Welded components remain difficult to reliably and effectively examine cause of their complex solidification process. Material grains orientations in welds usually generate waveform divergence and splitting due to the re-melting process after each welding pass. Generally flaw data visibility by various non destructive testing methods is corrupted by many noises as electric, pulse, ringing, and structure or spurious signals. Various processing techniques were investigated to extract useful data and interpret the resulted waveform for diagnostic and predictive purpose. However the complexity of the model order estimation carries on complicated modeling. Wavelet based auto regressive parametric model could be a successful processing technique able to withdraw non stable characteristics of data and to display the useful data. In this paper waveform multi scale analysis was investigated under a predictive approach for data extraction. This analysis based correlations, residuals and interpolations calculus seems to be a good tool for shaping the material micro structural dimension scales. This research illustrates a linear waveform micro structural distribution correlated to the matter phases and several residuals curved layers resulting on a Lorenz like modeling by inverse approach who could act as a predictive indicator of the matter scales.