Which planning problem the integration solves
In discrete and process manufacturing, the planning and accounting loop is almost always broken. ERP holds production orders, consumption norms, and financial accounting. MES manages execution on the shop floor. SCADA and controllers (PLCs) know the actual line readings every second. Quality systems record defects and deviations. Between these layers, data is most often transferred manually: at the end of the shift, an operator exports a report from MES to Excel, and a cost analyst loads it into ERP the next day.
The business result of this setup is predictable: the plan in ERP is always one shift to one day behind reality. Decisions on extra loading, order rescheduling, and raw material purchasing are made from outdated figures. The integration bus closes exactly this gap: actual output, downtime, and defects reach planning at the moment they occur, not after the fact.
Why ESB instead of point-to-point interfaces
The ISA-95 standard (international equivalent: IEC 62264) formalizes how business systems and the shop floor exchange data. It divides the manufacturing IT landscape into five levels: Level 0 is the physical process, Level 4 is the enterprise/ERP, and MES occupies Level 3 as the operational bridge. The levels operate at different speeds: Level 4 plans in months and weeks, Level 3 in days and hours, and Levels 0-2 react within seconds (Symestic, ISA-95).
This leads to two technical exchange channels. B2MML an XML implementation of ISA-95 for transactional ERP↔MES exchange (production schedules downstream, output reports upstream). OPC UA for continuous shop-floor signals from PLCs to the operations layer. The standard explicitly recommends building standardized connectors instead of custom point-to-point bridges to reduce coupling between heterogeneous systems (Symestic, ISA-95).
Talend ESB implements this approach on an open stack: Apache Karaf (OSGi runtime), Apache CXF (web services), and Apache Camel (routing and mediation) (Talend, Apache ESB). The high-performance bus provides intelligent message routing across the key Enterprise Integration Patterns, and Talend Studio is used to develop and publish mediation routes, data, REST, and SOAP services (Talend ESB Functional Architecture).
How mediation routes and ETL flows are structured
A mediation route in Talend ESB encapsulates Apache Camel capabilities: components `MessageRouter`, `LoadBalancer`, content-based routing, transformation, and a set of endpoints (JMS, file, REST/CXF, JDBC) (Talend ESB Mediation Examples, Manualzz). The key separation is: ESB is suited to real-time processing, while Data Integration (ETL) is suited to batch (Talend, Apache ESB). That is why these tools are used together in manufacturing integration.
A typical landscape looks like this:
- SCADA/PLC → bus. A Camel route subscribes to OPC UA / MQTT line topics, normalizes tags (output, speed, status), and places the event into a JMS queue. A content-based router separates "unit produced" events from "fault/downtime" events.
- Bus → MES. The route enriches the event with order context and writes it to MES via REST/SOAP, recording actual output by operation.
- MES → ERP (B2MML). A mediation route transforms the production confirmation into a B2MML/IDoc message and delivers it to ERP, closing the operation and writing off raw material based on actual consumption. This is a direct analogue of a scenario from an academic study of MES and ERP integration in an automotive supply chain, where bidirectional exchange is implemented through an intermediate document (IDoc) (Tehnički Vjesnik, 2017).
- Quality systems → ERP/MES. A separate flow collects inspection and measurement results; defects and deviation causes are sent to planning as an adjustment to the quantity available for shipment.
- ETL layer. Nightly Talend DI batch flows aggregate shift-level OEE metrics, raw material consumption, and reconcile standards where real time is unnecessary.
ISA-95 defines the semantics here: ERP sends `ProductionSchedule` downstream (a set of orders), and MES returns `ProductionPerformance` upstream with actual output and consumption (Symestic, ISA-95).
Operations and observability
For an industrial landscape, integration observability is critical. Apache Camel and Talend ESB provide JMX interfaces for monitoring routes, endpoints, and error handlers, and the stack is typically complemented by log collection in Elasticsearch/Kibana and the hawtio dashboard (Kai Waehner, Camel & Talend ESB monitoring). Dead-letter queues and retries in Camel address a typical shop-floor issue: temporary unavailability of MES or ERP. The message is not lost and is delivered after recovery.
Architectural principle: leave the core untouched
A key engineering principle of this solution is not to modify the ERP and MES cores. The integration business logic (routes, transformations, enrichment, retries) lives in bus services alongside the systems, not inside them. Mature international standards are used (ISA-95/B2MML, OPC UA, EIP), not custom exchange formats. Loose coupling provides replaceability: the integration team or MES vendor can be changed without rewriting the entire landscape.
What changes in the business process
The main result is not a "configured bus" but a changed planning and accounting process. Before integration, the cycle from "shop-floor fact to ERP figure" took one shift to one day and depended on manual exports. Afterward, an event from the shop floor passes SCADA → MES → ERP in seconds or minutes through standardized routes. The ERP plan reflects actual output, downtime, and defects in real time, dispatchers and cost analysts make decisions using current data, and the person-hours spent reconciling and moving Excel reports disappear from the process.
