|Ph.D Student||More Karin|
|Subject||Quality and Reliability Characterization of Multi-Parametric|
Physical Measurements Data Fusion
|Department||Department of Quality Assurance and Reliability||Supervisor||Professor Dov Ingman|
|Full Thesis text|
The methodology of the presented work is based on applications of the principles of Quality Assurance and Reliability (QA&R) to the data fusion process. The research suggests a set of QA&R indices for the evaluation of the extracted information i.e. the process performance assessment. The applications in the presented research are focused on the physical measurements in non-destructive testing.
The definition of relevant/useful information regarding the critical measurand was introduced and set as a goal to be achieved by the fusion to properly describe the object under consideration. Any data fusion process design should include this element and means to optimally fulfill the task. A global loss function is introduced as a weighted combination of the partial loss functions for raw data from each information sources which corresponds to the well-known Taguchi quality methodology in "mining" the relevant information. The weights in loss function cannot be predetermined and have to be found in the course of the fusion process (in principle) for each experiment. The research shows that optimal information sources should be of mutual negative correlation nature, i.e. the correlation coefficient of (-1) minimizes the previously introduced loss function.
The literature data of heating value of different natural gases served for demonstrating of combined classification and heating value assessment fusion. This example applies quantitative criteria for the fusion effectiveness and demonstrates the advantages of the approach. The estimator of reliability characteristic of the process is also suggested. An additional example shows the main difference in suggested and common fusion philosophy in the field of NDT.
Preliminary verification of the proposed fusion technique for NDT was performed by the benchmark simulation. The “deviation” of the restored parameters from the preset values in the simulation served as an index for the evaluation of the fusion success.
The two NDT experiments were performed for a case of three known defects. The instrument (blurring) functions of the measurement systems were restored. The fusion results were closed to the actual characteristics then those achieved by each of these methods separately.
The work presents the principles of the suggested data fusion approach, its advantages with the help of its implementation to the cases of the performed experiments, as well as, to the experimental data mined from the professional literature sources.