Why Breakdowns Take So Long to Fix — and What a Camera Shows

A breakdown has five timestamps: stop, report, arrive, fix, restart. Most Indian plants record only the first and the last, so the whole gap gets logged as "repair." Video timestamps all five — and in plants that measure it, the waiting half is often as long as the repair itself.

You run a plant of 200 to 1,000 workers somewhere between Rajkot, Coimbatore, Pune, Ludhiana or the Manesar belt. Your maintenance register says a machine was down 90 minutes. Your fitter says he fixed it in 25. Both are telling the truth. This article is about the 65 minutes nobody owns.

Key points

• Downtime is five events, not one. Stop → report → arrive → fix → restart. Maintenance is measured on the fix; the owner pays for all five. Standard OEE availability counts every minute of the chain against you, not just the spanner time — see OEE explained for Indian factories.
• The waiting half is invisible by design. Nobody logs "I looked for the supervisor for eleven minutes." A camera on the machine and one on the maintenance room door produce those minutes automatically, with no one filling a form.
• TPM already separates these losses. Total Productive Maintenance splits availability loss into breakdowns ("technical machine related failures") and waiting ("organisational interruptions"). Indian shop-floor registers usually merge them. Video un-merges them.
• The rupee arithmetic is simple enough to check yourself. On a 20-machine shop with 60 unplanned stops a month, cutting 25 minutes of pure waiting per event returns 25 machine-hours a month — ₹30,000 at a ₹1,200/hour contribution margin, ₹1,00,000 at ₹4,000. Your numbers, not ours; see the table below.
• Cameras diagnose response, never the fault. No camera tells you the bearing was misaligned. It tells you the operator stood there for fourteen minutes before anyone knew. Anyone selling you "AI predictive maintenance" from a CCTV feed is selling you a story.
• The law changed on 21 November 2025. The Factories Act 1948 no longer governs your plant — it was subsumed by the Occupational Safety, Health and Working Conditions Code, 2020, which came into force that day along with the other three labour codes (notification summary). Any camera policy you write in 2026 should reference the Code, not the Act.

The five timestamps

Your register gives you T1 and T5, rounded. Your maintenance MIS gives you T4 minus T3 and calls it MTTR. T2 and T3 — the report gap and the walk gap — belong to no department, so no department improves them.

That is the whole insight. It is not a technology insight. It is an accountability one.

What the three gaps actually look like

The report gap (T1→T2). An operator on a press will often not leave his station to find a supervisor — he is judged on output, and leaving looks like slacking. So he waits for the supervisor's round. Where one supervisor covers a whole bay, that round can be twenty or thirty minutes apart; on the C shift it may not come at all.

The walk gap (T2→T3). The technician is at the far end of the shed, or at the other unit, or at lunch, or already inside another machine. In the multi-shed plots typical of Gujarat and Tamil Nadu industrial estates, walking distance alone can be several minutes before anyone has touched a tool.

The part gap (inside T3→T4). The fitter arrives, opens the panel, and discovers the contactor is not in stores. Someone rides to the market. This is the tail that turns a 30-minute stop into a half-day one — and on video it is unmistakable, because the technician leaves the frame and comes back.

Each gap has a different fix. The report gap is an andon light or a WhatsApp group, not a hiring problem. The walk gap is zoning your technicians, not buying tools. The part gap is a stores problem masquerading as a maintenance problem. You cannot pick the right fix until you know which gap is eating your hours — and averaged monthly MTTR hides all three. The same logic applied to money is in the cost of factory downtime and how to measure it.

The rupee model: a 20-machine shop

Assumptions you should replace with your own — this is a structure, not a benchmark or a claim about your plant:

• 20 production machines, two shifts, 26 working days
• 3 unplanned stops per machine per month = 60 events/month
• Waiting (T1→T3) reduced by 25 minutes per event once it is measured and owned
• Recovered time: 60 × 25 min = 1,500 min = 25 machine-hours/month

Two honest notes. First, recovered machine-hours only become rupees if you are demand-constrained — if you are running at 60% of order book, saving 25 hours saves you overtime and nothing more. Second, the 25-minute reduction is an assumption, not a result we can promise: measurement creates it only because named gaps get owned. If nobody reviews the numbers weekly, you have bought a hard disk.

Weigh the annual figure against system cost in what a factory camera AI system costs in India and the payback logic in does camera AI pay for itself.

Why the gap tends to be longer in Indian plants

• Contract maintenance. Where the fitter is on a contractor's rolls, he is often not in your attendance system and not in your escalation chain. The OSH Code regulates contract labour through your principal-employer duties, but that is a compliance relationship, not a response-time one.
• Market runs for spares. Sourcing a contactor or a proximity sensor from the local electrical market — Ludhiana, Rajkot, Bhagirath Palace in Delhi — is normal practice. It is also a hole of an hour or more that no maintenance KPI captures.
• Night shifts run thin. Second and third shifts frequently have no maintenance head on site, so the report gap on C shift is the one worth measuring first.
• Supervisor rounds are the reporting system. Many mid-size plants have no andon, no call button, and no machine-level PLC alarm reaching a phone. The camera is often the first automatic signal the plant has ever had.

What cameras do not do — read this before you buy

They do not diagnose the fault. A camera cannot see a worn bearing, a drifting servo, or a hydraulic pressure drop. Predictive maintenance needs vibration, current and temperature sensors on the asset. Video answers how fast did we respond, never why did it break.

They are weak on stop detection for slow or enclosed machines. Motion-based stop detection works well on presses, conveyors, packing lines and open lathes. It is unreliable on ovens, tanks, enclosed CNC cabinets, and anything that looks identical running and idle. On those, a cheap signal tap off the machine's cycle lamp beats any camera.

They can make your people worse. If technicians learn that arrival time is on camera, some will arrive fast and idle at the machine. If you review footage to punish rather than to fix, operators stop reporting stops at all. The gain here is a process gain; used as a discipline stick it evaporates.

Timestamps drift. If NVR clock, camera clock and ERP clock are not NTP-synced, your five timestamps will not reconcile and the exercise loses credibility in the first review meeting. Sort this before day one.

Coverage is not free. You need a view of the machine and of the approach path. Cameras placed for security cover gates and aisles, not machine faces — see where to place cameras on a shop floor.

Rules before you switch it on

Two Indian regimes apply, and they are separate.

OSH and factory rules. The OSH Code 2020 has governed occupational safety since 21 November 2025, subsuming the Factories Act 1948 and a set of other central safety statutes. Your state's rules framed under the Code — not the old state Factories Rules — set your registers, safety committee and inspection obligations, and states have notified them at different speeds, so check yours. Nothing in the Code requires cameras, and nothing in it bans them in production areas. If your project touches guarded zones, pair it with proper machine guarding and danger-zone detection rather than treating one as the other.

Data protection. The Digital Personal Data Protection Act, 2023 received assent on 11 August 2023, and the DPDP Rules, 2025 were notified on 13 November 2025 with a phased runway — most substantive data-fiduciary obligations (notice, security safeguards, breach reporting, rights handling) bind from 13 May 2027. Footage of identifiable workers is personal data. Build the stated purpose, notice, retention limit and safeguards now rather than at the deadline; the practical version is in worker CCTV privacy notice under DPDP.

Practically: write the purpose as maintenance response measurement and safety, not "monitoring workers." That distinction is both legally cleaner and the reason your supervisors will cooperate.

How to start in two weeks, cheaply

1. Pick your five worst machines by downtime, not all twenty. Cameras on five machine faces plus one on the maintenance room door.
2. NTP-sync every clock. Verify by stopping a machine and checking the timestamp against your phone.
3. For 14 days, extract only three numbers per event: T1, T3, T5. Ignore everything else.
4. Put the median report gap and walk gap — by shift — on one sheet. Send it to the maintenance head and the production head together, in a daily shift report.
5. Fix the biggest gap with the cheapest tool: a call button, a zoning change, a stores min-max level. Re-measure.

If after 30 days the gaps have not moved, your problem is not visibility — it is capacity, and you need another technician, not another camera.

FAQ

Can a camera tell me why the machine broke down? No. Video shows the stop and the human response around it, not the failure mechanism. Root cause needs the technician's judgement plus, ideally, vibration or current sensors on the asset. Treat any vendor claiming fault diagnosis from CCTV footage as unserious. Cameras measure response time; sensors measure machine health. They are complementary purchases, not substitutes.

Is CCTV on the shop floor legal under the new labour codes? Nothing in the OSH Code 2020, in force since 21 November 2025, bans production-area cameras — and nothing in it requires them either. Your obligations come mainly from the DPDP Act 2023 and its 2025 Rules: a stated purpose, a worker notice, a defined retention period and reasonable security. Never place cameras in washrooms, changing rooms or crèches.

How long should I keep breakdown footage? Keep only what serves your stated purpose. For maintenance analysis, two to four weeks of full footage plus retained timestamp records is usually enough — the numbers are the asset, not the video. Longer retention may be driven by other purposes such as safety incidents, insurance claims or state women's night-shift conditions, which carry their own rules.

Will my maintenance team accept being timed? Only if the review is about gaps, not individuals. Present it as "the walk gap on C shift is 22 minutes" rather than naming the fitter. Where the first review turns into a scolding, reporting quality collapses and the data becomes useless. Frame it as a staffing and zoning question, which is usually what it is.

We already have an ERP with downtime entries. Why add cameras? ERP downtime entries are typed by a human after the fact, usually rounded and usually recorded by the person the number judges. They rarely separate waiting from repairing. Video gives you unrounded, unowned timestamps for the same events — useful precisely because nobody had to decide to record them.

Does this work on a single-shift plant with older machines? Often, yes. Older machines fail more often, giving you more events per month to measure, and single-shift plants have tighter dispatch pressure, so recovered hours convert to rupees more reliably. The real constraint is machine type, not machine age: open, visibly-moving machines are easy; enclosed or slow ones are not.