Multi-step prediction for coal-rock drilling rate of penetration using IVMD-CNN-BiLSTM

Accurate prediction of the drilling rate of penetration (ROP) prediction is an enormously important step to optimize the controllable parameters of anti-impact drilling robot (AIDR), and single-step forecasting cannot adjust parameters in advance. The feeding speed and rotational speed are crucial parameters that reflect the degree of obstruction presented by the geological environment to the AIDR. Furthermore, the vibration data can provide insight into the instantaneous impact of vibration on the drilling process. Therefore, accurate multi-step prediction for ROP is crucial for ensuring the efficient and safe propulsion of AIDR. This paper proposes a multi-step prediction strategy combining drilling multi-sensor signals and a bidirectional long-short-term memory combined with a convolutional neural network-bidirectional long short-term memory (CNN-BiLSTM) for AIDR drilling parameters. Initially, the vibration, feeding speed, and rotation speed signals are denoised using an improved variable mode decomposition (IVMD) and wavelet threshold approach. Subsequently, the feeding speed, rotational speed and vibration signals are weighted and normalized by self-adaptation. Given the basis, the CNN-BiLSTM neural network is utilized to make a multi-step prediction, ultimately realizing the multi-step feeding speed and rotation speed prediction. Compared with the recent approach, the result shows that the proposed strategy has superior prediction accuracy.