In this paper, we present a novel and completely di↵erent approach to the problem of scattering material characterization: measuring the degree of predictability of the time series. Measuring predictability can provide information of the signal strength of the deterministic component of the time series in relation to the whole time series acquired. This relationship can provide information about coherent reflections in material grains with respect to the rest of incoherent noises that typically appear in non-destructive testing using ultrasonics. This is a non-parametric technique commonly used in chaos theory that does not require making any kind of assumptions about attenuation profiles. In highly scattering media (low SNR), it has been shown theoretically that the degree of predictability allows material characterization. The experimental results obtained in this work with 32 cement probes of 4 di↵erent porosities demonstrate the ability of this technique to do classification. It has also been shown that, in this particular application, the measurement of predictability can be used as an indicator of the percentages of porosity of the test samples with great accuracy.Keywords: Ultrasonic Signal Modality, Ultrasonic NDT, Multiple scattering, Noise, Determinism, Chaos Theory, Higher order statistics
IntroductionWhen scattering materials are subject to ultrasonic non-destructive testing (NDT), the ultrasonic pulse undergoes some variations that are related to the internal grain microstructure of the specimen. Each grain behaves like a scattering center, producing an echo that when superimposed on other echoes coming from other grains can even hide the echoes produced by a possible defect. Similar situations are found in other related fields such as ultrasound B-mode scans (where the grain noise is called speckle) and in radar with clutter [1].In the literature, a wide range of solutions has been proposed to enhance the detection of small cracks or defects and to reduce (or even eliminate) the e↵ect discussed in each of the above situations. Some of these solutions are signal averaging, auto-and cross-correlation, matched filtering, frequency spectrum analysis [2], spectral correlation [3], and wavelet transformations [4]. The aforementioned analyses discard the information encoded in the grain noise; however, this information can be used to recognize potential di↵erences among materials, tissues, or surfaces. This approach has been employed to characterize materials by extracting temporal signal statistics [1], the resonance frequency [5], and even the penetration depth [6]. Our work continues with this line of thought and proposes a new approach that attempts to extract information about the nature of the signal, thereby characterizing the signal modality. Signal modality characterization is a key concept of a multidisciplinary research topic that includes di↵erent concepts such as signal linearity, stationarity, and stochasticity [7]. The applications of signal modality characterization are becoming mo...