Specs seduce buyers. They do it because a clean “CPH” number looks scientific, while feeder prep, nozzle changes, board handling, fiducial retries, reel shortages, and line imbalance look like the expensive chaos they really are. And what gets approved faster?
I will say the impolite thing. Theoretical placement speed is a lab ceiling, not a factory answer, and any buyer comparing pick-and-place machines, configuring mixed SMT lines, or pricing a turnkey SMT line solution without a board-level throughput model is still shopping by brochure, not by output.
The hard truth gets uglier once money enters the room. In the 2024 NIST Annual Report on the U.S. Manufacturing Economy, downtime is estimated at 8.3% of planned production time and $245 billion across discrete manufacturing; that alone should end the fantasy that nameplate speed survives contact with a live production floor.
The number that matters
Throughput modeling is just disciplined subtraction. You start with what the machine could do in a sanitized test, then remove every real production penalty your line will absolutely suffer, whether you admit it in the sales meeting or not.
Theoretical placement speed = maximum placements per hour under ideal machine test conditions
Actual placement speed = total placed components ÷ total attributed production time
Placement efficiency = actual placement speed ÷ theoretical placement speed × 100
That “attributed production time” is where adults separate themselves from catalog readers. I include board transfer, fiducial acquisition, vision alignment, nozzle swaps, feeder replenishment, micro-stops, changeover share, operator intervention, and any upstream or downstream blocking. If you ignore those, your placement speed calculation is not wrong by a little. It is wrong by design.
What usually destroys the model
The ugliest lie in this business is that speed loss comes from one big problem. It does not. It comes from seven medium problems that pile up quietly and then murder your shift output.
| Loss driver | What goes into the model | First-pass planning drag I usually assign |
|---|---|---|
| Component mix | 0402-only runs behave nothing like QFP, BGA, connector, or odd-form boards | 5% to 35% |
| Feeder slotting | Long travel paths, poor slot allocation, and reel fragmentation slow every cycle | 3% to 12% |
| Nozzle changes | Head changes and verification steps punish mixed products fast | 2% to 10% |
| Board handling and vision | Fiducials, centering, clamping, and transfer time are real time, not free time | 5% to 15% |
| Changeover allocation | Small batches get crushed when setup minutes are spread across too few boards | 5% to 30% |
| Line interaction | Printer, AOI, reflow, buffers, and conveyors can starve or block placement | 5% to 25% |
| Quality losses | Mis-picks, polarity errors, verification holds, and rework loops are hidden throughput taxes | 2% to 10% |
I do not trust “machine speed” discussions that ignore process quality or pretend the placer lives alone. On real high-speed mass production lines, the bottleneck often moves off the placer and into print stability, AOI false calls, reflow loading, or material handling. That is why I prefer real customer cases over showroom demos.
A blunt example
Here is the math I wish more buyers forced vendors to show. Not the poster. The math.
Assume one machine is rated at 80,000 CPH theoretical.
Now run a real batch:
Batch size = 200 boards Placements per board = 320 Total placements = 64,000
Now attribute the time honestly:
Pure run time = 62 minutes Changeover share = 12 minutes Feeder replenishment = 6 minutes Micro-stops and checks = 4 minutes
Total attributed time = 84 minutes = 1.4 hours
So:
Actual placement speed = 64,000 ÷ 1.4 = 45,714 CPH
Placement efficiency = 45,714 ÷ 80,000 = 57.1%
That is not a bad line. That is a real line.
And here is where buyers get trapped: they see 57.1% and assume failure. I see 57.1% on a mixed board, moderate batch, and honest time accounting, and I think, fine, now we can improve something real.
Why brochure speed keeps embarrassing people
Demand came back, and bad habits came back with it. In Spring 2024, WSTS projected the global semiconductor market at $611 billion, up 16.0% year over year, which means more factories have once again been pushed into fast capacity decisions, and fast capacity decisions usually mean somebody starts comparing CPH before they model board mix. (wsts.org)
The research says what the sales deck avoids. In a 2023 PCB assembly case study by researchers from the University of Calabria and York University, adding one operator cut average working time from 56 minutes to 43 minutes per hour, a 23% drop, and the authors said the right operator combinations could reduce processing times by up to 50%; that is throughput analysis for SMT line planning in plain English, because the machine did not become magical, the system simply stopped fighting itself. Simulation-Driven Analyses of Performance of PCB Assembly Operations: A Case Study
A 2024 study on spin-head surface mounters made the software side of the same argument. The authors tested nozzle assignment, feeder assignment, and component sequencing combinations in simulation, then explicitly recommended one combination, 2-2-1, for practical use after it minimized total processing time better than the alternatives; that should kill the lazy idea that programming is secondary after machine selection. Simulation-Based Hierarchical Heuristic for Printed Circuit Board Assembly Optimization in a Spin-Head Surface Mounter (mdpi.com)
So, yes, I am skeptical. I am skeptical of any comparison between a Yamaha YSM40R, Panasonic NPM-W2S, Hanwha XM520, Juki RS-1R, or Fuji NXT-class machine that does not force the same PCB, the same feeder constraints, the same batch size, the same operator assumption, and the same downstream limits into the model. Otherwise you are not comparing machines. You are comparing marketing departments.
The model I would actually use
I keep it simple first. Then I make it painful.
Step 1: Start with placements per board, board count, and theoretical CPH. Step 2: Add board-handling seconds per board. Step 3: Add vision and alignment time by component mix. Step 4: Add nozzle-change and feeder-replenishment penalties. Step 5: Allocate changeover time across the batch. Step 6: Add expected stops, verification holds, and operator response time. Step 7: Check whether printer, AOI, buffer, or reflow caps line output lower than the placer.
That final step matters more than most people want to admit. A fast placer inside a badly balanced line is just a very expensive way to create waiting inventory.
One more hard truth
Buyers ask, “What is the best machine?” I think that question is usually lazy. The better question is, “What machine gives me the best actual throughput on my board family, my lot size, my feeder discipline, and my staffing model?”
That is why throughput modeling beats showroom theater. Every time.
FAQs
What is throughput modeling in SMT?
Throughput modeling in SMT is the practice of converting machine specs, board data, feeder layout, changeover time, operator interaction, and line constraints into a realistic output rate, usually expressed as actual placements per hour or boards per shift instead of the maximum CPH printed in a machine brochure.
It is the difference between planning with production physics and planning with wishful thinking.
How do you calculate actual placement speed?
Actual placement speed is calculated by dividing the total number of placed components by the total attributed production time, where attributed time includes not just pure mounting time but also board handling, fiducial reading, nozzle changes, feeder replenishment, changeover share, micro-stops, and verification delays.
In short: count placements honestly, count time honestly, then divide.
Why is theoretical placement speed always higher than actual placement speed?
Theoretical placement speed is higher because it assumes near-ideal operating conditions, while actual placement speed absorbs the time losses caused by board transfer, vision alignment, part diversity, feeder travel, setup, operator intervention, downstream blocking, and quality-related interruptions that are unavoidable in normal SMT production.
Theoretical speed is a capability indicator. Actual speed is an operating result.
What is the best way to compare two pick and place machines?
The best way to compare two pick and place machines is to run the same PCB data, component mix, feeder rules, batch size, changeover assumption, operator model, and downstream line limits through the same throughput model so the only meaningful variable left is machine behavior.
Anything less is sales theatre with decimal points.
What is a good actual-to-theoretical placement ratio?
A good actual-to-theoretical placement ratio is one that is calculated consistently across the same product family, staffing model, and line conditions, because a single universal benchmark is misleading and often punishes honest factories while flattering demo-friendly setups that never resemble live production.
I do not trust floating percentage targets without context. I trust repeatable modeling against real boards.
If you want a real answer instead of a glossy one, send your centroid data, placements per board, batch size, and changeover target through the contact page. Or start by reviewing your turnkey automation options and current resource library before you spend one more dollar on brochure speed.
