Case Study · 06 / 10 Auto Parts Manufacturing Save as PDF ↓
Predictive maintenance · 10-week build

Machine 7 was going to fail. We told him on Tuesday.

How a Pune auto parts plant stopped buying servo motors at 2 AM.

ClientAnkur Polymers LocationChakan, Pune PublishedMay 2026
Industry
Tier-2 auto parts · injection molding
Scale
18 machines · 2 shifts
Team
240 people · ₹85 Cr ARR
Engagement
10 weeks · IoT + ML
Outcomes · Month 4 post-launch
−73%
Unplanned downtime per month — 67 hrs to 18 hrs
71hr
Avg failure prediction lead time, from 0 hours
62→79%
OEE moved 17 points · industry benchmark cleared
§ 01 · The company

A Chakan plant supplying Tata, Mahindra, Bajaj.

The company

Ankur Polymers is a Tier-2 auto parts manufacturer in Chakan, Pune, supplying interior plastic components — door handles, AC vents, dashboard trims — to Tata Motors, Mahindra, and Bajaj Auto. They run 18 injection molding machines across two shifts, employ 240 people, and clock around ₹85 Cr in annual revenue.

Plant Head Rohan Kulkarni spends most Mondays in OEM penalty review calls. What keeps him up: the 11 PM WhatsApp from a shift supervisor saying "Sir, Machine 12 is down again."

§ 02 · The problem

67 hours of unplanned downtime. Every month.

FY24 baseline

In FY24, Ankur logged 67 hours of unplanned downtime per month across the floor. Each downed machine cost roughly ₹52,000/hour in lost production plus expedited spare-part runs to Bhosari market at odd hours. A maintenance team of six was reactive by design: a barrel heater would burn out, a hydraulic pump would seize, and only then would someone call a vendor.

False operator alerts ("machine sounds weird") ate another 30% of their bandwidth.

Breaking pointNov 2024

A servo drive failure on Machine 14 killed a Bajaj front-panel order. Penalty plus a yellow card from the OEM's quality team. OEE sat at 62% against an industry benchmark of 75%. Rohan had budget approved for two new machines — then realised the old ones weren't dying. They were being killed by surprise.

₹11.3L penalty · OEM yellow card
§ 03 · The solution

Sensors on every machine. Anomaly model on top.

Architecture

Vynara retrofitted all 18 machines with vibration sensors on hydraulic pumps and servo motors, current clamps on the main supply line, and thermocouples on barrel zones. Sensor data streams to an industrial edge gateway on each machine, then to a central server running anomaly detection models trained on Ankur's own failure history.

When a vibration signature drifts or current draw shifts beyond baseline, the system files a work order in their existing maintenance app and pings the right technician on WhatsApp with the expected failure window. Diagnosis, repair, and the "take it offline now or after shift?" call stayed human. Build: 10 weeks.

Humans still own
diagnosis confirmation · repair execution · root-cause review
§ 04 · How it works

Surprise failure → planned intervention.

Before · Reactive 8.4 hr stoppage avg
  1. Machine runs · operator hears odd noise
  2. Calls supervisor · supervisor checks
  3. Maybe logs ticket
  4. Failure happens · production stops
  5. Vendor called · spare sourced from Bhosari
  6. Repair · restart
After · Predictive 45 min planned
  1. Sensors stream 24/7 · model detects drift
  2. Work order auto-filed
  3. Technician notified on WhatsAppavg 71-hr failure window
  4. Maintenance slotted into planned changeover
  5. Part pre-ordered · no rush sourcing
  6. Repair done · no production stop
§ 05 · The numbers

Six metrics. Month 4.

Unplanned downtime (hrs / month)−73%
Before67 hrs
After18 hrs
Mean time between failures+2.6×
Before280 hrs
After720 hrs
Maintenance spend / month−39%
Before₹14.2 L
After₹8.7 L
OEM penalty incidents / quarter−80%
Before5
After1
Avg failure prediction lead time+71 hr
Before0 hrs
After71 hrs
Overall equipment effectiveness+17 pts
Before62%
After79%
§ 06 · The timeline

10 weeks. All 18 machines.

Phase
W1
W2
W3
W4
W5
W6
W7
W8
W9
W10
Weeks 1–2Floor audit · failure history backfill · sensor spec
AUDIT
AUDIT
Weeks 3–4Pilot install · M7, M12, M14 · edge gateway
PILOT
PILOT
Weeks 5–6Baseline data · anomaly model training
MODEL
MODEL
Weeks 7–8Dashboard · WhatsApp alerts · work order integration
INTEGRATE
INTEGRATE
Week 9Rollout to remaining 15 machines
ROLLOUT
Week 10Maintenance training · shift handover · go-live
LIVE
§ 07 · What they said

Stopped chasing fires. Started planning weeks.

"

Last Tuesday we got an alert that the servo drive on Machine 7 would fail by Friday. We swapped it during Saturday's planned downtime. Six months ago, that same drive would have taken out a Tata order on a Monday morning.

RK
Rohan Kulkarni
Plant Head · Ankur Polymers
"

My team used to chase fires. Now they actually plan their week.

SP
Suresh Pawar
Maintenance Manager · Ankur
§ 08 · The takeaway

Teach old machines to talk.

You don't need new machines to fix unplanned downtime. You need to teach old machines to tell you they're tired.

Any plant with 10+ machines and a year of maintenance logs already has the raw material. If your night-shift WhatsApp group is louder than your day one, we should talk.

One-liner We cut a Pune auto parts plant's unplanned downtime by 73% in 10 weeks. Same machines. Better sensors.
Walk us through your floor →