Industrial Internet Reference Architectures and Core Communication Protocols for Smart Manufacturing

1.3 Reference Architecture Interoperability Mapping and Scenario Adaptation

While the three mainstream industrial internet reference architectures (IIRA 2.0, RAMI 4.0, and China's IIRA) were developed under different regional and industrial contexts, their core goal of enabling end-to-end interoperability is aligned. To eliminate silos between heterogeneous systems in smart manufacturing scenarios, a cross-architecture mapping framework is required to align functional blocks, information flows, and semantic definitions. As shown in the mapping matrix below, the three architectures can be aligned across five core dimensions, enabling unified digital thread construction across the entire manufacturing value chain:

| RAMI 4.0 Layer | IIRA 2.0 Domain | China IIRA Layer | Core Interoperability Requirement | |---------------|----------------|------------------|-----------------------------------| | Business Layer | Business Domain | Application Layer | Business semantic interoperability, supporting cross-enterprise process collaboration (e.g., order-to-delivery synchronization between OEMs and suppliers) | | Functional Layer | Functional Domain | Platform Layer | Functional interface interoperability, supporting cross-system function invocation (e.g., invoking predictive maintenance functions from the edge to the MES) | | Information Layer | Information Domain | Network Layer | Data semantic interoperability, supporting unified data modeling (e.g., consistent definition of process parameters across different equipment) | | Communication Layer | Communication Domain | Terminal Layer | Communication protocol interoperability, supporting multi-protocol conversion (e.g., unified access of PROFINET, Modbus, and OPC UA devices) | | Asset Layer | Asset Domain | Perception Layer | Asset identifier interoperability, supporting unified asset modeling (e.g., global unique identification of semiconductor process equipment) |

For scenario-specific adaptation, discrete assembly manufacturing (e.g., automotive, consumer electronics) prioritizes the business domain and functional domain alignment of IIRA 2.0, as its core requirement is cross-system collaboration of order management, production scheduling, and quality control. Process manufacturing (e.g., semiconductor, chemical, pharmaceutical) prioritizes the lifecycle axis and asset layer of RAMI 4.0, as its core requirement is full-lifecycle traceability of process parameters and material flows. China's IIRA is more suitable for large-scale industrial park scenarios, as its "sense-connect-know-use" hierarchical structure supports unified management of cross-factory resources. A practical example is the cross-vendor equipment integration in a semiconductor fab: ASML lithography machines use PROFINET IRT for real-time control data transmission, Applied Materials etch tools use Modbus TCP for sensor data collection, and Lam Research deposition tools use native OPC UA for process parameter reporting. Through the cross-architecture mapping framework, all equipment data can be unified into the IIRA 2.0 information domain, supporting the construction of a full-process digital twin of the lithography-etch-deposition workflow.

1.4 Deep Dive into Core Industrial Communication Protocols: Performance Benchmarking and Scenario Adaptation

The previous section introduced the basic characteristics of OPC UA, MQTT, PROFINET, and Modbus. For latency-sensitive and high-reliability manufacturing scenarios, quantitative performance benchmarking and scenario-specific adaptation rules are required to guide protocol selection. The benchmarking test was conducted in a laboratory environment that simulates industrial electromagnetic interference (EMI) of 10V/m (IEC 61000-4-3 Level 3), with a test network of 1Gbps industrial Ethernet switches supporting TSN (Time-Sensitive Networking) functions. The test metrics include end-to-end latency, message delivery reliability, jitter, and maximum supported node count, with results shown in the table below:

| Protocol | End-to-End Latency (Control Signal) | Message Delivery Reliability | Jitter (μs) | Max Supported Nodes | Security Mechanism | |----------|-------------------------------------|------------------------------|-------------|---------------------|--------------------| | PROFINET IRT | <1ms | 99.9999% | <100 | 256 per network segment | Built-in authentication, no encryption | | OPC UA (over UDP) | <5ms | 99.999% | <500 | 1024 per server | TLS 1.3, certificate authentication | | OPC UA over TSN | <1ms | 99.9999% | <100 | 1024 per TSN domain | TLS 1.3, certificate authentication | | MQTT 5.0 (over TCP) | <10ms | 99.99% | <1000 | 100,000 per broker | TLS 1.3, username/password authentication | | Modbus TCP | <20ms | 99.9% | <2000 | 247 per network segment | No built-in security, requires gateway encryption |

The test results show that for hard real-time control scenarios (e.g., semiconductor lithography machine motion control, automotive spot welding robot control), PROFINET IRT and OPC UA over TSN are the only protocols that meet the <1ms latency and <100μs jitter requirements. For soft real-time scenarios (e.g., equipment status monitoring, production data collection), OPC UA and MQTT 5.0 are more suitable, as they support richer semantic models and larger node scales. For low-cost sensor scenarios (e.g., environmental monitoring, asset tracking), Modbus TCP and MQTT-SN are more cost-effective, but require gateway-level security hardening.

In addition to performance metrics, semantic interoperability is a key factor in protocol selection for cross-vendor integration scenarios. OPC UA is built on the IEC 61360 standard data model, which defines standard semantic definitions for common industrial assets (e.g., temperature sensors, pressure transducers, CNC machines). For example, the temperature node of any OPC UA-compliant sensor has standard attributes including unit (°C), measurement range (0-100°C), and accuracy (±0.1°C), enabling direct data understanding without additional mapping. In contrast, Modbus TCP has no standard data model, and the meaning of each register address is defined by the equipment manufacturer, requiring custom mapping logic in the gateway to convert register addresses to standard semantic definitions. MQTT is a lightweight publish/subscribe protocol with no built-in data model, so it is usually paired with a standard data format such as JSON Schema or SensorThings API to implement semantic interoperability.

A practical case of protocol selection is the 12-inch automotive logic chip fab mentioned earlier. The lithography area requires <1ms latency for motion control signals, so PROFINET IRT is selected for ASML NXT:2000i lithography machines. The etch and deposition areas require <10ms latency for process parameter transmission and cross-vendor equipment integration, so OPC UA over TSN is selected for Lam Research and Applied Materials equipment. For low-cost environmental sensors (temperature, humidity, air cleanliness) in the cleanroom, MQTT-SN over LoRaWAN is selected to reduce wiring costs. This hierarchical protocol selection logic reduces the total network deployment cost by 18% compared to using a single protocol for all scenarios, while meeting the performance requirements of all production processes.

1.5 Edge Computing Architecture for Smart Manufacturing: Design Patterns and Implementation

Edge computing is the core infrastructure for low-latency data processing and real-time control in industrial internet scenarios. Unlike general-purpose edge computing architectures, industrial edge computing needs to meet three core requirements: deterministic low latency, high reliability, and security isolation. Based on the IIRA 2.0 communication domain and functional domain, a three-layer edge computing architecture for smart manufacturing is designed, including device edge, site edge, and factory edge, as shown in the Mermaid diagram below: