Issue 47

D. Rigon et alii, Frattura ed Integrità Strutturale, 47 (2019) 334-347; DOI: 10.3221/IGF-ESIS.47.25 337 peak of the tapered force amplitude signal is within the range of ±2.7 kN, i.e. below 5% of the force amplitude relevant to the fatigue test in all the acquisitions. Fig. 3b reports an example of t* evaluation. Figure 4 : Schematic representation of the 3-dimensional array T (a). Flowchart of the data analysis to obtain the energy distribution at the notch tip. (b) In order to obtain the energy distribution at the notch tip of the considered specimens, a numerical procedure was developed by using the Matlab code. First, the Altair video recording file (*.ptw) was converted into an ASCII film file (*.asc), which is readable in Matlab. Next, the ASCII file was input to a dedicated Matlab script, that converts it into a Matlab 3-dimensional array, named   , , T m n p , having dimension m-by-n-by-p. The m, n values are the dimensions of the frame expressed in terms of pixels, reduced to avoid vignetting (m=136 px, n=167 px), and p is the number of frames acquired by the infrared camera (i.e p is equal to 2000). A schematic illustration of   , , T m n p is shown in Fig. 4a. Let i, j and k be the indexes of   , , T m n p . An element of this 3D array corresponds to a temperature value of the n-th pixel having coordinate i and j for the k-th frame. In this way, fixing i and j and plotting T ij against the time (obtained from the division of the k index value by f acq ), the time variant temperature graph, commonly used for evaluated the Q parameter, is obtained for the n-th pixel. A tentative value of 0.1 s was assigned to a numerical variable, named “dt”, for evaluating the cooling gradient of the n-th pixel and it was kept constant for all pixels (see Fig. 4b). In practice, the numerical evaluation of the cooling gradient was performed by the polyfit matlab function, which returns the value of the slope of the linear fitting of the data within the time window “dt”. Finally, Q ij is evaluated applying Eqn. (1). This operation was routinely performed for all pixels by a for loop, resulting in a m-by-n matrix composed by Q ij values, called   , Q m n . Since the same dt value was fixed for all the pixels and a certain level of noise in the measurements was present, the cooling gradient might result meaningless for some pixels. Therefore, after plotting the   , Q m n matrix, a check was performed to single-out unrealistic spike-like values. Then, the dt value was iteratively modified in the range from 0.05 s to 0.15 s in order to eliminate the aforementioned unrealistic Q ij values. Once passed this control, the last step consisted in filtering the   , Q m n matrix, adopting the “imgaussfilt” matlab function, which uses a Gaussian smoothing kernel with specified standard deviation. For this application, a standard deviation ranging from 4 to 6 was adopted, by obtaining the   , flt Q m n matrix.