Project Award Number:
#0097438
Principal Investigator
First Name: Wesley
Middle Initial: W.
Last Name: Chu
Department: Computer Science Department
Institution:
Address line 1: UCLA Computer Science Department
City:
Zip Code: 90095-1596
Phone Number: 310-825-2047
Fax Number: 310-825-2273
Email: wwc@cs.ucla.edu
URL: http://www.cs.ucla.edu/~wwc/
Co-PI
First Name: Chih-Ming
Middle Initial:
Last Name: Ho
Department: Mechanical and Aerospace Engineering Department
Institution:
City:
Zip Code: 90095-1597
Phone Number: 310-825-9993
Fax Number: 310-206-2302
Email: chihming@ucla.edu
URL: http://ho.seas.ucla.edu/professor/
Collaborator
Daniel H. Whang: Civil &
Environmental Engineering Department, UCLA dwhang@seas.ucla.edu
MEMS sensors and
actuators Dynamic control Delta wing flight control
Temporal and spatial data
mining
This is an interdisciplinary project through the
departments of Computer Science and Mechanical Engineering. The aerodynamic
group has been focusing on data protocols, sub-baseline data, and the design of
the proposed experimental setup. During this period, we have completed the
automatic movement mechanisms simulating the transient conditions for the delta
wing aircraft in the wind tunnel. This consists of motion control hardware and
actuation mechanisms that will be mounted inside the wind tunnel.
In regards to data mining, we have focused on
developing algorithms for finding maximum frequent itemsets (MFI) to generate
targeted (attributes) association rules [ZCYK03], developing a framework for
the distributive aggregation of multidimensional objects [TC03] and the mining
of temporal spatial data sequences.
In the mining of temporal
spatial data sequences, the input/output variation may follow certain patterns.
We plan to extract a set of sequence patterns, which are formed by a series of
input/output pairs over a range of input values. Based on similarities among
the sequence patterns, we can cluster them into groups and organize them into a
cluster hierarchy via an inter-cluster error measure. Each node in the cluster
summarizes the input/output relationship of that group.
Mining temporal spatial patterns generated from
multiple sources (e.g. multiple sensors) is different from the mining of a
single time sequence. We need to capture the pattern generated from all the
sources from both temporal and spatial conditions. For the spatial dimension,
we use dynamic time warping (DTW) to match similar patterns in the multiple time
sequences. DTW [PCY00] may be used to stretch and shrink the time sequences to
increase their matching probability. For the temporal dimension, we use moving
windows on the temporal sequences to find patterns contained in the windows
that match with the patterns among the multiple time sequences [Oat02]. We plan
to use the aforementioned algorithm for mining the dynamic temporal spatial
data sequences from the wind tunnel experiments in the coming year.
[LCHFH 01] Z. Liu, Wesley. W. Chu, A. Huang, C. Folk,
C.M. Ho., "Mining Sequence Patterns from Wind Tunnel Experimental Data for
Flight Control", PAKDD, (2001), p. 270.
[HFHLC 01] A. Huang, C. Folk, C.-M. Ho, Z. Liu, Wesley
W. Chu, Y. Xu, Y.-C. Tai., "Gryphon M3 system: integration of MEMS for
flight control.", In MEMS Components and Applications for Industry,
Automobiles, Aerospace, and Communication, Proceedings of SPIE (The Int'l
Society for Optical Engineering), vol. 4559, (2001), p. 85.
[ZCJC 02] Q. Zou, Wesley. W. Chu, D.Johnson, and
H.Chiu, "A pattern Decomposition Algorithm for Finding all Frequent
patterns", Journal of Knowledge and Information System, vol. 1, (2002), p.
101.
[ZCL 02] Q.
[ZCCK 03] Q.
[TC 03] Meng-Feng Tsai and
Wesley W. Chu, "A Multidimensional Aggregation Object (MAO) Framework for
Computing Distributive Aggregations", 5th International Conference on Data
Warehousing and Knowledge Discovery, DaWaK 2003,September 3-5, 2003 Prague,
Czeck republic, (2003), Accepted.
We are developing techniques to automatically read information from massive distributed sensors, transfer the information to actuation rules and select the most applicable rule to derive the actuation schema. This process can be used in other real-time control systems, such as in the mining of medical clinical data to determine the success of certain surgical procedures based on the patient’s demographics, and pre and post surgery conditions. It can also be used for mining in Civil Engineering to discover the relationship of soil characteristics with settlement for certain seismic input data.
The goal of our research is to discover non-linear
relationships between the distributed sensor input and actuation schema output
by using data mining techniques. We have designed a set of experiments to
collect the data for dynamic system behaviors, and are extending the mining
algorithms to handle massive amounts of multivariate training data. We plan to
develop a methodology to generate temporal and spatial rules, as well as a
strategy to automatically select actuation rules for real time system control.
Data mining, dynamic control of macro-scale machine,
distributed sensors
[LCHFH 01] Z. Liu, Wesley W. Chu, A. Huang, C. Folk,
C.M. Ho, Mining Sequence Patterns from Wind Tunnel Experimental Data for
Flight Control. PAKDD 2001: 270-281.
[HT 98] C.M. Ho, Y.C. Tai, Micro-Electro-Mechanical-Systems
(MEMS) and Fluid Flows Annual Review of Fluid Mechanics. 1998
30:579-612.
[PCY 00] S. Park, Wesley W. Chu, J. Yoon, C. Hsu. Efficient Search for Similar Subsequences of
Different Lengths in Sequence Database,
16th International Conference on Data Engineering, San Diego
2000.
[Oat 02] Tim Oates. PERUSE: an unsupervised algorithm for finding recurring patterns in
time series, IEEE International Conference on Data Mining (ICDM'02),
Maebashi City, Japan 2002.
There are several temporal spatial data mining
projects (e.g. Christos Faloutsos, Lin Liu and Christopher Lee) in the IDM.
They can all be leveraged on each other’s research.
Data mining group website: http://www.cobase.cs.ucla.edu
Mechanical and Aerospace Engineering website: http://ho.seas.ucla.edu/