When we talk about predictive maintenance ROI with plant engineering teams, we hear the same pushback repeatedly: "The numbers in the case studies always seem too clean." Fair point. Most published PdM case studies are written by vendors whose interests are served by compelling results, and they tend to use the best-case pilot, quote the total project value over a multi-year horizon, and omit the cost categories that didn't cooperate.
This is a documented account of a 60-day pilot with a mid-size automotive stamping supplier in the Detroit area — a Tier-2 supplier running six mechanical presses between 400 and 800 tons, producing structural and closure stampings on two shifts. We're reporting what happened, including the parts that didn't go according to plan, and we're showing the actual numbers rather than extrapolated estimates.
The Starting Point: Baseline Downtime Before the Pilot
Before deploying any monitoring hardware, we spent a day with the maintenance supervisor reviewing 12 months of downtime records from their paper maintenance logs and their ERP work order system. Downtime events were categorized manually. The relevant figures for the press equipment specifically:
- Total unplanned press downtime over 12 months: 312 hours across 6 presses (weighted across the fleet)
- Events attributable to mechanical equipment failure (bearing failures, clutch/brake wear, connecting rod bushing wear, slide gibs): 61 events totaling 218 hours
- Average downtime per mechanical failure event: 3.6 hours
- Emergency parts procurement (next-day or overnight freight, markup versus standard lead time): estimated $12,400 over the 12-month period
The two highest-frequency failure modes on their floor were eccentric bearing wear (outer race spalling on the press eccentric — the main bearing that converts rotary crankshaft motion into slide linear motion) and clutch/brake assembly wear. We scoped the 60-day pilot to the three highest-downtime presses in the fleet: an 800-ton press designated P3, a 600-ton press P5, and a 400-ton press P1.
What We Monitored and How
Each press received accelerometer nodes on the eccentric bearing housing (top and bottom), the flywheel shaft bearings, and the gearbox output bearing. Thermal sensors were added to the clutch/brake assembly. Data was collected via edge computing nodes at the press station, communicating to the ingestion platform via the plant's existing Ethernet infrastructure. The edge nodes polled accelerometer data at 12.8 kHz sampling rate; the thermal sensors updated at 1-second intervals.
Installation on all three presses took a total of 11 hours of installation labor over one day, done during a scheduled weekend maintenance window. No production time was lost to installation. The edge nodes ran on 24V DC power available at the press control panel — no new power runs were required.
Baseline characterization took 10 days. During that period the platform learned the normal vibration signature for each press at its standard operating strokes per minute — including the expected impulse energy at press top dead center (TDC) and bottom dead center (BDC), which are normal kinematic events in a mechanical press cycle and should not trigger alerts.
What the Pilot Found — and What It Missed
Over the 60-day monitoring period, three meaningful findings emerged:
Finding 1 — P3 eccentric bearing, outer race wear: At day 23 of the pilot, the envelope spectrum on P3's eccentric bearing began showing a BPFO peak growing at approximately 4.2× press crankshaft frequency, consistent with the bearing geometry. Vibration RMS on the same sensor was within normal range. The plant maintenance team was notified and scheduled an inspection at the next planned maintenance stop (12 days later). The bearing was inspected and confirmed: visible outer race pitting on approximately 15% of the race circumference. The bearing was replaced during a planned 3-hour maintenance window rather than as an emergency. No production was lost.
Finding 2 — P5 clutch/brake thermal anomaly: The thermal sensor on P5's clutch assembly showed an average operating temperature 18°C above the P3 and P1 baselines, with an increasing trend over the first 30 days of the pilot. Investigation revealed the clutch air supply pressure was running low due to a partially closed manual valve in the supply line — a maintenance configuration error, not a failure. Correcting the air supply pressure normalized the thermal signature. This was a process issue, not a mechanical failure, but it was caught early enough to prevent the accelerated clutch wear that would have followed.
Finding 3 — What we didn't catch: On day 54, P1 experienced an unplanned 2.4-hour downtime event from a slide gib wear situation that had progressed unnoticed. Slide gibs are wear surfaces that guide the slide in the press column. The vibration signature from a worn slide gib is subtle and directional — it shows up in side-load vibration that was not well covered by the sensor placement in this pilot (sensors were optimized for bearing monitoring, not gib wear). This is an honest miss and worth reporting. Slide gib wear monitoring requires sensor placement on the slide guide surfaces rather than bearing housings, and it wasn't in scope for the initial deployment.
The Numbers
Over the 60-day period, comparing to the normalized baseline from the 12 months prior:
- Unplanned mechanical failure downtime on the 3 monitored presses: 2.4 hours (the P1 slide gib event). Normalized equivalent for the same 3 presses over a 60-day period from historical data: approximately 21.5 hours.
- Emergency parts cost: $0 during the pilot period for monitored presses. The P3 eccentric bearing replacement was sourced through standard procurement at standard lead time cost — approximately $340 versus an estimated $900–$1,100 on emergency overnight freight.
- Planned maintenance labor for P3 bearing swap: 3 hours at standard rate.
- Cost of the monitoring hardware + pilot deployment: We're not disclosing the specific invoice figure here, but it was in the range appropriate for a 3-press pilot deployment. The capital cost was recovered, at the hourly value of the recovered production time, within the 60-day window — primarily from the P3 bearing event alone.
We want to be direct about what these numbers do and don't show. They reflect a 60-day period on three presses. Sixty days is a short window and the 21.5-hour baseline estimate has meaningful uncertainty — mechanical failures don't distribute uniformly across time, and 60 days may have been above-average or below-average for this equipment. We are not claiming the results would extrapolate precisely over 12 months.
The Non-Financial Outcome That Matters Most to the Maintenance Team
The plant's maintenance supervisor made a comment at the end of the pilot that wasn't in any ROI model: "We spent two and a half days last year troubleshooting a bearing noise on P3 before we replaced it. This time we had the data showing us what was happening and where. The labor time alone on a reactive failure is significant even before you count the downtime."
Diagnostic labor on unplanned mechanical failures is a real cost that rarely shows up in formal downtime accounting. When a press goes down unexpectedly for a bearing failure, the maintenance team spends 30–90 minutes confirming the diagnosis, sourcing parts, and coordinating a repair plan before any wrench turns. On a planned maintenance event triggered by a condition monitoring alert, the diagnosis is already in hand — the team knows the bearing location, the suspected defect mode, and the replacement part specification. That labor efficiency is real, even if it doesn't appear as a line item in the maintenance department's P&L.
What Would Change the Calculation
There are scenarios where the ROI calculation looks worse than the pilot numbers suggest. If a plant's historical downtime is primarily driven by tooling failures, die crashes, or operator and setup errors rather than mechanical equipment failures, condition monitoring addresses only a small fraction of the overall downtime problem. The equipment monitoring costs are real and constant; the recoverable value depends entirely on what fraction of your downtime is attributable to the mechanical failure modes that vibration and thermal monitoring can detect.
There are also scenarios where the calculation looks better: a plant running three shifts, where the cost per downtime hour is higher because overtime labor and expedited customer shipments are in play, will see larger dollar values attached to the same recovered hours. The framework is consistent — the inputs are specific to your floor.
Running a 30-day or 60-day pilot on your specific equipment, with your specific baseline data, is the only honest way to generate numbers that reflect your situation rather than someone else's case study. The numbers above are from one plant, one pilot, one set of equipment conditions. They're a starting point for thinking about your own calculation, not a universal result to be cited without context.