English Views: 0 Author: Elsa Long Publish Time: 2026-11-12 Origin: Site
Most maintenance schedules on a placement line are inherited rather than designed. A manual lists dozens of items, someone assigns them to shifts, and within a few months the sheet is being signed without being done, because nobody can tell which of the forty lines actually prevents a defect and which exist to protect the supplier. A working JUKI pick and place machine preventive maintenance checklist is shorter than that and organized differently: by how fast each item fails, not by which part of the machine it belongs to, and with a written rule for what to do when a check does not pass. That structure is what makes the schedule survive a busy month.
The organizing principle in this guide is the failure timescale. A nozzle that clogs with flux dust can degrade pickup within a shift; a belt that is slowly losing tension degrades over weeks; a bearing that needs grease degrades over months. Checking everything daily wastes hours and trains operators to initial boxes; checking everything monthly lets fast failures run unobserved for weeks. The checklist below assigns each item to the layer that matches its failure speed, adds a three-outcome decision rule for every observation, and shows how to move an interval using your own failure data instead of a vendor default. Platforms and wear parts referenced throughout are drawn from the placement platforms this schedule is written for, and every calculation is labelled as an assumption so it can be re-run with your numbers.
A preventive schedule has one job: to find a degrading condition before it becomes a defect or a stop. Most published schedules do not do that, because they are written to describe the machine rather than to catch its failures — every subsystem gets equal weight, every item gets the same weekly box, and the operator is left to decide what "check" means. The result is a document that passes audits and fails production. Rebuilding it takes three steps: name the failures the line actually experiences, sort those failures by how quickly they develop, and force every observation into one of three outcomes. This section sets that structure up, and explains why the machine selection made months earlier still shapes how much maintenance the line will need.
Placement defects that maintenance can prevent fall into three groups, and each group fails on a different clock. Pickup failures come from a clogged or worn nozzle, a feeder presenting tape a fraction off the pickup point, or vacuum that is weaker than the gauge claims. Positioning failures come from a drifting axis, a loose rail, a camera that no longer recognizes a fiducial cleanly, or a worn belt that adds backlash at direction changes. Material failures come from tape splices that do not index, reels that run out mid-lot, and feeders whose mechanical condition has decayed enough to skip a pocket. None of these announces itself; each produces a small number of defects that get blamed on the process, the paste or the operator. A checklist earns its keep by making one measurement per mode early enough that the trend is visible before the defects are. The classification also decides who observes what: the operator sees pickup behavior shift during a shift, the technician sees positioning drift on a test board, and material problems usually surface as a feeder that has to be swapped mid-lot. Writing the sheet around the observer rather than around the subsystem is what keeps every line short enough to be completed.
Sort every candidate check by the time it takes for a healthy condition to become a defective one, then set the interval at roughly one third of that time. A nozzle that clogs in eight hours of running gets checked every shift; a belt that loses tension over six weeks gets checked every two weeks; a lubrication point with a six-month life gets serviced every two months. The one-third rule is a convention, not a law, and it is deliberately conservative because it assumes you find the condition somewhere in the middle of its degradation rather than at the start. Its practical value is that it converts an argument about frequency into a measurement: once you know how fast something deteriorates on your line, the interval chooses itself, and the checklist stops being a document that someone else's factory wrote. Two consequences follow. Fast-failing items belong to the people who are already at the machine every shift, while slow-failing items belong to whoever owns the maintenance calendar, because they need a planned stop and a measurement rather than a glance. When an interval cannot be stated in the same units as the failure it prevents, the item is not yet understood well enough to be scheduled.
A line that says "check nozzle" is not a check, it is an invitation. A line that says "inspect the bore of each nozzle on the head against a light, record the number that show visible contamination, clean and re-inspect, scrap any nozzle whose tip shows a nick" produces a number and a decision. The difference is not pedantry: the first version cannot fail, so it always passes, and a schedule that cannot fail teaches the operator that signing is the task. Write each item as an observation plus a threshold plus an action, and keep the observation physically small — a visual inspection under a light, a gauge reading, a counter value read from the HMI. Anything that needs a tool most shifts will not have should be moved to the weekly or monthly layer where the tool lives. The machine class you bought partly determines how much of this exists: the model shortlisting sequence that comes before configuration decides how many heads, nozzles and feeder positions you will be maintaining every day.
The daily layer exists to catch what can go wrong inside a single shift, and it should take minutes rather than hours. Everything here is either an observation the operator can make without tools or a number the machine already reports. The temptation is to add more — a full nozzle clean, a feeder audit, a rail wipe — and the result is a daily routine so long that it is skipped on the busy days when it matters most. Keep the daily list to pickup health, vacuum health, feeder presentation and the first board, and push everything else to the weekly layer where it can be done properly with the machine stopped. The table below shows the layer as it is usually written for a placement station, with the observation that has to be recorded for each line and the owner who records it.
Daily item | Observation recorded | Owner |
|---|---|---|
Nozzle tips, visual | Number cleaned, number pulled | Operator |
Pickup retries | Count at shift start and shift end | Operator |
Vacuum at nozzle | Reading against the manual band | Operator |
Feeder presentation, spot check | Slots checked, slots swapped | Operator |
First board of each product start | Presence, fine pitch, first-attempt fiducial | Operator and technician |
The daily nozzle check is a light and an eye, not a disassembly. With the head parked and the machine in a safe state, look at the tip of each nozzle for the two conditions that matter: contamination inside the bore, which shows as a dull or blocked opening, and a nick or flat on the tip, which shows as an irregular reflection. On a multi-nozzle laser head the count is the point — a six-nozzle head means six tips to look at, and a machine with a larger nozzle complement multiplies the routine, which is why head design changes the daily workload as much as it changes throughput. The published nozzle range for these machines is granular enough that a replacement is a part number rather than a rework decision: the codes cover a tip outer diameter from 0.7 mm to 9.5 mm with matching bores, listed with part numbers on the replacement nozzle part numbers and bore sizes page. Record how many tips were cleaned and how many were pulled; a rising pull count is the earliest signal that the cleaning interval is too long.
Most machines count pickup retries, and that counter is the cheapest early-warning instrument on the line. Read it at the start and end of each shift and write both numbers down; a shift that ends with an unusually high retry count has told you something before any defect appears. Pair the counter with a physical check that takes seconds: with the head at the pickup position and the machine in manual, confirm the vacuum reading at the nozzle is within the band the manual specifies, and listen for the pump. A slow decline across several shifts points at filter loading or a leaking hose; a sudden change points at a nozzle or a valve. Do not chase a single bad shift — the value of a daily counter is the trend, and trend only exists if the numbers are recorded rather than remembered. This is also where the feeder and nozzle configuration you chose shows up in daily work: an eight-nozzle chip shooter line carries more tips and more pickup events per hour than a single-head machine, so the same defect rate produces more retries to read.
The daily feeder check is a spot check, not an audit. Pick the feeders on the parts with the highest placement counts — usually the small chip positions — and confirm the tape sits flat in the pocket, the cover tape peels cleanly, and the component is centred at the pickup window when the feeder indexes. A feeder that presents a part a few tenths of a millimetre off will still place most parts, which is exactly why the defect rate creeps rather than jumps. Two conditions deserve an immediate swap rather than an adjustment: a splice within a few indexes of the pickup point, and a reel with less tape left than the next lot will consume. Both are predictable, and both are cheaper to fix at a natural break than during a run. If a feeder is swapped, tag it and put it on the bench queue rather than back on the cart. Keep that queue under a two-week aging rule: any unit not returned to service within two weeks is stripped for parts, which stops the queue from becoming a store of unknown feeders that reappear on a cart during a busy week.
The first board of each product start is the daily layer's closing check, and it should test the three failure modes rather than simply look at the board. Confirm placement presence on the smallest and the largest components, check one fine-pitch part under magnification if the product has one, and confirm that the fiducial was recognized on the first attempt rather than after retries. Write down four things: the retry count at start, the retry count at end, the number of nozzles cleaned or replaced, and any feeder swapped with its position. Those four fields take under a minute and they are what makes the weekly and monthly layers possible, because they are the only record of how fast things are degrading between the scheduled checks. A daily sheet without numbers is a signature; a daily sheet with four fields is data. Add the product and the program revision to the sheet as well, because a retry count without them cannot be compared with anything, and most first-board problems are product-specific rather than machine-specific.
Weekly work is where a placement line usually recovers the defects that the daily layer only flagged. It happens with the machine stopped, with the right tools within reach, and with enough time to look at something properly rather than glance at it. The four areas that repay the time are the feeder tape path, the conveyor and its sensors, the camera windows and lighting, and the mechanical condition of the parts that move the board. What does not belong here is anything requiring a full recalibration or a disassembly, because a weekly stop that overruns becomes a weekly stop that gets postponed, and a postponed stop is worse than a shorter one that always happens.
Once a week, take the highest-usage feeders off the cart and inspect the tape path end to end: the sprocket teeth for wear or debris, the cover-tape take-up for slippage, the pocket guide for a burr, and the splice area of any tape that has been joined. Run each one through a full index cycle empty and watch the tape advance; a hesitation or a double index is a feeder that will skip a pocket under load. Feeders are the most numerous wear item on the machine, so the width ladder matters for planning: a bank of 8 mm, 12 mm, 16 mm and wider tape feeders cannot be checked in the same time, because the wider bodies hide more of the tape path. The manufacturer listing for the tape feeder widths listed for JUKI machines runs from a narrow 4 mm class through 8, 12, 16, 24, 32, 44, 56, 72 and 88 mm, and it is worth splitting the weekly inspection into a fast pass on the narrow positions and a slower pass on the wide ones. Replace, do not repair, any feeder that fails the index test twice.
The conveyor is the part of a placement line that everyone assumes is fine and that quietly produces board-position variability. Once a week, with the machine in manual, run a board through and check four things: that the rails are parallel and at the width the program expects, that the clamp closes flat without lifting the board edge, that the board sits at the same height at the placement station as it did on the first article, and that every sensor flag returns freely. A flag that sticks will produce an occasional "board not present" stop that is cleared by cycling the machine, so it never gets investigated. Wipe the rails, inspect the belt for cracks and debris, and check the support pins or vacuum platens for a missing pin or a blocked hole. The transfer interface should also match the SMEMA convention the rest of the line uses, because a rail height or handshake mismatch between stations shows up as a placement symptom and is actually a conveyor condition. A single missing support pin under a thin board produces placement offsets that look like an accuracy problem and are actually a support problem — and no calibration will fix it.
Optics degrade gracefully and then fail suddenly. Weekly, clean the fiducial camera window and the head camera window with the approved method — a dry lint-free wipe first, then the specified solvent only if the contamination is adhesive — and inspect the ring light for dead segments. Then prove recognition rather than assume it: place a board with a known fiducial in the machine and confirm the camera finds it on the first attempt in the program's normal search window. A camera that needs a wider search window than the program specifies is telling you the fiducial contrast or the window cleanliness has changed. Check the nozzle-centering vision as well if the machine uses it, because a dirty lens there produces a centering offset that looks like a nozzle problem and leads to unnecessary nozzle replacement. Record the result as pass or fail against first-attempt recognition, not as a subjective "looks clean". Keep the cleaning materials at the machine and dated, because a lint-free wipe stored in an open box collects the particles it was bought to remove.
The fastest way to lose a weekly routine is to let it grow. Three things do not belong in it. First, full accuracy verification: measuring placement accuracy properly takes a test board, a measurement plan and a decision rule, and squeezing it into a weekly stop produces a number nobody trusts. Second, deep cleaning of the head or the vacuum path: disassembly at weekly frequency adds reassembly risk and consumes the time the routine needs. Third, changeovers and product setup: those are planned work with their own budget, and mixing them into maintenance means the maintenance is what gets cut when the changeover runs long. Keep the weekly stop to a fixed duration, publish that duration, and end it on time even if an item is incomplete — an unfinished item is a documented carry-over, while an overrunning stop is an excuse to skip next week. The changeover budget itself has its own arithmetic, described in the high-mix changeover budget that feeder planning feeds into.
Monthly work is where the line's long-term accuracy is protected, and it is the layer most often skipped because its absence produces no immediate symptom. Lubrication, calibration verification, belt and bearing inspection, and wear-part replacement all belong here, and all of them need a decision rule written in advance so the technician is not improvising at the machine. The monthly layer is also where the manufacturer's published figures become useful: a specification is only a maintenance tool if you can measure against it, so the numbers you verify against should come from the model-specific datasheet rather than from a family brochure. The monthly stop is also the right moment to reconcile the spares cupboard against the consumption record, so the next purchase is based on what the line used rather than on what someone expected it to use.
Lubrication is the item most often overdone. Every axis, ball screw and linear guide that needs grease has a specified quantity and interval, and adding more than the specification attracts dust, which turns grease into an abrasive paste and shortens the life of the very component it was meant to protect. The discipline is to follow the point list and quantity exactly, clean the fitting before attaching the grease gun, and record the quantity applied rather than just the date. Two practical rules repay themselves: never lubricate a warm machine just before a production run, because excess grease migrates onto rails and then onto boards; and treat any fitting that will not accept the specified quantity as a fault to investigate rather than a fitting to force. If a technician cannot say when a given point was last serviced and with how much, the lubrication schedule is not being run, it is being remembered. A fitting that has been painted over, or that sits behind a guard removed only during a major service, will be missed; mark it on the machine drawing and put it in the monthly route rather than in an annual one.
A calibration check compares what the machine does with what the datasheet claims, and the useful output is a margin, not a pass. Take a documented model class: a machine published at ±0.05 mm placement accuracy with laser alignment sets an effective budget of 0.05 mm, so a measured drift of 0.02 mm has already consumed 40 percent of that budget and is a watch item rather than a curiosity. The same discipline applies to vision-alignment figures, which are often quoted tighter than the laser figure on the same machine. Details like these are what make the platform documentation worth reading line by line, for example the KE-3010A platform and its published placement specification, where laser and vision accuracy are stated separately and component range differs between the two centering methods. Measure with a glass board or the approved calibration artefact, record the result as a fraction of the published budget, and act on the trend rather than on a single reading. Define the measurement the way the equipment standard does — IPC-9850 is the usual vocabulary for placement performance — so that your figure and the supplier's figure describe the same quantity rather than two convenient ones.
Mechanical degradation in the transport and head axes shows up as backlash, noise or a growing position error, and the monthly layer is the right frequency to look for all three. Check belt tension against the manual's deflection specification, inspect belts for cracks, glazing or a frayed edge, and listen at low speed for a bearing that has started to growl. On the vacuum side, read the pump hours, check the filter for loading, and confirm the reservoir drains as designed — a filter that is 80 percent loaded still passes the daily vacuum check and fails on a humid afternoon. Record belt tension and pump hours as numbers, because both are only meaningful as trends. Anything that requires replacing a belt or a bearing should be planned as a scheduled stop with a spare on the shelf, not discovered during a production run; the parts that wear are predictable, and the feeder and nozzle configuration behind the wear pattern tells you which ones to stock.
Wear parts need a replacement rule that is independent of how they look, because both nozzles and feeders can pass a visual inspection and still be out of specification. The rule that works is a count-based cycle: track placements per nozzle and indexes per feeder, and replace at the interval where your own failure data shows the defect rate rising. For nozzles, a practical starting point is to pair the count with a reference pickup test — a fixed set of positions, picked under fixed conditions, with the retry count recorded — and to replace when the reference result degrades rather than when the tip looks worn. For feeders, the equivalent is an index-accuracy check on the bench. Neither test needs to run monthly on every unit; sample the highest-usage positions monthly and the remainder quarterly, and let the sample results drive batch replacement. This is also how a spare-parts budget becomes defensible: it is built from counts and observed failure rates, not from a feeling that the line seems to be stopping more often.
A checklist that records observations without deciding anything just moves the judgement to whoever reads it later, usually someone who was not at the machine. The fix is a three-outcome rule attached to every line, agreed before the schedule is issued. Pass means the observation was inside the band and the line keeps running. Watch means it was outside the ideal band but inside a defined tolerance, and a named person must look at it within a stated interval. Stop means the condition can produce a defect or a safety risk and the machine is released to production until it is corrected. The value of writing this down is that it removes the conversation from the shift, when the pressure is to keep running, and puts it in a document agreed when nobody was under pressure.
The authority to stop a machine is a decision a factory makes once, and it should be written into the schedule rather than left to seniority. The usual arrangement that survives contact with production is this: the operator may call pass or watch on daily items and must escalate any stop condition to the maintenance lead or shift supervisor; the maintenance technician may call stop on weekly and monthly items and record the reason; only the maintenance lead may override a stop and return a machine to production, and the override is recorded with a name and a reason. That last clause is the one that matters, because an override is sometimes correct — a customer expedite, a single board for a sample — and an override that is written down is a managed risk, while an override that is not is an invisible one. Review the override log monthly; a rising count is a signal that the thresholds are too strict or the capacity too thin.
The thresholds themselves should be numbers taken from the manual, the datasheet or your own baseline, never adjectives. The table below is a template with assumed values, shown so the shape is clear; replace every figure with the one your documentation supports and your own two-week baseline produces. What matters is that each row has an observation, a band and a consequence, so that two technicians looking at the same machine reach the same conclusion. Keep the table to one page and post it at the machine, because a threshold document in a folder is a document that gets consulted after the argument rather than during it.
Observation | Pass | Watch | Stop |
|---|---|---|---|
Pickup retries per shift (assumed baseline 20) | Under 25 | 25 to 50, or a rising trend over three shifts | Over 50, or any repeat on the same nozzle |
Vacuum reading at nozzle | Inside the manual band | Within 10 percent of the lower limit | Below the lower limit after a filter check |
Nozzle bore, visual | Clear, tip reflection regular | Light contamination, cleaned and re-checked | Nick, flat or contamination that survives cleaning |
Feeder index test (empty tape) | Single clean index, no hesitation | One hesitation in ten cycles | Double index, skipped pocket, or a splice inside the pickup window |
Fiducial recognition | First attempt, standard search window | Recognized after one retry | Recognition only with a widened window |
Placement offset against published budget | Under 40 percent of the published figure | 40 to 70 percent | Over 70 percent, or moving in one direction over two months |
Conveyor rail width and clamp | Within 0.2 mm of program, clamp flat | Within 0.5 mm, or one support pin missing | Board lifts at the clamp, or width outside 0.5 mm |
Lubrication record | All points done to quantity, dated | One point missed, done within the week | Two or more points overdue, or quantity unknown |
A stop condition is not automatically a machine-down event, and confusing the two is how maintenance loses credibility with production. Three questions separate them: can the condition be corrected inside the current stop window, can the affected product be moved to another machine without breaking a commitment, and what is the cost of running one more lot against the cost of the stop. Answer them in that order, because the first often resolves it. A single failed nozzle is a two-minute swap; a rail that has drifted out of width is a stop; a camera that needs a widened search window is a watch item if the product has generous fiducials and a stop if it does not. The escalation path should be published with the schedule, including the phone numbers, because the decision usually has to be made at an hour when nobody senior is on site. Recording the decision — who called it, what was weighed, when the machine returned — turns the next identical situation into a precedent instead of a debate.
Vendor intervals are a starting point written for an average line that does not exist. The interval a factory should run is the one its own failure data supports, and the data needed to compute it is usually already being generated — it is just not being written down. Two numbers per item are enough: how often the condition occurs, and how much it costs when it reaches the machine instead of being caught early. With those, an interval becomes an economic choice rather than a tradition, and the maintenance budget becomes an argument with arithmetic in it. The calculations below use assumed values throughout, and they are deliberately simple so that a maintenance lead can redo them with plant data in a spreadsheet in an afternoon.
Start by expressing the failure rate in the unit the line already reports. If a head produced 120,000 placements last month and logged 30 pickup failures that required intervention, the rate is 0.25 interventions per thousand placements. Expressing it that way makes comparisons possible between machines with different speeds and between months with different volumes, and it makes the effect of a schedule change measurable. Then split the rate by cause, because the aggregate number hides where the value is: if 22 of the 30 came from three specific nozzle positions after a product change, the schedule is not the problem, the cleaning interval for that product is. A monthly review of this single table — interventions per thousand placements, split by cause — is more informative than a maintenance report that lists activities completed, because it measures the outcome instead of the effort. Keep the table by machine rather than by line: a shared line number hides which head is producing the interventions, and the corrective action is almost always at a specific position rather than at line level.
Feeder count turns a maintenance intention into a time budget, and the arithmetic is unforgiving. Take a machine with a large bank — the KE-3010A class publishes up to 160 positions with electric double tape feeders — and assume a 20-second visual pass per feeder. That is 160 × 20 seconds, or about 53 minutes, before any feeder is actually pulled and tested. A daily full-bank inspection is therefore not a daily task on that machine; it is a weekly one at best, or a rotating one where a quarter of the bank is inspected each week. Two conclusions follow. First, the daily layer must rely on counters rather than on physical inspection, because the physical inspection cannot fit. Second, the bank size is a maintenance variable, so a specification decision about feeder capacity becomes an operating cost — the same logic that turns placement capacity into a production plan, as in the capacity conversion for the high-speed rotary class.
The last comparison is the one that settles arguments. Assume, as an illustration only, a line running 20 hours a day at an effective 15,000 placements per hour, a board carrying 400 placements, and a contribution margin of USD 3 per good board. One hour of unplanned stop costs about 37 boards, or roughly USD 111 in margin, before any scrap or expediting. A weekly inspection that takes 45 minutes of technician time at USD 25 per hour costs under USD 19, and it is scheduled outside production. The check therefore pays for itself if it prevents one unplanned hour every six weeks — a threshold most lines cross easily on nozzles and feeders alone. Run the same arithmetic with your own numbers, including the rework cost of a wrong-part event if you have had one, and the schedule stops being a cost centre in the budget meeting and becomes the cheaper of two options. Two refinements make the comparison honest: include the technician's travel, setup and documentation time rather than the inspection minutes alone, and include the cost of the false stops a schedule creates when its thresholds are too strict.
Records are the part of maintenance that everyone resents and that decides whether the schedule is real. A completed sheet has three jobs: it proves to an auditor that the work happened, it gives the next technician the history of a specific nozzle or feeder, and it produces the data that sets the next interval. Most paper systems do the first badly and the other two not at all, because they record that a box was ticked rather than what was observed. A usable record is small, numeric and tied to an object — the machine, the position, the part number — and it is kept long enough to cover a warranty claim and a customer audit. This section covers the fields, the link to spares and warranty, and how to prove a schedule ran when the person who ran it has left.
Whatever the medium, five fields make a record useful. The object: which machine, which head position, which feeder slot, which nozzle code. The observation: a number or a state, not a comment. The outcome: pass, watch or stop, with the action taken. The identity: who recorded it and when. The next action: what the watch item requires and by when. A record without the object cannot be turned into a repair, a record without a number cannot be trended, and a record without a next action turns a watch item into a forgotten one. Everything else is optional. If the plant runs a paper route, print the five fields as columns so that a comment has nowhere to hide; if it runs a tablet or a CMMS, configure the form so that the outcome field is mandatory and the observation field rejects free text where a number is expected. The discipline is easier to hold when the form itself refuses vague entries.
A maintenance record becomes commercially useful the moment it is connected to a part number and a date. When a nozzle is replaced, the record should carry the code and the part number, so that consumption can be counted per machine rather than per purchase order. When a feeder is withdrawn from service, the record should carry the serial or tag, so that a repeated failure points at a unit rather than at a slot. When a warranty claim is raised, the records are the evidence that the failure was not caused by neglected maintenance — and their absence is the evidence that it was. The same principle applies across a mixed fleet, where a four-head machine such as the LED tubelight four-head placement platform accumulates nozzle hours on a different curve from a high-speed rotary machine, so a shared spare-parts plan needs per-platform consumption data rather than a fleet average. Retention follows the same logic: keep daily sheets long enough to cover a seasonal product cycle, and keep monthly records and overrides for the life of the machine, because warranty and audit questions arrive years after the technician who signed the sheet has moved on.
An audit question is rarely "did you maintain the machine", because the answer is always yes. It is "show me the records for the last twelve months, and show me what you did when a check failed". That second half is where most plants are thin, and it is the half that a well-built record answers automatically: every watch outcome has a next action and a date, every stop has an override entry or a repair reference, and every interval change has a reason written next to it. Prepare for the audit by making three extracts possible: records by machine for twelve months, watch items closed within their stated interval, and interval changes with justification. If those three extracts can be produced in an hour, the schedule is auditable. If producing them requires reconstructing what happened from memory, the records exist for the audit and not for the machine, which is a distinction an experienced auditor recognizes quickly.
Layer | Recorded by | Retention | What it must answer |
|---|---|---|---|
Daily | Operator on shift | 3 months | Was the pickup and vacuum check done, and what were the counters? |
Weekly | Maintenance technician | 12 months | Which feeders were tested and what failed the index test? |
Monthly | Maintenance technician and lead | Life of machine | What was the measured accuracy against the published budget? |
Overrides and stops | Maintenance lead | Life of machine | Who released the machine to production, and on what reasoning? |
Interval changes | Maintenance lead with production | Life of machine | Why was the frequency changed, and what data supported it? |
Build the schedule from your line, not from the manual's table of contents
If you are setting up or rebuilding a maintenance routine for a placement line, the fastest route to a schedule that survives is to start from data you already have: the intervention log from the last three months, the nozzle and feeder consumption from the stores system, and the unplanned stop reasons from the line report. Send those three, together with the machine models, the head configuration and the feeder count, and the schedule can be built as a first draft with intervals already tied to your own failure rates instead of to a generic table.
The draft that comes back assigns every item to a layer, states an observation and a threshold, names who may call a stop, and lists the spare parts the intervals imply. Where your data is too thin to set an interval, the draft will say so and propose a two-month measurement period rather than inventing a number. That is the honest version of a maintenance plan, and it is the one that holds up when someone asks why a check is done weekly and not monthly.
These are the questions that come up when a maintenance schedule is being rebuilt, usually from someone who has inherited a sheet nobody trusts. They are answered in the order the decisions get made: how much time the routine needs, which items are worth doing at all, how to handle a line that cannot stop, and how to keep the whole thing honest over a year. Where an answer depends on the machine model or the product mix, the answer says so and gives the measurement that settles it rather than a number pulled from another factory. Anything that reads like a performance promise has been left out on purpose: what this guide can tell you is how to decide, not what your line will achieve.
Start from the layers rather than from a target. A daily routine of four numeric checks should take under ten minutes per shift. A weekly stop of feeder index tests, rail and clamp checks, and optics cleaning should fit in 45 to 90 minutes on a machine with a moderate feeder bank, and longer on a wide bank because the count drives the time. Monthly work is a two-to-four hour stop for lubrication, accuracy verification and mechanical inspection.
If your current routine takes far more than that, the problem is usually that items are in the wrong layer — a deep clean being done daily, or a calibration squeezed into a weekly stop. Move each item to the layer that matches how fast it fails, and the total time usually falls while the defects found go up.
You need to look at them every day and clean them on evidence rather than on habit. The daily action is a visual pass under a light and a comparison of the pickup retry count against the previous shift. That takes seconds per tip on a multi-nozzle head and it catches the two conditions that produce most pickup defects: contamination in the bore and a nick or flat on the tip.
The cleaning interval itself should come from data. If the retry count climbs by the end of the second shift after a clean, clean every shift. If it stays flat for a week, a daily clean is wasted labor and, worse, extra handling that puts nozzles at risk. Record the count, then let the numbers set the frequency.
Split it. The feeder index tests and optics cleaning can be done on a rotating subset during planned changeovers, so that the whole bank is covered every four to six weeks without a dedicated stop. The rail, clamp and support-pin checks need the machine empty but not cold, and they can run at the end of a shift while the next product is being staged.
What cannot be split is the accuracy verification, because it needs a test board, a measurement plan and an undisturbed machine. Schedule that as a short dedicated stop once a month and treat it as production time in the plan, not as something to fit around it. A maintenance routine that is designed around the absence of stops will be abandoned the first busy week.
No, and the difference is mostly in counts and access rather than in principles. Head configuration changes how many nozzles you inspect; feeder capacity changes how long a bank inspection takes; whether the machine uses laser or vision centering changes which camera surfaces need cleaning and which accuracy figure you verify against. A four-head LED-class machine and a rotary high-speed machine produce different wear profiles on nozzles even when both run the same product.
Keep one master structure — the three layers, the three outcomes, the five record fields — and then write per-model annexes with the counts, the access steps and the accuracy budget. That way the schedule stays comparable across the fleet while remaining accurate for each machine.
In the currency management already uses: unplanned stop hours and scrap. Count the unplanned stops from the last quarter, group them by cause, and identify how many would have been caught by a check that does not currently exist or is being skipped. Multiply by the cost per stop hour, add the scrap and rework from the defects those conditions produced, and compare that total with the labor hours and spare parts the schedule requires.
Most lines find that a handful of causes account for the majority of the cost. The argument is strongest when it is narrow: not "we need a maintenance program", but "four nozzle positions and two feeders produced 60 percent of our unplanned stops last quarter, and a weekly index test plus a cleaning interval change removes them".
Lubrication to specification, closely followed by accuracy verification. Lubrication gets skipped because nothing happens when it is, and it gets done badly because more grease feels safer than the specified quantity — which attracts dust and wears the guide faster than no grease at all. Accuracy verification gets skipped because it needs a test board and a quiet machine, and because a drift of a few hundredths of a millimetre produces no symptom until a fine-pitch product arrives.
The cost is deferred rather than avoided. A neglected guide shows up as backlash and position error months later, at which point the repair is a mechanical job rather than a lubrication one. A skipped verification shows up as a product that suddenly will not pass, usually on the tightest board in the mix.
Paper works if the form enforces the five fields and if someone transcribes the numbers monthly into something that can be trended. The failure mode of paper is not the medium, it is the free-text comment box, which absorbs the observation and leaves nothing measurable. A printed sheet with columns for object, number, outcome, name and next action is entirely sufficient for a single line.
Software earns its place when there are several machines or when spares consumption has to be tracked per unit. Whichever route is chosen, the test is the same: can you produce twelve months of records for one machine, the watch items that were closed late, and the reasons for any interval change, in under an hour. If not, the system is recording activity rather than outcomes.
