Case study · Program management · Vendor management · Reliability

Pump Fleet Conversion

A fleet of process-vacuum pumps kept dying in infancy. I ran the program that converted 98 of them to a modular design — they now last about twice as long — and built the business case that paid for it.

Work: Modular Pump Conversion Program
Role: Program lead — equipment reliability engineer
Timeline: 12-month cross-functional rollout
Vendors: 6 suppliers evaluated and managed
Outcome: ≈$330K/yr savings, ≈$1.2M cumulative; ≈1.9× MTBF
Breakeven: ≈2.8 years on conversion capital

The fleet’s dry pumps were failing early — corrosion-driven, well before their design life. Each unplanned failure risks scrapping everything the process tool is working on and takes the tool down for hours. The failure data pointed to a modular pump architecture that isolates the corrosion-prone stage into a replaceable block.

My job was everything between that insight and 98 converted pumps: the failure analysis, the capital justification, selecting among six vendors, the FMEA that gated each rollout phase, PO and spares tracking, and the warranty program that kept suppliers accountable afterward.

01.1The reliability problem

New pumps should fail like old machines: rarely, and only at the end. These failed at the beginning — the left wall of the bathtub curve. Early-life failure is a program problem, not a maintenance problem: no PM schedule can fix a unit that dies before its first PM.

DiagramFailure-rate curve, redrawn. The shaded zone is where the legacy fleet lived; the program’s goal was to push the population into the flat middle.

01.2The money chart

The boxplot below is the entire program in one image — the same paneled-boxplot view the real analysis used. Before: median life around 8,000 hours. After conversion: around 15,600. Every hour in between is a failure that didn’t happen and product that didn’t get scrapped.

Recreation · synthetic dataSame chart form as the production analysis — paneled boxplots with individual units jittered behind them; distributions and counts re-rolled synthetically.

01.3The business case

Reliability arguments don’t fund programs; cashflow arguments do. The proposal priced each conversion at roughly $17K per pump and weighed it against avoided failures, avoided scrap risk, and reduced rebuild spend. It cleared at about a 2.8-year breakeven and six-figure annual savings — then beat the estimate in production.

Line item	Value	Basis
Conversion cost per pump	~$17K	hardware + labor + qualification
Initial program capital	~$200K	first conversion phase
Annual savings at full fleet	~$330K/yr	avoided failures, rebuilds, scrap-risk hours
Cumulative to date	≈$1.2M	tracked against baseline failure rate
Breakeven	≈2.8 yr	conservative case as approved

Recreation · rounded figuresCost model rebuilt with rounded, resume-consistent numbers.

01.4An FMEA-gated rollout

Ninety-eight pumps is too many to convert on faith. A failure-mode and effects analysis scored each risk at three stages — current state, at install, and post-implementation — and each rollout phase only proceeded when its high-RPN items had owners and closure evidence.

Failure mode	Effect	S	O	D	RPN	Mitigation	Post-RPN
Corrosive attack on exposed stage	Early-life pump failure	9	7	4	252	Modular corrosion-resistant block	72
Wrong block variant installed	Repeat failure, lost conversion	8	4	3	96	Variant matrix by tool type; install checklist	32
Spares gap during phase-in	Extended downtime on failure	7	5	2	70	Dual-sourced spares; rebuild loop sized to fleet	28
Vendor rebuild quality drift	Reduced converted-unit life	7	4	4	112	Warranty tracking + quarterly KPI reviews	42
Install-window overrun	Production schedule impact	6	3	2	36	Phased by tool availability; standard work	18

Recreation · generic wordingRepresentative rows with invented scores; the production FMEA’s structure, not its content.

DiagramThe twelve months, compressed. Warranty and spares workstreams ran past the last install by design.

01.5What I’d do differently

Start the spares and rebuild-loop conversation with vendors before the pilot, not after — the phase-in spares gap was the closest the program came to a schedule slip. And publish the vendor KPI scorecard from month one; suppliers behave differently when they know the chart exists. That scorecard became its own tool, and it lives in Fleet Analytics.

The discipline I hold hardest came from a call I got wrong. A long-down event: we swapped components, exchanged the pump, exchanged it again — and the pump was never the problem. A recipe upstream was flowing more gas than the pump could handle, and my team wasn’t aware of it, because we’re a subsystem and nobody tells the pump crew about flow changes. The fix wasn’t hardware; it was a meeting with the adjacent process team to lessen the load. Next time — and every time since — I ask better questions first and work to understand the issue externally from my direct system.