Aplicación del Procedimiento de Evaluación de la conformidad al Procedimiento de gestión

Surface cracks with branches are more difficult to characterise than simple defects due to the complexity in the geometry. However, by scanning the area around the crack, a B-scan can be generated. In the B-scans a number of Rayleigh wave reflections can be observed to arrive at different times, and they are well separated from each other. The key to detecting the presence of a branch on the crack is to look for earlier arriving low frequency reflections from the branch, as a portion of the low frequency component of the incoming Rayleigh wave will be reflected directly from the branch before the wave enters the crack wedge section. This causes this part of the low frequency component to be reflected much earlier than the other frequency components. In terms of imple- mentation in RCF crack detection, B-scans are sufficient to show the low frequency reflection, as described in Section 7.3. In addition, the spectrogram can also display the low frequency reflection clearly in the time-frequency domain. Detection of the presence of the branch is essential for analysing the integrity of that section of rail. Thus, as soon as the branch is detected and rail breakage may occur, the section of the rail needs to be replaced [3,4].

The total depth of branched cracks, d1+d2, can be estimated by calculating Ct and

referring to the inclined crack depth calibration. This has been shown in Section 7.4, where the actual depth of branched cracks are close to the calibration made from he tinclined crack. Once the total depth is calculated, the length of the branch on the crack can be calculated as following; From the B-scan, the low frequency reflection interferes with the incoming Rayleigh wave at the position of the branch along the scan line. The position of the crack opening can be identified through the prominent enhancement pattern. The distance between these two positions is the length of the inclined part of the crack on the surface (horizontal projection on the surface). Let us say that this

distance is lcrack. Using the knowledge that the RCF crack has θ of approximately

20°-30°, the value of d1 can be calculated by considering the geometry between lcrack ,

d1, and θ. Thus, d1 can be calculated following the relationship in Equation 7.16. The

length of the branch d2 can then be calculated by subtracting d1 from the total depth

gained from the transmission measurement.

d1 =lcracktanθ (7.16)

Equation 7.16 gives a good estimation of d1 and d2 provided that the typical value

of θ can be assumed, such as in the case of RCF cracks. For a more general case where

the value ofθ is not known, for example in SCC,θ can be estimated by calculating the

enhancement factor ratio FER of the in-plane to the out-of-plane component (Equa-

tion 4.17). Details of the enhancement factorFE measurement for each component are

described in Section 4.5, and θ can be estimated using the plot of FER against θ in

Figure 4.22.

The characterisation steps for branched cracks explained here can be incorperated into the crack characterisation procedure described in Section 5.2 for more detailed characterisation of surface cracks. A new category for the branched cracks can be added to the initial classification using B-scans, based on the low frequency reflection feature. Following this classification, the Rayleigh wave amplitude measurements lead

to the calculation of Ct and FER. For branched cracks, the calculation of the total

7.6

Conclusions

The interaction of Rayleigh waves with inclined cracks with vertical branches as illus- trated in Figure 7.1, has been studied through the use of a two-dimensional FEA model. The presence of a branch at the tip of the inclined crack caused a portion of the low frequency incident Rayleigh wave to be reflected earlier than the other frequencies. A spectrogram was generated from the signal recorded at 40 mm distance relative to the crack opening. The frequency content in the time domain shown in the spectrogram confirmed that there is a separation between the low frequency reflection of Rayleigh waves from the branch, and the later reflections from the inclined section. In terms of practical applications, a B-scan generated from scanning a branched crack can be used as an initial indication to detect the crack. The arrival time calculations for different mode conversions during the interaction have been made, and show a good agreement with the B-scan. A characterisation method has been proposed for estimating the crack length, and can be incorporated with the procedure described in Section 5.2 for characterising full-width cracks.

Chapter 8

Conclusions

This chapter draws together the conclusions of the main findings described in this thesis, and discusses how they are relevant to the characterisation of surface cracks, particularly RCF-like cracks. Suggestions are made for potential future work.

8.1

Main findings

The comparison of the Rayleigh wave transmission coefficient measurements made be- tween cracks growing normal to sample surface, and cracks growing inclined to the surface, proved that the transmission coefficient is different for these types of crack. This has consequences for the crack depth sizing method previously reported, which assumed that surface cracks grow normal to the surface [23]. Depth calibrations made using the calibration for normal crack should not be used for sizing inclined cracks, as it will cause underestimation of the depth. For accurate sizing of crack depth, a separate calibration is needed for inclined cracks. 45° calibration works well for most inclined cracks, and 90° calibration is best for near 90° cracks. This was explained in Chapter 4. The B-scan images generated from scan data give an indication of whether the crack is normal to or inclined to the surface, through the enhancement pattern. Inclined cracks have a stronger enhancement compared to normal cracks, with an alternating black and white pattern that exists for a significant amount of time. The stronger and prominent enhancement pattern for inclined cracks is caused by Rayleigh wave mode conversion to Lamb-like waves at positions close to and within the near-field of the crack. Inclined cracks form a wedge section that has varying local thickness. As the Rayleigh wave propagates from the section that has the full thickness of the aluminum sample, to

the section within the wedge, the frequency-thickness product changes rapidly causing mode conversion to Lamb-like waves. This was confirmed through the arrival time calculations of the Lamb wave fundamental anti-symmetric mode (A0). The signal enhancement caused by this mode conversion can be used as a fingerprint to identify inclined cracks in the B-scans (Chapter 4 and 5).

The enhancement factor in the in-plane and out-of-plane components can be used to estimate the crack inclination, with a reasonable accuracy, giving an idea of which calibration to use. Using the information from B-scans and Rayleigh wave amplitude analysis, the distinction can be made between normal and inclined cracks. Crack orien- tation can then be determined from the enhancement factor and estimation of the crack angle and better sizing of crack vertical depth can be made. An analysis procedure has been created for characterisaton of surface crack (Chapter 5).

A machine learning program has been developed through a collaboration with ex- perts in Genetic Programming for pattern recognition. The program learns the specific pattern in the B-scan asscociated with a certain crack type, and use the information to classify a given B-scan image as defective/non-defective or normal/inclined. Input from an ultrasonic expert was required, in particular on what kind of pattern features can be used for the association with each crack. The main attention was given to the difference between the enhancement pattern for normal and inclined cracks. Other dif- ferences include the changes in the greyscale, wave arrival time, and the position of reflected waves. After training was conducted using a small set of data from experi- ments and FEA models, the program reached more than 90% accuracy in classifying image (Chapter 5). The level of accuracy and the running speed of the program show potential for use in automated recognition of B-scan images. In theory, this can be in- tegrated with the amplitude based analysis, and can provide real-time analysis results. More work would be required to put this into realisation in practical terms.

EMAT measurements can be affected by the size of the crack width. This has been confirmed experimentally with measurements made on two cracks of widths smaller or comparable with the width of EMAT coil. For these cracks, Rayleigh waves incident on the crack are diffracted at each crack edge. Transmission of the Rayleigh wave is different compared to the transmission for full-width cracks, therefore the calibration made using either normal and inclined full-width cracks might not be accurate. Sug- gestion has been made to use a smaller width EMAT to detect small cracks, such that the crack width is comparable or larger than the width of EMAT (Chapter 6).

most dangerous case is when the branch grows vertically into the bulk of the rail, and can cause the rail to break. It is therefore important to detect the branch at the earliest opportunity before the condition get worse. A set of FEA models were computed to see the interaction of Rayleigh wave with a branched crack (Chapter 7). The results confirm that the presence of the branch can be detected from the low-frequency reflection from the branch. This is best viewed in B-scans, where a low-frequency reflection arrives earlier than the rest other reflections. Confirmation of the frequency content can be made using a spectrogram. The findings on the branched cracks model can be included in the procedure for characterising surface cracks presented in Chapter 5.