English Views: 0 Author: Elsa Long Publish Time: 2026-11-10 Origin: Site
Planning a JUKI mounter for high mix low volume production is not a machine selection exercise with a longer feature list. It is a changeover budget exercise. The machine that wins on one product at a time is rarely the machine that wins on forty products in a month, because the cost of a mixed schedule lives in the transitions between products — the feeder swap, the nozzle change, the program load, the first article, the re-qualification of a station that is suddenly the constraint. A configuration that ignores those transitions will look excellent on the quotation and behave badly on the floor.
This guide works through five steps in the order a project actually meets them: counting changeovers before counting components, building the feeder plan around the changeover rather than the machine, deciding how many program layers the product list really needs, sizing the nozzle set that has to be shared across the mix, and finding the bottleneck when every product has a different one. Each step produces a number that the next step consumes, and the last section assembles those numbers into a configuration budget that can survive a review. The platforms referred to throughout are the ones in the placement machine range this configuration plan is built on.
Most configuration conversations start with the machine: how many heads, how many feeder positions, what placement rate, which vision system. In a single-product line those questions are close to sufficient, because the machine runs one program for weeks and the only thing that changes is the reel. In a mixed schedule the same questions describe less than half of what the line actually does. The machine places components for a few hours and spends the rest of the shift being prepared, verified, emptied and refilled, and the configuration decision that matters is whether the machine was bought to place components or bought to survive transitions. Those are different purchases with different line items, and the second one is the one a high-mix plant needs. Getting this wrong does not show up as a broken machine. It shows up as a line that is technically capable of 40,000 components per hour and practically delivers eleven good boards in an afternoon.
The reason the mistake is so common is that the specification sheet is written in the language of steady state. Feeder count, placement rate, component range and accuracy all describe a machine that is already loaded and already running. None of them describes the forty minutes before that state is reached. A configuration review therefore has to add a second column to the specification — one that counts transitions, not components — and the two columns are frequently in conflict. More feeder positions usually means more pre-loaded work; more heads usually means a larger nozzle set; a wider component range usually means more program variants. The job is not to maximize either column but to find the point where the sum of placement time and transition time is lowest for the actual product mix. That point is rarely the cheapest machine and rarely the fastest one.
Before any feature is discussed, calculate how many times the line will stop being one product and start being another. Take the annual board volume, divide it by the average batch size, and divide again by the number of production days. A plant building 240,000 boards a year in batches of 600 across 240 working days performs roughly 1.7 changeovers per day; a plant building the same 240,000 boards in batches of 150 performs 6.7 per day. The two plants have identical output and completely different configuration requirements, and the second one will fail with a machine specified by the first one's logic. This single ratio explains most of the disappointment that follows a high-mix purchase, because it changes which cost dominates: at 1.7 changeovers a day, placement speed sets the day; at 6.7, changeover time sets the day and placement speed merely fills the gaps between changeovers.
The calculation is easy to get wrong in a subtle way. Batch size in an ERP system is the release quantity, not the run quantity. If a work order for 600 boards is split across two shifts because of material availability, the line sees two changeovers, not one. Ask for the number of distinct program loads per week from the shop floor rather than from the planning system; the two numbers usually differ by 20 to 40 percent, and the floor number is the one the machine will experience. Once that figure is in hand, the configuration conversation changes character immediately — feeder preparation, nozzle logistics and program handling stop being support functions and become the primary design constraints.
A line configured for one product embeds four assumptions that nobody writes down. First, that the feeder set is stable: the same reels sit in the same slots for weeks, so feeder count can be sized to the BOM and no more. Second, that the nozzle set is stable: one or two nozzle types cover the board, so a fixed head configuration is fine. Third, that the program is stable: it is created once, validated once, and then only edited when the design changes. Fourth, that the bottleneck is stable: whichever station limits output today will limit it next month, so improvement effort can be aimed and left in place. Every one of those assumptions is false in a mixed schedule, and each fails in a different way.
The feeder assumption fails first, because a mixed schedule needs more feeder positions than any single product's BOM. If the plant runs forty products with an average of thirty-five unique part numbers, the installed feeder inventory has to cover the union of the frequently used parts, not the average. The nozzle assumption fails second, because a mixed BOM spans 0201 chip components through connectors and LEDs, and the head has to carry a set that covers all of them without a full teardown at every changeover. The program assumption fails third, because forty products means forty programs, forty parameter sets and forty opportunities for a version to drift. The bottleneck assumption fails last and hurts most, because in a mixed schedule the constraint moves with the product — the station that limits a dense digital board is not the station that limits a sparse LED board. A configuration plan that treats any of these four as fixed will be re-planned within six months, usually at the buyer's expense. The screening logic that decides which platform is even worth configuring this way is set out in the platform screening that comes before any configuration decision.
Most configuration reviews open with a machine comparison: three models side by side, columns for heads, slots, placement rate, component range and price. That format is inherited from single-product purchasing and it quietly decides the answer before any question about the product mix has been asked, because the columns reward the machine that is best at steady state. A plant that fills in such a table has usually already accepted the premise that the configuration is a machine choice, and everything that follows is a defence of that premise.
The alternative opening is a two-page document that contains no machine names at all: the product list grouped into families, the annual volume per product, the batch size actually run, the changeovers per week measured on the floor, and the changeover duration split into its four phases. That document is uncomfortable to produce, because it exposes scheduling habits and measurement gaps that the plant would rather not discuss. It is also the only document that can tell a supplier what to configure, rather than asking a supplier to guess. In practice the review that starts here takes longer to prepare and much less time to conclude, because the machine shortlist falls out of the numbers instead of being argued. A JUKI mounter for high mix low volume production is specified by transitions, and transitions are counted before they are bought.
Changeover time is usually quoted as a single figure, and that figure hides four distinct costs that respond to different investments. The first is material time: physically removing reels, returning them to stock, pulling the next kit, loading it and confirming that every position matches the program. The second is program time: loading the new program, checking that the revision is the released one, and setting the parameters that the program does not carry. The third is qualification time: running the first article, measuring it, and deciding whether the line may continue. The fourth is recovery time: the ramp back to steady state, in which the first few boards run slower and the operator watches the stations that misbehaved last time. Only the first of these is reduced by buying more feeder positions. The second is reduced by program discipline, the third by metrology and a defined acceptance rule, and the fourth by documentation that tells the operator what normal looks like.
Treating changeover as one number leads to a predictable purchasing error: buying feeder capacity to solve a problem that is actually program time or qualification time. A plant that adds eighty feeder positions but still spends twenty minutes hunting for the correct program revision has bought storage, not speed. The practical way to avoid this is to time the four phases separately for ten real changeovers and record them in a simple table. The distribution almost always surprises the team — material time is often shorter than expected because operators are good at it, while program and qualification time are longer, because both depend on information that is not on the floor. That measurement, not a brochure, is the input to the configuration budget described later in this guide.
The first step produces a single number that every later decision depends on: changeovers per week, split by product family. It sounds trivial and it is routinely wrong, because the number that planning reports and the number the line experiences are computed from different things. Planning counts work orders; the line counts program loads, and a work order that is paused and resumed, or split across two shifts, becomes two program loads. Before the number can be trusted, it has to be built from the floor's own record — the program load log, the setup sheet, or the line leader's notebook — and reconciled against the planning figure. The gap between them is itself useful information, because it usually points at a scheduling habit that the configuration could absorb rather than fight.
This step also decides the shape of the answer. A plant with eight products running twice a week needs a configuration optimized for preparation depth: pre-kitted carts, an offline setup area, and generous feeder inventory. A plant with eighty products running once a month needs a configuration optimized for changeover speed and program retrieval, because there is no learning curve to ride. Those are different machines with different accessories, and choosing between them without the changeover count is guesswork. Once the count exists, the rest of the steps become arithmetic rather than opinion, which is the point of doing this first.
The bridge between BOM data and changeover data is the batch size, and the bridge has to be built deliberately because neither system holds both facts. Start from the annual volume per product, divide by the batch size actually run, and you get runs per year. Divide by working days and you get runs per day, which is the changeover count if every run is a distinct product. Where consecutive runs share a product, or where two products share a feeder kit, the count drops; where a run is split, it rises. The adjustment matters more than the base figure in most plants, because the split runs are invisible in planning data and dominate the floor's experience.
Worked example, with assumptions stated: a plant ships 180,000 boards a year across 36 products, runs an average batch of 500, and works 250 days. Runs per year are 360, or 1.44 per day. If the floor's log shows 2.1 program loads per day, the difference is split runs and re-runs, and the configuration must be sized for 2.1 rather than 1.44 — a 46 percent difference in changeover demand. At an average of 55 minutes per changeover, that is 1.9 hours a day versus 1.3, and across three shifts the gap is a full shift of lost placement time every week. Those figures are illustrative; replace them with the plant's own before using any of them in a business case.
Every mixed-schedule plant has a batch size below which the changeover cost exceeds the value of the run, and that floor is almost never written into a procedure. It can be estimated: if a changeover costs 55 minutes and the line is billed at an internal rate, the changeover has a cash cost, and dividing that cost by the contribution per board gives the minimum batch that pays for its own setup. For a line whose changeover costs the equivalent of 90 minutes of capacity and a board contributing a small margin, the floor can easily land at 150 to 250 boards. Runs below the floor are not automatically wrong — prototype and sample orders have their own justification — but they should be recognized as subsidized runs rather than planned as normal production.
Why this matters for configuration: the floor sets a practical limit on how far batch size can be reduced to serve the mix, and therefore a practical limit on changeover frequency. If marketing wants weekly deliveries of twelve products, the arithmetic may demand a changeover capability the proposed machine cannot deliver, and the honest answer is either more feeder preparation depth or a different batching policy. Discovering this after installation is expensive, because the remedy is usually additional carts, additional feeder inventory and additional offline setup space — items that could have been in the original order at a fraction of the retrofit price. The floor calculation is therefore a configuration input, not a production policy debate, and it belongs in the same document as the machine specification. The clauses that carry these numbers into a purchase are listed in the purchase checklist that converts these figures into clauses.
The two records disagree for reasons that are worth understanding rather than correcting. Planning counts releases, so a work order that is paused for material and resumed later still counts once. The floor counts program loads, so that same order counts twice, and a run that was split to serve an urgent shipment counts three times. Neither record is wrong; they measure different events, and the configuration has to be sized for the event that consumes machine time, which is the program load.
A one-week reconciliation exercise is usually enough to establish the ratio. Ask the line leader to note every program load for five consecutive production days, with the product, the start and end times, and whether the load followed a full kit change or a partial one. Compare that list with the work orders released in the same window. Plants typically find 20 to 40 percent more program loads than releases, and the extra ones cluster around two causes: material shortages that split a run, and quality holds that force a re-run after a partial batch. Both are scheduling realities that will not disappear because the configuration plan ignores them.
The reconciled figure becomes the input to every later step, and it should be written down with its source and date so that it can be challenged later. A configuration budget built on a figure nobody can trace will be reopened the moment the first month's data looks different, and the reopening usually costs more than the original analysis. The same reconciliation also produces a useful by-product: the split between full kit changes and partial ones tells the plant how much of its changeover load is shallow, which is exactly the load that carts and offline preparation can remove most cheaply.
Forty products is not forty problems if they cluster. Group the product list by three attributes that drive setup: board size class, component mix class, and stencil or panel format. Products that share all three can usually share a feeder kit with minor swaps, and that changes the changeover count dramatically — six families with an average of seven products each means the deep changeover happens six times a week rather than forty. The grouping has to be done on real data, not on product names, because names follow customer accounts and the setup follows geometry.
The output of this grouping is a two-tier changeover model. Within a family, changeovers are shallow: a few feeders swapped, the same nozzle set, the same program skeleton, a fast first article. Between families, changeovers are deep: full feeder exchange, possibly a nozzle change, a different program, and a full first article. Configuring for the average of the two is the classic error; the right approach is to make the shallow ones very fast and accept that the deep ones are rare. That is why feeder carts and offline preparation pay back so well in mixed plants — they convert deep changeovers into shallow ones by moving the kit change off the machine. A plant that has never grouped its products will usually find that half its apparent changeover load is self-inflicted, because products that could share a kit were never scheduled together.
Feeder planning is where a high-mix configuration is won or lost, because feeders are the only part of the line that scales with the product list rather than with the output. A mounter has a fixed number of slots; the product mix has an open-ended number of part numbers; the gap between those two facts is the entire engineering problem. The common response is to buy the machine with the most slots, which helps and is not sufficient, because slots on the machine are only useful if the reels in them match the program about to run. The configuration that actually works in a mixed plant separates feeder positions from feeder inventory: positions live on the machine, inventory lives on carts and in a preparation area, and the changeover becomes a cart exchange rather than a reel-by-reel rebuild. That separation costs money in carts and floor space and saves far more in changeover time, and the trade is worth quantifying before it is dismissed.
There is a second reason to plan feeders before anything else: the feeder decision constrains the machine decision, not the other way around. Feeder width mix determines how many effective positions a given slot count provides, and in a mixed BOM the effective count is usually far below the nominal one. A machine advertised with a large slot count can lose a third of that capacity to wide tape, sticks and trays once a real product is loaded. Discovering that in a simulation is cheap; discovering it when the first dense board is loaded is not. The numbers below are the ones to compute before the quotation is signed, and they are the reason a JUKI mounter for high mix low volume production is specified feeder-first rather than head-first.
A machine with 112 positions does not give a plant 112 products' worth of preparedness. It gives 112 positions, of which perhaps 80 are loaded for the current product, 20 hold parts common to the family, and 12 sit empty because the program does not need them. The inventory question is different: how many reels must be available, loaded and verified so that the next three changeovers can happen without a trip to the warehouse. That number is driven by the union of part numbers across the family, not by the slot count, and it is almost always larger than the slot count. In a plant with six families of seven products, a family may share 60 percent of its part numbers and differ on the remaining 40, so the family kit needs enough reels to cover the union plus a margin for reorders.
The practical rule is to size feeder inventory to the family, not the product, and to size carts to the number of simultaneous kits the schedule requires. If a plant changes over twice a shift and wants one kit prepared ahead, it needs three kits in circulation: one on the machine, one staged, one being built. Three kits at 80 reels each is 240 reels of loaded inventory, which is a different purchase from a machine with 112 slots and no carts. Plants that skip this arithmetic usually end up building kits on the machine during the changeover, which converts a twenty-minute cart exchange into a ninety-minute rebuild and quietly destroys the business case for the whole project.
The offline setup area is the least glamorous and most valuable square meters in a mixed plant. It needs a bench at working height, a place to park two or three carts, a reader to verify reel part numbers against the kit list, and light good enough to read a label without leaning in. Its function is to move the slow part of the changeover — finding, checking and loading reels — off the machine and into a window where the machine is still producing. Done well, the machine-side work shrinks to releasing the old cart, engaging the new one, confirming positions and running the first article.
There is a discipline cost that has to be planned for, because offline preparation only works if the kit is right. A mis-built kit moves the error from the machine to the first article, where it is more expensive to find. The countermeasure is a verification step at the bench: scan each reel, compare against the program's bill of materials, and record the result against the cart number. That record is also what makes the next improvement possible, because it turns "changeovers take too long" into "changeovers take too long because three of the last ten kits were rebuilt on the machine". Without the record, the plant cannot tell whether it needs more carts, better verification or a different scheduling rule. The feeder hardware that this preparation model assumes — widths, pitch and bank layout — is described in the feeder widths and bank layout this plan assumes.
Slot counts are quoted in 8 mm equivalents because that is the narrowest common tape width, and every wider feeder consumes more than one equivalent position. The exact ratio depends on the feeder family and the machine's bank design, but a simple width ratio is accurate enough for planning: an 8 mm feeder takes one position, 12 mm takes two in most bank layouts, 16 mm takes two, 24 mm takes three, and 32 mm takes four. A product with 40 lines of 8 mm tape, 14 of 12 mm, 8 of 16 mm, 4 of 24 mm and 2 of 32 mm therefore consumes roughly 104 equivalent positions even though it has only 68 part numbers. That gap between 68 and 104 is the number that decides whether a given machine can hold a whole family kit in one load.
The planning consequence is that the width mix, not the part count, should drive the slot specification. Two products with identical part counts can differ by 40 percent in equivalent positions if one is dominated by 8 mm chip tape and the other by 16 and 24 mm tape for connectors and larger parts. In a mixed plant the union across a family makes this worse, because the family kit must cover the widest product in the family, not the average one. The working method is to compute equivalents for the three densest products in each family, take the maximum, add a margin for reels that arrive at a wider width than the drawing says, and specify the machine and cart count against that figure. Plants that specify by part count instead are the ones that discover, on the first dense build, that the "112 position" machine holds their 90-line board only by leaving twenty positions empty.
Feeder width to 8 mm equivalent, and what a family kit actually consumes (illustrative) | |||
Tape width | Equivalent positions (assumed) | Typical content | Planning note |
|---|---|---|---|
8 mm | 1 | 0201 to 0603 chip components, small diodes | The baseline; most mixed BOMs are dominated by this width |
12 mm | 2 | Larger chip components, SOT packages | Punched tape often wider than the part suggests |
16 mm | 2 | SOP, small connectors | Embossed tape; check pocket pitch against feeder spec |
24 mm | 3 | QFP, medium connectors | Usually the first width to force a second cart |
32 mm and above | 4 or more | Large connectors, shields, LED arrays | Verify bank clearance; some layouts lose adjacent positions |
Tape is the easy case. A mixed kit almost always contains something else: a tray of QFNs, a stick of connectors, a bulk bag of LEDs, or a part that arrives in a matrix tray because the supplier will not tape it. Each of these consumes machine or cart capacity in a way the slot count does not describe, and each adds a preparation step that is invisible until the first product that uses it arrives. A tray feeder occupies a slot position and usually a wider footprint than its tape equivalent; a stick feeder occupies a position and needs its own magazine; bulk parts need a vibratory or custom feeder and a different nozzle.
The planning consequence is that the feeder specification should list odd-form requirements separately rather than folding them into the tape count. For each family, record which products use trays or sticks, how many tray positions they need simultaneously, and whether the tray feeder can be left mounted between products or must be exchanged. Plants that skip this step discover during commissioning that their densest product needs a tray feeder in a position the bank cannot reach without removing two tape feeders, which converts a planned changeover into a bank rebuild. Since the number of families using trays is usually small, the extra specification work is a few hours, and it prevents the single most disruptive surprise in a mixed configuration.
Forty products means forty programs, and forty programs means forty chances for the line to run the wrong revision. Program management is usually treated as an IT topic until the first mixed-schedule audit, at which point it becomes the explanation for a scrap event, a wrong-part build or a first article that passed twice with different results. The configuration question is not which software package to buy but how many layers of program information the plant intends to maintain and who owns each layer. A plant with one layer — the file on the machine — is one operator absence away from losing its process knowledge. A plant with three layers can rebuild a program from documentation after a disk failure, and can prove to a customer which parameters built which shipment.
This step also determines how much of the changeover is spent at the screen rather than at the feeder. In a well-run mixed plant, program load is a two-minute confirmation; in a poorly run one, it is a twenty-minute search through folders named after customers who no longer exist. The difference is not the software, it is whether the naming, versioning and parameter baseline were decided before the fortieth product arrived. Since the fortieth product always arrives, the decision is worth making while there are still only ten.
The first layer is the program on the machine: the placement data, the feeder assignment and the head sequence that the mounter executes. It is the layer everyone protects and the layer that is least durable, because it lives on a disk that can fail and in a state that can be edited by anyone with access. The second layer is the exported file set: the program in a format that can be re-imported, together with the feeder list, the nozzle assignment and the mark definition. It is the layer that makes recovery possible. The third layer is the document baseline: a written record of the parameters that are not carried in the file — conveyor width, placement height, vacuum and blow settings, vision thresholds, and the reason each value was chosen.
Most plants have the first layer and assume they have the second. A useful test is to ask for a specific product's program, feeder list and parameter baseline without touching the machine. If the answer requires someone to log in and export, the second layer does not exist in practice. The third layer is almost always missing, and it is the one that matters most in a mixed schedule, because a parameter that was tuned for a thick board will silently misbehave on a thin one unless someone wrote down what it was tuned for and why. Building the third layer takes a few hours per family, not per product, because parameters usually cluster by board class rather than by part number.
A naming rule that survives a mixed schedule has to encode four things: customer or product family, board revision, program revision, and the date the program was released. Anything less and the operator has to open the file to know what it is, which is exactly the step that produces wrong-revision builds. The rule should be enforced by the export process rather than by memory, and it should be printed on the setup sheet so that the person loading the program can compare the name on the screen with the name on the paper without interpretation. Version control does not require a server; a read-only archive folder with a weekly backup and a rule that released programs are never edited in place is enough for most plants of this size.
The parameter baseline is the part that gets skipped, and it is the part that pays. For each family, record the settings that are not in the program file, with the board class they were validated on and the date. When a new product joins the family, the baseline is the starting point and the deviation is documented rather than invented. This is also the mechanism that makes a first article meaningful: if the baseline is known, a deviation during first article is a signal; if it is not, the operator has no way to distinguish a real problem from the normal scatter of a machine that has been adjusted by five people. The acceptance gates that prove a program set was really delivered, rather than merely demonstrated once, are described in the acceptance gates that prove the program set is delivered.
Offline programming is the single largest changeover lever available to a mixed plant, because it moves program creation and feeder assignment into a window where the machine is producing. The risk it introduces is specific and predictable: a program built offline has never seen the actual board, so its fiducial definition, its component orientation and its feeder assignment are all assumptions until the first article proves them. The countermeasure is not to avoid offline programming but to structure the first board around its known failure modes — check mark recognition first, then orientation on the parts most likely to be reversed, then the placement of the largest components, and only then the full run.
The hidden cost of offline programming is the data it depends on. It needs a component library that matches what the plant actually buys, a footprint set that matches the stencil, and a BOM that matches the kit. When those three agree, offline programming is fast and reliable. When they drift — a supplier changes the tape orientation, a footprint is edited on the machine but not in the library — the offline program becomes a source of first-article failures that look like machine faults. The configuration implication is that offline programming needs an owner and a review cadence, not just a workstation. A plant that installs the software and assigns nobody to the library will find that its changeover time improves for six months and then reverts, as the library drifts away from reality.
Two things happen at the end of every changeover that determine whether the next two hours are productive: the first article is judged, and the line discovers which station is now the constraint. Both are usually handled informally, and both are expensive to get wrong in a mixed schedule. The first article is treated as a quality gate when it is really a changeover step with a quality consequence — its duration is part of the changeover, and shortening it without weakening it is a configuration opportunity. The bottleneck is treated as a fixed property of the line when it is really a property of the current product, which means the constraint moves several times a week and any improvement aimed at a fixed station will be aimed at the wrong place half the time.
These two are grouped because they share a mechanism: both depend on knowing what normal looks like for this product on this line. Without a recorded reference, the first article becomes a judgement call and the bottleneck becomes whoever complains loudest. With a reference, the first article becomes a comparison and the bottleneck becomes a measurement. The configuration requirement is therefore not a better inspection machine, although that helps, but a defined set of reference values per family and a way to compare against them quickly. That is cheap to build and almost never built.
The first article exists to prove three things before the line commits: that the program matches the product, that the material in the feeders matches the program, and that the process parameters suit this board. Those three checks can be sequenced so that failures are found cheaply. Mark recognition and board datum come first, because everything downstream depends on them. Component presence and orientation on the highest-risk parts come second, because a reversed connector is the most expensive single defect to find later. Placement accuracy on the finest pitch comes third, and only then is it worth running the board through reflow for a full inspection.
The time cost of this sequence is usually dominated by transport and inspection, not by the checks themselves. A three-board first article that travels to an offline microscope, waits for an inspector, and returns to the line can easily consume twenty-five minutes, of which perhaps six are actual verification. That is a configuration problem, not a discipline problem: an inspection point at the line, a defined sampling rule for what gets measured, and a reference board for visual comparison can cut the elapsed time substantially without reducing what is checked. The trap is to shorten the first article by checking less. The right move is to shorten it by removing waiting, and to keep the checks that have historically caught defects. A plant that does not know which checks have caught defects cannot make that trade, which is why the first article record matters more than the first article procedure.
In a steady-state line the bottleneck is found by comparing cycle times and watching the buffers. In a mixed line that comparison has to be redone per product family, because the station that limits a dense digital board and the station that limits a sparse LED board are usually different. Placement is the constraint on a board with 900 placements and a modest print; printing becomes the constraint on a board with a fine-pitch stencil and 80 placements, because the print cycle and its cleaning interval dominate. Reflow can become the constraint on a board with a heavy thermal mass, where the profile has to slow down to avoid delamination or incomplete reflow.
Worked example, assumptions stated: a family runs print at 22 seconds, placement at 38 seconds and reflow at 30 seconds per board. Placement is the constraint at 38 seconds, and the line's practical output is about 95 boards an hour before any allowance. After a changeover to a fine-pitch product, print rises to 34 seconds because of an extra cleaning cycle and slower squeegee speed, placement falls to 31 seconds because the board is sparser, and reflow stays near 30. The constraint has moved to printing, and any improvement plan still aimed at placement is now wasted. Those numbers are illustrative only; the method — recompute per family, with the print cleaning interval included — is the transferable part. A plant that recomputes finds that its apparent improvement projects were aimed at the wrong station in roughly half the cases.
A first article without a reference value is an opinion. The inspector looks at a joint or a placement and decides it looks acceptable, and the decision varies with the inspector, the lighting and the hour of the shift. A reference value converts that into a comparison: the placement offset for this family was 0.06 mm at the last accepted run, this run measures 0.09 mm, and the drift is either inside the family's tolerance or it is not. The reference does not have to be sophisticated. Three numbers per family cover most of the value — a placement offset range, a pick error rate per thousand placements, and the print or paste height range if the line includes printing.
The record that holds those values is the part that gets skipped, because it looks like paperwork and because the first article already produces a result. What it should contain is short: family, date, the three reference values, the actual measured values, the operator, and the disposition. Ten lines per event. Its value appears three months later, when a product that used to run cleanly starts failing first article and nobody can say whether the machine changed, the material changed or the acceptance rule changed. Placement performance standards such as IPC-9850 give a vocabulary for accuracy and repeatability measurement, and using that vocabulary in the record makes the numbers comparable across machines and across years; the standard describes how to measure, while the plant's own record supplies the history that makes the measurement meaningful.
Line balancing in a mixed plant is not the classic exercise of equalizing station times for one product. It is the exercise of choosing a schedule that keeps the constraint in the same place for as long as possible, so that the improvement effort and the operator attention stay useful. Running three dense products back to back keeps placement as the constraint for a shift; alternating dense and sparse products moves the constraint every changeover and makes both stations look permanently inadequate. That scheduling insight is worth more than most hardware upgrades and costs nothing to implement once the family cycle times are recorded.
The configuration consequence is modest but real: a buffer between print and placement, and between placement and reflow, buys schedule flexibility. With one board of buffer, a constraint that moves by a few seconds can be absorbed without stopping the line. With no buffer, every constraint change becomes a stop, and the changeover time effectively includes a line restart. Plants that specify a fully linked line with no buffering for a mixed schedule often discover this in the first month and add conveyors afterwards at retrofit prices. A single buffer position is cheap at order time and expensive later, which makes it one of the few configuration items that is almost always worth adding.
Where the constraint lands by product class, and what to record at first article (illustrative) | |||
Product class | Usual constraint | What moves it | Reference value to record |
|---|---|---|---|
Dense digital, many 0201 parts | Placement | Feeder count, nozzle set, head sequence | Placements per board, actual cycle time, pick errors per 1,000 |
Fine-pitch, few placements | Printing | Stencil quality, cleaning interval, squeegee speed | Print cycle, cleaning frequency, paste height range |
Mixed with connectors | Placement (wide feeders) | Feeder width mix, bank reconfiguration | Equivalent feeder positions used, changeover duration |
Heavy thermal mass | Reflow | Profile length, zone count, board support | Profile duration, peak temperature, soak time |
LED or panel arrays | Placement or depaneling boundary | Panel format, nozzle choice, conveyor width | Panel cycle time, placement accuracy at panel edge |
The machine class that suits a mixed build best is usually the flexible single-gantry type rather than the highest-throughput rotary head, because the flexible class tolerates feeder and nozzle variety without a full reconfiguration. The trade-off between those two classes, and the point at which a plant outgrows the flexible one, is described in the flexible single-gantry class used for the mixed build.
Nozzles are the smallest line item in the configuration and the one most likely to be under-specified, because the quotation usually lists a standard set and the plant accepts it without mapping it to the product list. In a single-product line that is fine: one or two nozzle types cover the board and the set is stable. In a mixed line the nozzle set is the second-most-frequent changeover item after feeders, and an incomplete set forces either a machine stop or a compromised placement that will not show up until an inspection finds it. The configuration question is how many nozzle families the mix actually requires, and whether the head can change between them automatically or whether every change is a manual stop.
The reason this is under-planned is that nozzle variety is invisible in the BOM. Two products can both list "0402 capacitor" and still need different nozzles if one is placed at high speed and the other requires a soft landing on a fragile dielectric. The mapping from part to nozzle is a process decision, not a purchasing decision, and it has to be made before the set is ordered. Plants that skip the mapping usually buy the standard set, discover two gaps during commissioning, and spend the next year ordering nozzles one at a time at expedited prices.
The practical method is to sort the product list by component size and type, then assign each group to a nozzle family rather than to a specific part number. Most mixed plants in the consumer and industrial electronics range land on four to six families: a small chip nozzle for 0201 and 0402, a general chip nozzle for 0603 through 1206, a larger nozzle for SOT and small SOP, a multi-purpose nozzle for connectors and odd shapes, and often a dedicated nozzle for LEDs or for parts that must not be marked. That is a smaller set than most teams expect, and the constraint is not the number of families but the number of nozzles per family that must be mounted simultaneously.
The simultaneous requirement is what drives head configuration. If a product needs three small-chip nozzles and one connector nozzle at the same time, the head must carry at least four positions with that mix. Across a family, the worst-case simultaneous combination sets the head requirement, not the average. The working figure is therefore not "how many nozzle types do we own" but "what is the largest simultaneous set any product in the family requires", and that figure should be checked against the head's positions before the machine is ordered. A plant that computes only the total number of types will buy a set that is complete in the cupboard and incomplete on the head, which is the most common nozzle mistake in mixed production. The part numbers and orifice sizes that make up a practical shared set are listed in the nozzle part numbers that cover a mixed BOM.
An automatic nozzle changer removes the manual exchange from the changeover, which sounds like a complete solution and is not. It solves the mechanical problem — pulling a nozzle, seating another, and confirming the seat — and it does not solve the decision problem, which is knowing which nozzle each product needs and whether the currently mounted set is adequate before the first board runs. A machine with an automatic changer and a wrong nozzle library will confidently mount the wrong nozzle and produce a first article that fails for a reason the operator cannot see.
The configuration implication is that the changer should be paired with a nozzle assignment that lives in the program, not in an operator's memory. When the assignment is in the program, the changeover becomes a confirmation, and a missing nozzle is caught before the first placement rather than after. When it is in memory, the changer accelerates a process that is already unreliable. The second implication is capacity: an automatic changer holds a limited number of nozzles, and a mixed plant's simultaneous set may exceed it. Checking the changer's station count against the worst-case simultaneous set is the same arithmetic as the head check, applied to a different component, and it is equally easy to skip.
Nozzle condition drifts slowly enough that no single board looks wrong. Pick errors rise from a fraction of a percent to one or two percent over weeks, and because the machine keeps running and the boards keep passing, nobody investigates until a customer complaint or an audit. In a mixed schedule the drift is harder to see than in a single-product line, because each product has a different baseline and the operator has no stable reference. The result is that a nozzle problem can affect one product family for a month before it is attributed correctly.
The countermeasure is a recorded baseline per family: pick errors per thousand placements, vacuum level at the nozzle, and the visual condition of the tip at a defined interval. Those three numbers, taken at first article and compared at every changeover, turn a slow invisible drift into a visible trend. The cleaning interval itself should be set from that trend rather than from a generic schedule, because a plant running abrasive parts will need a shorter interval than one running clean chip components. This is the cheapest item in the whole configuration and the one most often left out, because it looks like a maintenance task rather than a specification item. It becomes a specification item the first time a mixed plant loses a week to a nozzle problem it could have seen coming.
The instinct in a mixed plant is to buy a complete nozzle set for every product, which is both expensive and unnecessary. Nozzles wear, and a plant holding forty complete sets is holding forty sets of wear out of which only a few are actually cycling. The alternative is a shared inventory sized to the worst-case simultaneous set plus a rotation margin: enough nozzles that any product can be mounted without waiting, and no more. In practice that is two to three times the worst-case simultaneous set, not forty times.
The inventory rule has to be paired with a condition rule, or the shared set becomes a quality problem. Each nozzle should carry an identifiable position and a recorded service interval, so that a nozzle removed during a changeover can be traced to the boards it placed. Without that trace, a drift in pick errors cannot be attributed to a specific nozzle, and the plant ends up replacing the whole set instead of the one that is worn. That is a small administrative burden which pays back quickly in a mixed schedule, because the condition problem is diluted across products — a worn nozzle affects the family that uses it most, and that is rarely the family anyone is watching. A JUKI mounter for high mix low volume production gets the cost benefit of a shared set only if the set has identities and intervals attached to it.
The five steps produce a set of numbers that are individually convincing and collectively useless until they are added up. A configuration budget does that: it converts each decision into a cost and a benefit, so that the review can compare a cart against a feeder bank, or an automatic nozzle changer against a buffer conveyor, on the same basis. Without it, every item is argued on its own merits by whoever wants it most, and the items that pay back across the whole mix — offline preparation, program discipline, reference values — lose to the items that are visible in a demonstration. The budget is also what makes the specification defensible afterwards, because each line can be traced to a changeover number computed from the plant's own volume.
Two cautions before building one. First, the budget only works if the changeover count is the floor's number, not planning's; a budget built on 1.4 changeovers a day when the line sees 2.1 will understate every preparation item by a third. Second, the benefits should be expressed in the same unit as the costs — hours of line time, or boards of capacity — rather than in percentages, because percentages hide the base and make comparison impossible. Once both cautions are respected, the arithmetic is simple enough to do in a spreadsheet and honest enough to survive a challenge. It is also the document that turns a JUKI mounter for high mix low volume production from a preference into a specification, because every line in it can be traced back to a changeover the plant actually performs. A budget that cannot be traced that way is a wish list with totals attached.
Assumptions, explicitly illustrative: 40 products in 6 families, 180,000 boards a year, 250 working days, 2.1 program loads per day measured on the floor, current average changeover 55 minutes, three shifts. Current changeover load is 2.1 × 55 = 115 minutes a day, or about 1.9 hours, which is roughly 480 hours a year. A configuration that adds three pre-kitted carts, an offline verification bench and a program archive is assumed to reduce the average changeover to 35 minutes, a reduction of 20 minutes per event, or 42 minutes a day. Across 250 days that is 175 hours a year returned to placement.
Converting that to boards requires the constraint cycle time, not the rated rate. If the line's practical cycle for the dominant family is 38 seconds per board, 175 hours is about 16,500 boards of additional capacity, which at a modest contribution per board is a figure large enough to justify several carts and a bench. The same 175 hours bought as placement speed instead would require a machine upgrade costing far more. That comparison — hours recovered versus capital spent — is the core of the budget, and it is only available if the changeover count and the changeover duration were measured rather than assumed. The numbers here are placeholders; the point is that the arithmetic is short enough that there is no excuse for skipping it.
Configuration budget structure: what each item buys and how to judge it (illustrative values) | |||
Item | What it buys | Cost driver | Judge it by |
|---|---|---|---|
Pre-kitted feeder carts | Moves kit build off the machine | Number of carts × reels per kit | Minutes removed per changeover versus cart cost |
Offline verification bench | Catches kit errors before first article | Bench, reader, floor space | Rework events avoided per month |
Program archive and baseline | Removes program search and revision doubt | Owner time, not capital | Wrong-revision incidents, first-article repeats |
Automatic nozzle changer | Removes manual nozzle exchange | Changer stations, nozzle set breadth | Changeovers requiring a manual stop |
Buffer conveyors | Absorbs constraint movement between products | Positions × conveyor length | Line stops per shift |
Second feeder bank or cart parking | Keeps the next kit staged on the floor | Bank or parking positions | Changeovers that wait for a cart |
Four trade-offs dominate mixed-schedule configuration, and each one is usually presented as a free choice when it is not. The first is feeder positions against floor space: every additional bank or cart parking position consumes floor area that a mixed plant also needs for staging, and plants routinely run out of space before they run out of budget. The second is placement rate against nozzle and feeder variety: the highest-throughput rotary heads are optimized for a narrow, stable component set, while flexible single-gantry machines tolerate variety at a lower peak rate. Choosing the peak rate for a mix that never reaches it is one of the most common high-mix purchasing errors.
The third is automation against visibility: an automatic nozzle changer and an automatic feeder exchange reduce manual work and also reduce the number of moments at which a human sees the machine, which makes certain failures quieter. The fourth is buffer against linkage: a fully linked line looks efficient and transmits every constraint change directly to the next station, while a buffered line absorbs variation at the cost of floor space and a little handling. None of these has a universally correct answer, but each has a correct answer for a given changeover count and family structure. A useful discipline is to state, in one sentence per trade-off, which side the plant has chosen and why — because the choice will be challenged in six months by someone who was not in the room. Where the budget points toward a machine that is already in service rather than a new one, the checks that apply are set out in the checks that apply if the budget points at a used machine.
A mixed-schedule specification differs from a standard one mainly in what it refuses to leave to the supplier. Feeder count should be stated in 8 mm equivalents against the densest family, not in raw positions. Nozzle requirements should be stated as the worst-case simultaneous set per family, not as a total quantity. Program handover should be stated as three deliverables — machine file, re-importable export, and written parameter baseline — with the baseline reviewed at acceptance. Changeover support items such as carts, parking positions and offline bench should appear as line items rather than as options, because options get dropped at the last negotiation and are the items that make the difference afterwards.
Two clauses are worth adding even though they look like process rather than equipment. The first requires the supplier to demonstrate a changeover on the buyer's own product family during acceptance, timed and recorded, rather than a demonstration on a prepared board. The second requires the feeder and nozzle lists to be delivered as data files that match the machine's own naming, so that they can be loaded into the plant's archive without retyping. Both are cheap to ask for before the order and awkward to obtain afterwards. At the top end of the mix — the products that justify a high-throughput machine alongside the flexible one — the class that covers that volume is described in the high-speed rotary class that covers the volume end.
Need a configuration reviewed against your actual changeover count?
A configuration review is only as good as the numbers it starts from, and most of the numbers live in the plant rather than in the quotation: the product list grouped into families, the annual volume per product, the batch size the floor actually runs, the number of program loads per week measured at the line, and the changeover duration split into material, program, qualification and recovery time. Send those five items and we will return a configuration budget in the same structure used above — feeder equivalents calculated per family against the densest product, the worst-case simultaneous nozzle set, the program layers that should be delivered at handover, and the changeover support items that pay back at your volumes rather than at a generic plant's volumes.
If the batch sizes or the changeover figures are not yet measured, say so and we will work from the planning data with the gap stated explicitly, so that the recommendation can be re-run once the floor numbers arrive. That is a more honest starting point than a configuration built on assumptions nobody has tested.
These are the questions that come up in almost every mixed-schedule configuration review, answered with the same assumptions used above. Where a number appears, it is illustrative and should be replaced with the plant's own measurement before it is used in a business case.
Start from the densest product family, not from the average product. Convert every line of the family's bill of materials into 8 mm equivalents, take the maximum across the products in that family, add a margin for reels that arrive at a wider width than the drawing, and specify to that figure. A family with 68 part numbers can easily consume 100 or more equivalent positions once 12, 16 and 24 mm tape is counted.
Then add the staging question, which the slot count does not answer: how many kits must be in circulation at once? Two changeovers a shift with one kit prepared ahead means three kits — on the machine, staged, being built — and that is a cart purchase, not a machine purchase. Plants that size only the slot count end up building kits on the machine, which converts a cart exchange into a rebuild and removes most of the benefit of buying a larger machine in the first place.
Usually as one machine in a pair, not as the only machine. A high-throughput rotary head is optimized for a stable component set and a long run, so it does well on the volume end of a mixed product list and poorly on the tail. A flexible single-gantry machine tolerates feeder and nozzle variety and carries the changeover load. Pairing them lets the fast machine run the two or three products that justify it while the flexible machine absorbs the rest of the schedule.
The decision point is arithmetic rather than preference: if the products that justify the high-speed machine consume less than about half of the available hours, the second machine is idle capacity dressed as capability. In that case a single flexible machine with better preparation support usually beats two machines with one underused. The rated placement rate is the least useful figure in this comparison, because a mixed schedule rarely reaches it.
Most consumer and industrial mixes land on four to six families: a small chip nozzle for 0201 and 0402, a general chip nozzle for 0603 to 1206, a larger nozzle for SOT and small SOP, a multi-purpose nozzle for connectors and odd shapes, and often a dedicated nozzle for LEDs or for parts that must not be marked. The number of families is rarely the constraint; the simultaneous set per product is.
Compute the worst-case simultaneous combination for each family — for example three small-chip nozzles and one connector nozzle mounted at the same time — and check it against the head's positions and the changer's station count. A complete cupboard with an incomplete head is the most common nozzle failure in mixed production, and it is invisible until a first article fails for a reason nobody can see.
Without preparation support, 60 to 120 minutes is common, and most of that is material handling and waiting rather than machine work. With pre-kitted carts, an offline verification bench and a program archive, 30 to 45 minutes is achievable for a family change and considerably less for a within-family change. The range is wide because the definition is wide: some plants count only machine time, others count from last good board to first good board.
Use the wider definition, because it is the one that consumes capacity. Then time ten real changeovers and split the total into material, program, qualification and recovery phases. The split tells you which investment to make; the total alone does not. A plant that measures only the total usually buys feeder capacity to fix what is actually program search time.
For more than about fifteen active products, yes — not because the software is magical but because program creation on the machine competes directly with production. Offline programming moves that work into a window where the machine is running, and it makes the feeder assignment reviewable before the kit is built rather than during the changeover.
It comes with a specific risk: a program built offline has never seen the board, so its mark definition, orientation and feeder assignment are assumptions until the first article proves them. That is manageable if the first board is sequenced deliberately — marks first, then the highest-risk orientation, then the finest pitch — and if the component library is owned by someone with a review cadence. Plants that install the software without assigning a library owner see changeover time improve for six months and then drift back.
The limit is not the machine's program capacity, it is the plant's ability to keep feeder kits, program revisions and reference values current. In practice, a single mounter with good preparation support handles thirty to fifty active products comfortably, provided they cluster into a handful of families. Beyond that, the administrative load grows faster than the production load, and changeover time rises even though nothing about the machine has changed.
The practical test is whether a new product can be added without a special project. If every new product requires a meeting, a new feeder purchase and a manual program build, the plant is at its limit regardless of how many programs the machine can store. Family clustering is the lever: six families of seven products is a very different administrative problem from forty unrelated products.
Four things, recorded per changeover rather than per day: the changeover duration split into material, program, qualification and recovery; the number of changeovers that required an unplanned trip to the warehouse; the first-article result and whether it repeated; and the pick error rate per thousand placements by product family. Those four cover the material, program, qualification and nozzle dimensions of the configuration.
Thirty days of that record is worth more than any acceptance report, because it shows which part of the configuration is actually working. Plants that measure only output in the first month learn that the line makes boards and nothing about why it stops. The record also becomes the baseline for every later improvement, and it is the only way to tell a real constraint from a habitual complaint.
