Domain Generalization for Time Series: Enhancing Drilling Regression Models for Stick-Slip Index Prediction

TL;DR

The paper tackles predicting the Stick-Slip Index (SSI) from surface time-series data under cross-domain shifts in drilling wells. It compares Adversarial Domain Generalization (ADG) and Invariant Risk Minimization (IRM) to a baseline, plus a Transfer Learning (TL) extension, using 60-second, 1 Hz sequences. Results show ADG and IRM outperform the baseline by about 10% and 8% respectively in generalization, with ADG slightly better, and TL further boosts performance; severe events are detected much more reliably (60% vs 20% baseline). These findings demonstrate the practicality of DG approaches for drilling and suggest ADG as the most effective strategy for robust SSI regression across wells. The work indicates a path toward real-time, sensor-light monitoring by exploiting domain-invariant representations across diverse wells.

Abstract

This paper provides a comprehensive comparison of domain generalization techniques applied to time series data within a drilling context, focusing on the prediction of a continuous Stick-Slip Index (SSI), a critical metric for assessing torsional downhole vibrations at the drill bit. The study aims to develop a robust regression model that can generalize across domains by training on 60 second labeled sequences of 1 Hz surface drilling data to predict the SSI. The model is tested in wells that are different from those used during training. To fine-tune the model architecture, a grid search approach is employed to optimize key hyperparameters. A comparative analysis of the Adversarial Domain Generalization (ADG), Invariant Risk Minimization (IRM) and baseline models is presented, along with an evaluation of the effectiveness of transfer learning (TL) in improving model performance. The ADG and IRM models achieve performance improvements of 10% and 8%, respectively, over the baseline model. Most importantly, severe events are detected 60% of the time, against 20% for the baseline model. Overall, the results indicate that both ADG and IRM models surpass the baseline, with the ADG model exhibiting a slight advantage over the IRM model. Additionally, applying TL to a pre-trained model further improves performance. Our findings demonstrate the potential of domain generalization approaches in drilling applications, with ADG emerging as the most effective approach.

Figures (14)

• Figure 1: Simplified schematic representation of the drilling system
• Figure 2: Example of a sequence with severe stick-slip.
• Figure 3: Architecture of ADG model. The generator is composed of a sequence of LSTM and LN layers; the SSI predictor and domain classifier are fully connected neural networks; a Gradient Reversal Layer (GRL) is used to reverse the direction of the gradients during backpropagation.
• Figure 4: Average MSE of SSI validation data predictions for the three tested cases (Table. \ref{['Validation data selection for the three tested Cases']}) with varying regularization and weighting coefficient over three different intializations: The ADG model is trained for each combination of hyperparameters using three distinct initializations. The validation SSI error is calculated for each one, and the average error is then computed across the three initializations.
• Figure 5: Average MSE and DTW of SSI validation data predictions for the three tested cases with varying regularization and weighting coefficient in the ADG model.