Introduction to Fault Diagnosis

and Isolation(FDI)

By
Hariharan Kannan
Fault Detection & Isolation
– An Overview
 Goal of FDI:
 To meet the requirements of reliability, Safety
and low cost operation for today’s engineering
systems.
 To accurately isolate problems and make
control changes to bring system behavior back
to desired operating ranges or at least a safe
mode of operation.
Diagnosis- The Bigger Picture
Idea of Model Based Diagnosis
 A set of variables called observations
are measured.
 Residuals r, are computed as the
difference between the observations y,
and the predicted normal behavior ŷ.
 Non-zero residuals imply that there is a
fault in the system and this triggers the
diagnosis algorithm.
Types of Faults
 Incipient Faults
 Occur slowly over time
 Linked to wear and tear of components and drift in
control parameters.
 Intermittent Faults
 Present only for very short periods in time
 Could have disastrous consequences in time
 Abrupt Faults
 Dramatic and persistent
 Cause significant deviations from steady state
operations-Transients
Steps in Fault Diagnosis
 Fault Detection- signaled by a non zero
residual
 Fault Isolation
 Qualitative Fault Isolation
 Hypothesis Generation- Back Propagation Algorithm
 Generating Fault Signatures- Forward propagation
Algorithm
 Progressive Monitoring
 Quantitative fault Isolation
 Parameter Estimation
Modeling For Diagnosis
 The models should describe both normal and
faulty system behavior.
 The model should generate dynamic behavior
under faulty conditions, so fault transients
can be predicted by the model.
 The model should incorporate sufficient
behavioral details so that deviations in
observed variables can be mapped back to
system components and parameters.
Temporal Causal Graph
 Dynamic Characteristics of system behavior
derived from the bond graph are represented
as a temporal causal graph
 Algorithms for monitoring, fault isolation and
prediction are based on this representation.
 It is derived from the bond graph model.
 Incorporates cause effect relationship among
the power variables shown in the bond graph.
 Component parameters and temporal
information are added to individual causal
edges.
Transient Analysis
Our approach analyze measurements
individually.
Transient Response of a signal (can be
approximated by Taylor series of order k)
y(t) = y(t0) + y'(t0)(t- t0)/ 1! +
y''(t0)(t- t0)2/ 2! + …… +
y(k)(t0)(t- t0)k/ k! + Rk(t),
where Rk(t) is the remainder term based on
y(k+1)(t).
Signal transient due to a fault at t0 can be
expressed as discontinuous magnitude
change, y(t0), plus first and higher order
derivative changes, y'(t0), y''(t0), ….., y(k)(t0).
2 Tank System- Example
 Derivation of TCG from Bond Graph

•Effort and flow variables are vertices

•Relation between variables as directed edges
•=implies that two variables associated with the
edge take on equal values, 1 implies direct
proportionality,-1 implies inverse proportionality.
•Edge associated with component represents the
component’s constituent relation.
Backward Propagation
+ Above Normal
- Below Normal
0 Normal
 Fault Prediction-Establish Signature for system
variables

The prediction module uses the system model to compute the dynamic, transient
behavior of the observed variables and the eventual steady state behavior of the
system under fault conditions.
Future behavior is expressed in qualitative terms:magnitude(0th order), slope(1st order)
The algorithm used propagates the effects of a hypothesized fault to measure a
qualitative value for all measured system variables.
Forward propagation along temporal edges implies an integral effect, the cause
variable affects the derivative of the effect variable.

Algorithm stops when signature of sufficient order is generated.

Order depends on set of chosen measurement variables & desired level of “diagnosability”.
 Monitoring Implementation
 Progressive Monitoring to track system
dynamics after failure
 Higher-order derivatives as a predictor of
future behavior (justified by Taylor’s series)
 Activated when there is a discrepancy
between predicted and observed value.
Diagnosability of a system
 Diagnosability is a function of the number of possible
faults that can be uniquely identified by a fault
isolation system.
 Completely Diagnosable system- A system which can
uniquely isolate all possible hypothesized faults.
 Depends on selected observation set and chosen
order of their signature.
 Consideration of higher order variable effects is likely
to result in greater diagnosability.
 same diagnosabilty can be achieved- by considering
higher order signatures but smaller number of total
observations or using a large number of observations
with lower order signatures.
 Two Tank System
Response to Faults
f5:
Faults:
Rb 2
Rb1, Rb2, R12

Discontinuity

Faults: C1, C2

Discontinuity

It seems one measurement is enough but not really….

(especially if analysis is qualitative)

& discontinuities not reliably detected...

Progressive Monitoring
 Monitoring involves comparing predicted signatures of the
hypothesized faults to actual measurements as they change
dynamically.
 Choice of monitoring time step is vital-neither too low or too
small
 Transient characteristics at the time of failure tend to change
over time as other phenomena in the system affect the
measured variables.
 Ex: A fault may have no effect on initial magnitude(0th order) of
a variable but it may affect its 1st derivative(slope), predicting
that it will be above normal.
 Therefore immediately after fault occurs, variable value will be
observed to be normal , but as time progresses, the derivative
effect will cause the variable to go above normal.
 This notion of employing higher order derivatives – Progressive
Monitoring.
Progressive Monitoring..Contd
Progressive Monitoring-Contd..
Limitations of Purely
Qualitative Schemes
 For the case where a signal does not undergo abrupt
change, higher order derivatives beyond the first
non-zero derivative have no discriminatory power.
 Consider 2 faults with second order signatures-
(0,+,+) and (0,+,-) for a particular measurement.
 signal shows no discontinuous change at point of
failure, matches (+,+,.)
 Even if signal slope is measured to be -, the (+,+,+)
cant be eliminated as a higher order derivative not
captured in the second order signature could be -. So
faults cant be isolated.
 Solution- Quantitative diagnosis.
 Parameter Estimation

Consider system defined by

C1- and R12 + are the fault

candidates.

Estimate the parameter

by substituting the
nominal values values
for the variables in the I-
O model of the system.
If the error e converges to 0,
for a particular parameter,
that parameter is the fault