Panel Questions

1.

What is missing to enable this new platform

for CPS? Is there a universal platform?

2.

What are the business opportunities and

their requirements?

3.

How do we train talent for new domain?

Rajesh K. Gupta, Computer Science and Engineering, UCSD

Pla$orm: Instrumented Buildings for

measuring occupancy, energy use

BuildingDepot
Zodiac
ZonePAC
BuildingSherlock
Sen9nel
MCC
Built in 2004, 145,000 sq ft, 5 floors!
HVAC : VAV with reheat coil, 237 zones!
Occupants : Faculty, staff and students!
more than 1 year data, 17+ sensors per zone !
100s of Air Handler (AHU) sensors!
every ~5 minutes!

Complicated Informa:on
Flows in CP Systems
Many busy users

Infrequent,
Quick

Low processing
at receivers
Amplify Informa9on
Many, somewhat
interchangeable

Limited
aKen9on

Few overloaded (and

oB ignored) operators

LOW
BANDWIDTH

Ad-hoc, intermiKent,
diverse

Frequent, slow
interac9ons
Focused
aKen9on

Cyber Customers

REDUNDANCY

Filter and compress

requests for aKen9on

Data deluge

CP Opera9ng System

Wealth of data, poverty of aKen9on, Herb Simon

Value in numbers,
redundancy

Constrained
Physical world

Challenges for CPS Development

Smart Home

HVAC

Developer

Water

Security

Lack of infrastructure to collect data

Needs to acquire domain exper9se
Hard to extract informa9on from data

Mul:media
4

Find Architectural Support To

Structure Informa9on Flows
HVAC
Control based on weather
Use of solar panels and solar
heaters
Adapt to user comfort

Security
Biometric locking system
Video surveillance
Smoke detector
Alarm system

Smart Home as an example CPS

Water
Cooking, drinking & washing
Lawn irriga9on system
Solar hea9ng system
Pool ltra9on & hea9ng

Mul:media
Television & Radio
Video Games
Internet
Recording systems
5

Pu^ng Things Together

HVAC Model

Smart Home
Architecture

Security Model

Does it really work?

How does it work?

Developer

Water Model

Ease of access to
informa9on
Minimum domain
exper9se

Mul:media Model
6

Compose This!

Many thousands of lights!

Energy hogging elevators!

Strong requirements!

many 10s of large fans!

10s of large pumps!
Complex HVAC system!
with many moving parts!
many 100s of dampers!

PEOPLE ARE INTEGRAL PART OF CPS

1. HOW DO WE DIFFERENTIATE, INTEGRATE AND

COMPOSE CPS CONSISTING OF HUMANS?
CONTEXT ?
8

2. Business: Exci:ng New Pla$orm with new

Apps
Apps
Visualize Maintain

Analyze

Control

Next genera9on
building applica9ons
via standardized API

REST/ Na9ve API

Scalable, distributed data storage

Metadata and contextual tagging
Access control across users
REST API for app development

Data management
system for sensors
and actuators

Data Connectors

Building A

Building B

Building C

Building D

Large amount of data

generated in modern
buildings

[1] Agarwal, Yuvraj, et al. "BuildingDepot: An extensible and distributed architecture for building data storage, access and sharing."
Proceedings of the Fourth ACM Workshop on Embedded Sensing Systems for Energy-Eciency in Buildings. ACM, 2012.
9

Models can extend reasoning

methods
Rule based

If

but s9ll

then

Model based

If

then I expect

but I got

so

Change based

The last 9me

I did

sensors told
me
I got

but
this 9me

so
1
0

Building Data Models and Compara:ve

Data Mining

Working directly with sensor readings tends to nd

extremes in sensor readings

Large dierences between zones even on the same day

Models that capture on the inter-rela9onships between

sensors and parameters of interest

Cluster rooms with the same characteris9cs

Sensi9vity to confounding parameters (human ac9ons)

Compare rooms that have the same confounding parameters

[1]. Narayanaswamy, Balakrishnan, et al. "Data driven inves9ga9on of faults in HVAC systems with model, cluster and compare (MCC)."
Proceedings of the 1st ACM Conference on Embedded Systems for Energy-Ecient Buildings. ACM, 2014.
11

BeXer Models, Improved

Diagnosis

Build learning models from

sensor data

Cluster model parameters

Compare groups

Design Rules for fault
detec9on and diagnosis.

Model, Cluster, Compare: Enables us to detect

and use rela:onships

Conditioning more when unoccupied!

Actuation and Sensing not in sync!

High Airflow !

Causes overcooling!

High set point causes over-cooling!

1
3

Setpoints never change!

High Airflow all the

time !

Set points and actuation not in sync!

High Airflow ! Even when temperature

within setpoints!

(Unnecessary) high flow when within setpoints!

13

Needs, Requirements
Models
Encapsulate domain knowledge
Standardize representa9on of disparate systems
Abstrac:ons
Simplify access to domain exper9se
Facilitate communica9on across systems
Architecture

Allow models to co-exist to create a system of models

Provide mechanisms for protec9on, communica9on and
consistency
14