PSI - Issue 33

K. Kaklis et al. / Procedia Structural Integrity 33 (2021) 251–258 Author name / Structural Integrity Procedia 00 (2019) 000–000

257 7

(c) 1538 0 1000 2000 3000 4000 5000 6000 7000 8000 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 Average cumulative amplitude (dB) Load (kN) Predicted load 20% of max. cum. amplitude 0 1000 2000 3000 4000 5000 6000 7000 8000 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 Average cumulative amplitude (dB) Load (kN) Predicted load 20% of max. cum. amplitude 1538

1538 0 1000 2000 3000 4000 5000 6000 7000 8000 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 Average cumulative amplitude (dB) Load (kN) Predicted load 20% of max. cum. amplitude

(d)

0 1000 2000 3000 4000 5000 6000 7000 8000

Measured load Predicted load 20% of max. cum. amplitude 75% of max. load

0.9 kN

1538 dB

0.0 Average cumulative amplitude (dB) 0.2 0.4 0.6

0.8

1.0

1.2

1.4

Load (kN)

(e)

(f)

Fig. 5. Variation of the average cumulative amplitude for specimen #9 with respect to the predicted load, up to (a) 60%, (b) 70%, (c) 75%, (d) 85%, (e) 90%, (f) 100% level of the maximum load. 6. Conclusions In this study, several machine learning algorithms were investigated to predict the load variation in three-point bending tests based on acoustic emission signals. The best performing algorithms included the artificial neural networks, the random forests and the decision trees. The overall best performing technique identified by this analysis was the ANN model #2 with a 6-15-1 network architecture and a tansig transfer function. This model shows the highest coefficient of determination (0.996) and the lowest root mean square error (0.022) compared to other models. As a result, this ANN model seems to provide an appropriate prediction of load variation under TPB tests on Nestos marble. The comparison of the predicted load under TPB tests with the experimental data confirms the prediction capability of the proposed ANN model. The cumulative amplitude vs predicted load curve verifies the experimental critical AE and load levels, as well as the rapid slope increase of this curve. The ANN model can potentially be an appropriate tool to predict rock behavior, based on the experimental critical levels. During the TPB test, utilizing only the recorded AE signals, the critical load level could be identified and be used as a failure index of the material. Based on this study, the proposed methodology could be confirmed and applied in various laboratory experimental procedures of rock mechanics and rock engineering. In the context of the new "Mining 4.0" era, a modified and properly designed ANN algorithm could predict the pillar and roof behavior/failure in several underground mining methods, based on acoustic or seismic recorded signals.