1.2 Interoperability and Governance
A system architecture without governance is not an architecture; it is a "spaghetti topology" of fragile point-to-point connections. In a high-volume manufacturing environment, interoperability is the discipline of defining boundaries and contracts. If System A writes directly into the database of System B, you have violated the boundary. If System A changes a message format and crashes System B, you have violated the contract.
This chapter establishes the "Constitution" for how systems in your landscape (ERP, MES, SCADA) coexist. These rules are non-negotiable architectural constraints.
Architectural Topology Rules
Stop building fragile bridges. Enforce Decoupling.
Rule 1: Abolish DB-to-DB Integration.
- Prohibition: Never allow an external system to execute INSERT, UPDATE, or DELETE directly on another system's SQL database.
- Why: This bypasses business logic validation (e.g., checking if a Part Number exists before creating a Work Order). It creates "Zombie Records."
- Mandate: All integration must occur via an Abstraction Layer (API, Message Broker, or Enterprise Service Bus).
Rule 2: The "Hub" vs. "Mesh" Decision.
- Constraint: Avoid direct mesh connections (System A ↔ System B, A ↔ C, B ↔ C). This scales with N(N-1)/2 complexity.
- Mandate: Use a Hub-and-Spoke or Unified Namespace (UNS) pattern. Systems publish events to a central broker (MQTT/Kafka) or call a central API Gateway.
- Benefit: When you replace the ERP, you only update one connector at the Hub, not 50 distinct point-to-point scripts.
The Interface Control Document (ICD)
The ICD is the legal contract between two architectural blocks. No integration code is written until the ICD is signed.
Mandatory ICD Components:
- Transport Protocol: (e.g., HTTPS REST, MQTT, TCP Socket).
- Directionality: Who initiates? (Push vs. Pull).
- Authentication: API Key, OAuth, or Certificate.
- Schema Definition: The exact payload structure (JSON/XML).
- Strict Typing: Define quantity as Integer, not String.
- Unit of Measure: Define temperature as Celsius, not just 240.
- Error States: How does the system signal failure? (HTTP 400 vs 500).
Pro-Tip: Store ICDs in a Git repository alongside the code. They are living documents, not PDFs buried in SharePoint.
Semantic Governance: Naming & IDs
If you cannot uniquely identify an object, you cannot control it.
Naming Strategy: The ISA-95 Hierarchy
Do not invent names. Use the physical hierarchy to create logical namespaces.
- Format: Site/Area/Line/Cell/Device
- Example: MEX01/SMT/Line04/PickPlace02/Feeder12
- Why: This allows you to aggregate data logically. A query for MEX01/SMT/* returns all SMT performance for the site.
Identity Strategy: The Immutable UID
- The Problem: Vendor Serial Numbers are not unique globally. A resistor reel from Vendor A might have the same ID as a capacitor reel from Vendor B.
- The Mandate: Generate an Internal Unique Identifier (UID) at the point of entry (Receiving).
- Implementation: UUID (e.g., 550e8400-e29b...) or a prefixed integer (UID-999999). Use this Internal UID as the Primary Key in all database relations.
Temporal Governance: Time Synchronization
Distributed systems essentially function as a time-machine. If clocks drift, cause-and-effect logic breaks.
The NTP Mandate
- Master: Deploy a local Stratum 1/2 NTP Server in the OT Network.
- Drift Tolerance: ±500ms max.
- UTC Standardization:
- Storage: All timestamps in databases and logs must be UTC (ISO 8601).
- Display: Convert to Local Time only at the UI layer (Operator Screen).
- Risk: If you store Local Time, a Night Shift crossing Daylight Savings Time will duplicate or lose one hour of production data.
Message Resilience & Versioning
Assume the network will fail. Assume the API will change.
Versioning Policy
- Rule: Never break the contract.
- Implementation: Use Semantic Versioning in the Endpoint.
POST /api/v1/work-order(Legacy)POST /api/v2/work-order(New Feature)
- Deprecation: Maintain support for
v-1for a minimum of 6 months.
Error Handling & Idempotency
- Scenario: MES sends a "Consumption" message to ERP. ERP receives it but the Ack is lost. MES retries.
- Risk: ERP deducts materials twice.
- Mandate: The Receiver must be Idempotent. It must check the
Message-ID. If it has already processedMsg-101, it simply returns "Success" without re-processing the transaction.
Store-and-Forward (Buffering)
- Constraint: Network partitions are inevitable.
- Mandate: All Edge Gateways and MES interfaces must buffer messages locally (Disk/Queue) if the upstream connection is lost.
- Recovery: When connection restores, flush the buffer in FIFO (First-In, First-Out) order to preserve sequence.
Final Checklist: Governance Readiness
Governance Pillar | Control Point | Mandatory Standard | Engineering Consequence |
Contract | ICD | Signed & Versioned | Prevents "Tribal Knowledge" integrations that are unmaintainable. |
Topology | Decoupling | No Direct SQL Access | Protects data integrity and allows independent system upgrades. |
Time | NTP Sync | UTC + Local NTP | Guarantees accurate sequence of events for genealogy. |
Naming | Namespace | ISA-95 Structured | Enables scalable analytics and clear asset management. |
Resilience | Retry Logic | Idempotent Receiver | Prevents double-counting inventory during network jitters. |
Versioning | API Lifecycle | Explicit (v1, v2) | Prevents "Big Bang" deployments; enables safe rollouts. |