Drilling Steering Model Calibration With Probabilistic MCMC Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automated drilling systems face challenges in calibrating steering models due to non-linearity, sensor uncertainties, and model-system discrepancies, leading to unstable and underdefined parameter estimation, which affects the accuracy of drill-bit position and orientation control.
Innovation Solution
A method using Markov Chain Monte Carlo (MCMC) sampling with measurement uncertainty estimation and statistical bagging to calibrate the steering model, incorporating accelerometer and magnetometer data, and a probabilistic framework to update model parameters and improve real-time feedback control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If traditional parameter estimation methods are used for steering model calibration, then the calibration process is simple, but the estimation becomes unstable and underdefined with infinitely many solutions due to non-linearity
Solution Approach 1:
The patent transforms the traditional deterministic parameter estimation into a probabilistic framework by introducing prior distributions and likelihood functions. This changes the nature of parameters from fixed values to probability distributions, allowing the system to handle non-linearity and uncertainty while maintaining calibration feasibility through Bayesian inference.
Solution Approach 2:
The patent introduces an intermediary probabilistic model that mediates between the steering inputs and measurements. This probabilistic framework acts as a bridge that incorporates measurement uncertainties and model discrepancies, transforming the ill-posed estimation problem into a well-defined Bayesian inference problem with unique solutions.
2Measurement precision
If model calibration is performed continuously to maintain accuracy, then the steering control precision is improved, but sensor malfunctions and model-system discrepancies produce poor models that should be bypassed
Solution Approach 1:
The patent implements feedback through the Bayesian inference process that continuously updates parameter estimates based on new measurements. The system uses posterior distributions from previous iterations as priors for the next iteration, creating a feedback loop that maintains accuracy while the probabilistic framework naturally handles bad data through uncertainty quantification.
Solution Approach 2:
The patent applies beforehand cushioning by incorporating prior distributions that encode existing knowledge about system parameters before new measurements are processed. These priors act as a cushion against poor measurements or sensor malfunctions, preventing them from completely derailing the calibration process and allowing the system to bypass poor models gracefully.
3Productivity
If automated drilling systems use accurate steering models for control, then drilling efficiency is improved, but numerous unknowns in the downhole environment require continual model updates
Solution Approach 1:
The patent applies preliminary action by pre-defining the probabilistic framework, prior distributions, and likelihood functions before actual drilling operations begin. This preparatory work establishes the calibration methodology in advance, so that during drilling, the system only needs to perform straightforward Bayesian updates rather than designing the entire calibration process from scratch, reducing operational complexity.
Solution Approach 2:
The patent creates a universal probabilistic calibration framework that can handle multiple sources of uncertainty (sensor malfunctions, vibrations, model-system discrepancies) through a single unified approach. This multi-functional framework addresses various downhole unknowns simultaneously, reducing the need for separate calibration procedures for different types of uncertainties and simplifying the overall process.
Data Source
AI summary
A method for calibrating a steering model may comprise estimating an initial condition for one or more variables in the steering model and calibrating the steering model with a Markov Chain Monte Carlo Simulation (MCMC). A drilling system may comprise a bottom hole assembly, a drill string connected to the bottom hole assembly, and an information handling system connected to the bottom hole assembly. The information handling system may be configured to process one or more measurements from the bottom hole assembly, calibrate a steering model based at least in part on the one or more measurements, adjust a control logic based at least in part on the steering model, and adjust the bottom hole assembly based at least in part on the control logic.