Time-Series Database in Manufacturing

What is a Time-Series Database?

A Time-Series Database (TSDB) is a database specifically optimized for time-series data—measurements that always consist of two parts:

• A timestamp (when was it measured?)
• A value (what was measured?)

Examples in production environments:

• Temperature, pressure, torque, spindle load at second or millisecond intervals
• Machine states (RUN/STOP/FAULT), counter values, cycles
• Energy and utilities consumption
• KPIs like OEE, throughput, or defect rate in short intervals

Unlike traditional SQL databases, a TSDB is designed to performantly store and analyze many measurement points at high frequency over long periods.

Why Do You Need a Time-Series Database on the Shop Floor?

In the digital shop floor context, three hard requirements converge:

1. High Write Load: Many machines, sensors, and lines continuously write data
2. Long History: Data should remain available for months/years for analysis, audits, and optimization
3. Time-Axis Analysis: You want trends, progressions, comparisons ("today vs. last week," "before/after retrofit")—not just individual values

A Time-Series Database is designed exactly for this:

• Compression of time series (less storage, yet precise progressions)
• Fast queries over time windows ("last 24h," "shift 2," "batch 4711")
• Aggregations (min/max/averages, resampling to 1s/1min/5min) directly in the database

TSDB vs. Historian, MES & Traditional SQL

Process Data Historian: A historian is essentially a specialized time-series database with industrial focus (OT protocols, tag management, visualization). → Every historian engine is technically a TSDB, but with additional industrial functionality.

MES / Cloud MES: MES typically stores context-related events (order started, failure, defective part) and aggregated metrics. Raw process values are often kept in the historian/TSDB and connected as needed. → MES = "What happened, with which order/part?" → TSDB = "What did the process curves look like exactly, second by second?"

Relational SQL Database: Can theoretically store time series too, but scales significantly worse for industrial workloads (many tags, high frequency, long history) and requires extensive tuning.

Typical Use Cases for Time-Series Databases in Production

1. Process Monitoring & Troubleshooting

• Trend curves during failures: "What happened just before the breakdown?"
• Overlay of curves ("good" vs. "bad" cycles) for pattern recognition

2. OEE & Performance Analysis

• Derive load profiles, micro-stops, speed losses from signals
• Compare shifts/lines based on real running profiles

3. PAT & Predictive Quality

• Multivariate analysis on process curves (pressure/temperature progressions, force-displacement curves)
• Detect drifts and deviations before quality problems become visible

4. Predictive Maintenance

• Analysis of vibration, current, and temperature progressions to predict wear and failures

5. Energy & Resource Monitoring

• Time series of energy, compressed air, water, gas as foundation for energy efficiency projects and compliance reporting

How Does a TSDB Fit into Digital Factory Architecture?

Typical architecture:

1. Shop Floor / OT: PLCs, sensors, robots, test stands → OPC UA, MQTT, fieldbus, gateways
2. Time-Series Database / Historian:
   • Captures high-frequency data
   • Stores it compressed as time series
   • Provides API/query interfaces
3. MES / Cloud MES & Analytics:
   • Uses aggregated metrics and selected signals for OEE, quality, maintenance, dashboards, alarms, reports
   • Data science/BI tools access the TSDB directly or replicated data (data lake)

Result: Raw data remains in a specialized TSDB while MES and analytics work with processed data. This keeps the system performant and scalable—even when additional use cases (PAT, predictive, digital twin) are added later.