One of my the rare pictures with my chaotic bubbling column. Notice the period-2 formation at the top of the column.

This was how my graduate research at ORNL and University of Tennessee-Knoxville looked like. I havent had time to update this page since after graduating.  First of all, thank you for stopping by my research page. I encourage you to read through. My odyssey began with a core chemical engineering research topic, and with successive bifurcations through chaos, non-linear time series analysis, multi-variate statistical process control and finally applied artificial intelligence. On the way I picked up skills in automation, data acquisition and real-time data processing, OLAP, SQL and a plethora of hands-on skills which have really enabled me to develop a creative multi-disciplinary approach to problem solving. May I also add that I was also required to be proficient in advanced topics in process dynamics and control for the backbone of my research topic!  I find myself at the threshold of a brilliant career, in which I look forward to apply my engineering and advanced control knowlege, integrate it with software technology and finally envision innovative solutions to be developed. My project involves working at Oak Ridge National Laboratory. I am presently working with Dr.David DePaoli with the Chemcial Technology Division. At The University of Tennessee I am working with Prof.Duane D. Bruns with the Chaos Research Group. I shall be happy to collaborate on any of my research topics. Please send me an email if you need copies of any publications.  My resume may be accessed from here. Chapters in my research sojourn 1. MS thesis topic 2. Automation of experiments 3. Data analysis tools 4. Regression models 5. Neural network models 6. Chaos control

My Masters thesis is titled, "Real time regime identification and non-linear control of chaotic bubbles under the influence of electrostatic fields".  We have identified electrostatic fields as an additional bifurcation variable and this gives a whole dimension to the control of chaotic bubbling (which has not been successfully reported, as yet). We shall be attempting a control strategy, which involves multi-variate control using both flow and voltage as the two variables.  The results thus far have proved to be extremely promising. If successful, this will prove to be a landmark paper in the field of chaos and potentially could be published in a science journal instead of engineering. 
[top]

Automation of experiments…

If done the conventional way, this project was extremely time-consuming. Each run took approximately 14 days to complete, and then data analysis took another 2 weeks. LabView was used to automate the entire experimental set-up and reduced the duration of each run to 4 days. Now in the same time period, more runs could be carried out, which meant huge volumes of data. Using MATLAB and data compression techniques, automated routines were generated to preprocess and compress data from each run by a factor of 4, reducing the file size from about 0.8GB to just 140 MB.. 

.

Labview automated DAQ screen shot [top]

Data analysis tools…

With enormous amounts of data on hand, efficient data analysis tools were required to analyze each data set. Those were created in MATLAB using the GUI utility extensively to create a toolbox for chaotic and non-linear time series analysis. This included file-handling capabilities and used a database to store all the required runs in a systematic orderly manner. Without the toolbox, several aspects of the data, which could only be observed with large sets of data, would have been lost to study. The data visualization tools often provided the Aha! needed for further research.  This got my first publication out in just 1-½ months after the experiment was started. This also has gotten me interested in OLAP and database management.

Finally, a primer to the bubble toolbox is out. Click here..
 

The Bubble package for data visualization & analysis [top]

Regression models…

To further improve the efficiency of my experimental results, a real time method was needed to identify the regime of the bubbling. This was where MSPC came in. I took a course in advanced monitoring and diagnostic techniques in the summer ’99. Using the data analysis techniques used in that course, a regression model was developed. Non-linear PLS with neural networks was found to be the most successful method. But the models’ success deteriorated as the regime went from period one into chaos, i.e. as the non-linearity increased. Also, the generalization properties were poor. This prompted research into neural network models.
[top]

Neural network models…

Because of only one measured variable, time embedding of the time series by delaying the signal was done for extracting information about other system variables. PCA carried out and this was then fed to a neural network, which identified the periodicity of bubbling. This neural network model allows the operator to know the exact time when the fault, (change in bubbling periodicity occurs). Now, we can explore every regime to its exact limits and establish the critical boundaries that will be needed for designing a controller for controlling chaos.

At the same time, a self-adapting neural network was developed based on a Kohonen map. The highlight of this was that the model was self-learning and can be placed online with one clean training data set. It would then pick up any new patterns in the bubbling and remember them for future occurences. 

Though this started with the aim of improving efficiency for the experimental runs, this has now taken the form of a paper, which will soon be ready for publication.
[top]

Controlling chaos...

Currently I am exploring avenues on fuzzy logic, pattern recognition and data mining. Coupled with this I am also trying to market myself to potential employers. So if you happen to know of somebody who wants to hire a dynamic, enthusiastic and dedicated engineer, please click here. If you are looking for someone like me, all the better! 
 [top]