| 1) Develop a
comprehensive approach to represent faulty processes using multiple models.
|
|
The basic premise is the ability to model the fault process as a
Markov chain with the fault determining a state transition. The team evolved
from a fixed set of models to a time varying, hierarchical structure of models
that can be activated as necessary.
|
| Top of Page
|
|
|
| 2) Develop
multiple-model algorithms with a variable structure that can detect and
identify faults.
|
|
The approach provides early detection and identification of
deteriorated situations. In particular the algorithm based on Interacting
Multiple Models with Maximum Likelihood Decision rules (IM3L) appears a
extremely efficient for detecting sensor faults and is being tested for the
detection of actuator faults.
|
| Top of Page
|
|
|
| 3) Establish a
commonality with target tracking.
|
|
The team determined that the multiple model approach is an
effective tool to solve the problem of detecting an evasive target in a random
environment.
|
| Top of Page
|
|
|
| 4) Develop a
unified representation for the various sensor linear data fusion environments.
|
|
The team was able to present in a common framework such
varied approaches as centralized fusion and distributed fusion with partial
estimation. This unified framework permitted us to develop a general theory to
determine optimal fusion rules.
|
| Top of Page
|
|
|
| 5) Use temporal
redundancy to develop detection algorithms based on Sequential Probability
Ratio Tests.
|
|
The original Wald’s SPRT was improved and the team developed
tests with guaranteed bound on the decision time. The tests have been applied
to target tracking and are being evaluated for fault detection.
|
| Top of Page
|
|
|
| 6) Developed
enhanced fusion rules.
|
|
This is a derived result. They are using Dempster-Schafer theory of
evidence to develop enhanced fusion rules.
|
| Top of Page
|
|
|
| 7) Establish
commonality with network traffic.
|
|
This is a derived result. They have established that the multiple
model approach provides an effective tool to represent congestion in Internet
traffic and the team have been able to generate estimates of delay of service.
|
| Top of Page
|
|
|
| 8) Contributions to
particle filtering theory and recursive estimation.
|
|
The team has developed general recursive estimation
algorithms that can even account for data received out-of-sequence.
|
| Top of Page
|
|
|
| 9) Extend the
concept of pseudo power signatures.
|
|
The team verified the applicability of the pseudo power
signatures for the detection of faults, and developed the subspace
signatures and evaluated the concept with models of an F16 and spectral energy
distribution.
|
| Top of Page
|
|
|
| 10) Use the energy
density function derived from the continuous wavelet transform.
|
|
The team has implemented efficient algorithms to compute the
continuous wavelet tansform and its scalogram. They are developing
enhanced fault signatures by performing intensive simulation studies to
evaluate sensitivity to various faults for the twelve state variables of a B747
model at various resolution scales. The initial results are encouraging.
|
| Top of Page
|
|
|
| 11) The dual-loop
configuration has been expanded.
|
|
To incorporate multiple loops for creating structured residuals
that can be used to isolate faults at various nodes of the closed-loop system.
The approach incorporates real-time system identification in the absence of a
priori-available models. The team has explored system identification
methods for both linear and nonlinear systems; the linear systems are
identified using AutoRegressive Moving Average (ARMA) models that are estimated
in real time by a modified version of the Kalman filter. Nonlinear systems are
identified using neural networks. For this case, the team has developed a new
architecture for neural nets, called the ANARMA (Additive Nonlinear
AutoRegressive Moving Average) structure. The advantage of this model is that
it can be realized in classical state space form, allowing us to use various
control design tools that already exists in the literature. For neural
networks, the team developed a fast on-line training technique.
|
| Top of Page
|
|
|
| 12) Use
sensitivity analysis theory to investigate the quality of residuals.
|
|
Obtained at various nodes of the closed-loop system (MS thesis of Dilip
Vutukuru , ULL Dec. 2003). The team demonstrates that in the presence
of smart controllers, the system output is not a sensitive indicator of faults;
rather, residuals generated at the controller output provide the fastest
detection, while other residuals can assist in fault isolation; the results,
both theoretical and simulation experiments.
|
| Top of Page
|
|
|
| 13) Develop fault
indicators for noisy systems with slowly developing faults.
|
|
The team has shown that using AutoRegressive modeling of
the residuals at the controller output node, fast detection (early warning) of
faults can be achieved much before the fault turns into a failure. This
approach is independent of the actual distribution of the raw residuals, and
significantly reduces the rate of false alarms: the team obtained excellent
results from simulation study of aircraft and also from experimental data
supplied to us from closed-loop fault experiments at NASA Langley Research
Center. The LaRC data resulted in severely non-Gaussian residuals for the
throttle command, which implied a higher-than anticipated level of either false
alarms or missed detections; their approach of AR-modeling of the raw residuals
essentially discards the distribution, and thus, is a much better indicator of
faults than the raw residuals.
|
| Top of Page
|
|
|
| 14) The team
expanded their work to include observer-based fault estimation and
accommodation.
|
|
With this approach it is possible to implement fault-tolerant
controllers. This result is also applicable to nonlinear models.
|
| Top of Page
|
|
|
| 15) The
team developed a “composite fault indicator.”
|
|
In an effort to create a simple fault indicator capable of detecting plant and
controller faults under a general closed-loop configuration. The composite
indicator is a continuous blend of the residuals at the system output and
controller output. The main goal of Fault Tolerant Control is to design control
structures that can perform safely even when the aircraft undergoes sudden
changes in its characteristics. The main focus was in the application of the
Generalized Internal Model Control approach. The results can be summarized
as follows.
|
| Top of Page
|
|
|
| 16) Verification
of the Generalized Internal Model Control.
|
|
The team performed simulation and experimental studies of
the GIMC paradigm.
|
| Top of Page
|
|
|
| 17) New design
tools for decentralized H-infinity control design.
|
|
The team continued the development of alternative schemes
for fault tolerant robust control.
|
| Top of Page
|
|
|
| 18) Develop a
probabilistic risk analysis and control design approach.
|
|
The team developed efficient algorithms to compute a
robustness degradation function and effectively design controllers that
guarantee stability with a certain probability. The team established,
for the first time, that fault tolerant control based on worst-case design
could have a higher probability of failure than a design based on probabilistic
criteria. The apparent paradox arises due to imperfect modeling of
uncertainties.
|
| Top of Page
|
|
|
| 19) Develop
efficient space time codes for wireless systems.
|
|
This is a derived result from the team's work in probabilistic risk analysis
and it looks so promising that the team is preparing a patent disclosure on the
method. The team have been able to propose a new paradigm for the
design of unitary code constellations for differential space-time modulation
for multiple-antenna systems over Rayleigh-fading channels with unknown fading
coefficients. Extensive simulations show that the new codes have significant
better performance than existing codes. They have compared the performance
of their codes with differential detection schemes using orthogonal design,
Cayley differential codes, cyclic group codes and FPF codes and for the same
bit error rate, their codes allow smaller SNR by as much as 10dB.
|
| Top of Page
|
|
|
| 20) The
team is developing a complete fault tolerant control approach for
over-instrumented systems.
|
|
This class of systems is almost non existent in the process
control industry but covers well most of the commercial aircrafts. The class is
characterized by having enough redundant sensors so that the system state can
be actually measured. So far the team has explored some of the properties of
linear models and the results are exciting. For example, the team has derived a
simple detector for conventional actuator faults and they can detect and
identify multiple faults. Robust, fault tolerant control, capable of
stabilizing the system for all partial actuator faults, is a simple state
variable feedback. The team has derived and tested switching
strategies that maintain performance under partial faults by switching state
variable feedbacks. The strategy does not have the limitations of dynamic
switching and does not introduce uncontrollable transients.
|
| Top of Page
|
