Weight on drill bit prediction models: Sugeno-type and Mamdani-type fuzzy inference systems compared

- WOB models constructed from real well data for: rate of penetration, mud weight, formation type, pump pressure, pump flow rate and rotation rate.
- Models tested with multiple well data from large gas and oil fields in Iran.
- Sugeno-type FIS models significantly outperform Mamdani-type FIS models in the prediction of WOB using the same data input information.
- Sugeno-type FIS models for predicting WOB have the potential to be cheaper and more accurate than conventional rate of penetration methods.

Drilling optimization aims to optimize controllable variables during drilling operations, such as weight on bit (WOB), in order to improve drilling rate of penetration and reduce well costs. Prediction models of Weight on Bit (WOB) are developed using two widely-applied fuzzy inference systems (FIS), Mamdani-type and Sugeno-type, for the Ahwaz oil field and Marun gas field formations; two large producing fields in Iran. The FIS are constructed based on field data involving multiple wells; six wells from Ahwaz field, and two wells from Marun field. The controllable input variable data for the FIS includes: rate of penetration, mud weight, formation type, pump pressure, pump flow rate and rotation rate. The key difference between these two FIS techniques is the manner in which they calculate their crisp output values. The Mamdani-type FIS requires defuzzification, whereas the Sugeno-type FIS applies a constant weighted-average technique avoiding defuzzification. The results for the two field cases evaluated convincingly demonstrate that the Sugeno-type FIS is superior to the Mamdani-type FIS for WOB prediction using the same input data and membership functions. There is scope for further refinement of FIS models for WOB prediction (e.g., by adding bit type information) and the Sugeno-type FIS methods should reduce the time and cost associated with the conventional rate of penetration methods for predicting WOB.