Model Based Approach for non Destructive Testing of Pipes with Application to Microwave Scattering

The problem of non destructive testing of pipes has been addressed. Non destructive inspection of not accessible pipes consist in the detection, classification and measurement of possible flaws. The emission of a signal with specific characteristics and the analysis of the return of the emitted pulse allows flaw detection, and eventually classification and measurement. Such techniques use different preprocessing and processing techniques, and often use neural network approaches for classification. Inspection can be conducted both as the transient analy sis response and with a steady state analysis of the system under test. In the first case the "time of flight" represents a fundamental parameter to detect the presence and localize the position of the flaw in the inspected pipe. In the pipe inspection case, even if the pipe characteristics are known, the problem may be complicated by the presence of a highly dispersive medium characteristics. In some cases the high speed of the signal in the system under test makes the transient analysis response inapplicable. It's the case of the electromagnetic inspection. In such cases the application of a steady state analysis is mandatory. A reverberating phenomenon takes place, due to the unavoidable multiple reflection of the input signal both at the source location and at the pipe end, and of course of the possible presence of a flaw in the middle of the structure giving information about the presence and eventually the typology of the flaw. In this paper, a model of the flaw has been i ntroduced, to take care of the pipe inspection technique for sensor and transmitter lying at the same spatial position. The proposed model is general and takes care of the real reverberating phenomenon in the inspected pipe. As an application of the proposed flaw modeling, the application to microwave scattering parameters for an asymmetric bifilar transmission line has been used to acquire data to be used in the defect modeling, the pipe to be tested being one of the two wires in the line, and a normal copper wire the other. The proposed data preprocessing to be use d in a defect classification system is the focus of the paper. A linear time invariant convolving model of the defect is presented to take care of the experimental data acquisition setup. The assumption is that proper modeling can help in distinguishing different defects in the measurement system. The technique to extract the model from the acquired data is presented, together with real acquired data experimental results.