Flexible automatic production management and control system of new energy high-power high-frequency transformer

Abstract

The invention discloses a flexible automatic production management and control system for a new energy high-power high-frequency transformer, and relates to the field of production management and control, the flexible automatic production management and control system comprises a production demand receiving module, a production resource management module, a production planning module, a production execution module and a quality control module, an order is received through an API or EDI, and a work order is generated; collecting equipment and material information in real time, and constructing a production resource association network; carrying out production scheduling and process matching by adopting an improved deep learning model; scheduling materials, procedures and equipment based on a multi-level analysis control model; the quality is evaluated by using a two-dimensional cloud model, and data recording is performed through a distributed database, so that the traceability of the quality problem is ensured. According to the system, the production efficiency and the quality management level are effectively improved, the problem that existing rigid automatic equipment lacks flexibility is solved, and efficient and accurate production of the new energy high-power high-frequency transformer is ensured through cooperative work of a plurality of modules so as to meet different production requirements.

Description

Technical Field

[0001] The present invention relates to the field of production control, and in particular to a flexible automated production control system for new energy high-power high-frequency transformers. Background Art

[0002] With the global emphasis on environmental protection and sustainable development, new energy industries such as solar energy, wind energy, and electric vehicles are experiencing explosive growth. As core equipment in ultra-high voltage direct current (UHVDC) transmission, photovoltaic inverters, and electric vehicle charging stations, new energy high-power, high-frequency transformers are experiencing a surge in market demand, with extremely high performance requirements. Traditional transformer production relies heavily on manual or semi-automated equipment, resulting in low production efficiency, high quality fluctuations, and unstable output, making it difficult to meet the demands of large-scale, high-quality production. Therefore, to improve production efficiency and precision, the transformer manufacturing industry is gradually moving towards automation.

[0003] However, existing production lines for high-power, high-frequency transformers for new energy sources are mostly rigidly automated, resulting in inflexible production processes and lengthy changeover times. This makes it difficult to quickly adjust production processes and product parameters based on order requirements, and to adapt to the demands of high-variety, small-batch production. This results in long production cycles and makes it difficult to increase production capacity. For example, in the production of photovoltaic inverters and electric vehicle charging stations, different transformer models have varying requirements for winding structures, insulation materials, and other factors. Frequent changeovers on traditional production lines lead to low production efficiency and increased costs.

[0004] Therefore, a flexible and automated production control system for new energy high-power and high-frequency transformers is needed to solve the above problems. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention discloses a flexible automated production control system for new energy high-power high-frequency transformers. Through the deep integration of deep learning and industrial control, it ensures the efficient, accurate and traceable production of new energy high-power high-frequency transformers, and has flexible adaptability to meet different production needs.

[0006] The present invention adopts the following technical solutions:

[0007] A flexible automated production control system for new energy high-power high-frequency transformers, including:

[0008] The production demand receiving module receives production orders for new energy high-power high-frequency transformers through the application programming interface (API) or electronic data interchange (EDI). It uses edge nodes to decompose and classify the received production orders to generate multiple production work orders. Each production work order includes at least the type of product to be produced, the quantity, the process parameter requirements, and the production time information.

[0009] The production resource management module uses the Internet of Things protocol to collect real-time data on the operating status of production equipment and the inventory quantity of production materials, and uses a graph database to build a network of associations among production equipment, production materials, and production products;

[0010] The production planning module uses an improved deep learning model to plan the production of multiple production orders. The improved deep learning model schedules multiple production orders based on production order information and production resource data, and matches production processes to each production order.

[0011] The production execution module uses a multi-level analysis and control model to generate production control execution instructions for production work orders, and controls the scheduling of production materials and production processes, as well as the operation of production equipment based on the generated production control execution instructions. The multi-level analysis and control model decomposes the production control of production work orders into multiple control levels through a hierarchical architecture;

[0012] The quality control module conducts production quality assessment of new energy high-power high-frequency transformers based on a two-dimensional cloud model, and records the production process data and finished product inspection results of new energy high-power high-frequency transformers based on a distributed database to locate and trace quality problems.

[0013] Furthermore, the edge node uses a natural language processor to perform text parsing on the production order to obtain the type, quantity, process parameter requirements and production time of the products to be produced, and uses a preset rule engine to classify the production order into multiple production work orders based on the dimensions of product type, process parameter requirements and urgency of production time.

[0014] Furthermore, the Internet of Things protocol includes the lightweight message transmission protocol MQTT, the Internet protocol CoAP or the industrial equipment compatible protocol OPC UA, and the graph database represents production equipment, production materials and production products as nodes in the graph, and represents the usage relationship between production equipment and production materials, the relationship between production materials participating in product production, and the relationship between production equipment participating in product production as edges in the graph, and updates the attributes of the nodes based on the real-time collection of operating status data of the production equipment and the inventory quantity of production materials.

[0015] Furthermore, the working method of the improved deep learning model includes the following steps:

[0016] S1. Input production work order information and production resource data into the model. The input production work order information is converted into a vector representation through the embedding layer, and the input production resource data is numerically normalized through category coding.

[0017] S2. Use the attention mechanism to fuse the embedded production work order vector with the normalized production resource data for multimodal feature fusion. Use a graph neural network to capture the relationship between production equipment, production materials, and products to be produced to construct a joint representation tensor.

[0018] S3. Define the objective function as the on-time completion rate of production orders and production costs, and define the constraints as production equipment capacity constraints, production material inventory constraints, and production process sequence constraints.

[0019] S4. Generate a production scheduling strategy based on a reinforcement learning policy network and ensure that the output meets the constraints through constraint violation penalties. The scheduling strategy includes production order priority sorting, production equipment allocation, and production time nodes. Then, an attention mechanism is used to match the process path of each production order with the production equipment to generate a process-level task sequence. The equipment load balancing strategy is dynamically adjusted based on the real-time production resource status collected.

[0020] S5, based on the decoder and the joint representation tensor constructed in S2, outputs the production process sequence, production order priority ranking, production equipment allocation and production time node of each production work order;

[0021] S6. Use historical production data to pre-train the node representation capabilities of the graph neural network and initialize the reinforcement learning strategy network. Then, use transfer learning to update the model parameters in real time and embed process rules in the scheduling results to correct the model output.

[0022] Furthermore, the multi-level analysis and control model includes a production planning layer, a scheduling layer and an execution layer. The planning layer formulates a global production plan based on the production planning results output by the production planning module. The global production plan includes at least the start time, end time and path process of each work order. The scheduling layer breaks down the global production plan into scheduling tasks. The scheduling tasks include at least production material distribution path planning, production equipment switching sequence and capacity balance between production processes. The execution layer converts the scheduling tasks into control instructions for execution by the production equipment, and sends process start signals and process parameters to the programmable logic controller (PLC).

[0023] Furthermore, the quality control module establishes quality assessment indicators based on the characteristics of new energy high-power high-frequency transformers, and collects production process data and finished product inspection results of current new energy high-power high-frequency transformers. The two-dimensional cloud model evaluates the qualified probability of the finished product inspection results of the current new energy high-power high-frequency transformers through a membership function, and determines the expectation, entropy and super-entropy characteristics of the two-dimensional cloud model based on historical data statistics, as well as the upper and lower limits of the qualified interval. When the qualified probability is within the qualified interval, it is judged to be qualified, otherwise it is unqualified.

[0024] Furthermore, the distributed database uses RFID radio frequency identification tags to generate product identifications for the produced new energy high-power high-frequency transformer products, and stores them in shards according to product batches. The product identifications are associated with production work orders, production equipment numbers, production material batches, production process sequences and production process parameters.

[0025] The beneficial effects of the present invention are:

[0026] 1. This invention uses edge computing to decompose and categorize production orders in real time, reducing cloud reliance and improving order response speed. Compared to traditional centralized order processing systems, it can more quickly adapt to the diverse, small-batch, and customized production needs of the new energy sector. Each work order contains information on product type, quantity, process parameters, and time, supporting flexible production and meeting the differentiated needs of different high-frequency transformer application scenarios.

[0027] 2. This invention uses a graph database to construct a dynamic network of production equipment, materials, and products, enabling visualization and real-time updates of resource relationships. Compared to traditional relational databases, it can more efficiently handle complex, multi-level relationships. Real-time access to equipment status and material inventory data through protocols such as MQTT and OPC UA, combined with the graph database's correlation analysis capabilities, allows rapid identification of resource bottlenecks and optimized resource scheduling.

[0028] 3. This invention uses an improved deep learning model for production planning, automatically scheduling production based on production work order information and production resource data, and matching the optimal production process. This is more intelligent and efficient than traditional production scheduling methods. It also automatically matches the optimal process path based on the process parameter requirements of different work orders, supporting flexible switching of multiple process routes for new energy transformers. The model uses an attention mechanism to fuse multimodal features and a graph neural network to capture the relationship between production equipment, production materials, and products to be produced, improving the accuracy and rationality of production planning.

[0029] 4. This invention achieves precise control of the production process through a multi-level analytical control model, effectively breaking down each production process and control level into multiple control levels. This enables refined and flexible production control, improves the responsiveness and stability of the production system, and ensures that every link from raw material preparation to final product is under control. Furthermore, based on the scheduling of control execution instructions, it can flexibly respond to any changes in production, ensuring smooth, timely, and high-quality production.

[0030] 5. The present invention performs production quality assessment based on a two-dimensional cloud model, which can handle the ambiguity and randomness in quality assessment, improve the accuracy and reliability of quality assessment, and use a distributed database to record production process data and finished product inspection results, thereby realizing the rapid location and traceability of quality problems, and helping to take improvement measures in a timely manner to improve product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a schematic diagram of the overall structure of the present invention;

[0032] Figure 2 Schematic diagram of the process of the improved deep learning model in the present invention. DETAILED DESCRIPTION

[0033] The following is a combination of the embodiments of the present invention Figure 1 To the attached Figure 2 The technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0034] The embodiment of the present invention discloses a flexible automated production control system for new energy high-power high-frequency transformers, as shown in the attached Figure 1 As shown in the figure, a five-layer architecture is used to implement closed-loop management from order receipt, resource scheduling, production execution, and quality control, including:

[0035] The production demand receiving module receives production orders from the customer system through an application programming interface (API) or electronic data interchange (EDI). These orders include production requirements for new energy high-power high-frequency transformers, such as order number, product type, quantity, process parameter requirements, and production time. Edge computing nodes are used to decompose and classify received production orders in real time to generate multiple production work orders. Each work order contains the necessary information to ensure that subsequent processing can be carried out according to specific requirements. Each production work order should include key information such as product type, quantity, process parameters, and production plan time to ensure efficient and accurate production.

[0036] The production resource management module uses IoT protocols (such as the lightweight messaging protocol MQTT, the internet protocol CoAP, or the industrial equipment compatible protocol OPC UA) to collect the operating status of production equipment (such as temperature, pressure, speed, etc.) and the inventory quantity of production materials in real time. Sensor devices monitor the equipment status of the production line in real time to ensure the stability of equipment operation. Graph databases (such as Neo4j) are used to build an association network between production equipment, production materials, and production products to facilitate dynamic adjustment of production plans. This database enables the system to easily query the status of equipment, the inventory quantity of materials, and their flow in the production process, thereby supporting efficient production scheduling and material management.

[0037] The production planning module uses improved deep learning models (such as reinforcement learning or optimization algorithms) to intelligently schedule multiple production orders. Based on the requirements of the production order and the availability of production resources (equipment and materials), this model automatically adjusts the production plan to achieve the optimal production solution. Based on work order information and resource data, it automatically matches the most appropriate production process to reduce production cycle time and improve production efficiency.

[0038] The production execution module utilizes a multi-level analytical control model, breaking down production control into multiple control levels through a layered architecture. Each level corresponds to a different production control task, such as material scheduling, equipment operation, and process control. Based on this model, it generates specific production control execution instructions to guide equipment and material scheduling on the production line, ensuring that production proceeds according to plan.

[0039] The quality control module uses a two-dimensional cloud model to assess quality during the production process. This model can handle uncertainty and ambiguity, making it suitable for quality assessment of complex products such as transformers. A quality score is generated through fuzzy evaluation of product data. A distributed database (such as Hadoop or Cassandra) records real-time data from each production process (temperature, humidity, operating speed, etc.) as well as finished product inspection results. This data can be used to locate and trace subsequent quality issues, helping to quickly identify and resolve them.

[0040] Connections between modules:

[0041] The production demand receiving module serves as the entrance to the system, converting production orders into production work orders and providing basic data for subsequent modules.

[0042] The production resource management module is indirectly connected to the production demand receiving module. It triggers the collection and management of resource data through the received production work order information, and provides resource status information for production planning.

[0043] As the core of the system, the production planning module receives data from the production demand receiving module and the production resource management module, performs production planning and process matching, and outputs production planning results.

[0044] The production execution module receives the output results of the production planning module, and converts the plan into specific control instructions through a multi-level analysis and control model to guide the execution of the production process.

[0045] The quality control module runs through the entire production process, receives production process data from the production execution module, conducts quality assessment and locates and traces quality problems, and provides a basis for continuous improvement of the production process.

[0046] These modules are interconnected through data flow and control signals to form a closed-loop production control system, realizing the automation, intelligence and flexibility of the production process of new energy high-power high-frequency transformers.

[0047] All modules exchange data via unified communication protocols (such as REST APIs and WebSockets), ensuring smooth information flow and real-time feedback to every stage of production. The system automatically adjusts production plans based on real-time data, the output of deep learning algorithms, and control instructions from the execution layer. It also optimizes production when necessary, improving efficiency and reducing unnecessary resource waste.

[0048] The hardware layer includes IoT sensors (temperature and humidity sensors, equipment status monitoring sensors, etc.) and edge computing nodes (for processing real-time data);

[0049] The software layer includes:

[0050] Deep learning framework: TensorFlow or PyTorch for model training for production planning

[0051] Graph database: Neo4j, building relationship graphs between equipment, materials, and products

[0052] Distributed database: Hadoop or Cassandra, for storing and querying large-scale data

[0053] Automated production control: PLC (Programmable Logic Controller) and SCADA system to execute on-site control instructions

[0054] Through such a flexible and automated production control system, the production of new energy high-power and high-frequency transformers can be ensured to be efficient, accurate, and traceable, and it has flexible adaptability to meet different production needs.

[0055] Edge nodes use natural language processors (NLP) to parse production order text to extract key information for optimizing production plans. This process is divided into the following steps:

[0056] 1. Order text analysis (NLP technology application)

[0057] The input production order data usually includes information such as product type, quantity, process parameter requirements and production time, which are presented in text form.

[0058] First, the order text is segmented to identify its components, such as product type and quantity. Using named entity recognition (NER) technology within natural language processing, key information, such as product name, quantity, and time, is extracted. This extracted information is then associated with specific production requirements, such as product type and process requirements, production time, and priority. The extracted information is then structured and converted into a machine-readable data format, such as JSON or XML.

[0059] 2. Preset rule engine for production order classification

[0060] The pre-set rules engine will classify production orders based on the following dimensions: Differentiating between different product types, which may include different production process requirements. Classifying according to each product's specific process requirements (such as temperature, material type, etc.). Classifying orders based on production time priority, such as urgent orders that need to be processed first.

[0061] Use priority rule classification algorithms, such as "orders with production time within 24 hours are prioritized" or "orders of product type A are prioritized"; adopt priority-based scheduling algorithms, such as Shortest Job First (SJF) or Earliest Due Date (EDD); for orders of different urgency, the system will intelligently classify the orders based on the rule engine and generate multiple production work orders.

[0062] 3. Generate production work order

[0063] Based on the categorized orders, the system automatically generates multiple production work orders. Each work order includes the product type, required quantity, production process parameters, production time requirements, urgency, and priority. Once generated, the system can send it to the production line, dispatch production resources, and schedule production based on priority.

[0064] 4. Production Scheduling and Execution

[0065] During actual production, edge nodes can dynamically adjust the production sequence based on the production schedule, ensuring that urgent orders are completed first while also taking into account the production schedule of other orders. Edge nodes can also monitor production status, such as production progress and work order completion, in real time and feed this data back to upper-level systems for further optimization of production scheduling and order management.

[0066] 5. Optimize and learn

[0067] By analyzing historical data, the system can continuously optimize its rules engine and improve production scheduling efficiency. For example, it can adjust priority rules or production work order classification methods based on past order processing results. Through continuous machine learning, the system can optimize text parsing and classification rules based on feedback from different production environments, improving processing efficiency and accuracy.

[0068] The core advantage of this solution is to automatically parse production orders through natural language processing technology to improve production scheduling efficiency, while ensuring on-time delivery through rule engines and priority management.

[0069] IoT protocols include the lightweight messaging protocol MQTT, the internet protocol CoAP, or the industrial equipment compatible protocol OPC UA.

[0070] MQTT (Message Queuing Telemetry Transport) is a lightweight messaging protocol designed for efficient, real-time communication between devices, particularly in bandwidth-constrained environments. MQTT is used to transmit the real-time status of production equipment and inventory data.

[0071] CoAP (Constrained Application Protocol) is an Internet protocol designed for low-power devices and networks, and is commonly used for communication between sensors and actuators. In smart manufacturing environments, CoAP can be used for direct data exchange between devices.

[0072] OPC UA (Open Platform Communications Unified Architecture) is a protocol designed specifically for industrial automation and device compatibility. It is widely used for data transmission and interoperability between devices and enterprise-level systems.

[0073] Graph databases use nodes to represent different entities, such as production equipment, production materials, and products. Each node represents specific information about an entity. Edges represent relationships between entities, such as the usage relationship between production equipment and production materials (e.g., which materials are used by the equipment); the relationship between production materials and product production (e.g., a material is a component of a product); and the relationship between production equipment and product production (e.g., a piece of equipment is used in the production process of a product).

[0074] Real-time data collected (such as the operating status of production equipment and inventory levels of production materials) is used to dynamically update node attributes. This enables the graph database to reflect the real-time status and changes in the production process, facilitating monitoring and optimization of production processes. This data can be obtained from equipment and material management systems via MQTT, CoAP, or OPC UA protocols and updated in real time by the graph database, supporting intelligent decision-making and production scheduling.

[0075] Production equipment transmits real-time operational status and material inventory data to a central system via MQTT, CoAP, or OPC UA protocols. Based on this incoming real-time data, the graph database updates node attributes, such as equipment status, inventory levels, and production progress. These updates provide a precise picture of production status. The graph database's query capabilities enable analysis of equipment utilization efficiency and material consumption, providing data support for production scheduling and equipment maintenance, helping decision makers make more precise adjustments and optimization plans. Specific embodiment 1:

[0077] On a production line, a device (node) may use a specific material (node), and the use of this device may affect the production of the final product (node). In the system, if this device fails or the material inventory is insufficient, the system can provide timely feedback and adjust the production plan or take repair measures.

[0078] Through this implementation plan based on the Internet of Things protocol and graph database, every link of the production line can be perceived in real time, improving the transparency and efficiency of the production process, while helping enterprises achieve intelligent management.

[0079] The improved deep learning model combines multiple advanced technologies, including deep learning, graph neural networks, attention mechanisms, and reinforcement learning strategies to optimize the production scheduling process. Figure 2 As shown, the method includes the following specific implementation steps:

[0080] S1: Data input and preprocessing

[0081] First, production work order information and production resource data are input into the model. The production work order information is converted into a vector representation via the embedding layer. The embedding layer maps the discrete production work order information into a high-dimensional continuous space, facilitating neural network processing. Production resource data is numerically normalized using category encoding. Category encoding converts different types of production resources (such as production equipment and material types) into numerical data that can be used in deep learning models, ensuring data consistency and standardization.

[0082] S2: Feature Fusion and Relationship Modeling

[0083] The embedded production order vectors are fused with the normalized production resource data using an attention mechanism. This mechanism effectively captures important correlations between individual production orders and resources, and weights this information to improve model performance. A graph neural network (GNN) is used to capture the relationships between production equipment, production materials, and the products to be produced. Each node represents a production factor (such as equipment or materials), and the edges between nodes represent their relationships. GNNs capture these complex dependencies through a message-passing mechanism, constructing a joint representation tensor that provides both global and local information.

[0084] S3: Objective function and constraints

[0085] The objective functions are the on-time completion rate of production orders and the production cost. This objective function reflects the performance of the production scheduling system. By optimizing these objective functions, the system can control costs while ensuring production efficiency. The constraints include:

[0086] Production equipment capacity constraints: Ensure that the load of each device during the production process does not exceed its maximum capacity.

[0087] Production material inventory constraints: Ensure that the inventory of materials required for production is sufficient to avoid material shortages.

[0088] Production process sequence constraints: There is a specific execution order between each process, which must be followed to ensure the smooth progress of the production process.

[0089] S4: Generate production scheduling strategy and dynamic adjustment

[0090] A policy network is designed based on reinforcement learning (RL) to generate a production scheduling strategy. This policy network uses reward signals to continuously learn the optimal scheduling strategy, including: prioritizing production work orders, allocating production equipment, and determining production time nodes. When generating the scheduling strategy, a constraint violation penalty is added to penalize results that do not meet the constraints. Through reinforcement learning, the model gradually adjusts during the exploration and optimization process to ensure that the scheduling strategy meets constraints such as equipment, materials, and sequence. Using an attention mechanism, the equipment load balancing strategy is dynamically adjusted based on the real-time collected production resource status, ensuring the appropriate distribution of equipment load during the production process and avoiding resource waste and production bottlenecks.

[0091] S5: Matching production output with process path

[0092] Using the joint representation tensor constructed in the decoder and S2, the model outputs relevant scheduling information for each production work order, including: production process sequence, production work order priority ranking, production equipment allocation, and production time nodes. Through the attention mechanism, it accurately matches the process path and production equipment of each production work order to ensure that each process in the production process has appropriate equipment support and can be smoothly executed according to the established time nodes.

[0093] S6: Model pre-training and parameter update

[0094] Pre-train the graph neural network using historical production data to improve its ability to represent the relationships between production equipment, materials, and work orders. This step helps the model obtain better node representations in the early stages. The initial parameters of the reinforcement learning policy network are set to the optimal policy obtained through pre-training to ensure that the policy network has a good starting point from the outset. Transfer learning methods are used to update the model parameters in real time to cope with the ever-changing production environment and resource status. Through transfer learning, the model can continuously adapt to new production demands and constraints, maintaining efficient production scheduling. Process rules are embedded in the scheduling results to further correct and optimize the model output, ensuring that the generated scheduling plan meets the actual production process requirements.

[0095] This improved deep learning model effectively addresses the complexity of production scheduling by integrating multimodal features, graph neural networks, reinforcement learning, and dynamic adjustment strategies. By fully considering factors such as production order priority, equipment load, material inventory, and process sequence, the model generates efficient and constraint-compliant scheduling strategies, optimizing production efficiency and reducing costs. Furthermore, the application of pre-training and transfer learning techniques ensures the model's adaptability and enables efficient dynamic adjustment in real-world production environments.

[0096] The multi-level analysis and control model can effectively coordinate different links in the production process, ensure the smooth execution of the production plan, and improve the efficiency and flexibility of the production process. It includes the production planning layer, scheduling layer and execution layer.

[0097] The production planning layer develops a global production plan based on the production planning output from the production planning module. This global production plan includes the start and end times of each work order, as well as the routing steps. When developing a global production plan, factors such as material availability, equipment availability, and process requirements must be considered to ensure that each work order can be successfully executed within the appropriate time and resources. The planning layer is the foundation of the entire production process, with the goal of maximizing resource utilization and rationally allocating production tasks.

[0098] The scheduling layer breaks down the global production plan into specific scheduling tasks, including material distribution routing, equipment switching sequences, and capacity balancing between processes. Material distribution routing ensures that raw materials are delivered to the production line in a timely manner via the optimal route, reducing logistics costs and time. During the production process, multiple pieces of equipment may need to switch processes. The scheduling layer optimizes the switching sequence to reduce equipment idle time and improve equipment utilization. The scheduling layer rationally arranges production tasks for each process based on production capacity requirements, avoiding production bottlenecks and ensuring capacity balance across processes.

[0099] The execution layer is responsible for translating scheduling tasks into actual production control instructions. These instructions are then sent to a programmable logic controller (PLC) to execute the corresponding production tasks. During execution, the PLC controls the equipment to initiate the corresponding process and sets relevant process parameters (such as temperature, pressure, and speed) according to process requirements. The execution layer must closely coordinate with actual operations on the production site to ensure that equipment completes tasks according to the scheduling layer's instructions and provide real-time feedback on production progress and status.

[0100] Ensure seamless information flow between the production planning, scheduling, and execution layers. The flow of information between these layers is key to ensuring the smooth implementation of production plans. By introducing intelligent algorithms (such as artificial intelligence and machine learning), we can optimize production planning and scheduling tasks, thereby improving overall production efficiency. Establish a real-time monitoring system to track execution at all levels, promptly identify and resolve issues, and ensure production progress and quality.

[0101] Through this multi-level analysis and control model, the level of refinement of production management can be effectively improved, making production planning more scientific and reasonable, scheduling more efficient, and execution more precise.

[0102] The quality control module assesses the quality of the production process for new energy high-power, high-frequency transformers. Specifically, quality assessment indicators are designed based on the characteristics of these transformers. These indicators may include parameters such as transformer efficiency, stability, and temperature rise. During the transformer production process, real-time data collection is performed on current, voltage, temperature, and vibration. This data helps to understand the quality control status of the production process. Finished product inspection results are also part of the data collection process to determine the final product conformity.

[0103] The two-dimensional cloud model is based on fuzzy mathematics and probability theory. It uses a membership function to assess the acceptance of finished product test results. The core of this model is to statistically determine characteristics such as expectation, entropy, and super-entropy based on historical data. The expected value of the two-dimensional cloud model represents the ideal standard value for a qualified product. Entropy represents the uncertainty or volatility of the data; lower entropy values ​​indicate greater stability in product quality. Super-entropy represents the scalability of the data and reflects the reliability of transformer quality assessments. Based on statistical analysis of historical data, a standard range for qualified products is determined; that is, products are considered qualified if their test results fall within this range. Based on the acceptance probability calculated by the two-dimensional cloud model, products are considered qualified if the acceptance probability falls within the preset acceptance range; otherwise, they are considered unqualified.

[0104] Specific implementation steps:

[0105] Design a data acquisition system that can collect various parameter data such as current, voltage, temperature, etc. during the transformer production process in real time and record the finished product inspection results.

[0106] Based on transformer quality control standards, a suitable quality assessment model is developed. The 2D cloud model can be adjusted to ensure it is suitable for transformer testing under different production batches and conditions.

[0107] Collect past production data and test results, conduct statistical analysis, and calculate parameters such as expectation, entropy, and super entropy to provide a basis for the design of the two-dimensional cloud model.

[0108] Embed the 2D cloud model into the production management system to achieve automated inspection and quality assessment, provide real-time feedback on the probability of passing, and automatically determine product eligibility on the production line.

[0109] Design an automatic feedback mechanism to immediately notify relevant personnel to make adjustments or improvements if the transformer inspection results fail to meet the standards to optimize production quality.

[0110] This method can effectively improve the accuracy and real-time performance of quality control during the production process, help reduce the production of defective products, and improve production efficiency and product reliability.

[0111] The distributed database system uses RFID tags to identify new energy high-power high-frequency transformers and stores them in shards by product batch. Each high-frequency transformer is affixed with a unique RFID tag during the initial production phase. These tags typically contain a unique ID code that can be read using RFID technology. RFID tags should contain basic product information, such as the production work order, production equipment number, and material batch, to quickly track the product's status and origin during the production process.

[0112] During the production process, key data of each product is associated with the RFID tag and stored in the database, including:

[0113] Production work order: The production work order number of each transformer can trace the production plan and related tasks of each product.

[0114] Production equipment number: records the information of the equipment responsible for producing the product to facilitate tracking of equipment status.

[0115] Production material batch: Indicate the batch of raw materials used to ensure the traceability of the raw materials of the product.

[0116] Production process sequence: record the production process of the product and indicate the execution order and execution status of each process.

[0117] Production process parameters: Record the specific parameters used in each process, such as temperature, pressure, time, etc., to ensure quality control of the production process.

[0118] Data can be sharded and stored according to product batches. The following methods can be considered: production data can be dispersed and stored on different database nodes by batch, which can effectively avoid data bottlenecks; if the product production cycle is long, it can be sharded by time (such as day or week), and the production data can be divided into time periods to facilitate load balancing of the distributed database; if there are multiple types of transformer products, they can be sharded according to product type, so that different types of product data are stored on different database nodes, reducing data conflicts and resource competition.

[0119] Use a distributed database architecture, such as Cassandra, MongoDB, or CockroachDB. These systems support distributed storage and high availability for large-scale data. The database should support data redundancy and automatic fault tolerance to ensure stable operation and data loss even in the event of a node failure. Use an RFID data reader module to compare and update the real-time scanned tag information with the database record to ensure data real-time and accuracy.

[0120] At the start of production, each transformer is assigned a unique RFID tag, marking the product's starting information and linking it to information such as the production work order, production equipment, and material batches. During the production process, relevant production process parameters and process sequence information are updated in the database with each step. Simultaneously, data such as production equipment and material batches are also updated. At the end of production, when the product is complete, the information on the RFID tag is linked to final production data (such as quality inspection results and shipment date) and stored, completing the product's full lifecycle tracking.

[0121] A management platform enables production managers to query product status and trace historical data in real time. RFID tag scanning allows for quick access to historical production processes, equipment usage, and other information. The data visualization interface helps decision makers comprehensively monitor, analyze, and optimize the production process. Distributed storage and computing strategies ensure the system can dynamically scale according to production needs and automatically recover from node failures, ensuring high system availability.

[0122] Through these steps, efficient production traceability and data management can be achieved, the transparency, efficiency and controllability of the production process can be improved, and data support can be provided for the quality management and production optimization of new energy high-power high-frequency transformer products.

[0123] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is clearly not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

Claims

1. A flexible automated production control system for new energy high-power high-frequency transformers, characterized in that: include: The production demand receiving module receives production orders for new energy high-power high-frequency transformers through the application programming interface (API) or electronic data interchange (EDI). It uses edge nodes to decompose and classify the received production orders to generate multiple production work orders. Each production work order includes at least the type of product to be produced, the quantity, the process parameter requirements, and the production time information. The production resource management module uses the Internet of Things protocol to collect real-time data on the operating status of production equipment and the inventory quantity of production materials, and uses a graph database to build a network of associations among production equipment, production materials, and production products; The production planning module uses an improved deep learning model to plan the production of multiple production orders. The improved deep learning model schedules multiple production orders based on production order information and production resource data, and matches production processes to each production order. The production execution module uses a multi-level analysis and control model to generate production control execution instructions for production work orders, and controls the scheduling of production materials and production processes, as well as the operation of production equipment based on the generated production control execution instructions. The multi-level analysis and control model decomposes the production control of production work orders into multiple control levels through a hierarchical architecture; The quality control module conducts production quality assessment of new energy high-power high-frequency transformers based on a two-dimensional cloud model, and records the production process data and finished product inspection results of new energy high-power high-frequency transformers based on a distributed database to locate and trace quality problems.

2. The flexible automated production control system for new energy high-power high-frequency transformers according to claim 1 is characterized in that: The edge node uses a natural language processor to perform text parsing on the production order to obtain the type, quantity, process parameter requirements and production time of the products to be produced, and uses a preset rule engine to classify the production order into multiple production work orders based on the dimensions of product type, process parameter requirements and urgency of production time.

3. The flexible automated production control system for new energy high-power high-frequency transformers according to claim 1 is characterized in that: The Internet of Things protocols include the lightweight message transmission protocol MQTT, the Internet protocol CoAP or the industrial equipment compatible protocol OPC UA. The graph database represents production equipment, production materials and production products as nodes in the graph, and represents the usage relationship between production equipment and production materials, the relationship between production materials participating in product production, and the relationship between production equipment participating in product production as edges in the graph. The node attributes are updated based on the real-time collection of production equipment operation status data and the inventory quantity of production materials.

4. The flexible automated production control system for new energy high-power high-frequency transformers according to claim 1 is characterized in that: The working method of the improved deep learning model includes the following steps: S1. Input production work order information and production resource data into the model. The input production work order information is converted into a vector representation through the embedding layer, and the input production resource data is numerically normalized through category coding. S2. Use the attention mechanism to fuse the embedded production work order vector with the normalized production resource data for multimodal feature fusion. Use a graph neural network to capture the relationship between production equipment, production materials, and products to be produced to construct a joint representation tensor. S3. Define the objective function as the on-time completion rate of production orders and production costs, and define the constraints as production equipment capacity constraints, production material inventory constraints, and production process sequence constraints. S4. Generate a production scheduling strategy based on a reinforcement learning policy network and ensure that the output meets the constraints through constraint violation penalties. The scheduling strategy includes production order priority sorting, production equipment allocation, and production time nodes. Then, an attention mechanism is used to match the process path of each production order with the production equipment to generate a process-level task sequence. The equipment load balancing strategy is dynamically adjusted based on the real-time production resource status collected. S5, based on the decoder and the joint representation tensor constructed in S2, outputs the production process sequence, production order priority ranking, production equipment allocation and production time node of each production work order; S6. Use historical production data to pre-train the node representation capabilities of the graph neural network and initialize the reinforcement learning strategy network. Then, use transfer learning to update the model parameters in real time and embed process rules in the scheduling results to correct the model output.

5. The flexible automated production control system for new energy high-power high-frequency transformers according to claim 1 is characterized in that: The multi-level analysis and control model includes a production planning layer, a scheduling layer and an execution layer. The planning layer formulates a global production plan based on the production planning results output by the production planning module. The global production plan includes at least the start time, end time and path process of each work order. The scheduling layer breaks down the global production plan into scheduling tasks. The scheduling tasks include at least production material distribution path planning, production equipment switching sequence and capacity balance between production processes. The execution layer converts the scheduling tasks into control instructions for production equipment execution and sends process start signals and process parameters to the programmable logic controller (PLC).

6. The flexible automated production control system for new energy high-power high-frequency transformers according to claim 1 is characterized in that: The quality control module establishes quality assessment indicators based on the characteristics of new energy high-power high-frequency transformers, and collects production process data and finished product inspection results of current new energy high-power high-frequency transformers. The two-dimensional cloud model evaluates the qualified probability of the finished product inspection results of the current new energy high-power high-frequency transformers through a membership function, and determines the expectation, entropy and super-entropy characteristics of the two-dimensional cloud model, as well as the upper and lower limits of the qualified interval based on historical data statistics. When the qualified probability is within the qualified interval, it is judged as qualified, otherwise it is unqualified.

7. The flexible automated production control system for new energy high-power high-frequency transformers according to claim 1 is characterized in that: The distributed database uses RFID radio frequency identification tags to generate product identifications for the new energy high-power high-frequency transformer products produced, and stores them in shards according to product batches. The product identifications are associated with production work orders, production equipment numbers, production material batches, production process sequences, and production process parameters.

Images