Getting Started with Manufacturing Data Collection: A Practical Guide from Sensors to Dashboard

Identifying Data Collection Targets

Data to collect in manufacturing depends on your objectives. For equipment condition monitoring, vibration, temperature, and current data are essential. For production management, you need production count, cycle time, and defect rate.

In an actual injection molding plant, mold temperature is collected every second, production counts at each product ejection, and vibration data at 100Hz+ for anomaly detection. Sampling frequency must match data characteristics to balance storage costs and analysis accuracy.

Selecting Sensors and Communication Protocols

Sensor Selection Criteria

• Vibration: MEMS accelerometers (cost-effective) or piezoelectric sensors (high precision)
• Temperature: Thermocouples (K-type), PT100 RTD
• Current: CT (Current Transformer) sensors
• Production count: PLC counters, proximity sensors

Communication Protocol Comparison

OPC-UA is the industry standard, most stable for direct PLC communication. Supported by Siemens, Mitsubishi, and most PLCs, offering excellent security and data structuring.

MQTT is a lightweight protocol ideal for IoT sensors and cloud connectivity. Used in architectures where edge gateways aggregate data before sending to brokers (Mosquitto, AWS IoT Core).

REST API is used for MES/ERP system integration and advantageous for batch data transfer.

Building Data Collection Architecture

[Sensors/PLC] → [Edge Gateway] → [Message Broker] → [Time-series DB] → [Dashboard]

Edge Gateway can be built using Raspberry Pi or industrial IPCs running Node-RED or Ignition Edge. It preprocesses data on-site and buffers during network outages.

In an actual automotive parts factory, data from 10 CNCs is collected via OPC-UA, aggregated to 1-minute averages at the edge, then transmitted to cloud via MQTT. This reduced network traffic by 80%.

Choosing Data Storage

Time-series Databases

InfluxDB is open-source, easy to start with, and provides SQL-like query language. Suitable for small to medium factories.

TimescaleDB is PostgreSQL-based, allowing management of both relational and time-series data, facilitating MES integration.

AWS Timestream and Azure Time Series Insights are managed services with no infrastructure burden, but require cost optimization.

Storage Strategy

• Hot storage: Last 30 days, fast queries
• Cold storage: 6+ months, long-term analysis
• Downsampling: Aggregate old data from 1min→10min→1hour intervals

Data Quality Management

Essential Validation Items

1. Missing data handling: Caused by sensor failure or communication loss. Apply forward-fill or linear interpolation
2. Outlier detection: Flag when temperature exceeds physical limits or shows abrupt changes
3. Timestamp synchronization: Sync all devices with NTP server (error < 1 second)
4. Data integrity: Detect packet loss using checksums and sequence numbers

Practical tip: Collect data quality metrics separately. Monitoring collection rate, latency, and error rate enables early problem detection.

Creating Your First Dashboard

With Grafana, you can build your first dashboard within 30 minutes.

3-Stage Dashboard Configuration

1. Real-time monitoring panel: Current equipment status, temperature/current gauges
2. Time-series graphs: Last 8 hours trend, highlighting anomaly periods
3. Aggregate statistics: Daily production, average cycle time, utilization rate

First project recommendation: Collect only temperature and production count from a single machine and create a simple dashboard. Lower complexity enables visible results within 2 weeks, providing a foundation for gradual expansion.

Success Checklist

• [ ] Is data collected continuously for 5+ minutes?
• [ ] Is collection rate above 95%?
• [ ] Does dashboard display real-time data?
• [ ] Do shop floor operators understand the dashboard?

Manufacturing data collection prioritizes continuity over perfection. Start small, learn fast, and improve based on shop floor feedback.