Documente Academic
Documente Profesional
Documente Cultură
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
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.
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