| Distributed
Diagnostics
Sensor-rich
networks that track
the performance
of components inside
electro-mechanical
devices are leading
to a new generation
of machines that
can diagnose and
repair themselves.
Distributed sensors
monitor multi-modal
data, measuring
such physical phenomena
as vibration, noise,
electrical current,
and signature – characteristic
changes in a signal
over time. PARC researchers
are developing
scalable, model-based
techniques for
processing the
information from
these distributed
sensors to achieve
highly distributed
diagnosis and rapid
device reconfiguration
and repair.
Researchers are
exploring problems
of complex control,
self-repair, and
efficiency in developing
self-diagnosing
and self-maintaining
machines. These
devices embed diagnostics
intelligence close
to the sensor signals.
When the system
identifies a problem,
it can direct components
to repair themselves
or trigger an error
message calling
for a repair technician
to fix a part that
is wearing out
or that has broken
down.
Synergistic
Integration
A unique feature
of this research
is its synergistic
integration of
low-level signal
processing and
high-level, model-based
computation. Scientists
are using multiple
concurrent diagnostic
hypotheses at varying
levels of abstraction
to detect machine
malfunctions and
abnormalities.
Researchers are
going beyond traditional
diagnosis and fault
detection and isolation
(FDI) to address
large-scale networked
systems with many
components, complex
hybrid dynamics,
and rich, noisy
sensing capabilities.
They are focusing
on the distributed
and hybrid diagnosis
of networked devices
that exhibit complex
discrete and continuous
dynamics and operate
in a failure-prone
environment.
Applications
Researchers
first applied distributed
diagnostic capabilities
in PARC’s
signature analysis
tool, which predicts
the breakdown of
hardware components
by sensing and
analyzing their
vibrations. Signature
analysis can predict
the remaining lifetime
of used parts,
enabling components
from used machines
that are not in
danger of wearing
down to be used
in rebuilt machines.
Xerox Corporation
has used PARC’s
signature analysis
tool in its remanufacturing
process.
Distributed diagonistic
research is focused
on two projects:
Online Hybrid
Modeling
The Online
Hybrid Modeling
for
Fault-Adaptive
Real-Time Control
project is
aimed at developing
systems for
on-line
failure
diagnosis and
compensation
for fault-tolerant
operation of
smart
sensor systems.
These systems
are being developed
for use in
aerospace applications.
The
project is focusing
on a
hybrid approach
that combines
signal estimation
with model-based
reasoning
and distributed
qualitative
diagnostic
techniques
to
enable rapid
diagnosis
and recovery.
Scientists
are planning
to
scale experimental
networks
to
larger, more
complex
distributed
systems through
multiple
hypothesis management
(MHM),
which allows
the computer
to create
many hypotheses
and track
each
of them.
The Online
Hybrid Modeling
is funded,
in part,
by the government
through
the Defense
Advanced
Research
Projects
Agency (DARPA).
Features
- High-resolution diagnosis suitable for
control
- Computationally tractable
Data
Mining
In the
Data
Mining project,
funded
by
Xerox Corporation,
distributed
diagnostics
systems
collect
data
from
inside
machines
in
the
field
and
apply data
mining and
machine
learning
algorithms
to
understand how
their
parts perform
in
the
field. With
data
about
how parts
inside
its
machines wear
out in actual
use,
Xerox can
improve
service
by
scheduling maintenance
and
parts
replacement
in
advance of
a
breakdown.
|
