Neodymium iron boron production process management method based on intelligent and data technology

Abstract

The invention relates to the cross technical field of intelligent manufacturing, industrial artificial intelligence, a big data technology and high-end material engineering, in particular to a neodymium iron boron production process management method based on artificial intelligence and a big data technology. According to the method, a'big data platform + AI algorithm engine 'double-wheel-driven intelligent management system is constructed, artificial intelligence technologies such as machine learning, deep learning, knowledge graph and digital twinning are deeply fused with a mass multi-source heterogeneous big data processing technology, and the method runs through the whole process of neodymium iron boron material production. The core comprises the steps of constructing a unified process data lake based on a big data technology, and realizing fusion treatment of multi-source data; establishing a quantitative correlation model of process parameters and product performance based on a machine learning algorithm; digital expression and intelligent reasoning of process knowledge are realized based on the knowledge graph; constructing a virtual process simulation optimization platform based on a digital twinning technology; and finally, a continuous evolution closed loop of data acquisition, feature extraction, model training, intelligent decision making and feedback optimization is formed. The technical problems that in traditional neodymium iron boron production, value mining of mass data is insufficient, process decision depends on artificial experience, quality control lags behind, and the process optimization period is long are effectively solved, and fundamental transformation of the production process from experience driving to data and intelligent double-wheel driving is achieved.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary technical field of industrial big data, artificial intelligence, intelligent manufacturing, and advanced materials engineering. Specifically, it relates to an innovative method for the production process of high-performance permanent magnet material neodymium iron boron (NdFeB), which achieves intelligent and data-driven management of the entire complex and precision production process by deeply integrating artificial intelligence algorithms and big data processing technology. Background Technology

[0002] Neodymium iron boron (NdFeB) is an indispensable key functional material in modern high-tech industries. Its production process involves multiple complex steps, including smelting, powder preparation, molding, sintering, and post-processing. The precision of control at each step directly affects the magnetic properties and consistency of the final product. With the development of technologies such as sensors and the Internet of Things (IoT), NdFeB manufacturers have accumulated massive amounts of process data (time-series data such as temperature, pressure, vacuum, and current), quality data (performance test results), and various business data. However, under traditional management models, the value of this data has not been fully realized, mainly due to the following technical bottlenecks: 1. The phenomenon of "data silos" is serious: data is scattered across different systems and departments, with different formats and a lack of unified standards, making it difficult to conduct effective correlation analysis and global insights.

[0003] 2. Over-reliance on experience in process decisions: The setting and adjustment of key process parameters mainly rely on the personal experience of engineers, lacking the support of quantitative analysis based on massive historical data, resulting in strong subjectivity and poor repeatability in decision-making.

[0004] 3. Lagging and passive quality control: Quality problems are usually only discovered during final inspection, making it impossible to achieve early warning and proactive intervention during the production process, resulting in waste of materials and time.

[0005] 4. Low efficiency in process optimization: The development of new processes or the improvement of existing processes rely on expensive physical experiments and a lengthy "trial and error" process. There is a lack of effective virtual simulation and intelligent optimization tools, resulting in long innovation cycles and high costs.

[0006] 5. Difficulty in knowledge transfer and reuse: Core process knowledge exists mostly in the minds of a few experts in the form of tacit experience, or is scattered in various documents, making it difficult to systematically accumulate, share and transfer.

[0007] While existing Manufacturing Execution Systems (MES) or Statistical Process Control (SPC) tools have achieved partial data collection and monitoring, they still have significant shortcomings in processing massive, high-dimensional, and unstructured industrial big data, as well as in applying advanced artificial intelligence algorithms for in-depth analysis, prediction, and optimization. Therefore, there is an urgent need for a new method for NdFeB production process management that can deeply integrate artificial intelligence and big data technologies to systematically solve the aforementioned problems. Summary of the Invention

[0008] The technical problem this invention aims to solve is to provide a systematic and intelligent solution to the core issues existing in traditional NdFeB production process management, such as insufficient data value mining, reliance on experience for decision-making, lagging quality control, difficulty in knowledge transfer, and low optimization efficiency.

[0009] To address the aforementioned technical challenges, this invention proposes a method for managing the NdFeB production process based on artificial intelligence and big data technologies. The core idea of ​​this method is to use "data" as the foundation and "intelligence" as the engine. First, by constructing a unified big data platform, data silos are broken down, forming high-quality process data assets. Then, based on these data assets, artificial intelligence algorithms (machine learning, knowledge graphs, digital twins, etc.) are systematically applied to construct a full-chain intelligent application covering process design, process monitoring, quality prediction, anomaly diagnosis, and continuous optimization. Ultimately, this forms an intelligent management closed loop capable of learning from data and guiding continuous improvement in production practices.

[0010] Another objective of this invention is to provide an integrated hardware and software intelligent management system for implementing the above-described method.

[0011] The beneficial technical effects of this invention are as follows: 1. Achieve data-driven scientific decision-making: Shift process decisions from being based on personal experience to being based on quantitative analysis of massive historical data, thereby improving the scientific nature, objectivity, and consistency of decisions.

[0012] 2. Enhance the initiative and foresight of quality control: Through real-time intelligent monitoring and prediction models, early warnings can be issued before quality problems occur, and the root cause can be quickly located, transforming passive inspection into proactive prevention.

[0013] 3. Significantly improve the efficiency of process research and development and optimization: By utilizing digital twins and intelligent optimization engines, a large number of simulation experiments can be conducted in virtual space to quickly find the best solution, significantly reduce the number of physical trial and error attempts, shorten the research and development cycle, and reduce innovation costs.

[0014] 4. Achieve digital accumulation and intelligent reuse of process knowledge: Make tacit knowledge explicit and structured through knowledge graphs to form a queryable and reasonable enterprise knowledge base, effectively solving the problem of knowledge inheritance and supporting intelligent recommendations based on similar cases.

[0015] 5. Building a continuously self-evolving intelligent system: Through a closed-loop mechanism of "data feedback - model update", the system can continuously learn from new production data, enabling the model and knowledge base to continuously iterate and optimize, adapt to changes in raw materials, equipment and other conditions, and achieve dynamic improvement in process capabilities. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall architecture of the intelligent management system for neodymium iron boron production process in one embodiment of the present invention.

[0017] Figure 2 This is a core workflow diagram of the intelligent management method described in one embodiment of the present invention.

[0018] Figure 3 This is a schematic diagram of data interaction and closed-loop learning for intelligent process optimization and real-time monitoring in one embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0020] Overall System Architecture refer to Figure 1 The intelligent management system of this invention adopts a layered and modular design to ensure the system's openness, scalability, and high performance.

[0021] Data Acquisition and Edge Layer: Deployed on the production site, it collects the operating parameters of various devices and instruments in real time through industrial IoT gateways, protocol conversion modules, smart sensors, etc., and performs preliminary filtering, compression and edge computing to reduce the pressure on the cloud and ensure the real-time performance of key monitoring.

[0022] Big Data Platform Layer: This is the system's "data hub." Built upon distributed computing and storage frameworks (such as Hadoop and Spark), it is responsible for: Data Integration: Aggregating all relevant data from edge layers, LIMS, ERP, MES, and other systems. Data Governance: Cleaning, denoising, harmonic processing, and standardizing raw data to ensure data quality. Data Storage and Management: Employing a data lake architecture, it stores raw data, cleaned data, and derived feature data, and establishes a unified data catalog and metadata management system.

[0023] AI Algorithm and Model Layer: This is the system's "intelligent engine." It provides: Algorithm Library: Integrates various machine learning, deep learning, time series analysis, and graph computing algorithms suitable for industrial scenarios. Model Factory: Supports full lifecycle management of models (MLOps), from data preparation, feature engineering, model training, evaluation to service-oriented deployment. Core Model Services: Includes process parameter-performance prediction models, time series anomaly detection models, root cause analysis models, multi-objective optimization algorithms, etc. Knowledge Center Layer: Constructs a process knowledge graph based on a graph database, models entities such as materials, equipment, processes, faults, and expert experience, and their complex relationships, providing services such as semantic retrieval, relational reasoning, and impact analysis. Intelligent Application Layer: Provides specific business functions for different user roles. Intelligent Process Design Assistant: Assists engineers in developing new processes. Real-time Intelligent Monitoring and Early Warning Center: Provides operators with visualized production status dashboards and anomaly warnings. Process Quality Analysis and Optimization Platform: Provides in-depth analysis tools for process and quality engineers. Process Knowledge Portal: Provides a knowledge query and learning platform for all personnel. Interactive display layer: Provides users with a user-friendly interface and rich visualization charts through various forms such as web terminals, mobile apps, and industrial large screens.

[0024] Core processes of intelligent management methods refer to Figure 2In practical operation, this method manifests as the following interconnected and iterative core processes: Phase 1: Data-Driven Process Modeling and Knowledge Construction 1. In-depth Historical Data Mining: Utilizing massive amounts of historical production data from a big data platform, key features are extracted through feature engineering. For example, features such as heating rate, heat preservation stability, and cooling slope are extracted from the ten-segment temperature zone curves of the sintering furnace. 2. Building Predictive and Correlation Models: Machine learning algorithms (such as XGBoost and neural networks) are used to train models, establishing a mapping relationship from these process features to the final magnetic properties (Br, Hcj). Simultaneously, interpretable AI technology is used to quantitatively analyze the contribution of each process parameter to performance and the interaction effects between them, forming a quantitative understanding of the "process prescription." 3. Building a Process Knowledge Graph: The patterns discovered in step 2, equipment manuals, operating procedures, expert interview records, etc., are transformed into structured knowledge and stored in the knowledge graph. For example, a ternary relationship can be established as "Raw Material A (High PrNd) --[Requirements]--> Process B (Low Temperature Sintering)--[Output]--> Performance C (High Coercivity)". The second stage: Intelligent process design and optimization. When developing new products or improving existing processes: 1. Target input and constraint setting: Engineers input target performance indicators (e.g., Br≥14.5kG, Hcj≥20kOe) and constraints such as cost and energy consumption into the system. 2. Knowledge retrieval and baseline generation: Based on a knowledge graph, the system automatically retrieves historically successful cases with similar performance as initial references for the optimization algorithm. 3. Virtual simulation and multi-objective optimization: The intelligent optimization engine calls a digital twin model (integrating physical simulation and the aforementioned prediction model) to search in the parameter space. It uses multi-objective optimization algorithms (such as NSGA-II) to simulate thousands of parameter combinations in a virtual environment, evaluate their predictive performance and cost, and ultimately recommend a "Pareto optimal" solution that achieves the best balance among multiple objectives. 4. Solution Decision and Small-Scale Trial: Engineers select 1-2 of the most promising solutions from the recommended options and conduct small-batch trials to verify the effectiveness of virtual optimization. Third Stage: Adaptive Real-Time Monitoring and Intelligent Diagnosis in Formal Production: 1. Real-Time Data Stream Processing: Edge and cloud systems process sensor data streams in real time. 2. Time-Sequence Pattern Intelligent Monitoring: Using trained deep learning models such as LSTM, the system compares the current process parameter curves with the "golden curve" (standard mode or optimal historical curve) in real time. The model can identify minute anomalies that are difficult to detect with the naked eye but may lead to quality problems. 3. Early Warning and Root Cause Analysis: Once an abnormal pattern is detected, the system immediately issues a warning. Simultaneously, the root cause analysis engine is activated, combining real-time data, equipment status (from the equipment twin), and fault trees in the knowledge graph to perform inference analysis, quickly locating possible causes (such as "heater power attenuation in temperature zone 5" or "high moisture content in raw material batches"), and providing remedial suggestions.Phase Four: Closed-Loop Feedback and System Evolution After each batch of production is completed: 1. Data Closed-Loop: The complete data package for this batch (process settings, process curves, final inspection results) is automatically linked and stored in the big data platform. 2. Model Evaluation and Iteration: The system compares the actual quality results with the output of the predictive model to evaluate the model's accuracy. Using new data, the predictive model and anomaly detection model are retrained periodically or through incremental learning to adapt to changes in the production process. 3. Knowledge Base Update: Validated process adjustments or new causal relationships discovered during this production run are treated as new knowledge fragments and updated to the process knowledge graph after review.

[0025] Through the continuous cycle of the above four stages, the system achieves the sublimation from "data" to "information" to "knowledge" and then to "intelligent decision-making," and finally generates new "data" through "decision execution," forming an ever-enhancing intelligent engine that drives the continuous upward climb of NdFeB production process management capabilities.

Claims

1. A neodymium iron boron production process management method based on artificial intelligence and big data technology, characterized in that, Comprise the following steps: S1: Construct a unified multi-source heterogeneous data governance platform, collect, clean, fuse and standardize the massive data generated in the whole process of neodymium iron boron production, such as equipment operation parameters, raw material detection data, process setting parameters, quality test results, environmental state information, etc., and form an enterprise-level process data lake; S2: Based on big data analysis technology, deeply mine historical process data, and use machine learning algorithm to establish multi-dimensional quantitative correlation model between key process parameters (such as sintering temperature, smelting time, magnetic field strength, etc.) and final performance indicators of products (such as Br, Hcj, squareness, etc.); S3: Construct process knowledge graph, digitize structured and unstructured knowledge such as raw material characteristics, equipment capacity, process route and expert experience, and form a process knowledge base that can be queried and reasoned; S4: Integrate digital twin technology with the above correlation model and knowledge graph to form an intelligent process optimization engine that can simulate and optimize process parameters in a virtual environment for specific performance targets and output recommended solutions; S5: In the production execution process, based on real-time big data stream processing technology and intelligent monitoring model, realize real-time sensing of process state, early warning of abnormal mode and intelligent diagnosis of root cause; S6: Establish a data-driven continuous optimization closed loop, feed back the actual production results to the data lake, and regularly update the correlation model, knowledge graph and digital twin, realize system self-learning and continuous improvement of process capability.

2. The method of claim 1, wherein, The "building a unified multi-source heterogeneous data governance platform" in step S1 specifically includes: collecting time series data, business data and document data from smelting furnaces, sintering furnaces, presses, detection instruments and other devices, and MES, LIMS, ERP and other systems through Internet of Things interface, industrial protocol adaptation, API connection and other methods; use data cleaning, outlier processing, data matching, format conversion and other technologies to preprocess the original data and ensure data quality; adopt data warehouse and data lake layered architecture to store and manage the processed data, and build an enterprise-level process data asset center.

3. The method of claim 1, wherein, The "establishing a multi-dimensional quantitative correlation model" in step S2 specifically includes: using feature engineering technology to extract key features from massive process parameters, such as statistical features, time series features, frequency domain features, etc.; using machine learning algorithms such as random forest, gradient boosting decision tree and neural network to establish prediction models for different performance indicators; applying explainable AI technology to analyze the influence weight and interaction of each process parameter on the final performance, and forming a quantitative understanding of process mechanism.

4. The method of claim 1, wherein, The "real-time sensing and intelligent diagnosis" in step S5 specifically includes: using streaming computing framework to process and extract features from sensor data in real time during production process; applying time series pattern recognition models based on LSTM, CNN and other deep learning technologies to monitor the deviation of process parameter curve from standard mode in real time; when an anomaly is detected, use graph computing and causal reasoning technology to analyze the root cause in combination with knowledge graph and historical case library, and generate a diagnosis report and treatment suggestion.

5. A Nd-Fe-B production process intelligent management system for implementing the method of any one of claims 1-4, characterized in that, Comprise: Big data governance and computing platform, responsible for multi-source data access, governance, storage and computing; Intelligent algorithm model platform, integrating various machine learning and deep learning algorithms, and providing full life cycle services such as model development, training, deployment and management; Process knowledge center, based on knowledge graph technology, realizes the modeling, storage, query and reasoning of process knowledge; Digital twin and simulation optimization platform, providing virtual simulation, parameter optimization and scheme verification functions for process; intelligent monitoring and decision support application, facing production site and management personnel, providing real-time monitoring, abnormal early warning, intelligent diagnosis and decision suggestion services.

Images