The SMT Rework Queue: What Boards Waiting After AOI and ICT Cost an Indian EMS Plant (₹)
Every board on the rework rack costs you twice: once in technician time, flux and parts, and again in the days it sits there while the line keeps making the same defect. In a mid-size Indian EMS plant the direct cost of a reworked board is often under ₹100. The expensive part is the queue, because it delays the feedback that would have stopped the defect at the printer or placer.
By the Mama Editorial Team. Mama scopes camera projects for Indian factories. Last verified: 13 October 2026, against the IPC table of contents, the Uttar Pradesh wage order and the industry papers cited inline. All other figures are worked assumptions and are labelled as such.
Most owners know their AOI reject rate. Fewer know how many boards sit on the "awaiting rework" rack right now, or how old the oldest one is. Those two numbers say more about an SMT line's health than the reject rate does.
Key takeaways
- Rework is a queue, and queues have a law. Average boards on the rack equals daily arrivals times average days waiting (Little's Law). Count the rack every morning and you know your feedback delay.
- Direct cost per reworked board is modest. In our worked example it comes to about ₹89, including a share of scrap. That is roughly ₹1.1 lakh a month at 48 boards a day.
- The queue is the expensive part. A two-day backlog means a process fault runs for two days before anyone reads the pattern.
- Most SMT defects start upstream. One AOI vendor's market research puts 51% of defects at paste printing and 38% at placement. A busy rework bench is usually a printer or program problem.
- Not every AOI call is a defect. Unreviewed calls inflate the queue, and touching a good joint adds a heat cycle for nothing.
- Cameras count the rack and the bench. AOI and ICT tell you why. You need both.
What IPC-7711/7721 actually covers
The reference document for this work is IPC-7711/7721, Rework, Modification and Repair of Electronic Assemblies. The current edition is Revision D, dated December 2023, which supersedes Revision C of January 2017. It is one document in two halves:
- 7711 (rework) covers component removal and installation: through-hole desoldering, land preparation and tinning, and BGA/CSP removal and installation, including reballing by fixture, paper carrier and stencil methods.
- 7721 (repair and modification) covers damage to the board itself: delamination, lifted conductors and lands, plated-hole repair and jumper wires.
The three words are not interchangeable, and the difference matters commercially. The full text is paywalled; the wording below is from earlier revisions (A and B), as quoted by an industry expert panel. Check Revision D's definitions (section 1.7) before you write a procedure:
- Rework reprocesses a non-complying board "in a manner that assures full compliance" with drawings and specifications.
- Repair restores function "in a manner that precludes compliance" with drawings and specifications (Revision A). Revision B says "does not assure compliance".
- Modification revises "the functional capability of a product in order to satisfy new acceptance criteria."
The practical rule for an owner is that rework restores a board to spec. Repair does not, so it is your customer's decision, not your technician's. A repaired board sitting on the rack is often waiting for a customer email, not for a soldering iron.
The IPC-A-610 class also shapes the queue. Class 1 is the most lenient, Class 2 is the usual choice for non-critical assemblies, and Class 3 means building to all of the criteria. A Class 3 job typically sends more boards to rework from the same process.
Why the rack is a queue, and what that tells you
Little's Law says the average number of items in a stable system equals the arrival rate times the average time each item spends in it. It needs no software, only two numbers you can count:
- Arrivals: boards sent to rework per day, from the AOI review station and the ICT or functional-test bench.
- Boards on the rack: count them at the same time every morning.
Divide the second by the first and you have the average number of days a board waits. If 48 boards arrive a day and the rack holds 96 most mornings, a defective board waits about two days. That means the rework technician, the only person who handles every failed board, sees the pattern two days after it started.
That delay is where the money goes. A partly clogged stencil aperture or a placement program saved with an offset keeps making the same defect until someone connects the dots, and the bench is where the dots sit side by side.
Rework cost per board: a worked ₹ example
Assume one SMT line in Noida running two shifts and making 1,600 boards a day. 3% of boards (48 a day) go to rework after AOI review and ICT. One rework technician works each shift. The average board value at this stage is ₹800 (use your own figure). Average hands-on time is 12 minutes per board. That is a blend: a solder bridge takes a few minutes, while a BGA swap with reballing can take most of an hour.
For wages, we use the Uttar Pradesh interim minimum-wage order for Gautam Buddha Nagar and Ghaziabad (Category I), No. 374/36-2-2026 of 17 April 2026, effective 1 April 2026 (state's English translation, third-party copy). It sets the skilled rate at ₹16,868 a month (basic ₹15,876 + VDA ₹992) and the semi-skilled rate at ₹15,059. These are minimums; a trained rework technician often earns more. Assume a 1.3× loading for statutory contributions and leave, and 208 working hours a month.
| Cost line per reworked board | Basis (assumptions marked) | ₹ |
|---|---|---|
| Rework technician, 12 min | Skilled ₹16,868 × 1.3 ÷ 208 h ≈ ₹105/h | 21 |
| Re-inspection or retest, 5 min | Semi-skilled ₹15,059 × 1.3 ÷ 208 h ≈ ₹94/h | 8 |
| Logging, handling, rack moves, 3 min | Semi-skilled rate | 5 |
| Flux, wire, wick, average replacement part | Assume ₹15 average, excluding BGAs | 15 |
| Scrap share | Assume 1 in 20 reworked boards is scrapped at ₹800 | 40 |
| Direct cost per reworked board | ≈ ₹89 |
At 48 boards a day and 26 working days, that is about ₹4,270 a day, or roughly ₹1.1 lakh a month. The cost of the queue sits on top:
| Same line | Rack holds ~1 day of arrivals | Rack holds ~3 days of arrivals |
|---|---|---|
| Boards on the rack (48/day × days waiting) | ~48 | ~144 |
| Value sitting unbilled at ₹800/board | ~₹38,400 | ~₹1.15 lakh |
| A print fault adds 2% extra fallout at 100 boards/h; production hours before the pattern is read (assume 16 h/day) | 16 h → ~32 extra bad boards | 48 h → ~96 extra bad boards |
| Direct cost of those extra boards at ₹89 | ~₹2,850 | ~₹8,540 |
| Overtime to clear the backlog | Rarely needed | Likely: overtime needs the worker's consent and is paid at twice the ordinary rate of wages (OSH Code s.27), so about ₹162 an hour on a ₹16,868 basic + DA |
A three-day rack holds more cash, lets every process fault run three times longer and pulls your best technician into overtime fixing symptoms. The risk no table captures is a customer escape from a lot shipped before the pattern was spotted.
When the queue is hiding a process problem
A full rack is a symptom. Before you hire a second rework technician, check which of these it is.
The same defect on the same reference designator. If the rework log keeps showing bridges on U7 or tombstones on C14, the fault is upstream. In one AOI vendor's market research on world-class lines, defects split 51% after paste printing, 38% after placement and 9% after reflow. This is a vendor figure, not an independent study, but the direction is widely accepted: fix the stencil, paste handling or program, and the rack shrinks without anyone working faster.
Unreviewed AOI calls. An older IPC/APEX conference paper comparing four AOI systems measured false-call rates of roughly 664 to 855 per million opportunities across systems. Its authors concluded there "really are no such things as 'false calls'": the machine flags image variation, and a human has to reconcile it. On a board with 1,000 inspection opportunities, those rates would mean roughly 0.7 to 0.9 flags per board. That is an illustration, and your boards will differ. If the AOI review station is rushed or skipped, flagged-but-good boards land on the rework rack. Reworking a good joint costs a heat cycle and adds risk without fixing anything.
Boards waiting for a part, not a person. A board that needs a replacement IC is held up by stores, not by the bench. Moisture-sensitive replacements add their own wait (see our piece on MSL components sitting open on the SMT floor).
Boards waiting for diagnosis. An ICT fail is a question, not a rework order. If one engineer debugs everything, the rack is really a queue at that desk.
Boards waiting for a decision. Repairs, which by definition do not assure compliance, need customer sign-off. They sit until someone writes the email.
A skill bottleneck. If only one technician is trusted with BGA work, every BGA board waits for that shift. Like the waiting on manual benches in our piece on where a mid-size EMS line loses hours, it shows up nowhere on the machine reports.
What a camera shows, and what only AOI and ICT data can tell you
| Question | Camera over rack and bench | AOI / ICT / MES data |
|---|---|---|
| How full is the "awaiting rework" rack, morning and evening? | Yes, as tray or slot counts on a fixed rack | Only if every move is scanned |
| Is the rework bench occupied or empty, and for how long? | Yes, as aggregate occupied minutes per shift | No |
| Did trays leave the rack and go back to the line? | Yes, as movement events | Only with scanning |
| Is the bench idle while the rack is full (parts, decision or debug wait)? | Yes, as a pattern | No |
| Which defect, on which reference designator? | No | Yes |
| Was the AOI call real or false? | No | Yes, at the review station |
| Did the board pass after rework? | No | Yes, on retest |
| Root cause: stencil, paste, program or footprint? | No | Yes, with engineering analysis |
A camera reads time and occupancy, not solder. CCTV is no substitute for AOI, SPI or a test jig. Our overview of cameras in EMS plants covers the rest of the floor.
Limits and cautions
- A camera cannot see board identity. It counts trays and slots, not serial numbers. Ageing by board needs a scan at rack-in and rack-out.
- Rack counts depend on discipline. Boards piled on a side table are not counted. Mark the rack first.
- Bench occupancy is not productivity. Treat it as a pointer to look closer, not a score.
- People, not names. Count aggregate bench minutes and rack levels, never rank individual technicians. Put up a worker notice before any camera goes live; our DPDP worker CCTV notice template covers the wording. Keep cameras angled away from screens showing customer artwork or gerbers.
- Our ₹ figures are illustrations. Replace the reject rate, board value, touch time and wages with your own two weeks of data before you act on them.
Where Mama fits
Mama watches the rework rack and bench through cameras you already have, or a couple you add, and each morning gives the owner a short note: how many boards waited overnight, how long the bench sat idle while the rack was full, and whether the backlog is growing. It looks at racks, trays and benches, not at named people, and it leaves defect analysis to your AOI and test data, where it belongs. If you want a view of where to point those cameras: Leave an email or phone in the form below, and Mama will get back to you with a camera plan.
FAQ
What does SMT rework cost per board in India? Using the UP Category I minimum wages from April 2026, a 1.3× loading, 12 minutes of hands-on time and a scrap allowance, our worked example comes to about ₹89 per board. A BGA replacement costs several times that. The bigger cost is how long boards wait, because a fault keeps running while they sit.
What is the difference between rework and repair under IPC-7711/7721? Rework brings the board back into full compliance with drawings and specifications. Repair restores function without assuring compliance, so it normally needs your customer's agreement. Modification changes the product to meet new acceptance criteria and should only be done when authorised.
How many boards on the rework rack is too many? Divide the morning rack count by daily arrivals to get days of waiting. A reasonable working target is under one shift; that is our suggestion, not an industry standard. If it is over a day, look upstream at printing, placement and AOI review before you add rework staff.
Can a camera tell me why boards are failing? No. A camera counts rack levels, bench occupancy and tray movements. Defect type, false calls and retest results come from AOI, ICT and your MES.
Should I just add a second rework technician? Only after checking the rework log. If the same defects repeat on the same parts, a second technician only fixes a process fault faster. Add capacity when the defect mix is random and the bench itself is the constraint.