A datacentric long term rockburst intensity prediction for underground excavations

Purpose Rockbursts are characterised by the sudden and violent ejection of rock fragments due to the spontaneous release of strain energy stored within the rock mass. They are more frequent in deep underground excavations and high-stress zones. Although, rockbursts have long been the subject of extensive research, and valuable insights have been gained over time, they continue to present a persistent challenge in underground excavation projects. In this context, the present study aims to establish a mapping relationship between rock burst intensity and contributing factors. Design/methodology/approach A data set was compiled of 254 rockburst instances from published experimental results. Each occurrence was characterised by six critical variables: maximum tangential stress (σθ); uniaxial compressive strength (σc); uniaxial tensile strength (σt); brittleness index B1 (σθ/σc); brittleness index B2 (σc/σt); and strain energy storage index (WET). Findings Multi perceptron autoencoder models were developed to assess the performance of the neural network, and the potential of the data set to support data centred long term rock burst intensity prediction. Originality/value The results were evaluated through confusion matrices and four accuracy measures for data mining multiclass problems including accuracy, precision, recall and F1-score and receiver operating characteristic curve, suggesting that perceptron autoencoder model achieved a promising success rate of 85.8%.