| dc.description.abstract | Real-time methods of characterising the downhole impact motions and the impacted downhole material for the vibro-impact drilling (VID) system have been investigated in this study. As an impacting rock fragmentation system, characterising the downhole impact motions is essential for avoiding detrimental and low performance impacts. On the other hand, as a resonantly vibrated system, characterising the impacted downhole material is essential for maintaining resonance in the system. Meeting these conditions further enhances the benefits of the VID system as a rotary-percussion drilling system in terms of increased rate of penetration, well stability and tools life-span. This thus makes the present study an optimisation investigation targeted at the VID system.
Conventionally, the real-time characterisation of downhole rock properties during drilling is often performed using Logging While Drilling tools. However, due to the gap between the location of the tool and drill-bit there are cases when the conventional method may prove inadequate. The VID system requires the property of the drilled rock just at the drill-bit area, thus making it one of those cases when the conventional logging may be inadequate.
In this study, a one-degree-of-freedom impact oscillator mimicking the bit-rock interaction of VID has been explored. Measurable impact dynamic variables including displacement, velocity and acceleration were analysed and utilised alongside intelligent network models to characterise resulting downhole impact responses and the impacted constraint (i.e. rock) stiffness. Utilised network models include Long Short-Term Memory networks, Multi-layer perceptron networks, Support vector machines and Gaussian Process networks. This was based on their ability to learn both linear and nonlinear transformations directly from data.
Simulation and experimental data have been used to validate the network models. Results show that extracting appropriate features and using a suitable network are essential for both characterisations. Aside attaining good performance, the proposed characterisation methods are well adaptable for on-line use in the VID system. Compared to previous studies, their methods of data processing require little or no human interaction, thus making them suitable for VID optimisation and automation. | en_GB |
