MAIC — Mechanical Asset Intelligence Core

Executive Overview & Research Foundations

MAIC (Mechanical Asset Intelligence Core) is an industrial-grade predictive health and asset performance management (APM) platform for critical rotating machinery—including centrifugal pumps, compressors, fans, and induction motors. Grounded in the Industrial Intelligence Platform research, MAIC disproves the "horizontal-data hypothesis": dumping raw historian tags into generic data lakes or BI tools fails reliability teams. MAIC delivers the prescribed Phase-1 vertical wedge: deterministic machine intelligence with explainability by construction, zero black-box guesswork, and an advisory-only posture.

The Operational Challenge

Unplanned rotating machinery failures in refining and petrochemical plants cost $250,000 to $2.3 million per hour. Plants face three systemic operational bottlenecks:

• Raw Telemetry Floods: Thousands of time-series tags lack mechanical context, burdening engineers with noisy charts and manual threshold tuning.
• Black-Box Alarm Fatigue: Generic ML anomaly detectors fire unexplainable alerts during routine ramp-ups and transients, leading operators to mute warnings.
• Manual Work-Order Conversion: Confirmed anomalies take days to convert into structured CMMS work orders with accurate ISO failure codes and parts lists.

Deterministic Architecture & Analytical Pipeline

MAIC replaces uninterpretable AI scoring with a deterministic 6-stage engineering pipeline:

1. 6-Step Data Quality Metric (DQM): Telemetry passes through physical bounds checking, frozen-sensor detection, Hampel outlier filtering, rate-of-change limits, and drift tracking. Questionable readings are quarantined (DQM 0.00–1.00), preventing instrument faults from masquerading as machine failure.
2. Operating-State Regime Machine: Deterministically classifies asset intervals into Offline, Ramp-Up, Steady-State Nominal, Partial Load, Controlled Shutdown, and Trip Excursion. Dynamic transient suppression suspends anomaly evaluation during shifts, eliminating startup false alarms.
3. Kinematic Vibration & Spectral DSP: Ingests waveforms to compute FFT and Hilbert envelope spectra. Tracks kinematic bearing fault frequencies—Outer Race (BPFO), Inner Race (BPFI), Ball Spin (BSF), and Fundamental Train (FTF)—alongside ISO 20816 vibration severity zones.
4. Robust Statistical Baselines: Evaluates windowed features against versioned median and MAD baselines trained on clean steady-state history, preventing contamination from past faults.

The Nine-Question Explainability Framework (MAIC-9Q-1)

Every anomaly automatically generates an audit-grade forensic investigation answering nine engineering questions:

1. What Happened: Concrete operational headline identifying the detected failure mode.
2. Why It Happened: Degradation mechanism matched to kinematic spectral harmonics.
3. Quantitative Evidence: Ranked telemetry contributions with observed values, baselines, and engineering units.
4. Operating Regime & Topology: Operational regime at detection within the ISA-95 asset hierarchy.
5. Historical Precedents: Automated matching against prior workspace incidents and imported maintenance history.
6. Component Localisation: Pinpointed defect location down to the lowest replaceable unit (e.g., DE/NDE bearing, impeller, seal).
7. Actionable Prescriptions: Ordered maintenance inspection checklist and required part numbers.
8. Calibrated Confidence: Statistically derived certainty score floored by the input DQM and historical false-positive rates.
9. Falsification Criteria: Explicit operating conditions or measurements that would refute the finding.

Closed-Loop Maintenance & Avoided Downtime ROI Ledger

• ISO 14224 Work Orders: Converts validated investigations into draft work orders pre-populated with ISO 14224 failure codes, priority, prescriptions, and parts.
• Duty-Separated Governance: Enforces strict role separation across six RBAC personas (Owner, Plant Manager, Reliability Engineer, Maintenance Superintendent, Operator, Auditor). Review and approval duties remain strictly segregated from dispatch.
• Advisory CMMS Adapters: Dispatches approved work packages into enterprise maintenance systems (SAP PM, IBM Maximo, or internal ledger).
• Verifiable ROI Tracker: Closes the loop with post-maintenance feedback (accuracy, downtime hours avoided, repair cost), calculating financial payback against plant downtime cost baselines.

OT Safety & Guardrails

• Advisory-Only Posture: Zero closed-loop control writes, zero setpoint adjustments, and zero write access to DCS or PLC networks.
• Outbound-Only Ingestion: Edge gateways push encrypted telemetry via workspace-specific hashed ingest keys.
• Audit Provenance: Complete audit logs for all state transitions, baseline retrains, and work-order dispatches.