PSI - Issue 66

Vivek Vishwakarma et al. / Procedia Structural Integrity 66 (2024) 381–387 Vishwakarma and Ray/ Structural Integrity Procedia 00 (2025) 000–000

384 4

crack, originating from the notch, showed rapid growth in this second linear zone. As the load approached its peak, crack growth slowed. Increased CMOD in the yielding zone (beyond 1 mm CMOD) resulted primarily due to rotation of the beam around the crack tip. The final failure occurred due to steel rupture after yielding. 3.2. AE data processing Acoustic emissions analysis of the reinforced concrete beam under monotonic loading revealed a strong correlation between crack propagation and AE activity. Of the total recorded AE events, 77% were recorded in the two initial zones, with a significant concentration (75.6%) occurring during the second linear region of the load-CMOD curve, corresponding to rapid crack growth across the reinforcement. The dominant AE source was the formation and propagation of the primary crack in the concrete. The AE data processing involved several key preparatory steps. First, filtering process was implemented to remove illogical data points. First hit data of an acoustic event is used for the analysis as it is the best available representation of the original signal. Outliers are rejected based on the z-score value. Finally, the filtered data is subjected to unsupervised clustering using GMM. The average frequency (AF) and rise angle (RA) were chosen as the main clustering parameters. These parameters studied by Ohno & Ohtsu, (2010) are widely used in concrete for crack classification. Before clustering, AE data was checked for clustering ability using statistical parameters like the DB

index, Pearson correlation coefficient, and silhouette index. Tensile crack has typically high AF values and low RA values. Cluster 1 is chosen for further analysis, best representing the tensile failure. Within the tensile cluster, only those events are selected which have a probability of 0.8 or greater to be in that cluster based on the posterior probabilities provided by the GMM model. Clustering results are shown in Figure 3. The range of AF identified as highly probable tensile crack events is 8 kHz to 312.25 kHz, and the range of RA is 0.01 ms/V to 2.16 ms/V. Figure 3 Clustering results for beam LRB33S