Pattern recognition using correlation analysis (Cij) method is useful for non-destructive testing of physical objects, including pipes. An evaluation of the technique based on Computer Simulation Technology (CST) models has demonstrated the advantages of using the technique to detect and classify pipe wall thinning (PWT) in pipes. Given enough increments, the technique can be refined to detect any possible combination of PWT attributes. For this research 71 different simulations were modeled for purposes of calibration of the system, based on five varied properties of the modeled PWT instances. These properties include: location (29 simulations based on distance from origin and two lengths of PWT, for a total of 58 simulations), width (standardized at 25.4mm), depth (four simulations as radius of PWT at 78.74mm, 81.28mm, 83.82mm, and 86.36mm), length (four simulations as percentage of circumference: 25%, 50%, 75% and 100% circumferential PWT) and type of defect (five simulations based on five discrete profiles).
Microwaves were simulated from port 1 and port 2, with a sweeping frequency range (0.5 GHz bandwidth), analyzed as S11 and S21 for measuring and calibrating the response to the standards. The resulting waveforms became the standard patterns against which 11 unknown simulations were compared, sometimes using S11 waveforms for comparison, and at other times S21.
The correlation analysis technique was able to distinguish parameters for the unknown test cases. The technique is able to determine the correlation between the resonance frequency peak (RFP) and waveform for an unknown case, and those of nearby calibration models, via pattern recognition. For example, 0.847 and 0.872 correlations to two standard patterns for an unknown RFP which appears midway between two standard RFPs, produces a peak for the unknown that is equidistant from the RFPs for the standards.