PSI - Issue 52
Jiri Dvorak et al. / Procedia Structural Integrity 52 (2024) 259–266 Author name / Structural Integrity Procedia 00 (2019) 000 – 000
264
6
evaluated using neural networks. These networks work on the principle of identifying shape anomalies in the signal, which are then systematically searched for. Several types of anomalies were identified from previous AE records, which are referred to as “T races ” (Fig.7).
Fig. 7. - Eight types of color-coded "Traces" in the AE signal recognized by the DAKEL - NN neural network
Fig. 8. Graphical representation of emission events (Traces) using star charts
Advanced evaluation algorithms (DAKEL NN neural network with SSD model) were applied to the measured AE diagnostic data, which output records of eight types of traces, pre-selected within the material study (Fig. 7). Each Trace has a permanently assigned colour and data about its number, length and energy are evaluated for it. A characteristic feature is that, during the diagnostic test, there are very sharp changes in the individual phases of the creep test, manifested by different representation of individual tracks. A quick overview of the nature of trace occurrence at a given stage in the test process is provided by the star charts – Fig. 8. These show the average Trace values per diagnostic measurement, in our case 15 minutes. These are relative values; the maximum value of each monitored variable is always normalized to 1. The colour identifies the different types of Traces. This enables to capture even changes that would be difficult to observe due to high values of other parameters and noise. Therefore, star charts give us a better and clearer overview of the nature of the occurrence of Traces in a given phase of the test. Each star chart shows for one measurement and for each trace the average number of traces, their average length and average energy. In the case of the implemented measurements, an increase or change in emission activity was detected at the beginning of the test and especially in the tertiary stage. Furthermore, different traces of emission events are active for individual material states. Changes in the ratio of trace energies related to the development of creep damage are clearly visible in the Fig 9. It is observed high activity in the primary stage shortly after loading - these are short-term fluctuations, and they always return to the normal appearance of Traces. Subsequently, emission events are attenuated during the primary and secondary stages. At the beginning of the tertiary stage, Traces 1,2,4 become dominant, which is apparently closely related to the development and coalescence of cavities. It can be concluded that by scanning AE in selected types of operation, undesirable changes in the material can be located well in advance and thus reduce the risk of a possible accident.
Made with FlippingBook Annual report maker