GB/T 44122-2024Industrial internet platform - Guidelines for the development of industrial mechanisms model (English PDF)

3 Terms and definitions

3.1 industrial internet platform: a system that answers the digital, networked and intelligent needs of manufacturing, that builds cloud-platform-based services for collecting, aggregating and analysing mass data, and that supports wide connection, elastic supply and efficient allocation of manufacturing resources. The definition is cited from GB/T 42569-2023, 3.1.

3.2 industrial application, industrial App: software that carries industrial knowledge and experience in the form of best practice, is aimed at the industrial field, and answers specific business needs in scenarios such as research and design, production, business management and operation and maintenance. The definition is cited, with modification, from GB/T 42562-2023, 3.1.

3.3 industrial mechanisms model: a mathematical model built with laws, theorems and principles, covering the processes, materials, equipment and other manufacturing technology methods involved in industrial manufacturing and management, able to reveal the inherent scientific rules and technical principles of the manufacturing process, and formed into a program through algorithm development.

5 Overview

The industrial mechanisms model is described as an element of the industrial internet platform that gathers the principles, laws and theorems of an industrial production process, combines them with practical production experience, forms a mechanism and builds it into an interpretable model embedded in the platform.

By refining and encapsulating industrial experience and knowledge, such models drive the accumulation of sector knowledge in the industrial PaaS layer of the platform.

The capability description of a model expresses the elements a model needs during development and the flow it follows. Development recommendations are offered across the whole course of requirements analysis, design, development, testing, deployment with operation and maintenance, and change management.

The model management platform is the digital carrier of the models: a one-stop service platform gathering model development, management, data processing, updating and assessment, which offers the application layer a unified calling interface in the industrial PaaS layer. Recommendations are offered on the service capabilities of that platform for integrated data management, model development, model management, model service and operations management.

For ease of management, ten classification dimensions covering sector, business scope and product life cycle, application scenario and others are given in Annex A.

6 Capability description of an industrial mechanisms model

The capability description should carry the following elements, as shown in Figure 1: model name, model code, model classification, model function, model principle, model ownership, model input data, the mechanism model itself, model output result, model version, running environment and the effect of the model in use.

The model input data should include the industrial source data that can be labelled for the application scenario or object: intrinsic geometry, dynamic response, material, process, quality, management, environment and scenario.

The mechanism model should include the mathematical model built from the theorems, laws, algorithms, principles and knowledge and experience involved in the industrial manufacturing process, and should take account of the accuracy, completeness, complexity, real-time behaviour, reliability, stability and extensibility of the model, so as to secure the effect of the model in use.

The model output result is the control parameter sent to an external output device, or the data output to an industrial scenario application, after calculation and analysis by the mechanism model.

7 Development steps of an industrial mechanisms model

7.1 Requirements analysis. The stage settles the object described, the scenario needed, the problem to be solved and the function to be delivered, so that developers and users reach a common understanding. The points listed include: making the described object clear, covering business needs and the purpose of the model object, with the mechanism relationships combed through and an initial model target set; making clear the actual state of the source data, the key performance indicators, the confirmed dimensions and the data domains; making clear the application scenario and functional needs, including the scenario, the problem solved, the expected benefit, the function, the mode of use with the degree of human involvement and interaction with the outside, the data types, the sampling frequency, the historical data available, data quality and availability, the maturity of the business, the range of application and the value of wider use, the software and hardware environment covering chip, storage, operating system and network, and the deployment environment covering temperature, humidity and noise; making clear the requirements on extensibility, performance and security; making clear the reusability and maintainability of the model, including reuse and secondary development, flexible configuration, scenario orchestration, collaboration and management needs; making clear system integration, module calls, layered interfaces, coupling and the degree of standardisation; making clear the development effort, development cycle and operating cycle; and producing a proper requirements analysis document.

7.2 Design. Working from the points of 7.1, the stage describes how the model will be realised. The points listed include: characteristic parameters, input and output variables, constraints, performance requirements and deployment conditions; the relationship between source data and the modelled object and the core mechanism the model needs, such as related principles, laws and theorems; the functions and algorithms to be implemented together with the technical justification; the soundness of the positioning, the reliability of the chosen data quality, the applicability and stability of the algorithm and the accuracy of the output; deployment and integration, including how the model is integrated and how the deployment environment is chosen and configured, with support for cloud resource scheduling and containerised deployment; a complete system architecture that is easy to optimise and extend and supports architectures such as DevOps and DataOps for continuous iteration and automated operation; and a proper design document.

7.3 Development. Working from the points of 7.2, the stage carries out efficient and disciplined coding. The points listed include: choosing a mainstream, widely applicable and stable development environment; making clear the public libraries depended on and the models linked or combined with; making clear whether deployment and use are on the cloud, the edge, the device side or in cloud-edge coordination; supporting the ingestion and storage of multi-source heterogeneous data with unified format conversion for model development and validation; rapid encapsulation of functions and classes and rapid iteration of the model; accelerated computation through hardware and algorithm optimisation; a visual modelling development tool; attention to problems common across industrial sub-sectors so that the model suits several scenarios and duplicated development effort is cut; secondary development of the model; and a proper development document.

7.4 Testing. The stage validates that all the requirements of 7.1 have been correctly implemented and that defects are identified and handled before deployment. The points listed include: testing for functionality, performance efficiency, compatibility, usability, reliability, information security, maintainability, portability and the user documentation set, with the content verified in accordance with GB/T 25000.51-2016; validating the multi-modal data adaptation of the model; validating the soundness and accuracy of the model in different application scenarios; validating the match and smooth running of the model with the models linked or combined with it; validating performance, including computing efficiency, throughput and resource overhead; validating the stability of the algorithm against data disturbance; and a proper test document.

7.5 Deployment, operation and maintenance. The stage aims at better use of resources and a disciplined operation and maintenance process. The points listed include: making clear the duties of provider, management platform and user; supporting cloud, edge, device-side and cloud-edge coordinated deployment; supporting application creation, continuous integration, continuous release, scale-out, scale-in, monitoring and version rollback; supporting different access control capabilities, including rate limiting, request size limiting, metering, bot detection and IP geographic range; supporting calling and downloading of the model; supporting container image deployment, fast container start-up and environment isolation; meeting the network and data security protection requirements of the system; offering an effective channel for user feedback so that the model is iterated over its whole life cycle, including feedback on model accuracy and on false positive and false negative rates; offering a complete and detailed operating manual with worked guidance; and a proper deployment and operation record document.

7.6 Change management. The stage tracks and releases new iterations of the model. The points listed include: setting up mechanisms for model name, version change, release, deployment, review and calling; setting up standards for model differential management and model information management; managing the data produced during validation and parameter adjustment; before a change begins, confirming how the model is used and what the change will affect, running a trial in the test environment, and executing the change under the internal change mechanism after approval so that production is not disturbed; supporting downward compatibility, with the external service address and interface unchanged on image upgrade so that the upgrade is imperceptible; recording all change information and change effects in full and quantifying the weight of each change factor on production and cost; and a proper change management document.

8 Industrial mechanisms model management platform

8.1 The core functions should cover five parts, shown in Figure 2: data management, model development, model management, model service and operations management.

8.2 Data management covers data ingestion, storage, processing, output and sharing. The points listed include: ingestion and management of relational databases, non-relational databases, object storage and static data sources; carrying a certain volume of data; online preview, editing and pre-processing; format conversion, de-noising, de-duplication, correction and completion of ingested data; cleaning of abnormal data; adaptation and conversion of common protocols; data version management covering the evolution of a dataset and the switching and deletion of versions; meeting the timeliness needs of different scenarios; simulation of multi-source heterogeneous data; different data output interfaces; multi-level classified management of data with adjustable classification; and data sharing protected by means such as privacy computing.

8.3 Model development covers formula editing, model component configuration, visual modelling, process orchestration, model testing and model submission. The points listed include: graphical development tools and a multi-language runtime; online and offline development with multi-person online collaboration; several modelling operators, with common ones such as set operations, feature engineering, physical and chemical formulas and signal processing pre-installed; user-defined development and editing of operators, such as creating and modifying formulas and parameters; general model components, single-model upgrade, multi-model combination and multi-model orchestration; visual modelling with process orchestration by dragging operators to build a complete mathematical model; scheduling of cloud computing resources; model debugging and assessment so that the model can meet the recommendations of 7.4; the whole development flow from code hosting, compilation and building through testing and application deployment to software release; and multi-level classified management with recommendation of models of the same class.

8.4 Model management covers model library management, model search, model review, model release, model modification and model deletion. The points listed include: model definition, upload, editing, tag management, release, status monitoring, assessment and version management; fast definition, creation, search and use of models; multi-dimensional search for intelligent retrieval and precise calling; management of multi-dimensional and multi-domain models and of a model knowledge base; release of models to the outside as APIs into the model service marketplace; monitoring of model running; flexible orchestration of several models, with the coordination and coupling between them taken into account when several are used in one scenario; one-click deployment and rapid iteration to cloud, edge and device side under unified control; and an overall rating of the model on function, performance and effect in use over different time periods.

8.5 Model service covers the calling or downloading of industrial apps and industrial micro-services through the model service marketplace. The points listed include: release, search and purchase or subscription of model services; provision of services to external programs through API interfaces with the availability state shown; flexible cross-platform calling; hosting of APIs over their whole life cycle, covering both platform APIs and third-party APIs brought in; manual and automatic model update services; and downloadable operating documents, including the operating manual and deployment help.

8.6 Operations management covers user management, permission management, portal management, and subscription and metering management. The points listed include: multi-user use with logical isolation of user resources and data security; user management and identity authentication services giving one-stop management of accounts, users and products; setting up and maintaining a multi-level departmental structure with ordering of departments at the same level, configuration of organisations and posts, and graded authorisation of operators and models to users at different permission levels; whole-life-cycle management of terminals, including authenticated access, deletion, association and disabling; and visual statistical analysis of model views, subscriptions, downloads and calls.

Annex A Classification of industrial mechanisms models (informative)

Table A.1 has three columns: serial number, classification dimension and model category. Ten dimensions are listed and the table reconciles row by row.

By sector: aviation and aerospace, machinery, automotive, electronics, petrochemical and chemical, building materials, light industry, energy, pharmaceutical, metallurgical and materials. By business scope and product life cycle: basic theory, research design and simulation, production process management covering process, materials, warehousing and production management, equipment fault diagnosis, product quality control, service efficiency improvement and logistics optimisation. By application scenario: research and design, manufacturing, quality control, business management, product service, operation and maintenance service, warehousing and logistics, and energy saving and emission reduction. By academic discipline: mechanics covering solids or fluids and multi-body dynamics, electromagnetics, thermodynamics, chemistry, acoustics, nuclear magnetism, computer vision and geometry.

By way of realisation: basic theory, simulation, process, flow logic and data driven. By industrial object: component, equipment, production line, workshop, enterprise and cross-enterprise. By mechanism solving algorithm: parameter table, non-linear equations, ordinary differential equations, partial differential equations, stochastic simulation and machine learning. By way of use: file type and interface type. By business and classification objective: analysis, deduction, prediction, decision and optimisation. By the principle on which the model acts: geometric, covering external shape, size, internal structure, spatial position, assembly relationship and kinematic principles; physical, covering mechanical, electromagnetic, fluid and thermodynamic characteristics; behavioural, covering dynamic function, performance degradation, periodic motion, random disturbance and response mechanism; and rule-based, covering data relationships, expert knowledge, domain standards, historical experience and related criteria.