Root Cause Analysis for Recurring Stone Fabrication Defects

A chip on a mitered edge gets filled, buffed and shipped. Two weeks later the same profile chips again on a different slab, and it gets filled again. Nobody writes anything down, because each individual chip took ten minutes to repair and never felt like a problem worth escalating. That is how a shop ends up donating a meaningful share of its finishing labor to the same defect for years without ever seeing it as a line item. Rework hides well. It happens late in the day, by whoever is standing at the station, and it rarely generates a document.

Root cause analysis is the discipline of treating a repeat defect as a question about the process instead of a task on a part. That distinction carries more weight in stone than in most trades: you cannot un-chip a slab, and you cannot re-run a scrapped island top out of the same block six weeks later. What follows is how to separate a one-off from a pattern, how to describe a defect precisely enough to group it, how to work through causes without falling in love with the first plausible story, and how to make a fix outlive the person who invented it.

One-Off or Pattern: What Actually Deserves an Investigation

Not every defect earns an investigation. A slab with a hidden fissure that opens on the saw is bad luck. A forklift that clips a corner is an incident. What earns a formal look is repetition — the same defect, on the same class of part, appearing across different jobs, different slabs and different days. You cannot see repetition from memory. Shop memory anchors itself to whatever happened most recently and to whoever complained loudest, and it systematically forgets the quiet, cheap rework that a polisher absorbs without mentioning it to anyone.

So the first piece of infrastructure is not an analysis method. It is a defect log. A clipboard by the bay works, and so does a shared spreadsheet, as long as an entry takes under a minute and someone actually fills it in. The log has one job: to let you group. Grouping is the entire point. Forty scattered notes that say "chipped" teach you nothing. Forty notes that say chipped, front outside corner, ogee, machine two, dark granite, point straight at something.

For grouping to work, each entry needs enough dimensions to slice on later. Record what the defect was in plain language, where it sat on the part, which machine or station produced it, which operator was running it, which material and which slab or lot it came from, which tooling was in the machine and roughly how much life was on it, the date and the shift, and what was done about it. That is nine fields. It sounds excessive until a missing field stalls an investigation three months later.

Then read the log in bulk rather than one line at a time. Once a month, sort by defect type and look for concentrations. A defect spread evenly across machines, materials and people is probably telling you something about your method or your specification. A defect that clusters hard on one machine, one profile or one shift has already narrowed itself for you. The clustering is a finding, not an explanation, but it tells you where to aim an hour of real attention.

Working the Investigation

Describe the Defect Before You Explain It

Most failed investigations fail in the first five minutes, at the moment somebody converts an observation into a theory and the room starts working on the theory. "The edges are chipping" is already a step removed from what is visible. Push the description back to physical fact: chipping on the top arris of the finished edge, outside face only, beginning roughly 40 mm from the corner and running toward the middle, on the ogee profile, present after the second wheel and not after the first.

That level of detail does two useful things. It tells you which parts of the process are even candidates, because a defect already present after the second wheel cannot be caused by anything downstream of the second wheel. And it gives you an inspection criterion. Without one, two people looking at the same part will disagree about whether the defect is present, which makes every before-and-after comparison worthless.

Write down what is not happening, too. The straight profile does not chip. The light granite does not chip. The other machine does not chip. Negative observations narrow a cause faster than positive ones, because any candidate cause has to explain the absence as well as the presence. If your theory is worn wheels but those same wheels run a clean straight profile all day, the theory has to account for that or it is simply wrong.

Five Whys, Without the Fiction

The five whys is a chain in which each answer becomes the next question, and it works only when every link is something you can verify. Chipping on the ogee, because the second wheel is loading up, because coolant flow at that station is low, because the nozzle is partly plugged, because the filter sits on no cleaning schedule. Every link there can be confirmed on the floor inside an hour.

The failure mode is the invented link. "Because the operator was rushing" is not a cause you can inspect. It is a judgement, and it terminates the chain at a point where the only available remedy is telling somebody to try harder. If a link cannot be confirmed by a measurement, a photograph, a physical part or a document, label it an assumption and keep that label visible. Assumptions are fine; assumptions quietly promoted to facts because three people repeated them are how shops buy equipment they did not need.

Cause Categories Worth Checking First

When the five whys stalls and everyone shrugs one link in, switch to a cause-and-effect map. Put the defect on the right and hang branches for machine, tooling, material, method, measurement, people and environment. The categories exist for coverage: they force experienced people to say something about seven areas instead of arguing about the two they already believed in.

Populate each branch with candidates, mark each candidate as cheaply checkable or not, and check the cheap ones first. Coolant pressure at the head takes five minutes to read. A theory about slab batch variation takes a week and a call to the supplier. Order the work by evidence per hour, not by how interesting the theory sounds. The table below maps common recurring stone defects to the branches worth walking first.

Proving a Cause Instead of Telling a Good Story

The moment a plausible cause appears, the instinct is to fix everything adjacent to it at once: new wheels, fresh coolant, slower feed, and a conversation with the operator. If the defect then goes away, you have learned nothing and you are permanently carrying four changes, one of which is the fix and three of which are pure cost. Change one variable, run enough parts that the defect would show itself if it were still present, record the result, and only then move to the next variable.

How many parts counts as enough depends on how often the defect appears. If it showed up on roughly one part in five, three clean parts prove nothing at all. The working rule is to run enough parts that seeing zero defects would be genuinely surprising if nothing had changed. When a defect is rare, a deliberate test on offcuts answers the question faster than waiting weeks for production volume.

Take the recurring chip on one edge profile. The log shows it on the ogee, on both machines, across three materials, always on the top arris. That spread rules out a single machine and largely rules out material, leaving tooling and method. The controlled comparison is simple: run the same profile on the same material with the current wheel sequence, then again with the grit somebody removed to save cycle time. If the chipping tracks the missing grit, the cause is an abrupt step in the profile chain, not bad wheels.

The haze that only shows on dark granite is a different animal, because dark stone reveals what light stone conceals. Before chasing a process cause, confirm the defect is real. Put a gloss meter on a light-colored part finished through the identical sequence. If the light part reads the same as the dark one, you do not have a haze problem, you have a visibility problem, and the correct response is to set a measured acceptance level rather than to send parts back for another pass. If the dark stone genuinely reads lower, walk method and measurement: dwell at the finer grits, pad condition at the point of use, water volume, and whether the operator is judging finish under shop lighting that flatters everything.

Seams that open weeks after install are the hardest class, because the feedback loop is long and the evidence sits in a finished kitchen. The only workable approach is to build the evidence at install time, before you know whether there will be a problem: photograph the substrate and the support under the seam, note whether the cabinets were shimmed level, record the adhesive and its lot, note the temperature and whether the pieces came in cold off the truck. Six months of that turns a mystery into a sortable table. Openings that concentrate on one crew point at method; on one season, at thermal movement; on one cabinet style, at support and deflection.

Inconsistent gloss on one machine is the case where measurement itself should be the first suspect. Two meters rarely agree exactly, and one meter used at different spots on a part will not agree with itself. Before condemning the machine, mark five locations on a single part and have two people read all five with the same instrument. If their numbers diverge by more than the machine-to-machine gap you were chasing, the reading method is the defect. Only if the readings hold up should you walk the machine branch: head pressure, spring or pneumatic condition at each head, rail and carriage wear, and whether the part is held genuinely flat.

Verifying the Fix, Then Making It Permanent

A fix is not verified because the defect stopped appearing for a week. It is verified when the defect rate after the change is clearly lower than before, over enough parts that ordinary variation cannot explain the difference, and when you can state the mechanism by which the change works. If you cannot explain why it works, you have a correlation, and the right posture is to keep watching rather than close the file.

Keep the log running after the change, using the same fields, so that before and after remain comparable. Resist the temptation to redesign the form at the same moment you implement the fix, because changing the measurement and the process together destroys the comparison you were trying to make. If the defect returns at a lower rate rather than vanishing, you likely found one of several contributing causes and the others are still live. Watch, too, for the defect simply moving downstream rather than ending.

Then write the change into the process, because a fix living in one head has a half-life measured in staff turnover. That means updating the document the operator actually follows: the setup sheet, the polishing sequence card, the tool change interval, the install checklist. Record the setting, one line explaining why it exists, and the date. That reason line matters, because undocumented rules get deleted by the next efficient person who spots a step that looks redundant.

Close the loop with the machine itself wherever you can. If the fix is a setting, lock it into the program instead of trusting recall. If it is a maintenance interval, put it on the shop schedule rather than a sticky note. If it is a hand technique, watch a new hire perform it before signing them off. Corrective actions that depend on daily discipline decay; those built into a program, a fixture or a checklist do not.

Stable inputs make root cause work dramatically easier, because consistent tooling removes an entire branch of the diagram before you start. If your investigations keep landing on wheel and pad variation, it is worth reviewing the full range of fabrication tooling and standardizing one sequence across every machine on the floor. Our shop technique guides cover the setup details behind several of the defects described above.

Free Tool

Free Guides & Tools — a working set of shop references, checklists and selection aids you can use to build defect logs, setup sheets and corrective-action records that survive staff turnover.