing the complex relationship between multiple process parameters and performance indexes; the molecular structure-activity relationship scatter plot is a visualization result of projecting molecular structure feature data as feature vectors onto a two-dimensional plane after dimensionality reduction.

Images