Planning Spare Parts Around Condition Signals and Lead Times

It is Tuesday morning. A vibration trend on the main gearbox of your only finishing line has been climbing for three weeks, and the analyst's note is unambiguous: outer race defect on the input shaft bearing, early stage. The planner agrees. The reliability engineer agrees. Then purchasing comes back with the number that ends the conversation: 14 weeks, imported, no stock on site, no distributor inventory in the region.

The gearbox fails in week nine. Everyone did their job. The sensor worked, the model worked, the planner acted, and the plant still ran to failure.

Here is the rule that would have prevented it, and the direct answer to the question this article exists to settle: a condition signal is only actionable when the expected remaining useful life is longer than the total response clock, which is signal confirmation plus planning plus requisition plus supplier lead time plus transit plus receiving plus access to a maintenance window. If the degradation horizon is shorter than that clock, the signal is not a scheduling decision. It is a stocking decision you needed to make before the trend started. The rest of this piece gives you the framework: signal confidence, failure mode to part number mapping, asset criticality, lead-time risk tiers, and the inventory policy each tier earns.

Why Usage-Based Reorder Points Break on Predictive Programs

Min/max levels and economic order quantity math assume demand that repeats. You consume 40 V-belts a year, you set a reorder point, the math works because next year looks like last year. Predictive maintenance does not generate that kind of demand. It generates lumpy, event-driven demand on parts that may move once every three or four years, and only when a specific failure mode on a specific asset starts to progress.

That is exactly the population most likely to be stocked out and most likely to stop a line. A part with annual usage of 0.3 units looks like dead inventory in any consumption-based report. It also happens to be the rotor element, the main bearing, the gear set, or the custom seal that carries the entire production consequence of the asset.

Plants compensate in two predictable ways, and both are expensive. The first is blanket stocking: anything expensive and scary goes on the shelf, capital sits idle, and the storeroom fills with items for assets that were decommissioned two years ago. The second is heroics: emergency freight, weekend machining, a local shop fabricating a sleeve that will not last. Both are symptoms of the same gap, which is that no one defined in advance which failure modes the storeroom is supposed to protect.

The reframe is simple. Stop planning inventory around annual consumption and start planning it around protected failure modes on critical assets. Reliability-centered maintenance already gives you the failure mode list. ISO 55000 style asset management thinking already tells you to connect asset criticality to resource decisions. Federal operations and maintenance guidance treats planning, documentation, and program discipline as core O&M practice rather than an add-on ^[5]. The missing link in most plants is that those failure mode lists never cross the wall into the inventory policy. Our spare parts optimization guide covers the stocking math; this article covers the decision trigger that sits upstream of it.

The Lead-Time Clock: What Actually Has to Fit Inside Remaining Useful Life

Purchasing lead time is one segment of the response clock, and usually not the longest. Break the clock into segments and assign each one an owner and a measurement, because the delays that kill predictive wins are almost always internal.

1. Signal review and confirmation. From alert raised to a human agreeing it is real. Owner: reliability engineer or analyst. Hidden delay: alerts sitting unreviewed in a queue nobody owns. Condition monitoring workflows are designed around this loop of measurement, alert, investigation, corrective action, and recorded feedback ^[3]. Measure hours from alert to first documented review. 2. Diagnosis and horizon estimate. From "it is real" to "we have roughly this long." Owner: reliability engineer, often with a follow-up route or an additional sensor channel. Hidden delay: waiting for the next scheduled route instead of pulling an on-demand reading. Measure days from confirmation to horizon estimate. 3. Work order creation and approval. Owner: planner and maintenance supervisor. Hidden delay: approval thresholds that route anything over a dollar limit to a manager who travels. Measure days in approval status. 4. Requisition and PO issue. Owner: procurement. Hidden delay: vendor requalification, sole-source justification paperwork, missing part master data. Measure days from requisition to PO acknowledgement. 5. Supplier manufacture or pick, transit, and customs. Owner: supplier and logistics. Hidden delay: quoted lead time versus actual. Measure variance, not the quote. 6. Receiving, inspection, and kitting. Owner: storeroom. Hidden delay: parts sitting in receiving unreported, or arriving incomplete. Measure days from dock to kit-complete. 7. Window access. Owner: operations. Hidden delay: the only acceptable outage is the quarterly shutdown eleven weeks out. Measure days from kit-complete to execution.

Use your own actuals for every segment. Export purchase order history from SAP PM, IBM Maximo, Fiix, or UpKeep and compare requisition date, PO date, and goods receipt date per part number. Published supplier averages are marketing. Your receiving dates are evidence.

Then apply the test plainly. Confirmed degradation horizon longer than the response clock means you schedule. Horizon shorter than the clock means the part belonged on the shelf, and the output of that signal is a policy change, not a work order.

Mapping Failure Modes to Part Numbers Before You Need Them

Procurement cannot buy "bearing degradation detected." It needs a part number, a quantity, and a required-by date. That translation work has to exist before the alert, because doing it under pressure is how plants order the wrong bearing twice.

The mapping exercise is bounded and finite. For each critical asset, list the dominant failure modes from your FMEA or from three years of work order history. For each failure mode, write down the repair as it is actually performed, then attach every part that repair consumes: the primary component, plus seals, gaskets, O-rings, lockwashers, lubricant, and the shaft sleeve someone always discovers is scored.

Kit-level thinking matters more than part-level thinking. A bearing change that waits on a seal with a six-week lead time is still a stockout, and the downtime cost is identical.

A worked mapping for one asset looks like this:

• Asset: Finishing line gearbox, GB-4120, single train, no redundancy.
• Failure mode: Input shaft bearing outer race spalling.
• Detecting signal: Vibration trend with temperature confirmation, the standard pairing in condition monitoring deployments ^[1].
• Repair kit: Bearing 6314-C3 (qty 2), input seal 45x65x10 (qty 1), gasket set (qty 1), 4 liters ISO VG 320 gear oil, locknut and tab washer.
• Lead-time tier: Bearing 14 weeks imported, seal 6 weeks, everything else 3 days. Kit lead time is 14 weeks, so the kit is Tier 1.

Note that the kit inherits the worst lead time in it. That single rule changes a surprising number of stocking decisions.

Technician notes are your best source for what a repair actually consumes versus what the engineering BOM claims. Past repair documentation, including the resolution recorded against each investigated abnormality, is how you learn that the published kit was never complete ^[8]. This is where a knowledge hub for technician findings earns its keep: the parts list in someone's head becomes a parts list in the system.

A Four-Tier Policy That Matches Stocking to Risk

Build tiers on two axes. The first is consequence of failure on that asset, which accounts for redundancy, production impact, safety, and environmental exposure. The second is lead-time exposure for the repair kit. Neither axis alone gives you an answer.

Excluding Tiers 3 and 4 from signal-driven procurement is deliberate. If every condition alert generates a purchasing event, buyers stop reading them inside a month. Governance frameworks for AI-enabled systems make the same point about mapping and managing which risks actually warrant action, so that attention lands where consequence is highest ^[2].

Wiring the Handoff: From Condition Alert to Reserved Part on a Planned Work Order

The planner workflow has to be boring and repeatable. Here is the sequence that holds up:

1. Alert arrives with failure mode, detecting channel, and a confidence indicator. Threshold logic and machine learning analysis both contribute to that assessment in modern condition monitoring architectures ^[7]. 2. Planner reviews within a defined service level and records a degradation horizon in days, not adjectives. 3. Work order created in planning status, tagged with the failure mode code and the originating alert ID. 4. Parts reserved against that specific work order. Not checked for availability. Reserved. 5. If short, requisition raised with a required-by date derived from the horizon minus kitting and window access time. 6. Schedule placed in the next window that fits inside the horizon, with the parts reservation as a release gate.

Reservation is the step most plants skip, and it is the one that quietly destroys predictive value. An unreserved bearing sitting on the shelf for a work order eleven weeks out will be consumed by a different job in week four, and nobody will notice until kitting day.

A usable work order record carries these fields:

• Failure mode code: BRG-OR-SPALL
• Detecting signal and date: Vibration, GB-4120 input bearing, trend confirmed 14 Mar
• Estimated horizon: 90 to 120 days, reviewed every 14 days
• Required parts: 6314-C3 x2 (reserved), seal 45x65x10 x1 (reserved), gasket set x1 (on PO 44021, due 02 May)
• Required-by date: 10 May, nine days ahead of the target window
• Verification step: Capture a post-repair vibration and temperature baseline once the asset is back at normal operating load, and record the technician's confirmation that the abnormality is resolved against this work order

Recording the resolution against the originating abnormality is what lets you audit whether the diagnosis was right ^[8]. This is also where CMMS integration and work-order automation matter in practice: the alert, the parts reservation, and the schedule all reference one record instead of three systems and an email thread. If you are still building that path, our write-up on moving from pilot alerts to planned work covers the handoff design in more detail.

Planner readiness checklist before release to schedule

• Failure mode code and originating alert ID populated
• Horizon estimate recorded with a review date
• Every kit line either reserved or on a PO with an acknowledged date
• Required-by date earlier than the window start by at least one week
• Craft hours and specialty tooling confirmed
• Operations has accepted the window in writing
• Verification step defined, with the measurement and the owner named

Two Plant Scenarios, Two Different Right Answers

Scenario A: single-train centrifugal compressor, 20-week rotor element. There is no redundancy and the asset gates the entire plant. A trending signal here cannot be answered by procurement, because 20 weeks of lead time will not fit inside any credible degradation horizon. The correct response is a stocking decision made long before any signal exists: an on-site spare, a consignment agreement, or membership in a rotable pool with a guaranteed exchange window. The signal's job is to trigger reservation and schedule placement, not sourcing.

Scenario B: four parallel process pumps, local distributor stock. Any one pump can be taken down without reducing line rate, and the bearing and seal kit is available in two days. Holding spares here is working capital with no risk reduction attached. The right policy is a standing requisition template, pre-approved pricing, and a two-day release path so a confirmed signal becomes a PO the same afternoon.

Same signal quality, opposite inventory answers. The difference is redundancy and lead-time exposure, which is why criticality scoring without lead-time data produces bad stocking policy. If you are scoring assets now, the method in our guide to risk-based backlog prioritization carries over directly.

For district managers and VP Operations, the governance view is comparing tier assignments across sites. One plant holds three of a slow-moving gear set while a sister plant two states away holds zero and runs the same asset. Multi-site industrial asset monitoring that reports tier decisions alongside on-hand quantities is how you find both the hoarding and the exposure, and how you decide which slow-movers should be pooled regionally instead of duplicated five times. Consistent definitions across sites matter here as much as the data itself, which is the same discipline OT architecture guidance applies to standardizing how plant systems are documented and governed ^[4].

What to Do This Week and What to Measure Next Month

Start narrow. Pick your five highest-consequence assets, the ones where a failure stops shipment. For each, list the top three failure modes from history, write out the full repair kit including seals and consumables, then pull actual lead time for every line from your purchase order history. Requisition date to goods receipt, not the vendor's quote. This takes an afternoon with a spreadsheet and an export.

Then assign each kit a lead-time tier using the table above, and write a named decision owner next to each one for signal-driven procurement. Not a department. A name.

Four metrics to begin tracking next month:

• Signal-to-requisition time by tier, with decision owner
• Lead-time variance against quoted, per supplier and per part
• Percentage of predictive work orders released with parts fully reserved
• Emergency freight spend attributable to condition alerts that arrived inside the response clock

Measurement discipline here follows the same logic as production machine learning practice: instrument the pipeline before you tune the model, because you cannot improve what you are not observing ^[6].

Translating those four numbers into avoided downtime, freed working capital, and reduced expedite spend is what makes the program defensible in a budget review. Our predictive maintenance ROI guide walks through that cost model, and the platform comparison covers which tools expose parts reservation and lead-time fields natively rather than through custom work.

Back to that gearbox. The fix was never a better algorithm. The algorithm was right in week one. The fix was a bearing and a seal sitting on the shelf before the trend started, because somebody had already decided that a single-train finishing line with a 14-week kit lead time was a Tier 1 asset.

Frequently asked questions

When should a condition signal change stock levels rather than trigger a purchase?

When the kit lead time exceeds the shortest credible degradation horizon for that failure mode. At that point procurement can never win the race, so the answer is a policy change.

Who owns the decision?

Reliability engineering owns signal confirmation and the horizon estimate. The planner owns the work order and the reservation. Procurement owns the PO. A named maintenance or reliability manager owns tier assignment and signs off on Tier 1 stocking. Write the names down.

What about low-confidence signals?

Do not send them to procurement. Send them back for additional evidence: a second channel, a lubricant sample, an on-demand reading. Technician feedback on whether an alert was real is part of how condition monitoring programs improve their own signal quality ^[1].

What if the horizon is already shorter than the lead time?

Protect the asset operationally. Reduce load, increase inspection frequency, prepare a containment plan, and run the stocking decision in parallel so the next occurrence does not repeat.

How do you estimate a degradation horizon in days rather than adjectives?

Use the trend across repeated measurements, not a single reading. Condition monitoring tracks measurements on an asset over time and flags when they move outside the established normal state, so the horizon comes from how fast the trend is moving against that baseline ^[8]. Anchor it with your own history for the same failure mode on similar assets, record the estimate as a range, and set a review date so the number gets revised as new readings arrive rather than quoted once and forgotten.

Does consignment stock or a rotable pool satisfy a Tier 1 requirement?

Only if the agreement guarantees an exchange or release window shorter than the shortest credible horizon for that failure mode, and the part is physically reachable inside that window. A consignment contract that still triggers a ten-week manufacturing run is a payment-terms arrangement, not failure-mode protection. Write the guaranteed window into the agreement and audit it against actual receipts the same way you audit a supplier's quoted lead time.

References

[1] AWS, "What is Amazon Monitron? - Amazon Monitron", AWS documentation. https://docs.aws.amazon.com/Monitron/latest/user-guide/what-is-monitron.html

[2] NIST, "AI Risk Management Framework | NIST", 2023. https://www.nist.gov/itl/ai-risk-management-framework

[3] AWS, "The Amazon Monitron workflow - Amazon Monitron", AWS documentation. https://docs.aws.amazon.com/Monitron/latest/user-guide/deployed-workflow.html

[4] NIST, "Guide to Operational Technology (OT) Security | CSRC", 2023. https://csrc.nist.gov/pubs/sp/800/82/r3/final

[5] U.S. Department of Energy, "Operations and Maintenance in Federal Facilities | Department of Energy", DOE guidance. https://www.energy.gov/cmei/femp/operations-and-maintenance-federal-facilities

[6] Google, "Rules of Machine Learning: | Google for Developers", Google developer guidance. https://developers.google.com/machine-learning/guides/rules-of-ml?hl=en

[7] AWS, "How Amazon Monitron works - Amazon Monitron", AWS documentation. https://docs.aws.amazon.com/Monitron/latest/user-guide/how-monitron-works.html

[8] AWS, "Understanding sensor measurements and monitoring machine abnormalities - Amazon Monitron", AWS documentation. https://docs.aws.amazon.com/Monitron/latest/user-guide/anom-monitoring-chapter.html