Sourcing · 9 min read · Updated November 2026
Last-Time Buy: How Many Spares to Order (With the Formula)
When a manufacturer announces a last-order date, the question is not whether to buy spares but how many. Too few and you are at the mercy of brokers; too many and you have tied up capital in parts that will be obsolete anyway. The calculation is simple once you have four numbers — failure rate, installed quantity, remaining life and acceptable risk.
We see two typical responses to an end-of-life notice. The first is to buy "one of each", which leaves the plant with a spare CPU and no spare output module — and output modules fail three times as often. The second is to buy a full duplicate set for every machine, which is a large cheque for parts that mostly go unused. Both are guesses. The method below is what we use with customers, and it fits in a spreadsheet.
The four inputs
- Installed quantity (n) per order number — from the audit, not from memory.
- Annual failure rate (λ) per unit. Manufacturers publish MTBF figures that are optimistic for field conditions; field data from your own maintenance records is better. Where you have neither, use these starting values from our experience with PLC-class hardware in tropical plants: CPUs 1–2 %/year, digital input modules 1 %, digital output relay modules 3–5 %, transistor outputs 2 %, analog modules 2–3 %, power supplies 3–4 % (electrolytic capacitors), HMIs 4–6 % (backlights, touch screens), communication modules 1–2 %. Double these in panels above 45 °C or with poor power quality.
- Remaining life (T) in years — how long you intend to keep this platform running before migration. Be honest: if the migration budget is "someday", use ten years.
- Acceptable risk — the probability you are willing to accept of running out of spares before T. For a line-stopping part 5 % is typical; for a non-critical part 20–30 % is fine.
The formula
Expected failures over the horizon: E = n × λ × T. For a fleet of 12 CPU 315-2 PN/DP (λ = 1.5 %) over 8 years, E = 12 × 0.015 × 8 = 1.44 failures. If you buy exactly E rounded up (2) you have roughly a 50 % chance of running short, because failures are random. To cover the acceptable-risk target you add a safety margin based on the Poisson distribution — the number of events in a fixed period with a known average. The table gives the spares quantity S for a given E and risk:
| Expected failures E | S for 20 % risk | S for 5 % risk | S for 1 % risk |
|---|---|---|---|
| 0.5 | 1 | 2 | 3 |
| 1 | 2 | 3 | 4 |
| 1.5 | 2 | 4 | 5 |
| 2 | 3 | 5 | 6 |
| 3 | 4 | 6 | 8 |
| 5 | 7 | 9 | 11 |
| 8 | 10 | 13 | 15 |
| 12 | 15 | 18 | 21 |
So for the 12 CPUs above (E ≈ 1.5) with 5 % acceptable risk you buy 4 spares, not 2 and not 12. In a spreadsheet the exact figure is the smallest S for which POISSON.DIST(S, E, TRUE) ≥ 1 − risk. Add one unit for a test bench if you commission replacements before installation — it pays for itself the first time a "spare" turns out to be dead on arrival.
Adjustments that matter
- Repairability. If a part can be repaired (power supplies, HMIs, many drives) and you have a repair partner, a repaired unit returns to the pool. Reduce S by the expected repair successes — typically 60–70 % for HMIs and drives, under 30 % for CPUs with failed flash.
- Cannibalisation. When a machine is retired or migrated, its parts join the spares pool. If you plan to migrate 4 of the 12 machines in year 3, the installed base drops and E falls; recalculate per year rather than for the whole horizon.
- Shelf life. Spares age too. Electrolytic capacitors in power supplies and drives degrade unpowered; drives stored more than two years should be reformed (powered at reduced voltage) before use, and the manufacturer may not warrant a drive stored longer than that. For drives, a smaller buy repeated every few years beats one large buy.
- Firmware and revision. Buy spares that match the installed firmware family or make sure the project can be upgraded to the spare's firmware. A mixed fleet may need two spare pools.
- Common modules across platforms. A 24 V power supply (QUINT4) or a managed switch is not platform-specific; do not include those in a last-time buy — they will be available for years.
Worked example: a packaging line on FX3G
Six FX3G-40MR main units, six FX3G-4AD-ADP adapters, three GOT1000 panels; migration planned in year 5. Rates: main unit 2 %, ADP 2 %, GOT 5 %. E(main) = 6 × 0.02 × 5 = 0.6 → 2 spares at 5 % risk; E(ADP) = 0.6 → 2 spares; E(GOT) = 3 × 0.05 × 5 = 0.75 → 2 spares, but GOT1000 is already discontinued and GOT2000 is backward compatible on the serial link, so buy 1 GOT1000 and plan GOT2000 at failure. Total last-time buy: 2 main units, 2 ADPs, 1 panel — a few thousand dollars against a line that earns that every hour.
Where to buy and how to store
Place the order with the manufacturer or authorised distributor before the last-order date; after that you are buying surplus (see our surplus guide). Store spares sealed with desiccant at 15–30 °C, log them in the CMMS as strategic spares with a "do not consume for non-critical use" flag, and power up CPUs and drives once a year. When a spare is consumed, recalculate — the pool has changed and so has the remaining life.
Need engineering help with this?
Our engineers apply this to real projects every week — selection, specification, commissioning. Send your question or a part number.
Quick answers
Is MTBF from the datasheet good enough for the failure rate?
Use it only as a floor. Datasheet MTBF (often 100,000–500,000 h) is a component-level calculation at 25 °C; field failure rates in warm panels are typically several times higher. Your own maintenance history, or the starting values in this article, give better numbers.
Should I buy spares for every module or only CPUs?
Every module type with meaningful failure rate, weighted by quantity. Output relay modules, power supplies and HMIs fail more often than CPUs; a plan that only covers CPUs is the most common mistake we see.
How do I handle spares when migration is only two years away?
Shorten T to two years in the formula — the quantities fall sharply — and prefer surplus or repair for anything beyond that. A short horizon is the strongest argument for setting a firm migration date.