A factory does not need help deciding whether steam looks impressive. It needs an answer to a narrower question: will this work on this machine, in a window the schedule will actually give, done by the people who will do it, without a bill that arrives next week as rust, a stiff way cover or a stopped spindle. That answer does not come out of a specification sheet, and it does not come out of a demonstration on somebody else's equipment.
The difficulty with machine tools is that they are not one surface. A machining centre carries painted sheet metal, machined tables, way covers that hold water behind their folds, guides running under an oil film that protects them, a spindle bore that tolerates no residue, cable glands that admit steam through entries nobody has ever seen wet, and a floor around the base where run-off slows, cools and drops what it was carrying. A trial that treats the machine as a single object produces a single verdict, and that verdict will be wrong in both directions somewhere on the machine.
The sequence below starts with the questions that cost nothing to answer, moves through the one test that costs a machine and a window, and ends with the checks and the record that turn a wash into something a plant can act on. Every step is worth doing, and each one also misleads when it is asked a question it cannot answer. That mismatch, more than any parameter, is why a factory trial so often fails to predict the production it was meant to predict.
Write the decision before you design the test
A trial that is not attached to a decision produces a visit report. Before anything is scheduled, write down which of the three questions the trial exists to answer, because the cheapest test differs for each one.
The first question is whether a particular residue can be removed from particular surfaces by this method. The cheapest test that can carry that answer is a small one, run while production continues. The second question is whether the task fits a downtime window the plant will really give. That answer only exists inside a real window, so the test has to be a whole machine. The third question is which surfaces must never be in scope, and it is usually the most valuable answer a trial produces, because it converts an argument into a boundary.
Then write the pass condition in language a person can stand in front of. Low-angle light shows no film across the tested area. The white cloth the team has agreed on comes away clean from two marked panels. A low-angle check inside the joints finds no water. Bare metal still looks bare when it is looked at again hours later. Words such as clean, effective or acceptable cannot be signed, and a trial judged on them will be re-argued by people who were not standing there.
Write the scope as surfaces, not as machine names. "Clean machining centre four" is a task that two people will perform differently. "The exterior guards, the chip conveyor hood, the coolant tank lid and the panel faces, with the way covers and the electrical cabinet excluded" is a task that can be timed, repeated and compared.
Start with a panel that is already off the machine
The cheapest test available to a factory costs nothing in production time, because it uses something already removed: a guard that came off for a repair, a spare panel in the same finish, the inside face of a cover nobody ever looks at, a scrapped enclosure section from stores. It is a screening test, and it is worth doing first, because it fails cheaply.
What it can honestly prove is how a finish responds to heat and dwell. Whether gloss changes at the distance and pass speed you intend to use. Whether a decal edge or a printed service label lifts as the adhesive softens, and whether it lifts only after cooling, which is why the panel should be looked at again later in the day rather than only at the moment the wand moves away. Whether the deposit releases under heat at all, or whether it is the kind that needs mechanical work first.
That last point decides more trials than any machine specification, so it is worth being precise about the deposit. Three states are enough, and each can be identified in under a minute with a torch, a cloth and a plastic scraper. A fresh film is glossy, transfers fully onto a cloth, and feels smooth between the fingers; heat and a short dwell usually take it. A polymerised film is dark, dull or tacky at the edges, resists the cloth, and only breaks at the edge of a blade; it needs sustained heat, several short passes and work on it while it is soft. A deposit bound with grit or swarf feels abrasive between the fingers and will not move until the solid is removed dry; washing it straight only relocates the solids and they re-deposit downstream. A trial run on the wrong state of deposit gives a wrong verdict about the method, and the wrong state is far more common than the wrong machine.
What the panel cannot prove is everything a panel does not have: joints, a drainage path, way covers, gland entries, a nameplate, an oil film that has to be replaced the same day, and a production schedule waiting behind it. A panel that survives is not a machine that passes. Treat the result as a screening and move on.
Then one machine, inside a window that already exists
A machine-level test is only a test when the conditions around it are the ones the real job will have. That means planned downtime that is already on the production schedule, not a favour asked of the planner on the morning. It means the operators who will do the work, not a vendor technician holding the wand. It means the deposit as it actually is, left to accumulate for a stated interval, rather than a surface cleaned for the visitor. And it means the pass wording written down before the machine is wet.
Two controls make the difference between an opinion and a result. On each representative surface, mask half of it with tape and treat only the other half; the untreated half is what tells you whether the change you are looking at came from the process or from the surface. And change one variable per pass: distance, dwell, pass speed or mode, never two together. A trial that changes three things at once can report success but cannot say which setting produced it, and an unrepeatable success is not a process, it is an anecdote.
Timing starts at lockout and ends at sign-off, not at the last pass of the wand. Cool-down and isolation, dry removal of chips, masking, the passes the deposit state actually needs, drying and drainage, re-oiling of anything the wash stripped, the checks, and the signature of the person who will release the machine. Those are the numbers that decide whether the task is viable, and none of them appear on a data sheet.
This test can decide whether the method works on your surfaces, what the real task footprint is, and what has to be masked. It cannot decide how the task behaves across a fleet, or how quickly the deposit returns once the machine goes back to cutting. Those need a second and a third machine and a calendar, and pretending a single trial settled them is how a good trial gets a bad outcome.
Judge surfaces one at a time, because a machine is not one surface
The record sheet should have one row per surface, not one per machine. "Machine four clean" is not an executable result; it hides the two surfaces that failed and the one that should never have been attempted. In machine tool steam cleaning, the useful output of a trial is usually a boundary drawn surface by surface, and each boundary has a different cost attached to getting it wrong.
Painted guards and panels are the least consequential and the most visible: a gloss change or a lifted decal edge there is noticed by every auditor and every customer who walks the line. Machined tables and stainless surfaces are not harmed by contact but by what is left behind, since water spots and flash rust appear on bare metal hours after the surface looked dry. Way covers and bellows feel dry on the outside and hold water behind the folds long afterwards, and a cover that has been washed and left wet stiffens as it dries. Linear guides, ball screws and racks run under a film that was protecting them, so a wash removes the protection along with the soil, and re-oiling the same day belongs inside the task rather than in a follow-up note. Spindle bores and precision surfaces should simply not be in scope of a first trial. Electrical cabinets, cable glands, nameplates and decals admit steam through entries that are never seen wet, and condensation forms inside where nobody is looking. Chip conveyors, coolant tank lids and the floor around the base are where the run-off goes, and the question of who owns oily water belongs to the site and to the local drainage authority, not to the equipment supplier.
From that list comes the arithmetic that keeps a trial honest: the area you can actually wash is the visible area minus the protected area, and both the masking and the un-masking take time that belongs to the job. A trial that reports only the washing minutes has hidden most of the work, and will report every method as unreliable.
Measure what the data sheet does not give you
A data sheet describes a boiler. The Fixed Steam Cleaning Machine PowerDe HW02 listing gives a steam temperature of 160 degrees Celsius, a vapour pressure of 0.8 MPa, a heating power of 6 kW, an 80 L SUS304 tank, a tap-water inlet with a 1/2-inch internal-thread fitting, dry and wet steam modes with a vacuum cleaner, and a working voltage of 220 V. Read the first two as one condition rather than two capabilities: on a saturated steam generator, temperature and pressure are two readings of the same state, and neither of them says what arrives at the surface once a hose and a wand are fitted.
So the first hour of a trial is spent measuring the things the page does not publish. How long the machine takes from cold to usable steam. Whether the trigger puts out water for the first second, which means the line was not dry at the start of a pass. Whether the steam column thins under a long trigger, which means the machine cannot sustain the duty you are planning. Whether the change between dry and wet mode is actually felt at the nozzle, because if it is not, it cannot be written into a procedure. And the surface temperature at your own working distance and pass speed, taken with a contact or infrared thermometer, because that number and not the boiler number is what the deposit is actually exposed to.
Ask for the rest in writing before anything is ordered: the nozzle-end temperature and pressure with the intended hose and wand fitted, the hose and wand lengths, the warm-up time, the working time per tank, and the recovery time after a refill. The 6 kW heater and the 80 L tank describe a working cycle and a walk to a tap, not coverage per hour. The 220 V supply needs a dedicated circuit confirmed by your own electrician. The protections the page names, covering leakage, short circuit, high temperature, high voltage, water shortage and high-voltage explosion, protect the machine; they do not protect the site circuit, the floor, or the person holding the wand. A stop caused by an empty tank is a scheduling signal, and a team that reads it as a breakdown will lose an hour and phone someone.
One more written item matters more than it looks. The same page states its uses as vehicle exterior cleaning at fuelling and parking areas, interior vacuuming and disinfection, and engine-compartment cleaning. Machine-tool surfaces are not on that list. The numbers do not extend the list, so the surfaces in your scope should be put to the manufacturer as a written question, and the answer kept with the purchase record.
Test the window as well as the surface
A factory can prove a method works and still fail to use it, because the task does not fit the time it will be given. So the trial records each phase with a stopwatch and compares the total against the window the schedule will actually provide, taken from the plan rather than from goodwill: cool-down, isolation, dry removal, masking, the passes the deposit needs, drying, drainage, re-oiling, checks and sign-off.
If the total does not fit, there are three honest outcomes. Reduce the scope to one face and repeat that face on a later window. Split the task into repeats that return to the same face at the same stage, so each session continues where the last one stopped. Or move the task out of the shift calendar altogether and give it a shutdown, where the time genuinely exists. Asking the crew to be faster is not an outcome; it is how a well-run trial acquires a bad reputation, and it is also why a cleaning task that appears on every weekly plan quietly stops appearing at all.
Two clocks run on different schedules from the window. Flash rust appears on bare metal hours after the surface dried, and water that migrated into a joint shows itself as a streak the following morning. The window contains the work. The checks belong to the next shift, and a plan that counts them as part of the same window will always finish late.
Check twice, because machine tools answer late
The most informative results of a machine-tool trial appear after the team has gone. Bare metal that looked fine at sign-off shows flash rust the same evening. A decal edge or a printed service label lifts after the adhesive has been heated and then cooled, not while it is being heated. A way cover feels dry to a hand and gives up water when it is flexed the next morning. Run-off that flowed past the machine re-deposits a film of oil where it slowed and cooled, usually at the floor near the base or under the conveyor discharge. And the lubricant film that a wash removed from a guide shows up as the first thing the operator notices on the next shift, in a machine that was running well the day before.
A surface that looked right when the wand left it has not passed. The next-morning check is part of the test, not a follow-up to it: a low-angle light across the tested faces, a look inside joints and behind folds with a torch, a check that the labels the trial needed are still legible and still attached, and a look at the guides to confirm they were re-oiled. Photograph the same reference points before the work and at both checks, so the comparison does not depend on anyone's memory.
Write the result so someone who was not there can act on it
One page per machine, one row per surface. For each row: the state of the deposit before the work, what was masked and how long masking took, the setting that produced the result (mode, distance, dwell, number of passes), the measured surface temperature, the timings including warm-up, refills and recovery, what each of the two checks found, and whether the source of the residue was touched. That last column is the one that prevents a false success. A leaking joint that was washed but not repaired means the trial reset a clock rather than changing a rate, and the same film will be back before the next planned stop, which is when people conclude that the method does not hold.
That record supports three decisions, and only three. Extend the task to more machines, using the settings that worked and the exclusion list the trial produced. Change the method for one class of deposit, because the trial showed that bound deposits need dry removal or softening before any heat is applied. Or change the form of the equipment, because the trial showed the work converging on one point with a drainage path, or conversely scattered across a shop floor that no fixed installation can reach.
Send that record, not a summary, to whoever you are comparing. A summary attracts a brochure; a record attracts an answer. When you put cutting oil removal equipment side by side, the questions worth asking are the ones your own trial left open: nozzle-end figures with the intended hose fitted, hose and wand lengths, warm-up and per-tank times, and written confirmation for the surfaces in your scope. Those can be put to the contact PowerDe team together with the surface list, the deposit description and the downtime window, which is a shorter conversation than the one a wrong machine starts.
A factory does not need one large trial. It needs the cheapest test that can carry the decision in front of it, run by the people who will do the work, judged in wording written before the machine is wet, and checked again the next morning. Done in that order, "can steam clean this machine" stops being a single unanswerable question and becomes a set of small ones, each with an answer you can act on and a boundary you can defend.
