Direct Answer
Equipment has always run on an annual overhaul — the model says a unit will fail in 8 weeks. Repair early or stick to the plan? High-safety/high-value equipment (transformers, UPS, critical busbars) should be scheduled by the prediction model — CUSUM gives a 4-12 week window, combined with an Arrhenius life model to plan the overhaul; low-risk, low-value equipment keeps time-based inspection; the hybrid is predictive for critical equipment plus periodic for general equipment. Landing = grade the equipment first, then decide what drives the schedule.
A scenario: A factory transformer has always run on an annual overhaul — the Taiyi model reports "this transformer's temperature trend is abnormal, expected to exceed limits in 8 weeks" — the equipment manager hesitates — early repair disrupts the plan, following the plan risks a failure — the solution: the transformer moves to model-driven scheduling — a maintenance window opens in 4 weeks — defect cleared early — other general equipment keeps the annual inspection. The decision logic is not "do I trust the model" but "is the equipment worth model-driven scheduling".
1. Time-Based Inspection: Calendar-Driven Maintenance
What time-based inspection is — calendar-driven: maintenance by fixed calendar cycles, inspect when the date arrives; The basis — experience cycles: annual, semi-annual and quarterly fixed; The advantage — strong planning: schedules are fixed and resources arrangeable; The drawback — over-maintenance: equipment inspected even when healthy, wasted spend; The essence — "inspect when the date arrives": no state inquiry, only the calendar.
In one sentence: time-based inspection inspects when the date arrives — strong planning, but prone to over-maintenance.
An example: A factory overhauls everything annually: transformers, motors and switchgear on one schedule — resources concentrated — but most equipment is healthy at inspection. The logic is that the fixed schedule is certain, but it ignores state — healthy units get inspected anyway.
2. Predictive Maintenance: State-Driven Maintenance
What predictive maintenance is — state-driven: maintenance by the model-predicted degradation trend; The basis — online data: continuous monitoring of temperature, vibration and partial discharge; The output — maintenance windows: CUSUM suggests a window 4-12 weeks ahead; The advantage — on-demand: repaired only when the equipment truly has a problem; The essence — "repair before failure": maintenance scheduled before the fault occurs.
In one sentence: predictive maintenance repairs before failure — state-driven, no waste and no missed repairs.
An example: A motor's temperature trend turns abnormal: Taiyi warns 6 weeks ahead — maintained within the window — bearing wear found — replaced early — outage avoided. The logic is that predictive value lies in the lead time — repair before failure, neither wasting nor missing.
3. Cost and Risk: The Two Strategies Side by Side
The time-based cost — labor and spares: fixed cycles, stable maintenance spend; The time-based risk — missed and over-inspection: faults before the cycle slip through, healthy units over-serviced; The predictive cost — monitoring investment: sensors, models and data analysis; The predictive risk — false and missed alarms: an inaccurate prediction shifts the window; The comparison conclusion — each has a boundary: predictive for high value, time-based for low value.
In one sentence: time-based is stable but wasteful; predictive is accurate but can misjudge — use each by equipment value.
An example: A factory compares: annual maintenance 800k — predictive monitoring 200k a year — total cost down 30% after high-value predictive. The logic is that the comparison must use the total — time-based saves monitoring money but wastes maintenance money; predictive spends more on monitoring and saves on maintenance and outages.
4. The Credibility Boundary of Predictive Maintenance
What credibility means — lead time and confidence: prediction gives a "window", not an exact date; The lead-time boundary — 4-12 weeks: CUSUM usually gives a 4-12 week window; The confidence boundary — trend grade: the closer to the limit, the higher the confidence; Using the boundary — window management: schedule within the window, not on an exact date; The boundary conclusion — prediction is a scheduling input: leave margin, do not replace judgment.
In one sentence: the credibility boundary is a 4-12 week window plus trend confidence — used for scheduling, not replacing judgment.
An example: A project schedules by prediction: model gives a 6-10 week window — maintenance arranged 2 weeks before the window start — margin left — no gambling on the exact date. The logic is that prediction is not fortune-telling — it gives a window and confidence; keep margin when scheduling.
5. High-Value Equipment: Let the Model Schedule
The high-value list — transformers, UPS and critical busbars: high failure cost; The high-value profile — large outage loss: one failure exceeds the monitoring investment; The model value — early window: CUSUM warns 4-12 weeks ahead; The life-model companion — Arrhenius: temperature-based life model sets the overhaul timing; The high-value conclusion — predictive first: let the model decide the window.
In one sentence: high-value equipment is predictive-scheduled — failure cost is high, the early window pays.
An example: A data center's UPS and transformers: all on Taiyi prediction — CUSUM plus Arrhenius sets windows — maintenance follows the windows — zero unplanned outages in three years. The logic is that high-value accounting is outage loss — one unplanned outage far exceeds the monitoring investment.
6. Low-Risk Equipment: Keep Time-Based Inspection
The low-risk list — general motors, lighting circuits and ordinary switchgear: low failure cost; The low-risk profile — small outage impact: failures can wait for planned repairs; The time-based fit — cost-optimal: no monitoring investment needed; The time-based boundary — no model connection: trend data has low value; The low-risk conclusion — time-based fallback: annual inspection suffices.
In one sentence: low-risk equipment keeps time-based inspection — small outage impact, monitoring investment not worth it.
An example: A factory's general lighting and auxiliary circuits: annual inspection — no monitoring — lowest cost. The logic is that low-risk accounting is monitoring investment — sensors cost more than the failure loss.
7. The Taiyi S Board: What the 20 Models Cover
What the S board is — Taiyi's state-prediction board: 20 models covering equipment; What S-02 covers — transformer models: temperature, oil temperature and life prediction; What S-04 covers — motor models: bearing, winding and vibration trends; What S-13 covers — busbar models: joint temperature and contact degradation; The S board value — full coverage: critical equipment has a matching model.
In one sentence: the Taiyi S board is 20 models covering critical equipment — transformers, motors and busbars each have one.
An example: A project selects models by equipment: transformer to S-02, motor to S-04, busbar to S-13 — all three high-value types connected. The logic is that models must match by type — different equipment degrades differently.
8. The Hybrid Strategy: Merging with the Annual Overhaul Plan
What the hybrid is — predictive for critical plus time-based for general: two drivers for two classes; Merge way one — windows into the overhaul: prediction windows aligned with the annual overhaul; Merge way two — dynamic adjustment: an early warning inserts maintenance into the schedule; Merge way three — resource reuse: one window serves multiple units; The merge conclusion — the plan is not overturned: prediction optimizes the annual plan, not replaces it.
In one sentence: the hybrid is predictive for critical and time-based for general — prediction windows merge into the annual overhaul.
An example: A factory's annual overhaul: concentrated in June — Taiyi reports a transformer exceeding limits in 8 weeks in April — merged into the June window early — plan unchanged, window optimized. The logic is that prediction does not overturn the plan — it optimizes it — windows merge into the overhaul schedule.
9. How the Maintenance Plan Lands
Landing step one, grade equipment — high value/low risk: grade the equipment first; Landing step two, connect models — high value connected: pick the matching model and monitor; Landing step three, generate windows — prediction output: CUSUM generates the maintenance window; Landing step four, merge plans — into the overhaul: windows merge into the annual plan; Landing step five, close the loop — verify after repair: results fed back, model calibrated.
In one sentence: landing is grade, connect, window, merge and iterate — a rolling closed loop.
An example: A factory lands: grade (transformer high, lighting low) — connect (S-02/S-04) — window (CUSUM 6 weeks) — merge (into annual) — iterate (post-repair feedback). The logic is that landing is a systematic push — the five-step loop is what makes the model truly schedule maintenance.
10. Back to YESLON: Taiyi plus Predictive Maintenance
The Taiyi position — the prediction engine: the S board's 20 models predict state; The CUSUM warning — trend window: a maintenance window 4-12 weeks ahead; The Arrhenius companion — life model: temperature-based overhaul timing; The FEXCloud landing — plan management: windows presented on the platform; The YESLON solution — predictive maintenance: high-value equipment scheduled by prediction.
In one sentence: Taiyi plus predictive maintenance means CUSUM gives the window, Arrhenius sets the timing and the platform manages the plan.
An example: A project deploys the predictive maintenance solution: transformers, motors and busbars on Taiyi — windows presented in FEXCloud — merged with the annual plan. The logic is that the solution lands as a combination — the model gives windows, the platform manages plans.
11. Self-Check: Five Questions to Fix the Maintenance Strategy
Question one, how valuable is the equipment — value: how large is the outage loss; Question two, is the fault predictable — predictable: can temperature/vibration trends be monitored; Question three, is there a matching model — model: does the Taiyi S board cover it; Question four, can the plan flex — flexibility: can the annual plan merge dynamically; Question five, is the data sound — data: is monitoring continuous and reliable.
In one sentence: the five questions are value, predictability, model, plan flexibility and data — all answered, the strategy is clear.
A self-check example: A factory runs the five: transformer high value (large outage loss), trend predictable (temperature), model exists (S-02), plan flexible, data continuous — all five pass — the transformer moves to predictive scheduling. The decision logic of predictive maintenance vs annual time-based inspection is to grade equipment first — high-value equipment is scheduled by the model (CUSUM gives a 4-12 week window plus Arrhenius timing), low-risk equipment keeps time-based inspection — the hybrid is predictive for critical plus periodic for general — the answer to "should AI schedule outages" is "yes, for high-value equipment".
Scope and Limitations
- The conclusions in this article are intended for LV distribution and electrical-safety sites. Criteria are compiled from public codes and FEXLINK product documentation, and product parameters follow the latest selection handbook and product knowledge base.
- Quantitative criteria (protection zone, discharge and residual voltage, earth resistance, thresholds and grading) hold under the preconditions stated in this article. Where the site voltage level, earthing arrangement, environmental category or load profile differs, the figures must be recalculated. Specific configurations involving Tianyan engine, S track (S-02/S-04/S-13) should be verified against actual circuit parameters.
- Specific product models and parameters are subject to the latest selection handbook and product knowledge base; this article does not constitute selection or procurement advice.
- Standard numbers cited here refer to the currently effective editions; when a standard is revised, the related quantitative conclusions must be re-checked.
- Deployment and implementation for specific projects must be carried out by qualified engineers based on site survey and design drawings.
Frequently Asked Questions
Predictive Maintenance vs. Time-Based Inspection
Equipment has always run on an annual overhaul — the model says a unit will fail in 8 weeks