2. Overview
• Finished secondary school in
Canada
• B.Sc. in Applied Mathematics
from SFU – Vancouver, Canada
• Worked in:
– Process control
– Fuel cell technology
– Diesel truck technology
3. Universal Dynamics
• Worked with an industrial controller software -
Brainwave
• Expanded single variable into multivariable
system
• The system was:
– Model based
– Adaptive
– Predictive
– Able to handle processes with time delays
4. Brainwave
• Time delay => poor
PID (Proportional
integral derivative)
control
• Brainwave uses a
model
• Based on Fourier-
like Laguerre series
5. Laguerre Series
Different rates of decay
Sinusoids combined with
Exponentials which decay to
zero
Each model is built by
Applying a weight function
And adding the coefficients
8. Brainwave MIMO
• Originally Brainwave was a SISO (single
input single output) controller
• My work was to expand it to MIMO
(multiple input multiple output)
9. A MIMO System
Output 1 Output 2 Output 3
magnitude magnitude
time
Input 1
time time
magnitude
magnitude magnitude time
Input 2
time time
magnitude
time magnitude magnitude
Input 3
magnitude time time
10. A MIMO System
Output 1 Output 2 Output 3
magnitude magnitude
time
Each input affects
Input 1
All the outputs
time time
magnitude
magnitude magnitude time
Input 2
time time
magnitude
time magnitude magnitude
Input 3
magnitude time time
11. A MIMO System
Output 1 Output 2 Output 3
magnitude magnitude
time
Input 1
Each output
Is affected by magnitude
time time
All the inputs magnitude magnitude time
Input 2
time time
magnitude
time magnitude magnitude
Input 3
magnitude time time
12. MIMO – how it works
• Not easy to control a MIMO system
• Cost function with a cost matrix is
implemented
• Importance of each input to each output is
specified
Channels 1-1 and 3-2: most important
10 1 5
2 5 3
Channels 1-2 and 3-1: least important
1 10 6
13. Brainwave – technical details
• Based on Laguerre series
• Recursive least squares learning algorithm
• Written in C and C++
• Doubly linked list class to store sparse
matrices
• Prototyped in Matlab and Simulink
14. Ballard Power Systems
• World leader in fuel cell research
• Proton Exchange Membrane (PEM) fuel
cells
• Automotive applications – Mercedes,
Daimler-Chrysler
• Stationary applications – Ibarra – Japanese
national energy company
15. Fuel Cell
• A fuel cell is a battery that
runs on hydrogen
• Hydrogen is split into
protons and electrons
• Protons travel through
PEM
• Electrons travel through
the circuit
• Both combine with
oxygen and form water
vapour
16. Fuel Cell Stack
• Cells are combined in series to produce more
power
17. Stack Cooling in a Vehicle
• Freeze start-up analysis
• Stack thermal management
• Cooling system optimization
20. Vehicle Startup
• The higher the
temperature:
– The better performance
– The more power
produced
– The more heat to reject
– The faster the coolant
flow
– The more power sinks
needed
21. NxtGen Emission Controls
• Technology firm designing a Syngas
Generator
• More strict environmental laws
• Diesel trucks need to reduce emissions
• Soot filter management
• Filter cleaning with Syngas
22. Exhaust System
Once the DPF (Diesel
Particulate Filter) fills
with soot, Syngas
generator is turned on
producing Syngas
which reacts with and
burns the soot in the
filter.
23. Exhaust System Model
Syngas Generator Diesel Particulate
Diesel Oxidation
Module Piping Filter Module
Catalyst Module
25. Diesel Oxidation Catalyst
Module
Heat transfer
Syngas oxidation model
Gas mixer
module
Gas flow pressure
drop and redistribution
26. Diesel Particulate Filter Module
Heat transfer
Soot management model
module
Gas flow pressure
drop and redistribution
27. DPF Cleaning
• DPF is divided into 4
segments
Segment 1 Segment 2 • During cleaning only one
segment receives Syngas
• This segment:
Segment 4 Segment 3 – Heats up
– Has its soot burned
– Receives less total gas flow
(due to pressure balance)
28. Gas Temperature/flow
Higher temperature leads to a reduction in
Gas flow in a given segment in order to
Balance out the pressure across the DPF
29. Soot Burning
As Syngas is directed into
each segment, its soot loading
goes down
30. Programming details
• Each module of the system was
programmed in C
• The modules were compiled and put under
the mask of Simulink s-functions
• All the modules were connected in
Simulink
• The system was run using Matlab and
Simulink
31. Skills Summary
• Ability to translate physical phenomena into
mathematical language and computer algorithms
• Programming in C and C++
• Differential equations programming
• Excellent knowledge of Matlab and Simulink
• Knowledge of:
– Process control
– Fuel cells
– Chemical engineering
– Signal processing