Craftsman Smart Factory cheat sheet

ISA-95 defines five levels: L0 process, L1 control, L2 supervision, L3 MES/operations, L4 ERP/business planning; L3-L4 boundary is the classic IT/OT divide.

The six IIoT reference layers are sensing/physical, connectivity (edge gateways), data normalization, analytics, application, and security management.

OEE = Availability × Performance × Quality, where Availability = Run Time / Planned Production Time, Performance = (Ideal Cycle Time × Total Count) / Run Time, Quality = Good Count / Total Count.

Six Big Losses: breakdowns, setup/adjustment, minor stops, reduced speed, process defects, startup rejects; the first two drive Availability, the middle two Performance, the last two Quality.

Purdue model levels 0-2 are deterministic control, level 3 is site operations, levels 4-5 are enterprise/cloud; conduits between zones must be explicitly brokered.

Unified Namespace (UNS) is an ISA-95-aligned topic hierarchy published over MQTT/OPC UA where each asset publishes its own contextualized data rather than point-to-point integrations.

Time-series characteristics: high write rate, append-only, timestamp-indexed, value+quality+timestamp triple (OPC UA DataValue) preserved end to end.

Determinism vs. throughput trade-off: closed-loop control needs bounded jitter (ms), analytics tolerates seconds; mixing both on one network requires traffic segregation (QoS, VLAN, TSN).

Edge vs. cloud decision rule: latency <100 ms, data sovereignty, or bandwidth cost → edge; long-horizon model training and cross-site correlation → cloud.

Fieldbus/industrial Ethernet families: PROFINET (RT/IRT, isochronous), EtherNet/IP (CIP over TCP/UDP), EtherCAT (processing-on-the-fly), Modbus TCP/RTU (register-based, no native timestamp or quality).

Modbus addressing: coils 0xxxx (read/write bits), discrete inputs 1xxxx (read-only bits), input registers 3xxxx (read-only 16-bit), holding registers 4xxxx (read/write 16-bit); 32-bit values span two registers with configurable word order.

OPC UA provides information modeling (objects, variables, methods), subscriptions with deadband/sampling interval, and security via X.509 certificates and signing/encryption modes.

MQTT: publish/subscribe over TCP, QoS 0 (at most once), 1 (at least once, duplicates possible), 2 (exactly once); retained messages and last-will-and-testament support state tracking.

Sparkplug B adds a topic namespace, birth/death certificates, and stateful payloads (only changed metrics sent) on top of MQTT for SCADA interoperability.

Scan time budget: total poll cycle ≈ sum of per-device response times + network propagation; keep polling period ≥ 2-3× worst-case device response to avoid timeout storms.

TSN (IEEE 802.1AS time sync, 802.1Qbv time-aware shaping, 802.1Qbu preemption) enables bounded latency on converged Ethernet; requires synchronized clocks across bridges.

Data acquisition quality flags: good, uncertain (e.g., last-known, substituted), bad (e.g., comm failure); deadband and exception reporting reduce traffic but can hide slow drift.

Wireless in plants: ISA-100.11a/WirelessHART for process mesh, 5G/private LTE for mobile assets and AGVs; coexistence planning required in 2.4 GHz shared with Wi-Fi.

PLC scan cycle = input read → program execution → output write → housekeeping; scan jitter affects control stability more than average scan time.

PID: u(t) = Kp·e + Ki∫e dt + Kd·de/dt; integral windup mitigated by clamping or back-calculation; derivative term should be filtered and applied to measurement, not error.

Ziegler-Nichols ultimate gain method: increase Kp until sustained oscillation at Ku with period Tu; classic PID ≈ Kp=0.6Ku, Ti=Tu/2, Td=Tu/8.

Robot types by workspace: articulated (6R, flexible, large envelope), SCARA (fast planar assembly), delta (high-speed pick-and-place), Cartesian/gantry (large linear work volumes), collaborative (ISO/TS 15066 power-and-force limits).

Degrees of freedom: 6 DOF gives full position and orientation; singularities occur when wrist axes align, causing unbounded joint rates.

Safety standards: ISO 12100 risk assessment, ISO 13849-1 performance levels (PL a-e) from risk graph, IEC 62061 SIL; required PL derives from severity, frequency, and avoidance probability.

Safety functions: safe torque off (STO), safely limited speed (SLS), safe stop 1 (SS1); these are separate from the standard control program and must be independently validated.

Motion control: cascade loops (current → velocity → position); feedforward reduces following error; backlash and compliance limit achievable bandwidth.

Digital twin of a control loop requires actuator/sensor dynamics, dead time, and quantization, otherwise simulation-plant mismatch invalidates tuning results.

Digital twin fidelity levels: descriptive (as-is asset model), predictive (physics or data-driven simulation), prescriptive (closed-loop optimization); each requires bidirectional or unidirectional data sync defined explicitly.

Digital thread vs. twin: the thread is the traceable data lineage across lifecycle stages; the twin is a synchronized virtual instance of a specific asset at a point in time.

Asset Administration Shell (IEC 63278) is the standardized digital representation with submodels (identification, technical data, documentation) exchanged via AASX packages and APIs.

Time-series alignment: resample to a common grid before correlation; account for transport and PLC scan latency (often 100 ms-2 s) or lagged correlation is misread as causation.

Anomaly detection baselines: z-score (|x-μ|/σ > 3), IQR fences (Q1-1.5·IQR, Q3+1.5·IQR), EWMA for drift; static thresholds fail under regime changes such as product changeover.

Classification metrics under class imbalance: precision = TP/(TP+FP), recall = TP/(TP+FN), F1 = 2PR/(P+R); accuracy is misleading when defects are <1% of units.

Predictive maintenance data needs run-to-failure histories or well-censored survival data; remaining useful life models require degradation indicators validated against held-out assets, not random splits.

Feature engineering for vibration: time-domain RMS, kurtosis; frequency-domain envelope/Bearings faults appear at characteristic frequencies (BPFO, BPFI, BSF) modulated by shaft speed.

Data governance: ISA-95/B2MML context, unit-of-measure consistency, and immutable raw-layer storage (bronze/silver/gold) preserve reproducibility of models and audit trails.

IEC 62443 zones and conduits: group assets with common security requirements into zones, control all traffic between them through defined conduits; security level SL1-SL4 per zone.

IEC 62443-3-3 seven foundational requirements: identification and authentication control, use control, system integrity, data confidentiality, restricted data flow, timely response to events, resource availability.

Defense in depth: enterprise firewall → DMZ (historian mirror, patch staging) → cell/area firewall → controller; no direct L4-to-L1 traffic, no dual-homed engineering workstations.

Purdue-to-IEC mapping: L0-L2 = zones with SL3-4 typically, L3 = SL2-3, DMZ = conduit only, L4-5 = IT security controls (ISO 27001).

Legacy protocol risk: Modbus and many fieldbuses have no authentication or encryption; compensate with network segmentation, allow-listing, and one-way data diodes for outbound flows.

Safety instrumented systems must remain independent of the basic process control system and of cybersecurity remediation actions; patching an SIS requires a managed change with revalidation.

Remote access rule: brokered jump host in DMZ, MFA, session recording, time-bound accounts; vendor VPNs terminated directly into the control network are a top audit finding.

Incident response for OT: prioritize safe state over containment speed; preserve forensic images of PLC logic and network captures before rebooting; report under applicable critical-infrastructure rules.

Patch management: compensating controls (virtual patching, ACLs) when vendor validation lags; track firmware CVEs per asset inventory with SBOM coverage.

MES core functions (ISA-95 activity model): production scheduling, dispatching, execution tracking, quality management, maintenance management, and performance analysis.

Integration patterns: point-to-point (brittle, n×m growth), hub-and-spoke broker, and event-driven publish/subscribe (UNS); choose event-driven when more than ~5 systems share data.

Capacity planning: takt time = available production time / customer demand; line balance efficiency = sum of task times / (number of stations × cycle time).

Value stream mapping distinguishes value-added, necessary non-value-added, and waste (overproduction, waiting, transport, overprocessing, inventory, motion, defects).

Change management for digital rollout: phased pilot line → reference architecture → scale; each phase needs defined KPIs (OEE, first-pass yield, MTTR) captured before and after.

Brownfield integration: legacy PLCs without OPC UA need protocol gateways; map tag namespaces to the UNS and document ownership to avoid duplicate tag definitions.

KPI definitions must be frozen before go-live: OEE, MTBF = total operating time / number of failures, MTTR = total repair time / number of repairs, first pass yield = units passing first time / units started.

Project sequencing: network segmentation and asset inventory precede analytics; analytics without reliable time-synchronized data yields non-reproducible results.

Acceptance criteria for smart factory projects: data latency, availability (e.g., 99.5%), tag completeness, and validated model performance against a held-out production period.