Smart Manufacturing
Smart Manufacturing tests the data architecture behind connected plants: ISA-95 levels, industrial protocols, edge versus cloud, digital twins, analytics maturity, and OT security. Items usually ask you to place a function in the right layer or pick the right transport.
Architecture
ISA-95 levels: 0 physical process, 1 sensing and actuation, 2 control (PLC, SCADA), 3 MES for production, quality, and maintenance, 4 ERP for business planning. Data flows upward, commands downward; the Level 3/4 boundary is where execution meets scheduling.
Protocols: OPC UA provides secure, semantic, plant-to-enterprise data with an information model; MTConnect is a read-only, machine-tool-oriented standard; MQTT is lightweight publish/subscribe to a broker; Modbus is simple but neither secure nor self-describing. Edge computing filters, buffers, and runs inference when latency or bandwidth rules out a cloud round trip.
Digital twin: a model synchronized with a physical asset. A design twin (CAD, simulation) is not the same as an operational twin fed by live sensor data; the operational twin enables what-if scheduling and remaining-life estimates.
Worked example
A bearing defect excites vibration up to 2 kHz. Nyquist requires sampling above 4 kHz; practice calls for roughly 10× the highest frequency of interest, so sample at 20 kHz. At 16-bit resolution that is 40 kB/s per channel - trivial locally, but about 3.5 GB/day across 1,000 channels. Edge feature extraction (RMS, envelope spectrum) beats streaming raw waveforms to the cloud.
Analytics maturity
Descriptive dashboards, then diagnostic root cause, then predictive remaining useful life, then prescriptive action. Predictive maintenance needs labeled failure history or a physics model; unsupervised anomaly detection is used where failures are rare.
Security
Segment OT from IT with a DMZ, never expose PLCs directly to the internet, and patch in staged windows because downtime is expensive. IEC 62443 is the reference standard for industrial control system security.
Traps
Placing MES at Level 4 or ERP at Level 3. Treating OPC UA as a fieldbus; it is an information model plus transport. Confusing digitization (converting data) with digitalization (redesigning the process). Ignoring that sampling rate, not sensor cost, drives the analytics architecture.