Paper presented at the IADC/SPE International Drilling Conference and Exhibition, Galveston, Texas, USA, March 2026.
Paper Number: SPE-230802-MS.

While many different KPIs have been developed over the years for evaluating well construction performance, the definition and utilization of a wellbore quality KPI have remained challenging due to its subjective nature, data wrangling issues, and human resource demands. This paper introduces an automated method for calculating a Wellbore Quality Index (WQI) using 1 Hz drilling data, ML-generated performance outputs and physics-based models, to assist decision-making for casing running de-risking and clean-out requirements.

The WQI is calculated as a weighted summation of multiple features, where the weights are derived using Linear Discriminant Analysis (LDA). The WQI was validated for two applications: casing runnability prediction using 12 features derived from BHA runs prior to casing or liner installation, and overall wellbore quality post-analysis using 15 features including casing-related features. Most features focused on indicators related to pipe movement difficulty. The WQI was developed and evaluated using an initial population of 118 horizontal wells. The initial parameter list incorporated a study on Operator-defined Tortuosity Index (TI) to understand its possible impact on wellbore quality. NPT data associated with the WQI was considered during evaluation. Data quality checks identified missing or out-of-range values.

For casing runnability prediction, the WQI achieved an in-sample Area Under the Curve (AUC) of 0.971 with a Leave-One-Out Cross-Validation (LOOCV) AUC of 0.906, demonstrating strong and generalizable discriminative performance across 104 hole sections. For overall wellbore quality, the index achieved a test AUC of 0.923 across 230 hole sections. In both cases, statistically significant separation between High and Low NPT groups was confirmed using the Mann-Whitney U test (p < 0.0001). The results support robust risk evaluation and inform decision-making regarding wellbore conditions in relation to pipe movement. Parameter selection is adaptable, enabling use across various basins, well types, and business units. Real-time data can be processed down to a bit run, hole section, and at the end of the well. This paper presents an automated method for determining wellbore quality using physics-based models and drilling performance calculations generated using machine learning models (Bayesian network). This approach offers clearer, more usable results for prediction, root cause analysis, risk assessment, and performance benchmarking. The index, along with the various subcomponents, provides clear guidance on actions to be taken prior to pulling the BHA out of the hole for a casing run.