A method and system for lightweight deployment and real-time updating of industrial process models

Abstract

This application relates to a lightweight deployment and real-time update method and system for industrial process models. The method includes: an edge computing service collecting raw data packets from industrial equipment, extracting and generating packet slice features, and uploading them to the cloud to complete device authentication; the cloud issuing protocol parsing instructions and object model configuration instructions based on the authentication results, with the edge device generating a protocol parsing instance and constructing an object model instance accordingly; converting the raw packets into standardized process data and constructing a data feature distribution model; after cloud verification, the model is combined with a matching business rule engine and deployed as a running process model; this model performs business logic calculations on real-time data, monitors cloud configurations, and updates parameters seamlessly. The system monitors data status in real time, generating simulated data to maintain calculations in case of disconnection; if the difference between the packet structure features and the initial slice exceeds a threshold, the instance is automatically reset and reconstructed. This application ensures the continuity and high availability of business calculations under complex operating conditions.

Description

Technical Field

[0001] This application relates to the fields of industrial internet and edge computing technology, and in particular to a method and system for lightweight deployment and real-time updating of industrial process models. Background Technology

[0002] With the rapid evolution of Industrial Internet of Things (IIoT) technology and the deep transformation of smart manufacturing towards short-process production line models, the data value of industrial field equipment is being rapidly explored. Edge computing, as an architecture that pushes computing power closer to the data source, has become a key support for processing high-frequency industrial data and achieving low-latency business response. In complex industrial environments, various sensors and programmable logic controllers (PLCs) and other underlying devices are often produced by different manufacturers, and their communication protocols and data formats exhibit high heterogeneity and fragmentation. How to efficiently collect these multi-source heterogeneous data and transform them into business information with clear engineering semantics, and then build and deploy industrial process models that can directly serve production control, has become a core foundational link in realizing the digital transformation of production lines.

[0003] However, traditional industrial data processing and model deployment methods still have limitations when facing complex and ever-changing industrial environments. Existing edge computing gateways typically require pre-built-in large-scale protocol parsing libraries and object model templates to handle different types of devices. This not only consumes a significant amount of limited hardware storage and computing resources on the edge side, resulting in bloated systems that are difficult to adapt to the resource-constrained lightweight deployment requirements, but also necessitates manual reconfiguration of protocol parameters and business models by professional maintenance personnel when dealing with production line equipment changes or the physical migration of edge gateways to new production lines. This leads to high maintenance costs and a high risk of device access failure due to configuration incompatibility. Furthermore, traditional solutions often rely on cloud-based reconfiguration and restarting of edge service processes to adjust process parameters during production line operation, inevitably causing momentary interruptions in business computing and failing to meet the stringent requirements of high continuity and high availability in industrial control.

[0004] Furthermore, when faced with sudden network fluctuations or interruptions in underlying data acquisition, existing edge process models often lack effective fault tolerance and keep-alive mechanisms. They are prone to triggering a chain of false alarms or even unplanned shutdowns due to the instantaneous loss of data input sources, thus restricting the large-scale reliable application and flexible iteration of industrial process models in complex industrial scenarios. Summary of the Invention

[0005] To address the aforementioned technical issues, this application provides a method and system for lightweight deployment and real-time updating of industrial process models.

[0006] Firstly, this application provides a method for lightweight deployment and real-time updating of industrial process models, employing the following technical solution: Raw industrial data packets are collected from industrial equipment through edge computing services, packet features are extracted and sliced, packet slice features are generated and uploaded to cloud services for device identity authentication. Obtain the protocol parsing instructions and object model configuration instructions issued by the cloud service based on the message slice characteristics after authentication; Based on the protocol parsing instruction, the corresponding parsing algorithm is loaded to generate a protocol parsing instance, and based on the object model configuration instruction, an object model instance matching the current device is constructed; The protocol parsing instance and the object model instance are used to convert continuously input raw industrial data messages into standardized process data, and statistical algorithms are applied to the standardized process data within a preset time period based on a preset acquisition frequency to construct a data feature distribution model. The data feature distribution model is uploaded to the cloud service for verification and confirmation. The data feature distribution model and the business rule engine that match the data feature model are obtained after confirmation by the cloud service and deployed together as a running process model. The running process model is invoked to perform business logic calculations on the standardized process data, and when dynamic business configuration data is detected from the cloud service, the dynamic business configuration data is hot-loaded into the running process model to replace the original configuration parameters. The system monitors the input status and structural features of the original industrial data message in real time. If a data acquisition interruption is detected, the system extracts the data feature distribution model to generate simulated values ​​and uses the simulated values ​​as the input source for the business logic calculation. If the structural feature difference between the original industrial data message and the message slice feature exceeds a preset difference threshold, a reset signal is triggered to re-execute the step of acquiring the original industrial data message.

[0007] By adopting the above technical solutions, a closed-loop adaptive system based on feature self-awareness, cloud-driven on-demand operation, and cloud-edge collaborative hot updates was constructed. Through protocol-independent message slicing and feature extraction, unified access to multi-source heterogeneous industrial protocols was achieved without discrimination. Protocol parsing and object model instances were dynamically constructed using cloud-issued instructions, enabling lightweight deployment of the edge-side process model. Simultaneously, by combining the construction and verification of a data feature distribution model, the system was endowed with data disaster recovery simulation capabilities under line failure anomalies, as well as intelligent comparison and seamless reconstruction capabilities based on message feature differences during production line physical migration or equipment changes. This ensures high availability and scalability of the industrial process model across multiple protocols and scenarios while reducing maintenance costs, achieving maintenance-free deployment and real-time updates in short-process production line scenarios.

[0008] Secondly, this application provides a lightweight deployment and real-time update system for industrial process models, employing the following technical solution: A lightweight deployment and real-time update system for an industrial process model, specifically including a cloud service unit and an edge computing service unit that is communicatively connected to the cloud service unit; The edge computing service unit includes: The soft gateway access module is used to collect raw industrial data packets from industrial equipment through edge computing services, extract packet features and perform slicing processing, generate packet slice features and upload them to cloud services for device identity authentication. The cloud-driven instantiation module is used to obtain the protocol parsing instruction and the object model configuration instruction issued by the cloud service based on the message slice characteristics after authentication; to load the corresponding parsing algorithm based on the protocol parsing instruction to generate a protocol parsing instance, and to construct an object model instance matching the current device based on the object model configuration instruction; The feature modeling module is used to convert continuously input raw industrial data messages into standardized process data using the protocol parsing instance and the object model instance, and to perform statistical algorithm processing on the standardized process data within a preset time period based on a preset acquisition frequency to construct a data feature distribution model. The model deployment module is used to upload the data feature distribution model to the cloud service for verification and confirmation, obtain the data feature distribution model and the business rule engine that matches the data feature model after confirmation by the cloud service, and combine and deploy them into a running process model. The dynamic operation and update module is used to call the running process model to perform business logic calculations on the standardized process data, and when it hears the dynamic business configuration data sent by the cloud service, it hot-loads the dynamic business configuration data into the running process model to replace the original configuration parameters. The abnormal self-healing and reconstruction module is used to monitor the input status and structural features of the original industrial data message in real time. If the data acquisition interruption is detected, the data feature distribution model is extracted to generate simulated values, and the simulated values ​​are used as the input source for the business logic calculation. If the structural feature difference between the original industrial data message and the message slice feature exceeds a preset difference threshold, a reset signal is triggered to re-execute the step of acquiring the original industrial data message. The cloud service unit includes: The feature authentication and instruction delivery module is used to receive the message slice features to perform device identity authentication, and after successful authentication, generate and send the protocol parsing instruction and the object model configuration instruction to the edge computing service unit based on the message slice features; The model verification and synchronization module is used to verify and confirm the data feature distribution model, and after the verification is successful, encapsulate the data feature distribution model and the pre-stored business rule engine and synchronously send them to the edge computing service unit. The configuration management module is used to store and manage dynamic business configuration data, and actively push the dynamic business configuration data to the edge computing service unit.

[0009] Thirdly, this application provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as in any of the methods in the first aspect.

[0010] In summary, this application includes at least one of the following beneficial technical effects: By constructing a dynamic modeling mechanism based on message feature self-awareness and cloud-edge collaboration, it achieves "plug-and-play" and maintenance-free deployment of edge computing services in diverse industrial scenarios such as extrusion and calendering. In practical applications, this solution not only decouples the hardware and software binding relationship by utilizing multi-protocol soft gateways and object model instances, but also achieves numerical simulation disaster recovery during data acquisition anomalies and seamless adaptive reconstruction during production line changes through a data feature distribution model. Furthermore, combined with dynamic hot loading and cluster scaling capabilities, it reduces on-site operation and maintenance costs and computing resource consumption while improving the high availability, real-time response capability, and intelligence level of industrial process models in complex, short-process production lines. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the first process of a method for lightweight deployment and real-time updating of an industrial process model according to one embodiment of this application.

[0012] Figure 2 This is a schematic diagram of the second process of a method for lightweight deployment and real-time updating of an industrial process model according to one embodiment of this application.

[0013] Figure 3 This is a schematic diagram of the third process of a method for lightweight deployment and real-time updating of an industrial process model according to one embodiment of this application.

[0014] Figure 4 This is a schematic diagram of the fourth process of a lightweight deployment and real-time update method for an industrial process model according to one embodiment of this application.

[0015] Figure 5 This is a schematic diagram of the fifth process of a method for lightweight deployment and real-time updating of an industrial process model according to one embodiment of this application.

[0016] Figure 6This is a schematic diagram of the sixth process of a method for lightweight deployment and real-time updating of an industrial process model according to one embodiment of this application.

[0017] Figure 7 This is a schematic diagram of the seventh process of a method for lightweight deployment and real-time updating of an industrial process model according to one embodiment of this application.

[0018] Figure 8 This is a schematic diagram of the eighth process of a method for lightweight deployment and real-time updating of an industrial process model according to one embodiment of this application. Detailed Implementation

[0019] To make the purpose, technical solution, and advantages of this application clearer, the following description is provided in conjunction with the appendix. Figures 1-8 The present application will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the application.

[0020] This application discloses a method for lightweight deployment and real-time updating of industrial process models.

[0021] Reference Figure 1 A lightweight deployment and real-time update method for industrial process models, specifically including: Step S101: Collect raw industrial data packets from industrial equipment through edge computing services, extract packet features and perform slicing processing, generate packet slice features and upload them to cloud services for device identity authentication. In practical applications in industrial settings, edge computing services first need to address the data diversity and uncertainty brought about by heterogeneous devices. In this step, raw industrial data packets can be collected through a soft gateway, and these packets can be sliced ​​without regard to protocol. The core logic is to strip away the differences in the underlying communication protocols and extract only key packet features such as packet header identifiers, data lengths, and verification rules to form packet slice features that are uploaded to the cloud for device identity authentication.

[0022] This process not only enables remote verification of device legitimacy but also provides basic metadata for subsequent protocol parsing, avoiding access failures caused by protocol incompatibility in traditional solutions. For example, when dealing with PLCs using different communication protocols such as Siemens S7 and Mitsubishi FX, the slicing process can uniformly extract key fields that can be used to identify the device type without needing prior knowledge of the specific protocol details.

[0023] Step S102: Obtain the protocol parsing instructions and object model configuration instructions issued by the cloud service based on the message slice characteristics after authentication; After obtaining cloud authentication, the edge computing service receives protocol parsing instructions from the cloud based on the data length contained in the packet slice characteristics, as well as object model configuration instructions based on the packet header identifier contained in the packet slice characteristics. The logic of this step is to establish a cloud-edge collaborative configuration distribution mechanism. The protocol parsing instructions specify how to convert the raw byte stream into readable data, while the object model configuration instructions define the semantic structure and business attributes of the data. Together, they constitute the behavioral guidelines for edge-side data processing.

[0024] Understandably, the cloud service acts as a global intelligent scheduling center. When the cloud receives the packet slice characteristics and confirms the device's legitimacy, it uses its powerful computing resources to perform protocol family matching and device type identification on the slice characteristics. Then, it generates and distributes protocol parsing instructions and object model configuration instructions for the specific device. This cloud-driven on-demand distribution mechanism avoids the need to pre-install a large and redundant full protocol parsing library and object model template in the edge computing service. It realizes lightweight on-demand loading of edge-side parsing capabilities and reduces the consumption of hardware storage and computing power by the edge computing service.

[0025] Step S103: Based on the protocol parsing instruction, load the corresponding parsing algorithm to generate a protocol parsing instance, and based on the object model configuration instruction, construct an object model instance that matches the current device; Specifically, based on the aforementioned protocol parsing instructions, the edge computing service loads the corresponding parsing algorithm to generate a protocol parsing instance and constructs a device model instance matching the current device. This protocol parsing instance is a runtime entity specific to a particular device protocol, capable of identifying register addresses and data types, and converting the underlying byte stream into standard time-series data points with timestamps. The device model instance further maps these data points into triplet structure data containing device identifiers, attribute identifiers, and attribute values, thereby achieving semantic upscaling from physical signals to business data. For example, the holding register value in the Modbus protocol is mapped to the business attribute "extruder temperature," and a specific temperature value is assigned.

[0026] Step S104: Using protocol parsing instances and object model instances, the continuously input raw industrial data messages are converted into standardized process data, and statistical algorithms are applied to the standardized process data within a preset time period based on a preset acquisition frequency to construct a data feature distribution model. Specifically, during continuous data processing, the edge computing service performs statistical algorithm processing on standardized process data within a fixed time window according to a preset collection frequency. This process constructs a data feature distribution model, capturing the data fluctuation patterns of the equipment under normal operating conditions by calculating the mean, extreme values, and variance of numerical attribute values. This model not only reflects the real-time operating status of the equipment but also provides a benchmark reference for anomaly detection and simulation data generation. For example, in the extrusion and calendering process, the stable distribution range of temperature data can be accurately characterized by this model, providing a basis for subsequent quality control.

[0027] Step S105: Upload the data feature distribution model to the cloud service for verification and confirmation, obtain the data feature distribution model and the business rule engine that matches the data feature distribution model after confirmation by the cloud service, and combine and deploy them into a running process model. The data feature model initially constructed on the edge side may have deviations due to the equipment just starting up or unstable operating conditions. Therefore, the data feature distribution model must be validated by the process parameter boundaries preset in the cloud. If the validation passes, the authoritative data feature distribution model can be issued. If the validation fails, a model reconstruction signal is triggered to re-execute the steps of constructing the data feature distribution model.

[0028] Subsequently, the confirmed data feature distribution model is bound, encapsulated, and distributed with specific business logic judgment rules, i.e., the business rule engine. The edge side then deploys the combination of these two in the edge computing service to generate a runtime process model. This combined deployment process binds static data features with dynamic business rules, forming an executable and monitorable complete process model. This enables the edge side to independently perform complex business logic calculations without continuously relying on cloud commands.

[0029] Step S106: Call the running process model to perform business logic calculations on the standardized process data, and when the dynamic business configuration data sent by the cloud service is detected, hot-load the dynamic business configuration data into the running process model to replace the original configuration parameters. Once the running process model is in operation, the edge computing service will continuously call the model to perform business logic calculations on standardized process data. At the same time, it will listen for dynamic business configuration data sent from the cloud and write these configuration parameters directly into the memory space of the running business rule engine through hot loading technology. This will take effect without restarting the service. This design ensures that the edge side responds instantly and imperceptibly when the production line process parameters are adjusted, which greatly improves the flexibility and availability of the system.

[0030] Step S107: Monitor the input status and structural features of the original industrial data message in real time. If data acquisition is interrupted, extract the data feature distribution model to generate simulated values ​​and use the simulated values ​​as the input source for business logic calculation. If the structural feature difference between the original industrial data message and the message slice features exceeds the preset difference threshold, trigger a reset signal to re-execute the step of acquiring the original industrial data message.

[0031] Among them, the edge computing service will monitor the input status and structural characteristics of the original industrial data messages in real time to build a dual adaptive defense mechanism for the system to face complex interference in the industrial field.

[0032] Specifically, in abnormal operating conditions where data acquisition is interrupted, the system does not crash directly or issue blind alarms. Instead, it extracts the constructed data feature distribution model from the running process model and generates simulated values ​​that conform to historical fluctuation patterns based on the mean and variance of the model. This maintains the continuity of business logic calculations and avoids abnormal control commands due to data loss.

[0033] When an edge computing service is migrated to a new production line, if the structural features of the current message are found to differ from the initial slice features by more than a preset difference threshold, such as a continuous mismatch in frame header identifiers or data lengths, a reset signal will be triggered to destroy the existing instance and model, and the adaptive adjustment will start again from the data acquisition step. This enables seamless migration and adaptive protocol adaptation of the edge computing service between different production lines.

[0034] In the above embodiments, a closed-loop adaptive system based on feature self-awareness, cloud-driven on-demand operation, and cloud-edge collaborative hot updates is constructed. Through protocol-independent message slicing and feature extraction, unified access to multi-source heterogeneous industrial protocols is achieved without discrimination. Protocol parsing and object model instances are dynamically constructed using cloud-issued instructions, enabling lightweight deployment of the edge-side process model. Simultaneously, by combining the construction and verification of a data feature distribution model, the system is endowed with data disaster recovery simulation capabilities under line failure anomalies, as well as intelligent comparison and seamless reconstruction capabilities based on message feature differences during production line physical migration or equipment changes. This ensures high availability and scalability of the industrial process model across multiple protocols and scenarios while reducing maintenance costs, achieving maintenance-free deployment and real-time updates in short-process production line scenarios.

[0035] Reference Figure 2 As one implementation of step S101, the steps of collecting raw industrial data packets from industrial equipment through edge computing services, extracting packet features and performing segmentation processing, generating packet segment features, and uploading them to cloud services for device identity authentication include: Step S201: Receive raw industrial data packets collected from industrial equipment through the soft gateway built into the edge computing service; Specifically, in the complex and heterogeneous network environment of industrial sites, edge computing services first need to solve the problem of unified access and identification of data from multiple sources of devices. Therefore, this step uses a built-in software gateway to collect raw data packets from industrial devices. By leveraging the generic protocol compatibility of software-defined gateways, the physical interface and protocol barriers of traditional hardware gateways can be broken down.

[0036] In this step, the soft gateway acts as a unified access layer. It can simultaneously establish direct parsing channels for multiple types of industrial protocols (such as Modbus TCP, OPC UA, Profinet, etc.) and receive raw data packets from hardware gateways from multiple manufacturers (such as PLC communication modules from different brands) that have been converted but not yet standardized. This dual-channel access mechanism ensures that field devices can be seamlessly integrated into the edge computing system, regardless of whether they are directly connected or have undergone secondary conversion.

[0037] Step S202: Perform low-level slicing processing on the original industrial data message, extract the message header identifier, data length and verification rules as message features, and divide the original industrial data message into multiple data slice units; Traditional edge parsing typically involves deep unpacking of complete service messages, requiring the edge side to know the protocol type in advance and load the corresponding complete parsing library. This step deliberately avoids this heavy parsing mode and instead adopts protocol-independent low-level slicing technology. This is because, regardless of the industrial communication protocol, it must follow the basic rules of frame structure at the network transport layer or data link layer, which must include a message header identifier to identify the start of the frame, a length field to define the data boundary, and a verification mechanism to ensure transmission integrity.

[0038] Therefore, by extracting only these three most common structural attributes as message features and dividing long messages into multiple data slice units according to fixed or adaptive strategies, edge computing services can transform massive and unknown raw data streams into standardized data blocks with clear structures and controllable volumes without parsing business semantics, greatly reducing memory usage and computing power consumption on the edge side in the initial stage.

[0039] Step S203: Based on message features and data slice units, generate message slice features that include message header identifier, data length, verification rules and corresponding location information of data slice units; Specifically, in the computer's memory space, a simple slice of data cannot understand its context. Therefore, it is necessary to use the message header identifier, data length, and verification rules that can characterize the protocol framework as metadata, and to strongly bind and encapsulate them with the specific location information of the data slice unit in the original message, thereby generating a structured message slice feature with self-describing capabilities.

[0040] This feature packet actually constructs a lightweight digital mapping of the original complex message. It retains the key structural features that can be used to infer the protocol type, and records the physical topological order of the sliced ​​data. This allows subsequent cloud services or other processing modules on the edge to accurately restore the original context of the data based on this lightweight feature packet, without having to deal with the messy full raw byte stream again.

[0041] Step S204: Upload the message slice features to the cloud service, and use the cloud service to parse and compare the message header identifier and verification rules in the message slice features to perform device identity authentication.

[0042] In this process, after the edge computing service completes lightweight feature extraction, the identity verification step is moved to the cloud service, which possesses powerful computing capabilities and a global device ledger. Upon receiving the packet slice features, the cloud service does not perform complex business data parsing. Instead, it directly performs feature matching on the packet header identifier contained within. This is because the packet header identifier typically contains device manufacturer codes, network node identifiers, or fixed prefixes for specific protocol families. By comparing this identifier, the cloud can quickly identify the device in the global database. Simultaneously, it confirms whether the feature packet has undergone byte-level distortion or malicious tampering during transmission by recalculating or comparing verification rules.

[0043] This authentication method, based on underlying structural features, is not only highly secure, but also allows the cloud to quickly achieve secure access control for massive heterogeneous edge devices within a general computing framework that does not rely on any specific industrial protocol parsing plugins, because the authentication process is completely decoupled from business logic.

[0044] In the above implementation, the heavyweight mode based on deep parsing of specific protocols in the traditional solution is abandoned in the data access and authentication stage. Instead, a soft gateway is used to achieve the indiscriminate aggregation of multi-source heterogeneous data. A protocol-independent underlying slicing and feature extraction mechanism is introduced to extract message slice features with self-descriptive capabilities without touching business semantics. Then, lightweight structural features are handed over to the cloud for efficient security authentication. This not only opens up the data input channel in complex industrial sites with extremely low computing power, but also lays a solid and lightweight data foundation for subsequent edge computing services to achieve configuration-free migration and on-demand dynamic loading of parsing instances.

[0045] Reference Figure 3 As one implementation of step S103, the steps of generating a protocol parsing instance by loading the corresponding parsing algorithm based on the protocol parsing instruction, and constructing a device model instance matching the current device based on the device model configuration instruction include: Step S301: Receive the protocol parsing instruction issued by the cloud service. The protocol parsing instruction includes the protocol type identifier and the corresponding parsing rule template. In the initial stage of device authentication, the edge computing service adopts a protocol-agnostic blind switching mechanism. At this time, the edge side itself does not know what private or standard industrial communication protocol the currently accessed device is using. Therefore, the cloud service with a global device ledger and strong parsing capabilities needs to assume the central role of protocol identification and instruction distribution.

[0046] Specifically, the cloud service accurately identifies the corresponding protocol family by parsing the packet slice features uploaded in the preceding steps, and then generates and issues protocol parsing instructions containing protocol type identifiers and parsing rule templates. This mechanism shifts the computational burden of complex protocol identification to the cloud, ensuring that the edge side can still accurately obtain the parsing basis for specific devices without having to build a huge protocol identification engine.

[0047] Step S302: Based on the protocol type identifier, dynamically load the matching parsing algorithm module from the parsing algorithm library pre-built in the edge computing service; Industrial sites have extremely complex communication protocol systems. If all possible protocol parsing programs are permanently installed in edge computing services, it will consume a lot of valuable edge computing resources, which goes against the original intention of lightweight deployment.

[0048] Therefore, this solution adopts a lazy initialization strategy of on-demand dynamic loading at the edge. When a protocol type identifier with a clear pointer is received, the underlying scheduler of the edge computing service accurately extracts the parsing algorithm module corresponding to the identifier from the pre-built parsing algorithm library and loads it into the running memory. This dynamic plug-and-play module loading method not only avoids the occupation of invalid computing power, but also enables the edge computing service to have extremely strong horizontal scalability. If a new protocol needs to be adapted in the future, only the corresponding module needs to be added to the algorithm library, which completely solves the problem of long redevelopment cycle in the existing technology.

[0049] Step S303: Configure the running parameters of the parsing algorithm module based on the parsing rule template to generate a protocol parsing instance; wherein, the protocol parsing instance is used to convert the original industrial data message into a structured data stream; Understandably, a simple parsing algorithm module is usually just a general processing framework and does not have the ability to directly process data from a specific device. In this case, it is necessary to use the parsing rule template carried in the instruction to inject instantiation parameters into the general module.

[0050] Specifically, the parsing rule template specifies in detail the core operating parameters such as byte order rules, register address offset, and data type conversion coefficients for the current device message. By embedding these parameters into the memory space of the loaded parsing algorithm module, a protocol parsing instance specifically designed to process the data stream of that particular device is generated. This instance, as an independent operating unit on the edge side, can continuously and efficiently transform obscure raw industrial data messages into computer-readable structured data streams.

[0051] Step S304: Receive the object model configuration instruction issued by the cloud service, which includes the device attribute definition set and business logic mapping relationship; Since the aforementioned structured data stream lacks engineering semantics in actual industrial scenarios, it is necessary to assign it business attributes in order to achieve complex data processing. Therefore, the object model configuration instructions synchronously issued by the cloud service contain a set of device attribute definitions and business logic mapping relationships.

[0052] Specifically, the equipment attribute definition set is essentially a standardized blueprint for describing the measurement points, parameters, and states that such industrial equipment should possess. The business logic mapping relationship specifies how these standardized attributes are connected with subsequent processes such as specific business calculations and alarm judgments, providing a complete blueprint for building a digital mapping body with semantic understanding capabilities on the edge side.

[0053] Step S305: Extract device feature identifiers from the structured data stream obtained by protocol parsing instance conversion, dynamically match the device attribute definition set with the device feature identifiers, and generate a device attribute mapping table; This step enables the automated association from the underlying data address to the upper-layer semantic attributes. Since the edge computing service is designed as a stateless node that can move freely between different production lines, it cannot rely on manually pre-configured point mapping tables. Therefore, the system must have self-learning capabilities and accurately extract the device feature identifier that can uniquely identify the source of the data by intercepting the structured data stream output by the protocol parsing instance in real time.

[0054] Subsequently, the abstract attributes in the device attribute definition set issued by the cloud are dynamically aligned and mounted with these specific device feature identifiers, thereby automatically generating a device attribute mapping table in memory that indicates the correspondence between the underlying register address and the upper-layer business attribute name. This completely solves the pain point of manual mapping required due to the mismatch between edge data and cloud service data caused by the movement of field devices.

[0055] Step S306: Based on the business logic mapping relationship and the device attribute mapping table, construct a material model instance that matches the current industrial equipment; wherein, the material model instance contains an extensible configuration interface.

[0056] Specifically, after completing the dynamic binding of the underlying data address and the upper-layer semantic attributes, the system further uses the aforementioned generated device attribute mapping table as the data foundation, and deeply encapsulates it with the business logic mapping relationship carried in the object model configuration instructions. The business logic mapping relationship specifies which calculation rules or state transitions should be triggered when a certain attribute value changes. By integrating the static attribute mapping relationship with the dynamic business processing rules, an object model instance that is logically completely mirrored with the current industrial equipment is finally constructed in the edge computing service. This instance not only knows what data the equipment is currently reporting, but also knows how this data should be calculated and processed in the overall process flow, enabling the edge side to have a complete context environment for independently executing complex process model calculations.

[0057] It should be noted that industrial production is a dynamic process of continuous optimization. Business configuration data such as process thresholds and judgment rules are frequently adjusted as production needs change. By reserving an extensible configuration interface at the design level of the object model instance, the instance exposes a standardized parameter injection channel during operation. When the cloud service generates a dynamic update instruction based on the latest business configuration in the database, the edge side does not need to destroy and rebuild the entire object model instance. Instead, it directly hot-loads the new configuration parameters into the instance through this extensible configuration interface. This architecture design ensures the high efficiency of cloud-edge configuration collaboration from the underlying code level, and realizes real-time seamless parameter updates without interrupting the operation of the process model.

[0058] The above implementation breaks through the limitations of traditional edge computing, which relies on pre-configuration and heavy protocol stacks. By precisely issuing instructions from the cloud, it drives the edge side to dynamically load and instantiate parsing modules on demand. Combined with an automated semantic mapping mechanism based on data flow characteristics, it completes the configuration-free transition from underlying protocol messages to high-order semantic object models with extremely low computing power overhead. At the same time, it opens up the technical channel for cloud-edge collaborative hot updates by leveraging the scalable configuration interface of object model instances. This gives industrial edge computing services plug-and-play capabilities and dynamic business adaptation capabilities in the face of multi-source heterogeneous devices and frequent production line migration scenarios.

[0059] Reference Figure 4 As one implementation of step S104, the steps of converting continuously input raw industrial data messages into standardized process data using protocol parsing instances and object model instances, and performing statistical algorithm processing on the standardized process data within a preset time period based on a preset acquisition frequency to construct a data feature distribution model include: Step S401: Perform parsing operations on the continuously input raw industrial data messages through the protocol parsing instance, extract register addresses and data types, and generate structured data streams; During continuous operation in the industrial field, the edge computing service needs to process the continuous stream of raw industrial data messages in real time. Therefore, it first performs parsing operations on the continuously input raw industrial data messages through the built protocol parsing instance.

[0060] In this embodiment, the protocol parsing instance, as a runtime entity for a specific protocol specification, can accurately strip redundant communication layer information such as message headers, check bits, and control codes according to a predefined frame format, directly locate the effective data payload carrying actual process parameters, and extract key metadata such as register addresses and data types from these payloads. The original byte stream is then reconstructed into a structured data stream with clear field divisions and type definitions. This process realizes the first leap of data from the physical transport layer to the logical representation layer, enabling subsequent processing to operate based on clear address spaces and numerical types, rather than blindly processing meaningless binary sequences.

[0061] Step S402: Perform semantic mapping operation on the structured data stream through the object model instance, match the register address and data type with the pre-configured device attribute identifier, and generate standardized process data; After parsing the underlying data structure, the system further performs semantic mapping operations on the structured data stream through object model instances.

[0062] In this embodiment, the object model instance is pre-configured with a mapping table between device attribute identifiers and hardware registers. This table can dynamically link the cold register addresses and data types in the structured data stream to device attribute identifiers with clear engineering meanings. For example, it can map floating-point data at a specific address to "extruder barrel temperature" or "calender roll speed". This generates standardized process data that is free from the constraints of the underlying hardware protocol. This step realizes the essential transformation of data from machine-readable to business-understandable, providing a unified semantic interface for upper-layer applications.

[0063] Step S403: Based on the preset acquisition frequency within a preset time period, perform statistical algorithm processing on the standardized process data to calculate the data mean, data extreme values ​​and data distribution variance of the standardized process data, and generate a set of statistical feature parameters. Specifically, in order to capture the operating patterns of industrial physical quantities over time, the system performs statistical algorithm processing on standardized process data within a preset time period based on a preset acquisition frequency. The logic behind this step is that instantaneous data at a single point is easily affected by electromagnetic interference or mechanical chatter, resulting in noise and failing to truly reflect the steady-state operating conditions of the equipment. Therefore, through a sliding time window mechanism, data samples are extracted at a fixed preset acquisition frequency within a continuous preset time period, and statistical algorithms are used to calculate the mean, extreme values, and variance of the standardized process data.

[0064] Among them, the data mean represents the baseline operating level of the equipment under the current working conditions, the data extreme values ​​define the physical safety boundary of the equipment operation, and the data distribution variance keenly reflects the severity of micro-fluctuations caused by mechanical wear of the production line or external environmental disturbances. The statistical characteristics of these three dimensions are intertwined and together constitute a set of statistical characteristic parameters that can comprehensively characterize the operating status of the equipment.

[0065] It should be noted that the preset acquisition frequency and preset time period can be determined in conjunction with the signal response bandwidth of the industrial equipment. For example, in the extrusion and calendering process, for slow-changing signals such as temperature and fast-changing signals such as pressure, the preset acquisition frequency is limited to between 10 milliseconds and 1000 milliseconds to achieve a balance between accurately capturing changes in operating conditions and suppressing redundant overhead of edge computing. The preset time period is set to between 30 seconds and 300 seconds, covering several typical operating cycles of the equipment, to ensure that the amount of data sampled within this window has mathematical significance for calculating statistical characteristics such as variance and extreme values.

[0066] Step S404: Perform model building operation based on the statistical feature parameter set, combining the data mean, data extreme values ​​and data distribution variance according to the time series to generate a data feature distribution model.

[0067] Static statistical values ​​can only reflect a slice of the past at a certain moment, while industrial production is a dynamic evolution process. Therefore, it is necessary to connect discrete statistical features in the dimension of time. This combination method is essentially to build a multi-dimensional temporal feature matrix at the edge, so that the data feature distribution model not only records the normal fluctuation range of the equipment at the current moment, but also retains the evolution trajectory of this distribution feature over time, thus forming a unique behavioral baseline for the equipment in a specific production line environment. This data feature distribution model not only provides a data summary for cloud verification, but also provides a scientific mathematical basis for simulation and deduction when data acquisition is interrupted in subsequent steps and for mutation detection during production line migration.

[0068] The above implementation eliminates data silos and understanding barriers caused by the diversity of industrial protocols, and uses statistical algorithms to extract data feature distribution models that characterize the essential behavior of devices on-site at the edge. This model not only elevates the edge computing capability from traditional simple threshold filtering to the level of deep feature cognition, but also provides a solid data foundation to support the realization of configuration-free seamless migration and self-healing disaster recovery in the entire lightweight deployment architecture.

[0069] Reference Figure 5 As one implementation of step S105, the step of obtaining the data feature distribution model and the business rule engine matching the data feature distribution model after confirmation from the cloud service, and combining and deploying them into a running process model includes: Step S501: Obtain the current remaining hardware resource information of the edge computing service; and calculate the target memory allocation space based on the current remaining hardware resource information and the preset running resource requirements of the obtained business rule engine. Edge computing nodes are limited by the physical environment and hardware costs of industrial sites, so their computing and storage resources are usually extremely limited and fixed. Meanwhile, the business rule engines deployed from the cloud often contain complex judgment logic and a large number of rule trees. If an indiscriminate direct deployment strategy is adopted, it is very likely that memory overload will squeeze out the survival space of other core tasks on the edge side (such as high-frequency data acquisition and protocol parsing), which will lead to the system-level collapse of the entire edge computing service.

[0070] Therefore, before executing the deployment action, a dynamic evaluation mechanism based on real-time resource awareness was introduced. By accurately obtaining the current remaining hardware resource information and performing comprehensive calculations on the difference and ratio with the inherent preset running resource requirements of the business rule engine, a memory boundary can be defined for the engine that just meets its bottom line for smooth operation and will never cross the boundary and infringe on the system security red line. Thus, the target memory allocation space is derived in the most scientific numerical calculation method, laying a physical resource-level security guarantee for subsequent reliable deployment.

[0071] Step S502: Configure a memory isolation region based on the target memory allocation space, create a business rule engine runtime container in the memory isolation region, load the business rule engine into the business rule engine runtime container, and generate a business rule engine instance; Specifically, in complex industrial multi-tasking concurrent environments, simply statically allocating memory size is insufficient to prevent runtime resource conflicts and fault propagation. By further configuring memory isolation areas and building runtime containers within the target memory allocation space, a robust sandbox barrier is essentially established at the process scheduling level of the operating system.

[0072] Once the business rule engine is loaded into the enclosed runtime container and transformed into a business rule engine instance, all variable allocation, object instantiation, and thread scheduling within this instance are strictly confined to this isolated area. This means that no matter how complex the process logic is being processed by the business rule engine, even if a serious memory leak or logical infinite loop exception occurs, its destructive power can never penetrate the container boundary to contaminate the main control process or underlying communication link of the edge computing service. This fundamentally ensures the absolute high availability and robustness of edge nodes when loading uncontrollable third-party cloud logic.

[0073] Step S503: Analyze the data feature distribution model issued after confirmation by the cloud service to extract the sub-statistical feature parameter set arranged in time series, calculate the deviation of the statistical parameters corresponding to each time segment in the sub-statistical feature parameter set from the global statistical mean, remove abnormal operating condition parameters whose deviation exceeds the preset outlier threshold, and re-aggregate and calculate the remaining sub-statistical feature parameters after removal to generate a lightweight data feature distribution model. Specifically, for the data feature distribution model issued after confirmation by the cloud service, a secondary purification is performed at the edge without relying on the original massive data. By parsing the data feature distribution model constructed according to the time series, the sub-statistical feature parameters in each small time segment are restored. Then, using the global statistical mean as the benchmark anchor point, the degree of deviation of the parameters in each time segment from the overall steady-state operation center of the equipment is quantitatively evaluated. In this way, those extreme deviation sub-features caused by temporary sensor glitches or occasional abnormal operating conditions of the equipment are accurately identified and removed. Finally, only the remaining sub-statistical features that can truly represent the health pattern of the equipment are used for re-aggregation and calculation. In this way, the occasional noise components carried in the model are directly removed at the edge, generating a lighter data feature distribution model with smaller volume and stronger anti-interference ability.

[0074] In this process, in order to achieve the optimal balance between effectively filtering out abnormal operating conditions and retaining normal process fluctuations, the preset outlier threshold is usually configured to be between 2 and 3 times the historical standard deviation of the corresponding attribute data in the data feature distribution model, so as to ensure that only extreme deviation data that seriously deviates from the boundary of the normal statistical distribution will be judged as abnormal operating condition parameters and removed.

[0075] Step S504: Configure the data communication mapping interface between the lightweight data feature distribution model and the business rule engine instance. Mount the lightweight data feature distribution model to the business rule engine runtime container through the data communication mapping interface. Encapsulate and combine the lightweight data feature distribution model and the business rule engine instance to generate the runtime process model.

[0076] After the independent processing of the preceding steps, there is now a computation execution subject and a data reference object in the system that are securely isolated. The two are physically separated in memory space. The purpose of configuring the data communication mapping interface is to cross this memory isolation barrier and establish a standardized contract specification for cross-sandbox data reading and calling.

[0077] In this embodiment, the lightweight data feature distribution model is seamlessly integrated into the business rule engine runtime container through the data communication mapping interface. This allows the engine instance to directly address and call the parameter features inside the model with extremely low latency when executing each business rule. This deep encapsulation and combination breaks the loose coupling between data modules and logic modules in traditional solutions. It constructs an organic unity with high cohesion and low coupling characteristics, and high synergy between data-driven and rule-based judgment within the limited resources on the edge side. The resulting running process model truly possesses a complete form that can operate independently, efficiently, and lightweightly in harsh industrial environments.

[0078] In the above implementation, a security protection system based on dynamic resource awareness and strict memory isolation is built in the edge computing service. In the combined deployment stage, a model lightweight purification and standardized interface mapping mechanism for edge-side resources is introduced. This not only blocks the potential risk of failure propagation of complex business logic to the underlying core link of the edge through containerization technology, but also significantly reduces the resource overhead of redundant data by eliminating outlier parameters. Under extremely limited edge hardware conditions, a highly secure, low-consumption, and tightly coupled process model cloud-edge collaborative deployment is achieved, which perfectly meets the stringent requirements of short-process industrial production lines for lightweight, highly reliable, and maintenance-free edge control nodes.

[0079] Reference Figure 6 As one implementation of step S106, the steps of calling the running process model to perform business logic calculations on standardized process data, and hot-loading the dynamic business configuration data into the running process model to replace the original configuration parameters when dynamic business configuration data is detected from the cloud service include: Step S601: Call the running process model to perform business logic calculation operations on the continuously input standardized process data and generate business calculation results; The process of injecting preprocessed and feature-modeled data into a computing engine with actual business decision-making capabilities is essentially a key step in transforming abstract data features into specific production instructions or diagnostic results. For example, using trained process parameters to perform calculations on real-time temperature and pressure data can output business calculation results such as equipment health index or process quality score, thereby enabling real-time perception and intelligent control of the production process at the edge.

[0080] Step S602: During the execution of business logic calculation operations in the running process model, a long MQTT connection channel with the cloud service is established to listen for dynamic business configuration data packets sent by the cloud service. In order to ensure that the model can continuously adapt to new process requirements in a complex and ever-changing production environment, the system needs to maintain a dynamic configuration channel with the cloud service during the execution of business logic calculations by the running process model. This can be achieved by establishing a long MQTT connection channel with the cloud service.

[0081] In this embodiment, the MQTT protocol, as a lightweight publish / subscribe message transmission protocol, is particularly suitable for industrial field environments with limited and unstable network bandwidth. By establishing persistent long connections and subscribing to specific configuration update topics, the edge computing service can listen to the dynamic business configuration data packets sent by the cloud service in real time with extremely low resource consumption. The packets are usually encapsulated in JSON format. This mechanism ensures that the cloud's control commands to the edge can reach the edge in milliseconds, realizing real-time synchronization of cloud and edge configurations.

[0082] Step S603: When a dynamic service configuration data packet is detected, a parsing operation is performed on the dynamic service configuration data packet to extract the configuration parameter set; The process of restoring the received serialized string into a structured set of configuration parameters involves the extraction and validation of JSON key-value pairs to ensure the integrity and legality of the configuration data. For example, new threshold judgment criteria, point mapping relationships, or alarm triggering conditions are parsed from the data packet. These parameters often correspond to adjustments in production processes or changes in equipment status. The parsed set of configuration parameters will directly determine the behavior pattern of the model in the next stage.

[0083] Step S604: Dynamically write the configuration parameter set into the memory object of the running process model, inject the configuration parameter set into the business rule engine instance to replace the original configuration parameters, and keep the running state of the running process model uninterrupted.

[0084] To achieve seamless updates of business rules, the system dynamically writes the extracted configuration parameter set into the memory object of the running process model. Through dynamic bytecode manipulation techniques such as Java Agent technology, the new configuration parameters are injected into the business rule engine instance to replace the original configuration parameters, while maintaining the uninterrupted operation of the running process model. The underlying logic is to bypass the traditional restart and loading mechanism and directly modify the business rule object loaded into memory during JVM runtime. This hot-loading technology enables edge computing services to complete algorithm iteration and parameter optimization without stopping the production line, eliminating the risk of production interruption due to system maintenance and ensuring the high availability of the industrial control system.

[0085] In the above implementation, a cloud-edge real-time communication link based on the MQTT protocol is built at the edge, and JavaAgent technology is used to realize hot loading of the business rule engine and dynamic replacement of memory objects. This enables the running process model to seamlessly accept dynamic configurations from the cloud while continuously providing services. This mechanism not only improves the agility and flexibility of the industrial process model in response to changes in production line conditions, but also fundamentally ensures the continuity and stability of the production process, achieving true zero-interruption operation and maintenance.

[0086] Reference Figure 7 As one implementation of step S107, if a data acquisition interruption is detected, the step of extracting a data feature distribution model to generate simulated values ​​and using the simulated values ​​as the input source for business logic calculation includes: Step S701: When a data acquisition interruption is detected, the deployed data feature distribution model is extracted from the storage module of the edge computing service; In complex industrial edge computing scenarios, the sudden interruption of real-time data streams is extremely common due to uncontrollable factors such as network fluctuations, brief sensor failures, or equipment polling gaps. If the downstream process model calculations are stopped at this moment, it will inevitably cause the entire production line control logic to stagnate or even trigger a system-level safety cascading shutdown.

[0087] Therefore, this step designs an active fault-tolerant takeover mechanism in the edge computing service. Once the underlying monitoring logic captures the physical signal of data acquisition interruption, the system does not adopt a passive blocking and waiting strategy, but immediately switches to the local storage module to extract the data feature distribution model that has been verified in the cloud and successfully deployed on the edge side during the early normal operation phase. This action essentially allows the edge computing node to instantly switch from a state that depends on external real-time input to a state that depends on local historical features.

[0088] Step S702: Analyze the set of statistical feature parameters in the data feature distribution model; The statistical characteristic parameter set includes at least one of the following: mean, variance, range, and piecewise parameters of the probability density function. These parameters are not merely static numerical values; together they define a multidimensional probability space. The mean determines the central tendency of the simulated data, the variance and range define the fluctuation range of the data, and the piecewise parameters of the probability density function more precisely describe the data clustering regions under non-normal distribution. By analyzing these parameters, the system can accurately reproduce the statistical distribution pattern of the original industrial data.

[0089] Step S703: Based on the data distribution pattern represented by the statistical feature parameter set, perform numerical generation operation within a preset time window to generate a simulated numerical sequence. Specifically, an approximate normal distribution feature is generated by fitting a set of statistical feature parameters. Within a preset time window, a numerical generation operation is performed by a random sampling algorithm based on the approximate normal distribution feature, thereby generating a simulated numerical sequence.

[0090] Specifically, in disaster recovery scenarios where edge data acquisition is interrupted, a set of statistical feature parameters (such as mean and variance) is used as the core parameters. Based on the statistical characteristics of industrial steady-state data approaching a normal distribution, an approximate normal distribution feature curve representing the normal fluctuation range of the equipment is fitted. Then, within a set preset time window, a lightweight random sampling algorithm on the edge side is called to continuously extract discrete values ​​according to the time series using the fitted distribution feature as the probability density benchmark. This allows for the rapid generation of a set of simulated numerical sequences that can highly simulate the historical healthy operating trajectory of the equipment with extremely low computational overhead, ensuring a smooth transition and continuous operation of downstream business logic calculations during the outage period.

[0091] Step S704: Use the simulated numerical sequence as the input source for business logic calculation to replace the standardized process data that is missing due to data acquisition interruption.

[0092] In particular, since the simulated numerical sequences generated by the aforementioned steps based on rigorous statistical principles are extremely similar to real physical data in terms of temporal continuity and numerical distribution characteristics, when the running process model receives these data, its internal data verification logic and business judgment logic cannot distinguish whether the source is real hardware sensors or algorithm simulation.

[0093] Therefore, by seamlessly integrating the simulated numerical sequence into the input interface of business logic calculation, the data gap caused by the interruption of data acquisition is filled. Since the simulated numerical sequence is consistent with the real data in terms of statistical characteristics, the business rule engine in the running process model cannot perceive the change in the data source. Thus, it can continue to perform operations such as process control, quality judgment or equipment warning without changing any calculation logic, ensuring the transparency of the upper-layer application to the underlying data anomalies and maintaining the continuity and stability of production line control.

[0094] The above implementation effectively solves the problem of data acquisition interruption caused by network fluctuations, sensor failures and other reasons in industrial sites. By using the data feature distribution model stored locally to generate simulated values ​​based on normal distribution, it realizes seamless takeover and smooth transition of edge-side business logic computing in the event of missing data sources. This disaster recovery mechanism not only ensures the continuous and stable operation of the production control system and avoids false alarms or downtime caused by data interruption, but also enhances the robustness and autonomy of the industrial process model under harsh working conditions.

[0095] Reference Figure 8As one implementation of step S107, if the structural feature difference between the original industrial data message and the message slice features exceeds a preset difference threshold, the step of triggering a reset signal includes: Step S801: Real-time acquisition of the message structure feature set of the currently input raw industrial data message; In complex scenarios such as industrial production line reorganization, equipment replacement, or network topology adjustment, edge computing services must be able to keenly perceive changes in the underlying communication environment. Therefore, it is necessary to first obtain the message structure feature set of the current input raw industrial data message in real time. The logical principle of this step is that the system does not rely on the upper-layer application logic, but goes down to the communication link layer to perform deep packet inspection on the incoming data packets, and captures the frame structure, byte order, and encoding format of the current communication flow in real time, thereby forming a feature snapshot that reflects the current real connection status.

[0096] Step S802: Read the stored message slice feature set from the storage module of the edge computing service; In order to determine whether there has been a substantial change in the current communication environment, the system reads the stored message slice feature set from the storage module of the edge computing service. This feature set represents a legitimate and valid device communication template. The system compares the real-time acquired message structure feature set with this benchmark template. The core logic is to identify abrupt changes in the communication protocol. For example, if the production line replaces an old model PLC with a new model, or if an IP address conflict causes a connection to the wrong device, these changes will produce significant differences in the message structure.

[0097] Step S803: Extract the message header field sequence from the current message structure feature set as the key comparison feature, and extract the corresponding benchmark message header field sequence and benchmark message length from the message slice feature set; Among them, the message header field sequence has extremely high stability and distinguishability in representing the inherent communication identity of the device. Therefore, establishing it as a key comparison feature can effectively eliminate the interference of dynamic fluctuations in business data on structural judgment. The message slice feature set is a snapshot of historical benchmark data that has been accumulated after the device is initially connected and authenticated by the cloud. The corresponding benchmark message header field sequence and benchmark message length are accurately extracted from it. In essence, an absolutely reliable reference system is established in the memory space on the edge side. This step accurately anchors the dimensions of the real-time collected dynamic features and the static historical benchmark, and locks in the most core comparison source for subsequent measurement of whether the device has undergone physical migration or protocol mutation, avoiding meaningless global traversal in the huge and redundant full message data.

[0098] Step S804: Perform equal-length alignment on the message header field sequence and the reference message header field sequence, and calculate the field sequence difference value between the two aligned sequences using the Hamming distance algorithm; In real-world industrial environments with multiple protocols coexisting, even for the same type of message header, there may be slight differences in byte length between the real-time message header and the historical baseline message header due to vendor-specific extensions or dynamic additions or deletions of optional fields. Therefore, it is necessary to introduce an equal-length alignment mechanism to eliminate the blocking obstacles to algorithm execution at the physical level by padding shorter sequences with zeros or truncating and aligning extremely long sequences.

[0099] After the equal length condition is met, the Hamming distance algorithm is called. Its essence is to compare the data differences between two sequences byte by byte or bit by bit at the same physical offset position. This position-sensitive and precise comparison mechanism can extremely accurately capture any subtle structural misalignment inside the message header, such as abrupt changes in function codes or drifts in device addressing addresses, thereby transforming abstract sequence structure differences into extremely accurate field sequence difference values ​​with very low computational overhead.

[0100] Step S805: Calculate the message length difference between the current message length and the reference message length, and perform weighted calculation on the field sequence difference value in combination with the message length difference to generate a normalized feature difference value; Among these, relying solely on the micro-field differences in the message header may still leave blind spots in judgment, because certain complex protocol tampering or production line switching will not only cause changes in the internal encoding of the message header, but also cause the entire message frame to expand or contract at the macro-physical scale. Therefore, it is necessary to introduce a comparison between the current message length and the baseline length to build a macro-dimensional defense verification.

[0101] In this embodiment, a message length difference is generated by calculating the absolute difference between the current message length and the baseline message length. This difference is then used as a penalty factor or weighting coefficient and fused with the microscopic field sequence difference value for weighted calculation. This mechanism ensures that when severe frame structure damage occurs, the difference value exhibits an exponential amplification effect, while applying a gentle weighting effect to small and reasonable length fluctuations. Finally, the weighted composite value is normalized and mapped to a unified dimensional range, eliminating the difficulty in setting thresholds caused by differences in the base length of different protocols. This makes the final output feature difference value a standardized health indicator that combines microscopic bit-level resolution with macroscopic frame-level perception and can be directly used by edge computing services for rigid threshold judgment.

[0102] Step S806: When the feature difference value exceeds the preset difference threshold, a reset signal is generated and the memory resources occupied by the protocol parsing instance are released, the configuration file corresponding to the object model instance is deleted, and the service process of the running process model is terminated to complete the instance clearing. When the calculated feature difference value exceeds the preset difference threshold, the system determines that the current communication environment has undergone irreversible changes, and the original protocol parsing instance, object model instance and running process model have all become invalid. Subsequently, a reset signal is generated and an instance clearing operation is performed.

[0103] Specifically, the system's computing power is reclaimed by releasing the memory resources occupied by the protocol parsing instance, incorrect semantic mapping relationships are eliminated by deleting the configuration files corresponding to the object model instance, the business rule engine running container is stopped, and the service process of the running process model is terminated to cut off the incorrect business output. This series of cleanup operations ensures that the system will not continue to run on the basis of incorrect configuration, and completely avoids production accidents caused by configuration drift.

[0104] It should be noted that the preset difference threshold applies to the normalized feature difference value after weighted calculation of Hamming distance and length difference. Its value range can be set between 0.15 and 0.35. This empirical range has been verified by a large number of heterogeneous protocol switching. It can effectively accommodate microstructure drift caused by network instantaneous jitter or dynamic addition or deletion of protocol private optional fields to prevent frequent system reset. It can also form an absolutely sensitive limit trigger for situations where the core identifier of the message header or the length of the data payload changes fundamentally due to the physical migration of the edge gateway to the new production line. This ensures the accuracy and robustness of the system's adaptive reconstruction.

[0105] Step S807 triggers the edge computing service to re-execute the step of collecting raw industrial data packets.

[0106] After completely eliminating the old instances, the system proactively triggers the edge computing service to re-execute the initial steps of collecting raw industrial data packets. This logic forms a complete closed-loop self-healing mechanism. The system will re-perform packet slicing, cloud authentication, instance building, and model deployment as if it were deployed for the first time. This enables the edge computing service to migrate seamlessly after changes in the physical production line and automatically adapt to the new industrial environment without human intervention.

[0107] The above implementation effectively solves the problem of edge computing service configuration failure and manual reset caused by equipment replacement, production line reorganization or network failure in industrial sites. By introducing a message structure feature difference detection mechanism based on Hamming distance and weighted calculation, it achieves accurate identification and quantitative evaluation of communication environment changes. Combined with atomic instance clearing and process restart logic, it endows the edge computing service with strong environmental adaptability and self-healing ability, ensuring that the industrial process model always maintains the correct mapping relationship and calculation logic in the ever-changing short-process production line scenario, reducing the complexity of system operation and maintenance and manpower costs.

[0108] The following uses an extrusion calendering process model as an example scenario in this application for specific illustration: Since extrusion and calendering processes typically involve dense data interaction among a large number of heterogeneous devices such as temperature, pressure, and rotation speed, a local area network can first be established on the production line based on Alibaba Cloud IoT APN cards to enable network communication between edge computing service clusters, between edge computing services and edge gateways, and between edge computing services and cloud servers.

[0109] Under this network architecture, the edge computing service utilizes a soft gateway to reconcile data from various industrial protocols across multiple scenarios and categories, including Siemens, Rockwell Automation, Omron, Mitsubishi, Modbus, OPC, IEC, and EtherNet / IP. It can also be compatible with data access via hardware gateways from different manufacturers, enabling unified parsing and seamless access to data from multiple manufacturers and various data formats.

[0110] When raw data packets from the extrusion and calendering production line enter the edge computing service, the edge computing service first extracts the features of the incoming data packets and performs segmentation processing, then performs data authentication with the cloud service. After successful authentication, the cloud service issues protocol parsing instructions and object model configuration instructions. The edge computing service, combined with business configuration data such as parsed data configuration and point mapping stored in MySQL, initiates the protocol parsing process and object model configuration process, completing the conversion of the underlying packets into standardized attribute data for extrusion and calendering.

[0111] Subsequently, standardized data continuously enters the edge computing service. Based on the collection frequency, the edge computing service constructs a data feature distribution model reflecting the actual operating conditions of extrusion calendering within a periodic range, and uploads this feature model to the cloud service for confirmation. After confirmation by the cloud service, it is synchronized to the edge computing service for deployment.

[0112] During the actual operation of the extrusion and calendering process model, the edge computing service uploads the calculation results to the cloud server in real time through the MQTT-based data transmission protocol. At the same time, it relies on the cloud-edge configuration collaboration mechanism to obtain the dynamic configuration in the cloud service in real time. For example, when the cloud changes the judgment threshold of extrusion temperature or calendering pressure, the edge computing service can obtain it in real time and directly use the new threshold to judge the data. Moreover, when the edge computing service needs to be moved to other extrusion and calendering production lines, it can directly access the new production line and calculate the data of the new production line without changing any configuration.

[0113] If a data acquisition anomaly occurs during operation, the edge computing service will activate the deployed data feature distribution model to provide simulated numerical services for algorithm applications. When the original messages received by the system change due to factors such as changes in production line equipment, the edge computing service will restart the process of extracting message feature slices and cloud authentication, and re-execute the deployment and application of the edge computing service.

[0114] Furthermore, in complex scenarios involving a large number of connected devices, a single edge computing server can provide real-time data access and processing for thousands of devices. When computing power is insufficient, it can be expanded by adding machines to form a cluster.

[0115] This application also discloses a lightweight deployment and real-time update system for industrial process models.

[0116] A lightweight deployment and real-time update system for industrial process models, specifically including a cloud service unit and an edge computing service unit that communicates and connects with the cloud service unit; Edge computing service units include: The soft gateway access module is used to collect raw industrial data packets from industrial equipment through edge computing services, extract packet features and perform slicing processing, generate packet slice features and upload them to cloud services for device identity authentication. The cloud-driven instantiation module is used to obtain the protocol parsing instructions and object model configuration instructions issued by the cloud service based on the message slice characteristics after authentication; it loads the corresponding parsing algorithm based on the protocol parsing instructions to generate a protocol parsing instance, and constructs an object model instance that matches the current device based on the object model configuration instructions; The feature modeling module is used to convert continuously input raw industrial data messages into standardized process data using protocol parsing instances and object model instances, and to perform statistical algorithm processing on the standardized process data within a preset time period based on a preset acquisition frequency to construct a data feature distribution model. The model deployment module is used to upload the data feature distribution model to the cloud service for verification and confirmation, obtain the data feature distribution model and the business rule engine that matches the data feature model after confirmation by the cloud service, and combine and deploy them into a running process model. The dynamic operation and update module is used to call the running process model to perform business logic calculations on standardized process data, and when it hears the dynamic business configuration data sent by the cloud service, it hot-loads the dynamic business configuration data into the running process model to replace the original configuration parameters. The anomaly self-healing and reconstruction module is used to monitor the input status and structural characteristics of the original industrial data message in real time. If the data acquisition interruption is detected, the data feature distribution model is extracted to generate simulated values, and the simulated values ​​are used as the input source for business logic calculation. If the structural feature difference between the original industrial data message and the message slice feature exceeds the preset difference threshold, a reset signal is triggered to re-execute the steps of acquiring the original industrial data message. The cloud service unit includes: The feature authentication and instruction delivery module is used to receive message slice features for device identity authentication, and after successful authentication, it generates and sends protocol parsing instructions and object model configuration instructions to the edge computing service unit based on the message slice features. The model verification and synchronization module is used to verify and confirm the data feature distribution model, and after the verification is successful, it encapsulates the data feature distribution model and the pre-stored business rule engine and synchronously distributes them to the edge computing service unit. The configuration management module is used to store and manage dynamic business configuration data, and to actively push dynamic business configuration data to the edge computing service unit.

[0117] The lightweight deployment and real-time update system for industrial process models according to an embodiment of this application can implement any of the above methods, and the specific working process of each module in the system can refer to the corresponding process in the above method embodiments.

[0118] In the several embodiments provided in this application, it should be understood that the provided methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for example, the division of a certain module is merely a logical functional division, and in actual implementation there may be other division methods, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0119] This application also discloses a computer-readable storage medium.

[0120] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described above in any of the methods for lightweight deployment and real-time updating of an industrial process model.

[0121] The computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device; the program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0122] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A method for lightweight deployment and real-time updating of an industrial process model, characterized in that, The method includes: Raw industrial data packets are collected from industrial equipment through edge computing services, packet features are extracted and sliced, packet slice features are generated and uploaded to cloud services for device identity authentication. Obtain the protocol parsing instructions and object model configuration instructions issued by the cloud service based on the message slice characteristics after authentication; Based on the protocol parsing instruction, the corresponding parsing algorithm is loaded to generate a protocol parsing instance, and based on the object model configuration instruction, an object model instance matching the current device is constructed; The protocol parsing instance and the object model instance are used to convert continuously input raw industrial data messages into standardized process data, and statistical algorithms are applied to the standardized process data within a preset time period based on a preset acquisition frequency to construct a data feature distribution model. The data feature distribution model is uploaded to the cloud service for verification and confirmation. The data feature distribution model and the business rule engine that match the data feature model are obtained after confirmation by the cloud service and deployed together as a running process model. The running process model is invoked to perform business logic calculations on the standardized process data, and when dynamic business configuration data is detected from the cloud service, the dynamic business configuration data is hot-loaded into the running process model to replace the original configuration parameters. The system monitors the input status and structural features of the original industrial data message in real time. If a data acquisition interruption is detected, the system extracts the data feature distribution model to generate simulated values ​​and uses the simulated values ​​as the input source for the business logic calculation. If the structural feature difference between the original industrial data message and the message slice feature exceeds a preset difference threshold, a reset signal is triggered to re-execute the step of acquiring the original industrial data message.

2. The method for lightweight deployment and real-time updating of an industrial process model according to claim 1, characterized in that, The steps involved in collecting raw industrial data packets from industrial equipment via edge computing services, extracting packet features and performing segmentation processing, generating packet segment features, and uploading them to cloud services for device authentication include: The edge computing service receives raw industrial data packets collected from industrial equipment through its built-in soft gateway. The original industrial data message is processed by low-level slicing, and the message header identifier, data length and verification rules are extracted as message features. The original industrial data message is then divided into multiple data slice units. Based on the message features and the data slicing unit, a message slicing feature is generated that includes the message header identifier, data length, verification rules, and the corresponding location information of the data slicing unit. The message slice features are uploaded to a cloud service, and the cloud service parses and compares the message header identifier and the verification rules in the message slice features to perform device authentication.

3. The method for lightweight deployment and real-time updating of an industrial process model according to claim 1, characterized in that, The steps of loading the corresponding parsing algorithm based on the protocol parsing instruction to generate a protocol parsing instance, and constructing a device model instance matching the current device based on the device model configuration instruction include: Receive the protocol parsing instruction issued by the cloud service, wherein the protocol parsing instruction includes a protocol type identifier and a corresponding parsing rule template; Based on the protocol type identifier, a matching parsing algorithm module is dynamically loaded from the parsing algorithm library pre-built in the edge computing service; Based on the parsing rule template, configure the running parameters of the parsing algorithm module to generate a protocol parsing instance; wherein, the protocol parsing instance is used to convert the original industrial data message into a structured data stream; Receive the object model configuration instruction issued by the cloud service, which includes the device attribute definition set and business logic mapping relationship; Extract device feature identifiers from the structured data stream obtained through the protocol parsing instance, dynamically match the device attribute definition set with the device feature identifiers, and generate a device attribute mapping table; Based on the business logic mapping relationship and the device attribute mapping table, a material model instance matching the current industrial equipment is constructed; wherein, the material model instance contains an extensible configuration interface.

4. The method for lightweight deployment and real-time updating of an industrial process model according to claim 3, characterized in that, The steps of converting continuously input raw industrial data messages into standardized process data using the protocol parsing instance and the object model instance, and then performing statistical algorithm processing on the standardized process data within a preset time period based on a preset acquisition frequency to construct a data feature distribution model include: The protocol parsing instance performs parsing operations on the continuously input raw industrial data messages, extracts register addresses and data types, and generates a structured data stream; The structured data stream is semantically mapped using the object model instance, matching the register address and data type with pre-configured device attribute identifiers to generate standardized process data. Based on a preset collection frequency and within a preset time period, statistical algorithm processing is performed on the standardized process data to calculate the data mean, data extreme values, and data distribution variance of the standardized process data, and to generate a set of statistical feature parameters. Based on the statistical feature parameter set, a model building operation is performed, combining the data mean, data extreme values, and data distribution variance according to the time series to generate the data feature distribution model.

5. The method for lightweight deployment and real-time updating of an industrial process model according to claim 4, characterized in that, The steps of combining and deploying the data feature distribution model and the business rule engine matching the data feature distribution model, which are issued after confirmation by the cloud service, into a runtime process model include: Obtain the current remaining hardware resource information of the edge computing service, and calculate the target memory allocation space based on the current remaining hardware resource information and the preset running resource requirements of the business rule engine. Configure a memory isolation region based on the target memory allocation space, create a business rule engine runtime container in the memory isolation region, load the business rule engine into the business rule engine runtime container, and generate a business rule engine instance; The data feature distribution model issued after confirmation by the cloud service is parsed to extract the sub-statistical feature parameter set arranged in time series, calculate the deviation of the statistical parameters corresponding to each time segment in the sub-statistical feature parameter set from the global statistical mean, remove abnormal operating condition parameters whose deviation exceeds the preset outlier threshold, and re-aggregate and calculate the remaining sub-statistical feature parameters after removal to generate a lightweight data feature distribution model. Configure the data communication mapping interface between the lightweight data feature distribution model and the business rule engine instance. Use the data communication mapping interface to mount the lightweight data feature distribution model to the business rule engine runtime container. Encapsulate and combine the lightweight data feature distribution model and the business rule engine instance to generate the runtime process model.

6. The method for lightweight deployment and real-time updating of an industrial process model according to claim 5, characterized in that, The steps of calling the running process model to perform business logic calculations on the standardized process data, and hot-loading the dynamic business configuration data into the running process model to replace the original configuration parameters when the dynamic business configuration data sent by the cloud service is detected, include: The running process model is invoked to perform business logic calculations on the continuously input standardized process data, and business calculation results are generated. During the execution of the business logic calculation operation by the running process model, the dynamic business configuration data packets sent by the cloud service are listened to by establishing an MQTT long connection channel with the cloud service. When the dynamic service configuration data packet is detected, a parsing operation is performed on the dynamic service configuration data packet to extract the configuration parameter set; The configuration parameter set is dynamically written into the memory object of the running process model, and the configuration parameter set is injected into the business rule engine instance to replace the original configuration parameters, while keeping the running state of the running process model uninterrupted.

7. The method for lightweight deployment and real-time updating of an industrial process model according to claim 1, characterized in that, If a data acquisition interruption is detected, the steps of extracting the data feature distribution model to generate simulated values ​​and using the simulated values ​​as the input source for the business logic calculation include: When a data acquisition interruption is detected, the deployed data feature distribution model is extracted from the storage module of the edge computing service; Analyze the set of statistical feature parameters in the data feature distribution model; Based on the data distribution pattern represented by the statistical feature parameter set, a numerical generation operation is performed within a preset time window to generate a simulated numerical sequence. The simulated numerical sequence is used as the input source for the business logic calculation, in order to replace the standardized process data that is missing due to data acquisition interruption.

8. The method for lightweight deployment and real-time updating of an industrial process model according to claim 1, characterized in that, If the structural feature difference between the original industrial data message and the message slice features exceeds a preset difference threshold, the steps to trigger a reset signal include: Real-time acquisition of the message structure feature set of the currently input raw industrial data message; Read the stored message slice feature set from the storage module of the edge computing service; Extract the message header field sequence from the current message structure feature set as the key comparison feature, and extract the corresponding baseline message header field sequence and baseline message length from the message slice feature set; The message header field sequence is aligned to the reference message header field sequence by equal length, and the field sequence difference value between the two aligned sequences is calculated by the Hamming distance algorithm. Calculate the message length difference between the current message length and the reference message length, and perform a weighted calculation on the field sequence difference value based on the message length difference to generate a normalized feature difference value; When the feature difference value exceeds the preset difference threshold, a reset signal is generated and the memory resources occupied by the protocol parsing instance are released, the configuration file corresponding to the object model instance is deleted, and the service process of the running process model is terminated to complete the instance cleanup. This triggers the edge computing service to re-execute the step of collecting raw industrial data packets.

9. A lightweight deployment and real-time update system for industrial process models, characterized in that, The system is used to perform a lightweight deployment and real-time update method for an industrial process model as described in any one of claims 1 to 8, the system comprising a cloud service unit and an edge computing service unit communicatively connected to the cloud service unit; The edge computing service unit includes: The soft gateway access module is used to collect raw industrial data packets from industrial equipment through edge computing services, extract packet features and perform slicing processing, generate packet slice features and upload them to cloud services for device identity authentication. The cloud-driven instantiation module is used to obtain the protocol parsing instruction and the object model configuration instruction issued by the cloud service based on the message slice characteristics after authentication; to load the corresponding parsing algorithm based on the protocol parsing instruction to generate a protocol parsing instance, and to construct an object model instance matching the current device based on the object model configuration instruction; The feature modeling module is used to convert continuously input raw industrial data messages into standardized process data using the protocol parsing instance and the object model instance, and to perform statistical algorithm processing on the standardized process data within a preset time period based on a preset acquisition frequency to construct a data feature distribution model. The model deployment module is used to upload the data feature distribution model to the cloud service for verification and confirmation, obtain the data feature distribution model and the business rule engine that matches the data feature model after confirmation by the cloud service, and combine and deploy them into a running process model. The dynamic operation and update module is used to call the running process model to perform business logic calculations on the standardized process data, and when it hears the dynamic business configuration data sent by the cloud service, it hot-loads the dynamic business configuration data into the running process model to replace the original configuration parameters. The abnormal self-healing and reconstruction module is used to monitor the input status and structural features of the original industrial data message in real time. If the data acquisition interruption is detected, the data feature distribution model is extracted to generate simulated values, and the simulated values ​​are used as the input source for the business logic calculation. If the structural feature difference between the original industrial data message and the message slice feature exceeds a preset difference threshold, a reset signal is triggered to re-execute the step of acquiring the original industrial data message. The cloud service unit includes: The feature authentication and instruction delivery module is used to receive the message slice features to perform device identity authentication, and after successful authentication, generate and send the protocol parsing instruction and the object model configuration instruction to the edge computing service unit based on the message slice features; The model verification and synchronization module is used to verify and confirm the data feature distribution model, and after the verification is successful, encapsulate the data feature distribution model and the pre-stored business rule engine and synchronously send them to the edge computing service unit. The configuration management module is used to store and manage dynamic business configuration data, and actively push the dynamic business configuration data to the edge computing service unit.

10. A computer-readable storage medium, characterized in that: The computer program is stored that can be loaded by a processor and executed as described in any one of claims 1 to 8.

Images