← All work

Product development

A dual-flush valve, modelled first and proven on the bench

Barros & Moreira, a Portuguese manufacturer of flushing systems, needed a dual-flush valve whose full and reduced flushes both land inside the EN 14055 acceptance windows. We modelled the valve's hydraulics, calibrated the model on our own test bench, iterated with 3D-printed prototypes and ran a designed experiment to find the setting that meets both.

The challenge

A dual-flush valve delivers two volumes from one mechanism, and EN 14055 sets an acceptance window for each. Because both flushes run through the same parts, a change that corrects one can push the other out of its window. Both also depend on hydraulic effects, such as buoyancy, surface tension and the float’s piston effect, that are hard to predict from first principles alone. Get them wrong and the valve fails the standard, or delivers the wrong volume on every flush.

What we did

  • Built a parametric hydraulic model of the valve and calibrated it against bench measurements, so each design change could be predicted before printing.
  • 3D-printed and tested successive prototype generations in water within days, until the full flush was on target with fewer, simpler parts.
  • Ran a root-cause analysis and a 2³ factorial experiment on our own bench, then used a regression model to optimise the setting against the standard.

Approach

1. Model the valve, then prototype it

We modelled the valve parametrically, covering volumes, flow rates, buoyancy, weight and timing, and calibrated the model against bench measurements. It shows how each geometric change, such as raising the float or narrowing a tube, moves the discharged volume, so design choices rested on physics rather than trial and error.

Complete flush: discharged volume vs time02468012345Time (s)Volume (L)
The hydraulic model (line) against bench measurements (points): discharged volume over a complete flush. Once calibrated, it predicted the effect of a design change before a part was printed. Figures simplified for publication.

We then 3D-printed each candidate (tube diameters, float volumes, seal cones, counterweights) and tested it in water within days. Several prototype generations later, the full flush was on target and the design was converging on fewer, simpler parts.

Bagged and labelled 3D-printed prototype parts from successive design iterations, next to two assembled valve mechanisms and a laptop.
Successive 3D-printed prototype iterations, each bag holding one configuration under test (pin diameter, hole diameter), labelled and tracked generation by generation.

2. Find what controls each flush

When early prototypes fell short of the target flush, we went through every contributor to the discharged volume, from flow, buoyancy and weight to surface tension and the float’s piston effect, and tested the suspects one by one. To measure what actually controls each flush, we ran a 2³ factorial experiment: three factors (tank water level, full-flush window, reduced-flush window), each at two levels, in eight runs, every flush timed and measured on our own bench rig.

A flushing valve mechanism mounted inside a clear acrylic enclosure on a wooden test stand, plumbed to a water supply line.
Our bench rig, with a clear enclosure around the valve so every flush could be watched and timed directly, plumbed to a live water supply.

The answer was simple. Water level moves both flushes by about the same amount; the full-flush window trims the full flush only; and, within the range tested, the reduced-flush window barely moves either. Interactions between the factors were negligible, so the valve behaves almost linearly.

Effect on discharge volume (L)Full flushReduced flushWater level+1.38+1.23Full window+0.53+0.02Reduced window+0.08+0.03
Main effects from the 2³ experiment, in litres. Water level drives both flushes; the full-flush window acts on the full flush alone.

3. Optimise against the standard

Because the behaviour is close to linear, a regression model fitted to the eight runs predicts the settings in between. We used it to run 3,000 virtual settings, a Monte Carlo search, scored each by its distance from the standard’s acceptance windows and kept the closest.

Normative space: full vs reduced discharge11.82.53.343456786 / 3 L4 / 2 LoptimumoptimumFull discharge (L)Reduced discharge (L)
Each flush plotted as full against reduced volume, over the standard's acceptance windows. From 3,000 virtual settings the model found the setting that lands inside both windows, full and reduced flush at once.

The result is a product decision: a single factory setting can satisfy both the 6/3 L and 4/2 L variants. That means fewer product codes, simpler assembly and a path to removing the adjustment windows altogether.

Outcome

Barros & Moreira has a prototype tested against EN 14055 and a calibrated model that lets its team explore new settings without going back to the bench for every change. The same approach, model first and then prove it on the bench, now runs through the other products we are developing with them.

Related work · Barros & MoreiraChoosing the product strategy before committing to toolingThe strategy side of the same product range: two routes compared on components, material, unit cost and tooling.Read case study

Facing a similar challenge?

A first conversation takes about 30 minutes and costs nothing.