Choose a TSDB for MES integration, covering OPC UA data collection, OEE calculation, quality traceability, relational data fusion, edge deployment, and historian alternatives.
As smart manufacturing accelerates, the operational data generated by production line equipment is becoming a key asset for factory digital transformation. Efficiently storing, processing, and analyzing this time-stamped data has become an important consideration for manufacturing IT architecture upgrades. Time-series databases, designed specifically for time-series data management, are becoming a strong data foundation for MES (Manufacturing Execution System) integration.
1. Data challenges in smart manufacturing
Modern smart factories face data challenges far more complex than traditional manufacturing. An automotive welding line may deploy hundreds of robots, PLCs, and sensors, with a single line generating hundreds of thousands of data points per second.
Key data characteristics:
High-frequency continuous writes. Equipment parameters are uploaded at millisecond or second-level frequencies. Traditional relational databases show severe performance bottlenecks when facing hundreds of thousands of writes per second.
Strong time correlation. Industrial data value depends heavily on the time dimension. Knowing precisely when anomalies, downtime events, or threshold deviations occurred is essential for root cause analysis.
Demanding real-time requirements. MES monitoring, alerting, and quality assessment functions require data latency from collection to availability to be controlled at the second or even sub-second level.
Varied data lifecycles. Hot data from the past 7 days requires frequent queries. Warm data up to 3 months supports statistical analysis. Cold historical archives need only occasional retrieval but must be kept for years.
2. Core MES requirements for time-series databases
MES sits between the ERP layer and the shop floor, and its time-series data needs concentrate in three dimensions:
2.1 Real-time equipment status monitoring
MES requires real-time awareness of each device’s operating status: running, idle, faulted, or under maintenance. The database must support high-concurrency writes of status data and efficient latest-value queries to power live dashboards and status-driven automation.
2.2 OEE calculation
OEE (Overall Equipment Effectiveness) is a standard metric for measuring production line efficiency, calculated as the product of availability, performance efficiency, and quality rate. This requires strong time-window aggregation capability across hourly, shift, daily, and weekly dimensions, with the ability to drill down into any time period.
2.3 Quality traceability and process analysis
When quality anomalies occur, MES needs to quickly retrieve the process parameter curves for the specific batch and production order involved. The database must support high-performance time-range retrieval and correlation with batch numbers and work order identifiers.
3. Key capabilities to evaluate
3.1 Native OPC UA protocol support
OPC UA is the de facto standard communication protocol in industrial environments. The database should provide a mature OPC UA data collection solution that can subscribe directly to data from SCADA systems or OPC UA servers without intermediate middleware.
3.2 Data model alignment with MES business logic
Industrial data is typically organized around device and measurement point units. The database should support a flexible tag system that allows metadata such as equipment IDs, line names, and process step codes to be attached to each measurement point for efficient filtering and grouping.
3.3 Real-time KPI calculation engine
MES requires extremely high real-time responsiveness for KPIs. The database should have built-in continuous query or stream processing capabilities that automatically execute aggregation calculations as data is written, so that dashboards read pre-computed results rather than scanning raw data.
3.4 Data compression and storage efficiency
Industrial time-series data has significant compression potential due to temporal continuity and value repetition. A capable database should achieve 10:1 or higher compression ratios using columnar storage and specialized encoding, reducing both storage cost and I/O overhead.
3.5 High availability and edge deployment
The system should support cluster deployment with data replicas and automatic failover. Lightweight deployment on edge gateways or industrial PCs at the production line is increasingly important for reducing latency and ensuring local operations during network interruptions.
3.6 Ecosystem integration and SQL compatibility
MES systems are typically built on SQL. A database that supports standard SQL significantly reduces integration cost and shortens the learning curve for the operations team.
4. Data fusion: joint querying of time-series and relational data
Two data types must coexist in an MES environment: time-series data (equipment parameters, sensor readings, energy consumption) managed by the time-series database, and relational data (work orders, BOMs, staffing records, quality inspection results) managed by the RDBMS.
Three integration approaches are available:
Application-layer correlation. The MES application queries both databases separately and joins results at the business logic layer. Simple to implement but increases application complexity.
External table and federation queries. Some time-series databases can map relational tables as external tables, allowing unified SQL queries across both systems while keeping physical storage separate.
Data synchronization and wide-table construction. Key relational dimensions are synchronized into the time-series database’s tag system to build wide tables internally. This approach can deliver the strongest query performance but requires maintaining synchronization mechanisms.
5. Comparison with traditional SCADA and real-time databases
Traditional real-time databases such as Wonderware Historian and AVEVA PI System face growing limitations:
Scalability bottlenecks. Proprietary formats and closed architectures limit horizontal scaling, with high costs for capacity expansion.
Limited analytical capability. These systems were designed primarily for fast storage and retrieval, not for complex analysis such as anomaly detection or trend prediction.
Weak cloud-native support. Mostly monolithic architectures that are difficult to adapt to containerized and microservices-based infrastructure.
Cost and vendor lock-in. Commercial licenses are often tied to tag counts, with costs rising sharply at scale, and proprietary data formats create vendor lock-in risks.
New-generation time-series databases use distributed architectures with horizontal scalability, support cloud-native deployment on Kubernetes, provide open APIs and SQL interfaces that avoid vendor lock-in, and are optimized for time-series analytical workloads. Open-source databases like TDengine offer a cost-effective alternative to traditional real-time database products.
6. Typical deployment architecture
A three-layer architecture is recommended:
Edge collection layer. Edge gateways or collection agents are deployed at the production line, collecting data from PLCs, CNC machines, and sensors via OPC UA, Modbus, and MQTT. The edge layer handles preprocessing including data cleaning, format conversion, local caching, and checkpoint resume to ensure no data gaps.
Time-series data layer. Edge-collected data converges to the time-series database cluster, which handles high-performance writes, compressed storage, and real-time queries. Tiered storage strategies can be configured to manage data across its lifecycle.
MES application layer. MES functional modules built on the time-series database include production monitoring dashboards (real-time equipment status, line takt time, production progress), alert and event management (threshold-based real-time alerts), OEE analysis (scheduled aggregation of equipment efficiency metrics), quality traceability (batch-based process parameter curve queries), and energy consumption management (time-series analysis of energy usage).
This layered architecture decouples data collection, storage, and business logic, allowing each layer to scale and evolve independently.
7. Conclusion
Time-series databases have become important data infrastructure for smart manufacturing MES systems. A five-step action plan is recommended: inventory your data requirements (equipment count, measurement points, collection frequency, retention periods), identify current bottlenecks (write throughput, query latency, storage cost), run POC validation with 2-3 candidate databases using real production line data, design a fusion plan for integrating time-series and relational data, and develop a phased migration plan that maintains business continuity.
The choice of data infrastructure affects a company’s technology options for years to come. For manufacturing IT leaders, the database decision is a critical one that deserves thorough validation.
