API 670 Explained: Machinery Protection Systems — Why 73% of Turbomachinery Failures Are Preventable (If You Apply These 5 Non-Negotiable Monitoring Rules from the Latest 5th Edition)
Why API 670 Isn’t Just Another Checklist—It’s Your Last Line of Defense
API 670 Explained: Machinery Protection Systems. Overview of API 670 standard for machinery protection systems including vibration, temperature, and position monitoring requirements. If you’re reading this, you likely manage rotating equipment—turbines, compressors, pumps, or motors—in oil & gas, power generation, or chemical processing. And if you’ve ever watched a $2.8M centrifugal compressor trip offline at 3 a.m. due to undetected subsynchronous vibration—or worse, seen bearing metal in lube oil analysis *after* a catastrophic failure—you already know: API 670 isn’t theoretical. It’s operational insurance. The latest 5th Edition (2022) tightened requirements on sensor redundancy, signal validation logic, and alarm hierarchy—but most plants still deploy legacy systems that pass ‘compliance audits’ while failing real-world protection. This isn’t about ticking boxes. It’s about engineering integrity.
What API 670 Really Is (and What It’s Not)
First, let’s dispel a myth: API RP 670 is not a mandatory regulation like OSHA 1910.119—it’s a Recommended Practice. But don’t mistake ‘recommended’ for ‘optional’. In practice, it’s de facto law. Major operators (ExxonMobil, Shell, Saudi Aramco) require full API 670 compliance in EPC contracts. Insurance underwriters demand it. And when an incident occurs, OSHA and PHMSA investigators treat non-compliance as evidence of negligence—especially if root cause analysis traces failure to inadequate monitoring.
The standard applies specifically to machinery protection systems (MPS) for rotating equipment with criticality ≥ Level 2 (per API RP 686). That means shaft speeds > 3,000 RPM, power > 1,000 HP, or process safety implications (e.g., H₂S service, high-pressure hydrocarbons). It covers three core monitoring domains:
- Vibration: Absolute and relative displacement (eddy current probes), velocity (seismic sensors), and acceleration (for high-frequency faults); requires dual-channel voting logic for trip decisions
- Temperature: Bearing metal, winding, seal gas, and thrust collar temps—with Class A RTDs mandated for critical points and cold-junction compensation traceability
- Position: Axial position (thrust) and differential expansion—where probe calibration must account for thermal growth differentials between rotor and casing
Crucially, API 670 doesn’t govern control systems (DCS/PLC)—only dedicated, hardwired or certified SIL-2/SIL-3 MPS hardware. As Dr. Elena Rostova, API Subcommittee Chair for RP 670, stated in the 2022 Foreword: “An MPS is not a data logger. It is a fail-safe decision engine with deterministic response times ≤ 100 ms.”
The 5 Non-Negotiable Rules Most Plants Get Wrong (Real Case Study)
Consider the 2023 failure at a Gulf Coast LNG train: A 22,000 HP main refrigerant compressor tripped repeatedly on ‘high vibration’ alarms—then catastrophically failed during startup after bypassing alarms for 72 hours. Root cause? Not sensor failure—but misapplication of API 670 Section 5.4.2.
Here’s what actually happened—and the five rules that would have prevented it:
- Rule #1: Dual-Channel Voting Isn’t Optional—It’s Physics-Based
API 670 mandates independent, redundant sensors with separate signal conditioning and voting logic (2oo2 or 2oo3) for all trip-class measurements. The LNG plant used single-channel eddy current probes on thrust bearings, citing ‘cost savings’. When probe drift occurred (±12 µm over 6 months), no cross-check existed. Result: false trips → operator distrust → alarm suppression → missed true fault onset. - Rule #2: Calibration Must Include Thermal Drift Validation
Section 6.3.4 requires calibration at operating temperature—not ambient. The plant calibrated probes at 25°C, but bearing metal temps ran at 112°C. Thermal coefficient errors introduced ±8.3 µm offset in axial position readings—enough to mask incipient thrust collar wear. - Rule #3: Alarm Setpoints Must Be Derived From Machinery-Specific Baselines—Not Generic Tables
Annex A provides guidance, but API 670 §4.5.1 requires site-specific baselines established during commissioning and updated after major overhauls. The LNG facility used ISO 10816-3 thresholds (‘good’/‘satisfactory’) for a compressor with flexibly coupled double-suction impellers—a configuration prone to resonance at 0.4× running speed. Their ‘satisfactory’ threshold was 4.5 mm/s; actual safe limit was 2.1 mm/s. - Rule #4: Signal Validation Logic Must Detect Sensor Degradation—Not Just Failure
New in the 5th Edition: Section 5.6.2 requires continuous health monitoring—e.g., checking for DC bias shift (>5% of full scale), noise floor increase (>3 dB), or impedance deviation (>10%). The failed compressor’s system only checked for open-circuit faults. It missed gradual insulation breakdown in the probe cable, causing intermittent signal dropout masked as ‘process noise’. - Rule #5: Trip Logic Must Be Hardware-Enforced—Not Software-Configurable
API 670 §5.7 prohibits trip decisions from residing in DCS logic solvers. The LNG train routed vibration trips through its DeltaV DCS—allowing engineers to ‘temporarily disable’ trips via password-protected logic blocks. Post-failure, investigators found 14 trip inhibits active during the final 48 hours.
API 670 vs. Competing Standards: Where Boundaries Matter
Confusion arises when teams conflate API 670 with ISO 10816 (vibration severity), ISO 20816 (machinery-specific vibration), or IEC 61511 (functional safety). They serve distinct purposes:
| Standard | Primary Purpose | Scope Overlap with API 670 | Critical Difference |
|---|---|---|---|
| API RP 670 (5th Ed.) | Design, installation, and verification of dedicated Machinery Protection Systems | Core scope | Focuses on trip reliability, sensor redundancy, and deterministic response—not severity assessment |
| ISO 10816-3 | General vibration severity evaluation for industrial machines | Referenced in Annex A for baseline guidance | Descriptive, not prescriptive; allows operator judgment—API 670 forbids subjective interpretation for trip decisions |
| IEC 61511 | Functional safety lifecycle for SIS (Safety Instrumented Systems) | Applies to MPS when SIL-rated (typically SIL-2) | IEC 61511 governs how to achieve SIL; API 670 defines what the MPS must monitor and how signals must be conditioned |
| IEEE 112 | Motor efficiency testing | No direct overlap | Irrelevant to MPS design—though motor winding temp monitoring falls under API 670 §5.5.2 |
Implementation Roadmap: From Audit to Assurance
Compliance isn’t achieved by buying ‘API 670-certified’ hardware. It’s engineered. Here’s how leading operators do it—step-by-step:
- Phase 1: Criticality Assessment — Use API RP 686’s machinery criticality matrix to classify assets. Only Level 2+ require full API 670 compliance. (Don’t waste budget on Level 1 pumps.)
- Phase 2: Sensor Architecture Review — Map every monitored parameter against API 670 Table 1 (Sensor Types) and Table 2 (Redundancy Requirements). Note: Differential expansion requires two independent probes (casing + rotor), not one ‘differential’ probe.
- Phase 3: Signal Path Validation — Trace each signal from probe tip to trip relay: shielding, grounding, cable routing (avoid VFD runs!), conditioner specs, voting logic type, and relay response time (must be ≤ 100 ms per §5.7.3).
- Phase 4: Baseline Establishment — Run machinery at rated load for 72 hours. Capture vibration spectra, phase, and orbit plots. Use Bently Nevada System 1 or Emerson DeltaV AMS to generate machine-specific alarm bands—not generic ISO tables.
- Phase 5: Independent Verification — Hire third-party auditors (e.g., ABS, DNV, or API-authorized training providers) to perform functional safety assessment (FSA) per IEC 61508 and API RP 670 Annex C.
Pro tip: The biggest ROI comes from Phase 4. A refinery in Rotterdam reduced unscheduled downtime by 68% after implementing machine-specific baselines—because their old ‘ISO-based’ alarms triggered on harmless resonances, desensitizing operators to real faults.
Frequently Asked Questions
Is API 670 legally enforceable?
No—it’s a Recommended Practice, not federal law. However, it carries weight in litigation, insurance claims, and regulatory investigations. Under the General Duty Clause (OSHA 5(a)(1)), employers must provide a workplace free from recognized hazards. Courts consistently rule that ignoring API 670 constitutes ‘recognized hazard’ for critical rotating equipment.
Does API 670 apply to electric motors alone?
Only if the motor is part of a critical rotating system (e.g., driver for a Level 2 pump/compressor) AND operates above 3,000 RPM or 1,000 HP. Standalone low-voltage motors under 250 HP fall outside scope—unless specified in site-specific safety standards.
Can I use wireless sensors for API 670 compliance?
Not for trip-class measurements. Section 5.3.1 mandates ‘hardwired, intrinsically safe or explosion-proof’ connections. Wireless is permitted only for non-safety monitoring (trend data, diagnostics) per API RP 553 Annex F—but never for trip initiation.
What’s new in the 5th Edition (2022) vs. 4th (2014)?
Key updates: (1) Mandatory thermal drift validation during calibration; (2) Expanded requirements for signal health monitoring (noise, bias, impedance); (3) Clarification that MPS must monitor seal gas temperature in dry gas seals; (4) Stricter documentation for voting logic architecture; (5) Alignment with IEC 62443 for cybersecurity of MPS controllers.
Do variable frequency drives (VFDs) invalidate API 670 vibration limits?
Yes—absolutely. ISO 10816 assumes constant speed. VFD-driven machines require spectral analysis across the entire operating range. API 670 §4.5.3 requires establishing speed-dependent alarm bands. A compressor running at 45% speed may show ‘normal’ broadband vibration at 2.8 mm/s—but exhibit destructive 1× subharmonic at 0.5× running speed. Your MPS must detect that.
Common Myths About API 670
Myth #1: “If our vibration system has dual probes, we’re compliant.”
False. API 670 requires independent signal paths—including separate cables, conditioners, and voting logic. Running two probes into one conditioner, then splitting the output, violates §5.4.2.
Myth #2: “API 670 only applies to new installations.”
False. Section 3.2 states: “Existing MPS shall be upgraded to comply with this RP when modifications exceed 20% of the original system cost or when major overhauls occur.” Many retrofits trigger mandatory updates.
Related Topics (Internal Link Suggestions)
- API RP 686 Machinery Reliability Guidelines — suggested anchor text: "API RP 686 reliability guidelines for rotating equipment"
- Vibration Analysis Certification Paths (ISO 18436 vs. ASME Vibration Analyst) — suggested anchor text: "vibration analyst certification comparison"
- SIL-2 vs SIL-3 for Machinery Protection Systems — suggested anchor text: "SIL-2 vs SIL-3 for turbine protection"
- Bently Nevada 3500 vs. Emerson CSI 6500: API 670 Compliance Comparison — suggested anchor text: "Bently 3500 vs Emerson CSI 6500"
- How to Perform an API 670 Gap Assessment — suggested anchor text: "API 670 gap assessment checklist"
Your Next Step: Stop Auditing—Start Validating
API 670 compliance isn’t about passing an auditor’s checklist. It’s about ensuring your MPS will act—correctly, decisively, and within 100 ms—when a thrust bearing begins to lift off its pad at 12,000 RPM. The LNG case study proves that gaps aren’t found in paperwork—they’re exposed in milliseconds during failure. Don’t wait for the next trip event to reveal your blind spots. Download our free API 670 Gap Assessment Tool—a 12-point field-validated checklist used by 37 refineries to identify hidden vulnerabilities in sensor architecture, calibration traceability, and voting logic before the next turnaround. Then, schedule a 30-minute engineering review with our API-certified specialists—we’ll map your critical trains against the 5th Edition’s 147 requirements and prioritize fixes by risk impact.
