Production processes optimization through machine learning methods based on geophysical monitoring data

The purpose of the article is to create an effective method for low-delay monitoring of the operating state of a drill string and a drill bit without interfering with the proper drilling process. For the drilling process to be continuously controlled, an experimental setup that operates by utilizing the phase-metric method of control was developed. Any movement of the bit causes a change in the electrical characteristics of the probing signal. To obtain a stable signal from a bit immersion depth of up to 250 m, a frequency of probing electrical signals of 166 Hz and an amplitude of up to 500 V were used; the sampling rate of an analog-to-digital converter (ADC) was 10101 Hz. To identify the state of the drill string and the bit based on graphs of time-dependences of changes in the probing signal electrical characteristics, the present authors investigated a number of deep learning methods. Based on the results of the study, a series of capsular neural network methods ( CapsNet ) was chosen. The authors developed modifications of 2D-CapsNet: windowed Fourier transform (WFT) - 2D-CapsNet and frequency slice wavelet transform (FSWT) - 2D-CapsNet. Both of these methods showed a 99% accuracy in determining the transition between two layers of rocks with different properties, which is 2–3% higher than the currently used measurement-while-drilling (MWD) and logging-while-drilling (LWD) rock surveys. Both of these methods unambiguously reveal self-oscillations in the drill string. When determining a fully serviceable bit in the case of self-oscillations, the (FSWT) - 2D-CapsNet method showed an accuracy of 99%.