HBG Model

1
Hybrid System Modeling, Control, and Fault
Diagnosis on a Three Tank Testbed
Simulink Model
Validation
HBG Model
Summary

• Models were validated by comparing Simulink model
  results and data collected from the testbed over
  identical control sequences.
• Average absolute error was calculated to validate
  the models accuracy. For one such experiment

For this project, we studied modeling,
simulation, control and fault diagnosis of a
hybrid system using a three tank testbed. Models
were constructed using hybrid bond graphs (HBG)
and implemented using Matlab/Simulink. Using the
model, we created a set-point controller for the
testbed using a limited-lookahead approach. The
controller's behavior was verified first on the
simulation and then on the testbed. Finally, we
integrated the testbed with the Fault Adaptive
Control Technology (FACT) Diagnoser to detect and
identify system faults. The results show that the
model-based controller effectively maintains the
system set-points, and FACT Diagnoser
successfully detects and identifies faults in the
system.
Three Tank Testbed
Limited Lookahead Control1

• Hybrid system consisting of three tanks.
• Eight valves control filling, draining, and
  transferring between the three tanks.
• A variable speed pump can fill tanks one and two.

• Developed a set-point controller utilizing a
  limited lookahead approach.
• Constantly predicts the possible future states of
  the system over a prediction horizon.
• Minimizes a cost function or maximizes a utility
  function when choosing best control action from
  possibilities.
• Results show the controller can successfully
  maintain the given set points using the limited
  lookahead approach.

For this experiment the controllers objective is
to maintain the water in tanks 1, 2 and 3 at
heights of 40 cm, 30 cm and 15 cm respectively.
The average shows how the controller was able to
sustain the water at the height specified once
transient was complete. The various valve
settings (shown to the right) show the pump speed
and the enabled valves over the course of the
experiment.

• Four nodes provide distributed monitoring and
  control for the system.
• Each node provides an HTTP-based API for
  commands and queries.
• The nodes are interfaced with the systems
  transducers using the IEEE 1451.2 standard.

Fault Diagnosis2
Tracking Tank 1 during Transfer

• FACT Diagnoser uses an annotated hybrid bond
  graph model to detect faults manifesting as
  unexpected system behavior.
• A hybrid observer tracks the system and detects
  faults. The faulty component is identified, and
  its quantitative change is estimated and applied
  to future tracking.
• Integrated Diagnoser with system and experimented
  with several faults.
• Tracking results before and after faults show
  successful diagnosis of the fault and estimation
  of the faulty parameter.

For this experiment, a fault was introduced by
increasing the resistance on the transfer pipe
between tanks 1 and 3. The Diagnoser estimated
the parameter well enough for normal operation
after the second attempt. The graph shows that
the observer (blue) approaches the actual data
(red) after the second estimation.
Data Collection
Future Work
References

• Wrote multithreaded libraries and applications
  for collecting, processing, distributing, and
  logging data.
• Framework created to simplify controller design
  on the system.
• Data distributed over UDP multicast to multiple
  consumers.
• Data collected, processed, and distributed at
  configurable rate as high as 3Hz.

• Incorporate other failure modes into experiments
  including the possibility of intermittent faults.
• Integrate controller with fault diagnosis to
  explicitly adapt control algorithms in response
  to system faults.

• J. Wu, G. Biswas, S. Abdelwahed, and E. Manders,
  A Hybrid Control System Design and
  Implementation for a Three Tank Testbed, in
  Proc. IEEE Conf. Contr. Applications, Toronto,
  Canada, Aug. 2005, pp. 645-650.
• P. J. Mosterman and G. Biswas, Diagnosis of
  Continuous Valued Systems in Transient Operating
  Regions, IEEE Trans. on Systems, Man, and
  Cybernetics, 29, 9, pp. 554-565, November, 1999.

SIPHER Students Nathan Allotey, Brian
Turnbull Graduate Student Advisor Jian Wu