Prognostics TestbedPrognostics TestbedPrognostics TestbedPrognostics TestbedBhaskar Saha and Kai Goebel

National Aeronautics and Space Administration

Problem

Bhaskar Saha and Kai Goebel Prognostics Center of Excellence, NASA ARC

Problem

Motivation

To facilitate research in prognostics, it is imperative to have

Requirements

The testbed shall:To facilitate research in prognostics, it is imperative to have

a hardware testbed that mimics the complexities and

issues encountered for a real system.

Such a system will support

• Algorithm development

• Testing and validation of prognostic tools

• Resemble a system that has real-world relevance

• Allow for repeated run-to-failure of components

• Perform run-to-failure in reasonable time

• Support monitoring of ground truth

• Collect data for state assessment• Testing and validation of prognostic tools

• Benchmarking of different approaches

• Development of metrics for prognostics

• Collection and dissemination of run-to-failure data

Goal

• Demonstrate ability to distinguish between components

• Collect data for state assessment

• Support demonstration of prognostic solutions

• Allow control of several operational and/or environmental

variables

• Allow quantification of uncertainty sources

• Support repeated run-to-failure within a finite budget

System

• Demonstrate ability to distinguish between components

at different health states having similar external

observables and then to predict the end of life

• Support repeated run-to-failure within a finite budget

• Support automated data collection during the aging

SystemTestbed – Data

CollectionExperimental setup

• A set of Li-ion cells

• Aging dynamics slow enough to be observable and fast enough for reasonable run-to-failure

times (~1 month)

• Low cost

BHM

• Low cost

• May be aged either inside or outside an environmental chamber

• Programmable Charger and Electronic Load

• EIS equipment for battery health monitoring (BHM)

• Sensor suite – Voltage, Current, Temperature

• Custom switching circuitry

• Data acquisition system

• Computer for control and analysis

Experimental Plan • Cells are cycled through charge and discharge under

different load and environmental conditions set by

the electronic load and environmental chamber

respectively

• Periodically EIS measurements are taken to monitor

the internal condition of the battery

Aging

• Computer for control and analysis

the internal condition of the battery

• DAQ system collects externally observable

parameters from the sensors

• Switching circuitry enables cells to be in the charge,

discharge or EIS health monitoring state as dictated

by the aging regime

Results

Algorithm Development

The algorithms considered so far include both

model-based as well as data-driven algorithms, for

Sample Results

Results

example

• Relevance vector machines (RVM)

• Gaussian Process Regression (GPR)

• Particle Filters (PF, RBPF)

• Neural Networks (NN)

• Random Forest Regression

• ARIMA models

• Kalman Filters

Further algorithms will be explored and results will

be published to disseminate findings on be published to disseminate findings on

advantages and disadvantages of each one.

www.nasa.gov

Battery PrognosticsBattery PrognosticsBattery PrognosticsBattery PrognosticsBhaskar Saha and Kai Goebel

Problem

Bhaskar Saha and Kai Goebel Prognostics Center of Excellence, NASA ARC

Problem

(AFRL, Artist's depiction)

Electric Propulsion Space Experiment (AFRL)

•Ammonia Arcjet onboard ARGOS

(1999)

•Gases released from electrolyte

Questions to be Answered

Can the current mission be completed?

• Given the health of the battery, is there enough charge left

for anticipated load profile (within allowable uncertainty

Beech A200 (Reg # N258AG)

•Gases released from electrolyte

decomposition resulted in a breach of

the battery case, releasing

superheated gas into the unit

•Onboard generators failed to activate as

starter was still engaged after ignition

(Courtesy: AFRL-PR-ED-TR-2001-0027)

for anticipated load profile (within allowable uncertainty

bounds)?

• Dominant metrics: state of charge (SOC), state of health

(SOH)

Can future missions be completed?

• Given the health of the battery, at what point can typical

future missions not be met?

(NASA/JPL, Artist's concept) Mars Global Surveyor

starter was still engaged after ignition

•Battery completely dischargedresulting in total electrical failure

disabling normal landing gear extension

capability

•Landing gear failed & the plane crashed (Courtesy: NTSB, ID #SEA00LA066)

•The MGS failed Nov 2006

future missions not be met?

• Dominant metrics: end of life (EOL), state of health (SOH)

end of charge ?

Mis

sio

n

Pro

file

Time (Short term)

Time (Long term)

SO

H

end of charge ?end of charge ?end of charge ?end of charge ?end of charge ?end of charge ?

Mis

sio

n

Pro

file

Mis

sio

n

Pro

file

Mis

sio

n

Pro

file

Mis

sio

n

Pro

file

Mis

sio

n

Pro

file

Mis

sio

n

Pro

file

Time (Short term)Time (Short term)Time (Short term)Time (Short term)Time (Short term)Time (Short term)

Time (Long term)Time (Long term)Time (Long term)Time (Long term)Time (Long term)Time (Long term)

SO

HS

OH

SO

HS

OH

SO

HS

OH

Approach

•The MGS failed Nov 2006

"We think that the failure was due to a

software load ... The radiator for the

battery pointed at the sun, the

temperature went up, and battery

failed...”John McNameeMars Exploration Program, NASA

Time (Long term)Time (Long term)Time (Long term)Time (Long term)Time (Long term)Time (Long term)Time (Long term)

GOAL: Develop a model that makes a prediction of

end-of-charge and end-of-life based on rapid

state of health (SOH) assessment

ApproachBattery Schematic

Lumped Parameter Model

Impedance = Resistance + Reactance

CDL

RCT RW

RE

CDL

RCT RW

RE+– +

–

–+

+–

e- e-II

Relevance Vector Machine

– State of the art in nonlinear

probabilistic regression

– Data driven learning

– Learn degradation mode

Particle Filter

RVM

Impedance = Resistance + Reactance

•The External Voltage (E) of a battery

is less than its Open Circuit Potential

(Eo) whenever it is in use.

•Losses are due to:

• Ro: Internal Resistance (IR drop),

• Rp: Polarization Resistance

DISCHARGE CHARGE

Electrochemical Impedance Spectroscopy (EIS)

•Carry out a frequency sweep

•Plot Capacitive (1/ωC) v/s Resistive (R)

component of the Reactance

•Response is different in presence of

passivation and corrosion, providing a

diagnostic for the health state of battery

– State of the art for nonlinear

non-Gaussian state estimation

– Uses model to predict and data

to correct prediction

– Sequential Monte-Carlo simulations

with Importance Sampling for

p-step ahead predictions

PF

• Rp: Polarization Resistance

• Rc: Concentration Polarization

•These losses tend to increase as the

current drawn from the battery

increases.

Concentration

polarization

IR drop

Activation polarization

Eo

Incre

asin

g V

olta

ge

E

diagnostic for the health state of battery

Re(Z)

Im(Z

)

Increasing Ohmic Resistance

0

0

Increasing Ohmic Resistance

Electrolyte Weakening

Feature

Extraction

RVM

Regression

Model

Identification

EIS

Data

Feature

Extraction

PF

Tracking

PF

EIS

Data

Results

Increasing Current I

Incre

asin

g V

olta

ge

Polarization Curve

Ohmic resistance (Ro) and polarization resistance (Rp) of a Li-ion battery gradually increase with aging

Re(Z)

Im(Z

)

Increasing Charge Transfer Resistance

0

0

Increasing Charge Transfer Resistance

Plate Sulphation

PF

Prediction

Impedance-

Capacity Mapping

Cause-Effect

Analysis

SOH SOC SOL

Results

0.015

0.02

0.025Particle Filter Output

E, R

CT (Ω

)

PF Tracking

PF Prediction

RE

Particle Filter Prediction α-λ Metric ( α = 0.2, λ = 0.5)

0 10 20 30 40 50 60 700

0.005

0.01

Time (weeks)

RE

Fault d

ecla

red

Measurements

RCT

0.75

0.8

0.85

0.9

0.95

1Particle Filter Prediction

C/1

ca

pa

city (

mA

h);

R

UL

pd

f (b

iase

d b

y 0

.7)

Real data

15

20

25

30

35

40

α-λ Metric ( α = 0.2, λ = 0.5)

RU

L

POC: Bhaskar Saha, (650)604-4379, bhaskar.saha@nasa.gov

0 10 20 30 40 50 60 700.6

0.65

0.7

0.75

Time (weeks)

C/1

ca

pa

city (

mA

h);

R

UL

pd

f (b

iase

d b

y 0

.7)

RUL threshold

RUL pdfs

30 35 40 45 50 55 60 650

5

10

Time Index ( i)