Dewaxing LLM Project Overview

Project Objective

• Existing job summaries for dewaxing operations often lack sufficient detail, with individual run information embedded as unstructured comments.
• While some jobs have manually created summaries and run details, most remain incomplete due to the effort required.
• Consistent data capture will facilitate extraction of insights from past jobs. Converting run details from unstructured text into structured tables enables machine learning model training and supports optimization of dewaxing scheduling.
• The goal is to leverage large language models (LLMs) to automatically generate job summaries and structured tables for run details based on job log comments.

Existing Process and Infrastructure

Dewax Program

The dewaxing process at ARC is a critical maintenance operation designed to ensure optimal well performance by mitigating wax buildup in production tubing. The current workflow integrates scheduled planning, field diagnostics, and adaptive intervention strategies to maximize efficiency.

Scheduling and Frequency:

Dewax operations are scheduled using a master well list maintained in Excel, which incorporates frequency guidelines based on historical performance. Typically, each well undergoes 4–6 dewax runs per job. Wells requiring more frequent intervention are revisited sooner, while those needing fewer runs are scheduled later. Currently, two rigs are dedicated to dewaxing, operating Monday through Friday and occasionally on weekends. The well servicing team is actively optimizing costs by reducing rig moves and refining dewax frequencies.

Operational Scale and Costs:

• Annual dewax well servicing costs exceed $4 million, with a slight year-over-year increase since 2022.
• Over 1,500 dewax jobs are completed each year.
• The number of wells serviced has increased by approximately 15 per year.
• Total dewax jobs have grown by about 100 annually.
• In 2024, ARC serviced 390 wells across 75 pads, completing 1,806 dewax jobs at a total cost of $4.7 million.

Triggers

• At the start of each month, a temperature survey identifies wells flowing below 30°C, indicating potential wax deposition. These wells are added to the dewax program for regular maintenance.
• Wells exceeding a 30-day dewax interval may be considered for plunger lift installation, with an associated cost of approximately $2,400 and a 12-month economic evaluation.

Mechanical Dewax Procedure

Standard Operating Procedure:
The full SOP is available here: ARC Routine Dewaxing SOP - V2.6 (PDF)

Tool Progression: Dewaxing begins with a spear tool to a depth of 750m, followed by knives until wax fill is reduced to 10%. Tool depth is determined by wireline tension feedback.

Chemical Intervention: If mechanical tools are insufficient, chemical solvents are introduced and allowed to soak. If ineffective, the process is repeated or left overnight.

Thermal Methods: Steam is applied as a tertiary method to the backside of the tubing to heat and mobilize wax.

Chemical Usage and Spend

• ARC spends between $4.6 million and $5.2 million per quarter on Chem Wax (605300), with the majority of expenditure concentrated in Kakwa.
• The dewax program relies heavily on paraffin inhibitors, delivered via dedicated pumps at the wellsite. ARC collaborates with vendors such as Secure and Baker Hughes to supply and manage these chemicals.
• Providing detailed information about the dewax program and production rates has been essential for optimizing chemical usage.
• A recent manual review led to the optimization of 117 wells, saving 6,000 liters of chemical and $33,500 per month.

For additional project background and detailed documentation, please refer to the Dewaxing LLM Project Charter.

Location

Model Feature Tables:

• prd_zone3.dewaxingllm.input_features

Model Results Tables:

• prd_zone3.dewaxingllm.job_summary_inference (inference results for job summary)
• prd_zone3.dewaxingllm.job_summary_prediction (predictions on labeled samples during model training for job summary)
• prd_zone3.dewaxingllm.job_summary_metrics (model results on labeled samples for job summary)
• prd_zone3.dewaxingllm.run_details_inference (inference results for run details)
• prd_zone3.dewaxingllm.run_details_prediction (predictions on labeled samples during model training for run details)
• prd_zone3.dewaxingllm.run_details_metrics (model results on labeled samples for run details)
• prd_zone3.dewaxingllm.monitoring_summary (model drift monitoring)
• prd_zone3.dewaxingllm.api_integration_log (log of inference data sent via API to WellView)

Repository:

• ARC-Data-Analytics/llm-wellview

Databricks Workspaces:

• PROD: arcPrdDbwDAL
• UAT: arcUatDbwDAL
• DEV: arcDevDbwDAL

Data Sources

• prd_zone2.wellviewetl.jobtimelog_v1
• prd_zone2.wellviewetl.job_v1
• prd_zone2.wellviewetl.wells_v1

Contacts

See the contacts page for project leads and support.