12.2 Rate-of-Change and Multi-Inspection Trending
Key Takeaways
- Rate of change of ΔT (or same-point temperature under comparable conditions) is often more predictive than a single snapshot severity band alone
- Normalize or qualify trends for load and ambient whenever possible; raw temperature plots without condition context produce false trends
- Multi-year route plots should use consistent ROI, geometry, and metadata so each point is comparable
- Condition mismatch (different load, ε, path, solar, process state) is a leading cause of false upward or downward trends
- Level II interprets trend slope, step changes, and scatter—not just the latest number—before escalating or clearing findings
A single survey answers a static question. Rate-of-change analysis answers a dynamic one: Is the asset stable, slowly degrading, or racing toward failure? Level II programs treat multi-inspection trending as a core analytical product—especially for assets that sit for years in mild Priority 4 bands while quietly worsening, or for unique equipment without peers.
Why ΔT Over Time Can Beat a Single Snapshot
| Insight | Example |
|---|---|
| Early faults grow | Lug ΔT_peer: 3 °C → 6 °C → 11 °C over three annual surveys at similar load |
| Snapshot band lags risk | Still “Priority 4” at 9 °C, but tripling in two years warrants earlier action |
| Post-event steps | Bearing housing jumps +15 °C after a lubrication miss, then stays high |
| Process cycles hide in one visit | Seasonal load changes look like “faults” without a multi-point view |
Exam theme: The slope and acceleration of thermal abnormality matter. Standards-style priority bands remain mandatory for snapshot classification, but Level II adds trend intelligence when history exists.
Define quantities clearly:
- ΔT_peer(t) = T_suspect(t) − T_similar(t) at survey t (NETA-style)
- ΔT_base(t) = T_ROI(t) − T_ROI(baseline) under comparable conditions
- ΔT_amb(t) = T_ROI(t) − T_ambient(t) (context; not a full substitute for peer/baseline when those exist)
Rate of change might be expressed as °C per month, per year, or per 1,000 operating hours—whatever matches the inspection interval and runtime data available.
Normalization: Making Points Comparable
Raw temperature is a function of fault resistance/friction, load/process, ambient/cooling, and measurement error. Plotting raw °C without context is how false trends are born.
| Factor | Effect on temperature | Level II practice |
|---|---|---|
| Electrical load (I) | Connection heating ~ I²R | Record amps/%; prefer surveys at similar % rated; note if not |
| Mechanical load / RPM / process rate | Bearing and drive heat rise with work | Same production state when possible |
| Ambient / cooling air | Shifts absolute surface T | Record ambient; consider rise above ambient or peer ΔT |
| Wind / enclosure ventilation | Changes convection | Outdoor and enclosure notes |
| Emissivity / RAT / window τ | Apparent T bias | Lock methods; document changes |
| Geometry / focus / distance | Spot-size errors | Standardize stand-off and ROI |
Practical normalization approaches
| Approach | When useful | Caveat |
|---|---|---|
| Survey only at similar load bands (e.g., 50–70% rated) | Best simple control | Plant may not cooperate every visit |
| Peer ΔT (same survey) | Cancels much common-mode ambient/load on three-phase gear | Fails if all phases degrade together or loads differ |
| Temperature rise vs ambient | Unique assets | Still load-sensitive |
| Engineering load correction (estimate full-load T from partial load) | Advanced programs | Document assumptions; do not fake precision |
| Qualifying notes (“↑ load this visit”) | Always available | Not a full numeric normalize but prevents false alarm |
Rule: If you cannot normalize, you must qualify. A report that says “ΔT rose 8 °C but load rose from 30% to 90%” is honest; a chart that screams “failure” without load is not.
Multi-Year Route Plotting
Predictive routes (switchgear rooms, motor lists, steam trap walks, roof grids) generate time series per asset or per ROI.
| Plot element | Good practice |
|---|---|
| X-axis | Date or operating hours |
| Y-axis | Primary metric: ΔT_peer or ΔT_base (label which!) |
| Secondary series | Load %, ambient (on second axis or companion table) |
| Events | Repair, outage, process change markers |
| Limits | Snapshot priority thresholds as horizontal reference lines |
| Missing data | Gaps shown; never interpolate a fake “smooth failure” without basis |
Consistency beats camera megapixels. The same bolt head, same phase, same window, same ε method, year after year, is worth more than a new camera model with wandering ROIs.
Example multi-year table (teaching model)
| Survey | Load (% rated) | Ambient (°C) | T_PhaseB (°C) | T_peer avg (°C) | ΔT_peer (°C) | Notes |
|---|---|---|---|---|---|---|
| 2023-04 | 60 | 22 | 38 | 35 | 3 | Baseline year |
| 2024-04 | 58 | 23 | 42 | 36 | 6 | Rising |
| 2025-04 | 61 | 21 | 48 | 37 | 11 | Crosses into P3 band |
| 2025-10 | 60 | 24 | 49 | 37 | 12 | Mid-cycle check; still climbing slowly |
Interpretation: rate of change ~ +4 °C/year in peer ΔT, then a slower rise. Snapshot classification in 2025 is Priority 3, but the trend justified mid-cycle inspection before the next annual only survey.
False Trends from Condition Mismatch
| Mismatch | False story | Correction |
|---|---|---|
| Load much higher this year | “Fault doubled” | Compare % load; use peer ΔT; reschedule at matched load |
| Load much lower | “Repair miracle” without work | Under-loaded survey hides I²R faults |
| ε set 0.95 then 0.60 on metal | Huge fake ΔT | Standardize surface preparation / tape |
| Different IR window or open door vs closed | Path bias | Same path + τ |
| Solar-loaded outdoor bus | Afternoon “hotspot” | Time-of-day control; shade; peer check |
| Cold start vs steady | Transient rise | Wait for thermal equilibrium |
| ROI moved from lug to cable jacket | Different object | Template ROIs |
| Different camera range/calibration status | Bias | Calibration program (Chapter 7) |
| Process fluid temperature change | Whole machine warmer | Normalize to process T or peer unit |
Level II skepticism: Any large year-to-year jump without a matching work order, failure mode story, or confirmed matched conditions is suspect measurement or condition change until proven otherwise.
Reading Trend Shapes
| Shape | Typical meaning | Response thinking |
|---|---|---|
| Flat low ΔT | Stable healthy or stable minor anomaly | Maintain interval |
| Linear slow rise | Progressive degradation (loosening, wear, fouling) | Tighten interval; plan repair before next band jump |
| Accelerating rise | Approaching failure | Escalate priority earlier than snapshot alone (Section 12.3) |
| Step change up | Event (torque loss, lube failure, blocked cooling) | Investigate cause now; do not wait for “next year” |
| Step change down after work | Successful repair | Re-baseline |
| High scatter | Uncontrolled conditions or intermittent load | Fix method before trusting slope |
Linking Trending to Other Severity Systems
| Domain | Snapshot tool | Trend add-on |
|---|---|---|
| Electrical | NETA-style peer ΔT bands | Same-connection ΔT_base and ΔT_peer history |
| Rotating machinery | ISO 18434-style zones vs reference | Bearing ROI vs baseline under same load |
| Steam traps | Failed vs live patterns | Repeat failures on same trap ID |
| Buildings | Pattern + envelope ΔT validity | Seasonal series only under valid weather |
Do not force electrical Priority numbers onto unrelated domains, but do carry the shared idea: history changes urgency.
Worked Scenarios
Scenario A — Normalized rise. ΔT_peer goes 4 → 9 → 14 °C at 55–60% load each year. Real degradation; schedule repair (P3) and consider shorter interval.
Scenario B — False rise. Absolute lug T up 18 °C versus last year, but load was 25% then and 95% now; peer ΔT unchanged at 2 °C. Not a failure trend—load mismatch.
Scenario C — False improvement. Peer ΔT falls from 16 °C to 5 °C, yet load collapsed from 80% to 15%. Do not close the work order as “healed”; re-inspect at load.
Scenario D — Multi-year route. Motor DE bearing ROI rise vs ambient climbs 2 °C/year for four years with stable production. Even if still in an “advisory” ISO-style band, plan bearing service before an exponential end-stage.
Scenario E — Step event. Gearbox surface ROI jumps +20 °C between monthly routes after oil leak. Treat as event-driven investigation, not a gentle annual slope problem.
Common Traps
| Trap | Better practice |
|---|---|
| Charting absolute T only across seasons outdoors | Use peers, rise, or matched conditions |
| Declaring trends from two points with mismatched load | Need comparable conditions or heavy qualification |
| Ignoring accelerating curvature | Late-stage risk |
| Changing ROI freely | Kills multi-year meaning |
| Assuming “Priority 4 forever” if still < 10 °C | Rising 2 → 9 °C may still need action planning |
Summary for Recall
Rate-of-change of peer or baseline ΔT often reveals risk that a single snapshot understates. Normalize or qualify for load and ambient (and lock radiometric method). Build multi-year route plots with consistent ROIs, event markers, and clear metric definitions. Condition mismatch is the chief source of false trends—Level II challenges dramatic slopes that lack matched conditions or plausible failure physics before alarming the plant.
Why can multi-year rate-of-change analysis be more important than a single mild snapshot ΔT?
Absolute temperature of a lug is 15 °C higher than last year’s survey, but load increased from 20% to 90% of rated and peer-phase ΔT is unchanged. Best Level II conclusion?
What is a primary purpose of standardizing ROI placement on multi-year IR routes?
Which situation most clearly illustrates a false trend from condition mismatch?