Infrastructure-Mediated Single-Point Sensing of Whole-Home Water Activity

design:use:build:

univ. of washington

computer scienceand engineering

Jon Froehlich1, Eric Larson2, Tim Campbell3, Conor Haggerty4, James Fogarty1, Shwetak N. Patel1,2

1Computer Science & Engineering, 2Electrical Engineering,3Mechanical Engineering, 4Community, Environment, and Planning

water scarcity

barcelona, spain

lake mead, nevada

average indoor household water usage per person/day (70 gpd)

Dishwasher

Baths

Other Domestic

Leaks

Faucets

Showers

Clothes Washer

Toilets

0% 5% 10% 15% 20% 25% 30%

what are the most consuming water activities in your home?

Vickers, 2001

• single-point pressure-based sensor of water usage

• identifies water usage activity down to fixture level (e.g., toilet)

• provides estimates of flow at each fixture

hydrosense

hydrosensetoilet

kitchen sink

shower

Pressure Sensor

16-bit ADC20 MHz

Microcontroller

Class 1 Bluetooth

Radio

3D-Printed Enclosure

the hydrosensor prototype

water tower

brief plumbing primer

water tower

incoming cold water from supply line

40 psi

100 psi

1 psi = 6.89 kpabrief plumbing primer

water tower

incoming cold water from supply line

pressure regulator

pipe layout

water tower

incoming cold water from supply line

thermal expansion

tank

laundry

bathroom 1hose

spigot

hot water heater bathroom

2

kitchen

dishwasher

pressure regulator

closed pressure system

incoming cold water from supply line

thermal expansion

tank

laundry

bathroom 1hose

spigot

hot water heater bathroom

2

kitchen

dishwasher

pressure regulator

water tower

hot water heater drain

valve

hot water heater

water tower

incoming cold water from supply line

thermal expansion

tank

laundry

bathroom 1hose

spigot

bathroom 2

kitchen

dishwasher

pressure regulator

hot water heater

some possible installation points

raw bathroom sink signal

stabilized pressure drop

openvalve

closevalve

time (t)

detecting water usage events

1. detect a water event

2. classify event as “open” or “close”

3. determine source of event (e.g., toilet, kitchen sink).

4. provide flow estimate

-2

0

2

44

46

48

0 5 10 15 20 25 30 35 40 4542

44

46

48

50

raw pressure

(psi)

smoothed pressure

derivative

event detection

time (s)

0 5 10 15 20 25 30 35 40 4542

44

46

48

50

44

46

48

-2

0

2

event detection

!

critical change in pressure

raw pressure

(psi)

smoothed pressure

derivative

time (s)

time(s)

0 5 10 15 20 25 30 35 40 4542

44

46

48

50

44

46

48

-2

0

2

event detection

time(s)

!!

critical stabilization point

raw pressure

(psi)

smoothed pressure

derivative

time (s)

0 5 10 15 20 25 30 35 40 4542

44

46

48

50

44

46

48

-2

0

2

event detection

time(s)

!!

raw pressure

(psi)

smoothed pressure

derivative = valve open event

pressure decreaseand

negative initial derivative

time (s)

automatically detected event

valve open event

0 5 10 15 20 25 30 35 40 4542

44

46

48

50

44

46

48

-2

0

2

event detection

time(s)

raw pressure

(psi)

smoothed pressure

derivative

!!

valve open event

0 5 10 15 20 25 30 35 40 4542

44

46

48

50

44

46

48

-2

0

2

event detection

time(s)

raw pressure

(psi)

smoothed pressure

derivative

!!

= valve close event

automatically detected event

valve open event

pressure increase

andpositive initial derivative

home 1 home 2 home 3

toile

tfa

ucet

show

er

example open events

signature dependent on:- fixture type- valve type- valve location in home

bath

tub

bath

fauc

et

kitc

hen

fauc

et

dish

was

her

toile

tsh

owe

r

unclassified open event open event library

fixture classification

detrendedunclassified

derivativeunclassified

cepstrumunclassified

bath

tub

bath

fauc

et

kitc

hen

fauc

et

dish

was

her

toile

t

unclassified open event

detrendedshower

derivativeshower

cepstrumshower

open event library

show

er

bath

tub

bath

fauc

et

kitc

hen

fauc

et

dish

was

her

toile

t

unclassified open event open event library

detrendedunclassified

derivativeunclassified

cepstrumunclassified cepstrumshower

detrendedshower

derivativeshower

possible events

matched filter

matched filter

matched filter

show

er

test similarity

bath

tub

bath

fauc

et

kitc

hen

fauc

et

dish

was

her

unclassified open event

detrendedunclassified

derivativeunclassified

cepstrumunclassified

possible eventscepstrumtoilet

detrendedtoilet

derivativetoilet

open event library

toile

t

bath

tub

bath

fauc

et

kitc

hen

fauc

et

dish

was

her

unclassified open event open event library

detrendedunclassified

derivativeunclassified

cepstrumunclassified

possible eventscepstrumtoilet

detrendedtoilet

derivativetoilet

matched filter

matched filter

matched filter

toile

t

possible events

unclassified open event open event library

detrendedunclassified

derivativeunclassified

cepstrumunclassified

bath

tub

bath

fauc

et

kitc

hen

fauc

et

dish

was

her

kitc

hen

fauc

et

dish

was

her

show

er

unclassified open event open event library

detrendedunclassified

derivativeunclassified

cepstrumunclassified

possible events

kitc

hen

fauc

et

dish

was

her

show

er

unclassified open event open event library

detrendedunclassified

derivativeunclassified

cepstrumunclassified

possible events

derivativeshower

derivativetoilet

show

er

dish

was

her

kitc

hen

fauc

et derivativekitchen faucet

unclassified open event open event library

derivativeunclassified

possible events

derivativeshower

derivativetoilet

show

er

dish

was

her

kitc

hen

fauc

et derivativekitchen faucet

nearest neighbor match

raw bathroom sink signal

stabilized pressure drop

open close

time (t)

this is ∆P

using ∆pressure to estimate flow

r

L

pipe

𝑄= ∆𝑃 𝜋 𝑟4 8 𝜇 𝐿 ≡ ∆𝑃 𝑅𝑓 poiseuille’s law:

𝑅𝑓 = ∆𝑃𝑄 ≡ 8 𝜇 𝐿𝜋 𝑟4 fluid resistance formula:

∆P = change in pressureL = length of piper = radius of pipeμ = viscosity of liquidQ = volumetric flow rate

using ∆pressure to estimate flow

r

L

pipe

∆P = change in pressureL = length of piper = radius of pipeμ = viscosity of liquidQ = volumetric flow rate𝑄= ∆𝑃 𝑅𝑓

water tower

incoming cold water from supply line

thermal expansion

tank

laundry

bathroom 1hose

spigot

hot water heater bathroom

2

kitchen

dishwasher

pressure regulator

acquiring Rf

𝑅𝑓 = ∆𝑃 𝑄 𝑄= ∆𝑃 𝑅𝑓

in-home data collection

home profiles

ten locations- 8 houses- 1 apt / 1 cabin

size- avg: 2,300 sq ft- min: 750 sq ft- max: 4,000 sq ft

install point:- 8 hose bib- 1 water heater- 1 utility faucet

experimental protocol

- controlled experiments- 2 researchers per site

- 5 trials per valve- e.g., 5 cold / 5 hot for bathroom sink

- for each trial, valve open for 5 seconds, then closed

collecting flow data

- 4 / 10 homes gathered flow data

- measure time to fill 1 gallon in a calibrated bucket

- this provides Q, allowing us to solve for Rf - Rf = ∆P / Q

data collection stats

- ten locations- 706 trials- 155 flow rate trials- 84 total fixtures tested

• cross validation experiment• learn similarity thresholds from test data• classify each event per home using a leave-

one out method

classification results across homes

fixture classification results across homes

H1 (12 valves)

H2 (8 valves)

H3 (6 valves)

H4 (5 valves)

H5 (9 valves)

H6 (8 valves)

H7 (8 valves)

H8 (6 valves)

H9 (7 valves)

H10 (7 valves)

Aggregate Overall

0%

20%

40%

60%

80%

100% 98.9% 97.9%96.8%

Open Events Close Events

fixture classification results across homes

H1 (12 valves)

H2 (8 valves)

H3 (6 valves)

H4 (5 valves)

H5 (9 valves)

H6 (8 valves)

H7 (8 valves)

H8 (6 valves)

H9 (7 valves)

H10 (7 valves)

Aggregate

Overall50%

60%

70%

80%

90%

100%

Open Events Close Events97.9%

Sinks Toilets Showers Bathtubs Clothes Washer

Dishwasher0%

20%

40%

60%

80%

100%98

%

99%

96%

100%

100%

100%

95%

98%

89% 10

0%

100%

Open Events Close Events

fixture classification results across fixtures

flow estimation results

HomeAvg Error

(GPM)Stdev Error

(GPM)Avg Error

(%)Stdev

Error (%)H1

(7 valves)0.17 0.13 7.3 6.7

H4 (6 valves)

0.19 0.17 5.6 5.3

H5 (8 valves)

0.13 0.11 4.5 5.5

H7 (8 valves)

0.67 1.47 22.2 46.0

flow estimation results

HomeAvg Error

(GPM)Stdev Error

(GPM)Avg Error

(%)Stdev

Error (%)H1

(7 valves)0.17 0.13 7.3 6.7

H4 (6 valves)

0.19 0.17 5.6 5.3

H5 (8 valves)

0.13 0.11 4.5 5.5

H7 (4 valves)

0.15 0.18 4.5 3.8

water tower

incoming cold water from supply line

thermal expansion

tank

laundry

bathroom 1hose

spigot

hot water heater bathroom

2

kitchen

dishwasher

pressure regulator

acquiring Rf

𝑅𝑓 = ∆𝑃 𝑄 𝑄= ∆𝑃 𝑅𝑓

water tower

incoming cold water from supply line

thermal expansion

tank

laundry

bathroom 1hose

spigot

hot water heater bathroom

2

kitchen

dishwasher

pressure regulator

acquiring Rf

𝑅𝑓 = ∆𝑃 𝑄 𝑄= ∆𝑃 𝑅𝑓

0

0.5

1

1.5

2

2.5

H1 H4H5 H7

Number of Samples

Aver

age

Erro

r (G

PM)

flow estimation using interpolated Rf

After 5 Rf samples, average error is 0.27 GPM

future work: hydrosense 2.0

longitudinal data collection

water feedback interfaces

Jon Froehlich Eric Larson Tim Campbell

Conor Haggerty James Fogarty Shwetak Patel

Infrastructure-Mediated Single-Point Sensing of Whole-Home Water Activity

design:use:build:

univ. of washington

computer scienceand engineering

Jon Froehlich1, Eric Larson2, Tim Campbell3, Conor Haggerty4, James Fogarty1, Shwetak N. Patel1,2

1Computer Science & Engineering, 2Electrical Engineering,3Mechanical Engineering, 4Community, Environment, and Planning

jonfroehlich@gmail.comhttp://sustain.cs.washington.edu/blog

flow estimation

traditional inline water meter

typical water meters

- only provide aggregate information on water usage

- require pipe modification for installation

matched filtering

save maximumsimilaritysi

mila

rity

matched filtering

new maximumsimilarity

sim

ilarit

y