Subsurface sensors, such as inclinometers and tools used to take measurements while drilling, are important to the mining and petroleum industries. The current sensor systems are susceptible to shock, vibrations, and magnetic disturbances. To overcome these challenges, we propose a subsurface sensor fusion system with two sets of redundant inertial measurement units (IMU) to protect against magnetic and shock disturbances that affect the performance of magnetometers and gyroscopes. Orientation information is obtained by multiple micro-electromechanical system (MEMS) based inertial sensors, which consist of three-axis accelerometers, gyroscopes, and magnetometers.

In this thesis we obtain angular displacements using two different approaches to improve sensor robustness to magnetic and shock disturbances;​ also, we discuss the pros and cons of these two different approaches. The first approach is the supervised learning filter (SLF) approach, and the second is the supervised learning-Kalman filter (SL-KF) approach. In SLF, azimuth angle errors obtained from different sensors (magnetometers, accelerometers, and gyroscopes) are compared under magnetic and shock disturbance conditions;​ then, we employ an adaptive neuro fuzzy inference system (ANFIS) to calculate the error models of the sensors. Based on these sensors’ error models, the proper weights of the azimuth angles obtained from different sensors are computed and applied to the azimuth angles to output a final azimuth angle. However, to achieve the best results of SLF, we assume that at least one magnetometer is not affected by interferences at the same time interval (two magnetometers are separated by a distance D, and D can prevent both magnetometers from being affected by a magnetic disturbance at the same time). Therefore, SL-KF combines SLF with a KF to further reduce the effect of disturbances on sensors. SLF computes the corrected rotational angles and angular velocities that are subsequently fed into a global filter KF, which performs further corrections.

The present subsurface positioning (directional drilling) relies on angular displacements and values of measurement depth (drill string length) to estimate a well path. However, these methods have limitations to apply in working conditions (for example drill string length maybe inaccurate caused by steel expands with increased temperature and stress). To deal with the drill string length inaccuracy problem, instead of using real external measurement signals (drill string length), we use correction signals designed based on the dual acceleration difference (DAD) method to correct the positions.

The proposed ideas of angular and position estimations are evaluated by experimental results. From the angular evaluation, based on a 59 second root mean square (RMS) calculation, the error of the proposed SLF approach is about 0.26 degrees, assuming one magnetometer is not disturbed by magnetic disturbances. When all sensors are disturbed by shock and magnetic disturbances, compared with SLF, the proposed SL-KF approach increases the performance by up to 56% using a 59 second RMS calculation. From the position evaluation, the proposed dual acceleration method reduces the error magnitudes caused by disturbances from meters to millimeters.