Robot Floor Scrubbers: How to Verify Cleaning Quality on Site
Author: SIBEN Publish Time: 2026-10-22 Origin: Site
Coverage data proves a robot drove somewhere. It does not prove the floor is clean. A map can show a completed path while a strip along every wall stays untouched, and a high percentage can hide a film that dries into visible marks an hour later. The gap is not a machine problem; it is a measurement problem. "Clean" is doing the work of a specification without any of its structure.
This guide replaces the adjective with something a supervisor, a client and a contract can all use: four measurable values, four soil classes, three field tests, coded sampling points, a shift-consistency routine, a cause-analysis order and contract columns that can survive a re-inspection.
Siben Thunder God robot scrubber working a marked channel in a power plant. The yellow boundary and the green lane are exactly the places where a coverage report and a visual check can disagree.
1. Why "clean" is not an acceptance criterion
Public tenders show the problem in plain language. One office contract asks for "no stains, water marks or sputum on floors". A hospital contract asks for compliance with national and provincial standards and for meeting the supervisor's acceptance standard. An equipment tender sets quality as "acceptable, meeting purchaser requirements".
None of these can be judged on site in a repeatable way, and none can be reviewed six months later. A word that cannot be measured cannot be enforced, priced, disputed or improved. The fix is not a stricter adjective. The fix is a method, a pass line, a location code and a re-inspection rule.
2. Split cleanliness into four measurable values
Most quality disputes are really four different questions wearing the same word. Separate them and each one becomes measurable.
| Value | Question it answers | How to measure | Example pass line |
|---|---|---|---|
| Coverage | Was the work done? | Share of cleanable surface actually reached, checked at coded points and subtracted from map exclusions | ≥ 95% of coded points reached; uncleaned edge strip ≤ 50 mm |
| Residue | How well was it done? | White-cloth wipe, raking light, residue grade against a reference scale | No visible film 30 seconds after the squeegee passes; wipe grade ≤ 2 |
| Consistency | Is the result stable? | The same test at the same points across shifts and days | Friday result within one grade of Monday at identical settings |
| Recovery time | What happens after a failure? | Hours from report to corrected patch, verified at the same point | Missed patch corrected and re-inspected within 4 hours |
Every complaint on site maps to one of the four. "You missed the corner" is coverage. "It is still sticky" is residue. "It looked better yesterday" is consistency. "Nobody came back" is recovery time. Once the complaint has a category, it has an owner and a number.
3. Audit what the map leaves out
Coverage reports look strong because the denominator is the planned path, not the floor. The path excludes exactly the places where complaints originate. Check five gaps before trusting any percentage.
| Gap | What happens | How to verify |
|---|---|---|
| 1. Virtual wall edge | The navigation boundary leaves a band along its edge that the robot never enters | Walk the boundary and measure the untouched band in three places |
| 2. Minimum edge distance | A fixed strip along walls, columns and machine bases is outside the brush path | Measure the strip at three locations and record the median, not the best case |
| 3. Low-clearance areas | Under racks, conveyor frames and machine bases cannot be reached at all | List them and give them a separate manual schedule with its own code |
| 4. Re-scrub triggers | Dark oil on a dark floor does not register as soil, so the robot skips a second pass | Keep a manual flag route for known oil zones and log the flagged points |
| 5. Transit time | Battery is spent moving between zones without cleaning anything | Read the hourly coverage curve, not the daily total |
Report two numbers instead of one: planned-path coverage and cleanable-surface coverage. The difference between them is the honest size of the gap, and it is usually the number the client has been asking for.
4. Classify soil before choosing a process
The most common process error is applying one recipe to four different soils. Water volume fixes one, chemistry fixes another, pressure fixes a third, and a stiff brush can make the fourth worse.
| Soil type | Typical source | Behaviour | Process that works | Settings to record |
|---|---|---|---|---|
| Particulate | Dust, sand, metal chips, packaging debris | Dry, moves with airflow and water; can be pushed rather than removed | Pre-sweep first, then water volume plus suction | Water volume, suction, pass count |
| Oil film | Coolant, hydraulic leaks, tyre marks, machining mist | Hydrophobic; water beads and the film survives a light pass | Chemical dwell time plus mechanical action, warm water | Chemical type and dose, dwell time, brush pressure, water temperature |
| Water-soluble | Sugar, syrup, soft drinks, some detergents | Becomes tacky when under-flushed and dries into a film that attracts dust | Enough water, two passes, avoid overdosing chemical | Water volume, pass count, squeegee condition |
| Adhesive | Chewing gum, tape residue, resin, labels | Sticks and smears; a stiff brush pushes it into joints | Spot agent with dwell time and targeted pressure | Agent, dwell time, spot tool, pressure |
Classify the dominant soil per zone, not per building. A lobby, a workshop and a loading bay rarely share one class, and a single setting applied to all three will fail in at least two of them.
5. Four settings you can record
Brush pressure, chemical type and concentration, water temperature, and speed with pass count decide the result. All four can be written down and repeated. A site that only increases dosage has not found the right setting. A site that only adds passes may be treating a suction problem as a speed problem.
Machine / ID: ____________
Zone / point prefix: ____________
Dominant soil class: ____________
Brush pressure: ____________
Chemical / concentration: ____________
Water temperature: ____________
Speed / passes: ____________
Squeegee check (film within 30 s): pass / fail
Card date / trainer: ____________
Fix the card to the machine where operators actually read it, not in a folder. The card is also the handover document between shifts, which solves most of the variation described later.
6. Three field tests beyond the eye
Visual inspection is not useless, but it is not repeatable unless the conditions are fixed. These three tests are cheap, fast and defensible.
| Test | Procedure | Record | What it misses |
|---|---|---|---|
| Raking light | Hold a torch almost flat to the floor; water marks, drag shadows and films appear | Lamp model, angle, distance and point code | Subjective unless lamp, angle and distance are fixed |
| Wipe test | Fixed white cloth, fixed 30 × 30 cm area, fixed stroke count; compare against a residue scale or weigh before and after | Point code, cloth grade or weight difference, timestamp | Transferable soil only; not film thickness |
| Wet friction | Measure at fixed points in wet and dry states with a portable meter | Value, method, point code, wet or dry state | Slip compliance is a separate judgement against a standard |
Whether a floor meets an anti-slip standard is a compliance judgement, not a field result. Keep the measurement, but do not turn a friction reading into a pass or fail on a cleaning contract without stating the standard and the test method.
7. Code the sampling points
Number each surface by zone, material and function. A stone lobby entrance becomes A-ST-01: zone A, stone, point 01. The code is what makes results comparable across shifts, operators and contract years.
| Part of code | Meaning | Example |
|---|---|---|
| Zone | Area and access level | A = main entrance and lift lobby |
| Material | Floor finish, because it decides the test method | ST = stone; EP = epoxy; CT = ceramic tile |
| Function | Traffic and soil load | 01 entrance, 02 corridor, 03 plant room |
Grade the zones so sampling effort follows risk. Grade A covers entrances and lift lobbies, every shift. Grade B covers circulation areas, daily. Grade C covers plant rooms, low frequency but never zero, because a plant room that is never sampled becomes invisible in the contract.
1. Who samples: not the operator on shift. Use a cross-shift lead, the client or a third party.
2. How it is recorded: always with point code, timestamp, method and pass line.
3. What counts as a failure: define it before sampling, not after seeing the result.
Siben Thunder God GT robot in a hospital corridor. Fixed points at the entrance, the corridor midpoint and the wall edge make a Monday and a Friday result directly comparable.
8. Hold consistency across shifts
Three causes explain most variation, and none of them is operator attitude.
Tank level: tanks that are nearly full lose suction before the alarm sounds. Dump by level, not by warning.
Squeegee wear: blades wear gradually and leave a film that dries into marks. Judge them by no visible film within 30 seconds, not by visible damage.
Filter clogging: reduced airflow shows up as the same passes no longer working. Check airflow before adding passes.
A written parameter card handles the handover. Without one, each shift re-invents the settings and the floor records the difference.
9. Assign cause with four questions
Ask in order, and stop at the first "no". This prevents the usual leap to "the operator did not try".
| Order | Question | If the answer is no |
|---|---|---|
| 1 | Was pre-sweeping done? | Sweeping debris with a scrubber only spreads it; restore the pre-sweep step first |
| 2 | Are brush, squeegee, filter and chemical mix in condition? | Replace or correct the consumable before changing settings |
| 3 | Are the settings right for that soil class? | Match chemistry, dwell time, temperature and passes to the soil |
| 4 | Is the machine degrading? | A wide morning-to-afternoon gap at identical settings points to suction, battery or brush wear |
Then close the loop in writing: log the failure with point code and timestamp, name an owner, set a deadline, re-inspect at the same point and record the result. A failure log without a re-inspection is only a complaint archive.
10. Write acceptance into the contract
Most cleaning schedules list frequency only. Add three columns and the schedule becomes enforceable.
| Column | What it contains | Why it matters |
|---|---|---|
| Point | Coded locations tied to each frequency, not "public areas" | Both sides know exactly where the result will be checked |
| Judgement | Method and pass line: lighting, permitted residue type and area, film recovery time, maximum uncleaned edge strip | Turns an adjective into a repeatable test |
| Re-inspection | Who checks, how often, and what a repeat failure triggers | Creates recovery time as a contractual value |
For outcome-based pricing, state which of the four values the invoice depends on. If payment is tied to coverage alone, the supplier is paid for motion. If it is tied to residue and recovery, the supplier is paid for the result.
After (example wording, numbers to be agreed locally): "After each scheduled pass, under 300 lux measured at floor level and raking light at 15° from 1 m, no visible film shall remain 30 seconds after the squeegee passes. Residue on a white cloth wipe over 30 × 30 cm at coded points A-ST-01 and A-ST-02 shall not exceed grade 2 on the attached reference scale. The uncleaned strip along walls and columns shall not exceed 50 mm. Any missed patch shall be corrected and re-inspected within 4 hours of report."
Siben Thunder God robot on a glossy epoxy warehouse floor. Reflections reveal film and water marks that a flat, front-on view will not show, which is why raking light and a fixed viewing angle belong in the acceptance method.
Frequently Asked Questions
Why does a high coverage report still leave missed patches?
The denominator is planned path area, not cleanable surface. Zone edges, fixed edge strips, low-clearance areas and manual exclusions are outside the path and therefore outside the percentage.
Four: coverage, residue, consistency and recovery time. Each answers a different complaint and each needs its own pass line.
Particulate, oil film, water-soluble and adhesive. Each needs different water volume, chemistry, temperature, pressure and pass count.
Raking light with a fixed lamp, angle and distance; a fixed white cloth wipe test over a fixed area and stroke count; and portable wet friction measurement at fixed points.
Partly. It detects transferable soil but not film thickness, dried marks or slip risk, so it cannot stand alone as an acceptance test.
Yes. A zone-material-function code such as A-ST-01 keeps results comparable across shifts, operators and contract years.
Not the operator on shift. Use a cross-shift lead, the client or a third party so the result is independent of the person being measured.
Usually tank level, squeegee wear or filter clogging, then inconsistent settings after handover. Check those three before blaming the operator.
Not necessarily. Film and residue are operational results; anti-slip compliance is a separate judgement against a standard, not a field result.
As method, pass line, point code and re-inspection: lighting conditions, permitted residue type and area, film recovery time, maximum uncleaned edge strip, and who re-checks what.
