.png){kind=icon}
.png){kind=icon}
Return
Optimization Model for Ultrasonic Vibration Rock-Breaking Parameters Based on PCA and an Attention-Mechanism-Optimized KAN: Insights from Limestone
.png){kind=icon}
PRE
AI
X
Y
Z
H
X
Z
D
Y
A
DOI：10.1007/s00603-026-05824-1.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Although ultrasonic vibration rock breaking is used for extracting deep mineral resources, quantitative mapping between the excitation parameters and rock responses remains underdeveloped. Conventional regression models often fail to capture complex nonlinear interactions among control variables. In this study, an intelligent parameter optimization framework that integrates principal component analysis (PCA) with an attention-mechanism-optimized Kolmogorov−Arnold network (Attn-KAN) is developed. A three-factor, three-level orthogonal experimental design with different vibration amplitude (VA), static load (SL), and indenter geometry (IG) was used to develop a three-dimensional damage assessment index system. Ultrasonic wave velocity (UWV), uniaxial compressive strength (UCS), and elastic modulus (EM) served as the damage variables. PCA was used to condense the multidimensional damage variables into a comprehensive indicator (PC1), accounting for 94.41% of the total variance. The Attn-KAN prediction model achieved a coefficient of determination (R2) of 0.973 on the test set and outperformed conventional backpropagation (BP) neural networks. Interpretability analyses using SHAP and LIME consistently identified SL as the dominant controlling parameter, and the feature importance rankings from the two methods were highly consistent (Pearson correlation coefficient R = 0.99). Entropy-weighted TOPSIS identified an optimal parameter combination of SL = 0.75 MPa, VA = 20 μm, and a wedge-shaped indenter. Because the excitation frequency was fixed at 20 kHz owing to transducer constraints, the frequency-dependent fatigue effects were discussed mechanistically rather than optimized experimentally. These findings provide a basis for the intelligent selection and optimization of ultrasonic vibration rock-breaking parameters.
Keywords:
Ultrasonic vibration rock breaking
Principal component analysis
Kolmogorov‒Arnold network
Attention mechanism
Parameter optimization
Explainable artificial intelligence
Journal
IF:
6.6
Papers:
6.0K
Citations:
3.0W
Organization
Cited Papers