|
|
 |
Automation
solutions often fail to make efficient use of the available
data. This usually results from the inherent complexity
of the data. Data analysis and pattern recognition tools
aim to help understand the data, discover correlations,
separate useful from the irrelevant, extract information
and respond to this information. Whether it is waveforms,
images, data taken from sensors of a monitoring system,
user-responses from an interface, quality control, business,
or raw data taken from files or databases, these methods
can help understand and make more "intelligent"
use of it.
While the field of application
for such tools is very broad, the algorithms used vary,
and include statistical, decision surface based, hierarchical,
agglomerative, neural networks etc. The methods usually
apply inherent strategies to discover particular types
of structures, optimize some given separation criteria,
discover a decision surface of a given type etc, making
a given method more effective on some data than on other.
Therefore a significant factor in the success of a pattern
recognition-based solution is the careful selection, fine
tuning, or even customization of the algorithm(s) to be
used.
|
|
|
| Below
is a very brief outline of the steps usually taken:
-
Define problem, obtain representative data.
- Preprocess, filter, and extract features.
- Use an unsupervised clustering approach to discover
correlations and structures in the (typically multivariate)
data, clusters that may be of particular significance
to the given problem or application. Evaluate the clustering,
repeating the process if necessary.
- Perform supervised training and testing cycles, teaching
a pattern recognition algorithm to identify the structures
of interest. Evaluate performance on new data, repeating
the entire process if necessary.
- Use the trained, stable method in the final application.
We
can provide:
-
Data analysis services.
-
Custom software development for analysis, control, monitoring,
alarms etc, using C++, C#, or Matlab, for Windows, Linux,
or the Web.
|
|
 |
 |
 |
|
|
