PREPROCESSING TECHNIQUES FOR PROGNOSTIC FRAMEWORK INPUTS
The core concept of the Prognostic Framework requires that it be flexible enough to accept inputs from a variety of different sources including prediction mechanisms, diagnostic test information, sensor data, etc. Additionally, user inputs in the form of variables regarding stresses and environmental conditions for a mission scenario will need to be incorporated. As a result, the development tools software design must include a generic methodology for not only accepting data inputs, but also preprocessing them for use inside the run-time framework. The addition of this kind of functionality will increase the framework’s ability to handle many types of input and perform supplemental prognostics analysis.
Two
types of preprocessing techniques will
be used in various combinations to achieve a generic approach capable of
handling all application instances.
The first type is a
mathematical expression. The
basic operations of addition, subtraction, multiplication, and division
will be supported. Other
types of function operations will also be added such as logarithms and
trending over time. Software implementation of this functionality will be
flexible so that other functions and operations can be added as needed.
Mathematical expressions offer better support for all types of
inputs and will greatly enhance the framework’s ability to incorporate
raw sensor data and user mission scenario variables.
The second type is a Boolean expression of ranges used to filter out noise within a system. Filters will use original raw input as well as the outputs of mathematical expressions to determine whether various measurement data should be used as an input to the framework matrix column. (A matrix column is test/measurement data that defines fault coverage in the fault/symptom matrix schema.) During runtime, the software will perform mathematical calculations first, apply filters second, and finally pass verified and preprocessed information into the matrix as a column. From a software point of view, this approach breaks each framework input into three data entities: raw data, refined data and column data.
Raw Data à [Math Calculations] à
Refined Data à
[Filters] à
Column Data
Raw data is the actual data
passed into the framework by the client program. It is described in terms
of how and where to retrieve the data during runtime, for example, index 4
into an array of floating point data elements.
Mathematical calculations use raw data.
Refined data can be either the
result of mathematical calculations or raw data passed through without
modification. It is an
intermediate format used by filters. Filters use refined data to indicate
whether information (column data) should be used inside the framework
matrix. They basically turn
on and off framework matrix columns.
Column data is the information
used inside the framework matrix. It
consists of a data value and an optional confidence level. Both are
described in terms of refined data. Therefore they can be derived via
mathematical calculations or unprocessed raw data passed through without
preprocessing. Column data
may or may not be used as a matrix column value based on the result of
existing filters.
The run-time sequence of raw
data, math calculations, refined data, filters, column data provides
optimum flexibility for applications.
This combination should provide support for just about any
application problem. Because mathematical calculations are performed first
and raw data inputs can be used in calculations as well as being passed
through to the refined data step unprocessed, the right kind of data
required for any application task is available at any level: raw, refined,
and column. Processed and unprocessed data is available for calculations,
filters, and column information.
The
new design concepts applied here include:
1.
Mathematical calculations will be used as a method of preprocessing
framework inputs.
2.
Filters will be used as a method of filtering out noise within a
system and between framework inputs.
3.
Each framework input can be transformed from raw data into refined
data and finally into column data.
4. Framework inputs do not have to pass through to the framework
matrix.
5. Each framework input may be used several times, for example, passed down to refined data and used in a mathematical calculation.
6. A matrix column has two pieces of data associated with it:: a
source data value and a confidence level.
Prognostics Framework development tool suite
Prognostics Framework Run-Time Software: Health Management System
Preprocessing Techniques for Prognostic Framework Inputs
Download Prognostics Framework Health Monitoring System Demonstration