Data Fusion by a Supervised Learning Method for Orientation Estimation Using Multi-Sensor Configuration Under Conditions of Magnetic Distortion and Shock Impact

Abstract

Accurate subsurface sensing during directional drilling is critical in the mining and energy extraction industries. One challenge is to measure the azimuth accurately. Azimuth measurements are hindered by magnetic disturbances such as iron debris, especially when magnetometers are used. Moreover, gyroscopes are susceptible to shocks during drilling surveys. To overcome these challenges, we developed a supervised learning filter (SLF) using a multi-sensor configuration (MSC) to accurately estimate the azimuth. The MSC consists of micro-electro-mechanical systems (MEMS) based magnetometers, gyroscopes, and accelerometers into two set of sensors, and the groups are separated by a known distance D to acquire additional rotational information using a dual acceleration difference (DAD) method. Also, D can reduce the negative effect of magnetic disturbances. A Kalman filter (KF) with known a priori noise information removes white noise; however, it is difficult to deal with unknown magnetic and shock disturbances. To reduce the effect of unknown magnetic and shock disturbances, we use the SLF to estimate orientation information. First, the SLF employs an adaptive neuro network fuzzy inference system (ANFIS) to build error models of each sensor; then the SLF calculates the proper weights of the sensors using the error models. Lab-scale experiments are performed on a test rig where the SLF is evaluated using one case with training and verified using two cases without training. The results showed an improvement in azimuth estimation.

Keywords

• Supervised learning filter
• directional drilling
• mwd
• anfis
• magnetic disturbance robustness
• dual acceleration difference