Pulsed Eddy Current signal processing using wavelet scattering and Gaussian process regression for fast and accurate ferromagnetic material thickness measurement

Testing the structural integrity of pipelines is a crucial maintenance task in oil and gas industry. This could be compromised by corrosions that occur pipeline wall. They cause catastrophic accidents are very hard to detect due presence insulation cladding around pipeline. corrosion manifests as reduction pipe wall thickness, which can detected quantified using Pulsed Eddy Current (PEC) state-of-the-art Non-Destructive Evaluation technique. The method exploits relationship between natural log transform PEC signal with material thickness. Unfortunately, measurement noise reduces accuracy technique particularly its amplified effect log-transform domain, inherent characteristics sensing device, non-homogenous property material. As result, requires averaging reduce improve prediction accuracy. Undesirably, this increases inspection time significantly, more measurements needed. Our proposed predict thickness without averaging. applies Wavelet Scattering log-transformed generate suitable discriminating feature then Neighborhood Component Feature Selection dimension before it train Gaussian Process regression model. Through experimentation ferromagnetic samples, we have shown our produce accurate estimation samples’ than other methods over different types materials layer thicknesses. Quantitative proof conclusion provided statistically analyzing comparing root mean square errors model those from inverse derivative approach well machine learning models.