ViscoSense AI – Viscosity Calculator
ViscoSense AI: The Dead-Oil Viscosity Agent is an AI- and machine-learning-enabled engineering product developed by the IPB&F Consortium. Authored and developed by Dr. Birol Dindoruk and Utkarsh Sinha, the tool transforms peer-reviewed dead-oil viscosity research into an intuitive conversational engineering application.
Dead-oil viscosity prediction is challenging because crude-oil viscosity can vary by nearly seven orders of magnitude, from approximately 0.1 to 1,000,000 cP, depending on crude-oil properties and temperature. ViscoSense AI provides a practical interface for applying validated physics-augmented correlations and machine-learning models to this complex fluid-property prediction problem.
Product Description: ViscoSense AI is directly grounded in the published SPE Journal research, “Physics-Augmented Correlations and Machine-Learning Methods to Accurately Calculate Dead-Oil Viscosity Based on the Available Inputs.” The underlying framework combines physics-augmented correlations with machine-learning methods, with XGBoost serving as a major component of the modeling ensemble.
The published methodology also includes an alternative equation that can be calibrated using a single reference-viscosity measurement to generate highly accurate viscosity-temperature relationships. This validated scientific methodology was subsequently implemented in the freely accessible IPB&F Dead-Oil Viscosity Calculator and now forms the computational foundation of ViscoSense AI.
How ViscoSense AI Works: Instead of requiring users to navigate multiple input forms or independently determine which viscosity model should be applied, users can describe the fluid properties available to them using common petroleum-engineering and PVT terminology. The conversational agent organizes the supplied information and determines the appropriate validated calculation workflow.
- Understands common petroleum-engineering and PVT terminology
- Retains supplied fluid-property information throughout the conversation
- Determines which validated viscosity model is applicable
- Identifies missing or inconsistent input parameters
- Guides the user toward a complete viscosity prediction
- Invokes the established numerical calculation engine to perform the prediction
AI-Assisted, Science-Grounded Computation: ViscoSense AI follows a clear design principle: language intelligence is separated from scientific computation. The large language model functions as an intelligent orchestration layer by interpreting user intent, organizing engineering inputs, checking completeness, and selecting the appropriate calculation workflow. The numerical viscosity prediction itself remains grounded in the published and validated scientific model and is not independently generated by the large language model.
This architecture combines the accessibility of conversational AI with the reliability of a validated engineering calculation engine. ViscoSense AI demonstrates how peer-reviewed petroleum-engineering research can be transformed into an intuitive, scalable, and deployable digital product, while providing a foundation for future expansion into broader PVT calculations and integrated engineering workflows.
📘 Product & API Details:
Sinha, U. (August 30, 2026). ViscoSense AI: The Dead-Oil Viscosity Agent.
Authors: Birol Dindoruk and Utkarsh Sinha
CO₂ Solubility in Brine Calculator
Product of Interaction of Phase-Behavior and Flow (IPB&F) Consortium.
Calculates the CO₂ solubility in brine for different salt types at specific pressure and temperature.
Product Description: Estimates CO₂ solubility in brine systems for different salt compositions under specified pressure and temperature conditions, supporting CCUS, EOR, and reservoir fluid analysis workflows.
Dead Oil Viscosity Calculator
Product of Interaction of Phase-Behavior and Flow (IPB&F) Consortium.
Calculates the Viscosity (cp) of dead oil using Molecular Weight of Stock Tank Oil (MW), API, and Temperature of Interest (°C) using XGB Method.
Dr. Birol Dindoruk
|
Utkarsh Sinha
|
CO₂ Minimum Miscibility Pressure (MMP) Predictor
A product of Interaction of Phase-Behavior and Flow (IPB&F) Consortium.
Developed by – Utkarsh Sinha, Dr. Birol Dindoruk and Dr. M.Y. Soliman
|
Dr. Birol Dindoruk
|
Utkarsh Sinha
|
Product Description: Calculates the Minimum Miscibility Pressure (psia) for pure CO₂ injection.
HC-Gas-MMP App: A Physics-Guided Tool for Rapid Miscibility Screening
The HC-Gas-MMP App is an interactive Streamlit-based web application developed to support screening and design of hydrocarbon gas injection and Enhanced Oil Recovery (EOR) projects. The tool enables rapid estimation of Minimum Miscibility Pressure (MMP) and Minimum Miscible Enrichment (MME)—two critical parameters used to evaluate the feasibility and effectiveness of miscible gas injection. By transforming advanced thermodynamic and data-driven modeling into a lightweight, accessible interface, the app allows engineers, researchers, and students to quickly analyze reservoir and injection-gas scenarios without the need for complex compositional simulation workflows.
|
Dr. Birol Dindoruk
|
Utkarsh Sinha
|
The application is built on the physics-guided, data-driven methodology presented in Physics-Guided Data-Driven Model to Estimate Minimum Miscibility Pressure (MMP) for Hydrocarbon Gases (Sinha, Dindoruk, and Soliman, Geoenergy Science & Engineering). At its core is a Light Gradient Boosting Machine (LightGBM) model enhanced with physics-informed feature engineering to ensure predictions remain consistent with the governing principles of multi-contact miscibility. The workflow requires only readily available inputs such as reservoir temperature, oil and gas compositions, and key fluid properties, while incorporating derived parameters (e.g., characterization factors and pseudo-critical properties) to capture the functional dependence of miscibility.
Beyond technical prediction, the HC-Gas-MMP App supports broader operational, economic, and environmental decision-making. By enabling rapid evaluation of miscible injection feasibility using locally available hydrocarbon gases, the tool helps operators optimize recovery, reduce flaring, and minimize greenhouse-gas emissions. It also supports early-stage screening, injection-program design, and future CCUS planning by providing a practical bridge between academic research and field deployment.
Q&A Chatbot on Physics-Informed Machine Learning
The Q&A Chatbot on Physics-Informed Machine Learning (PIML) was developed as an interactive companion to the review paper
“Review of Physics-Informed Machine Learning (PIML) Methods Applications in Subsurface Engineering”
(Sinha and Dindoruk, 2025, Geoenergy Science & Engineering).
🔑 Access key: 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
|
Dr. Birol Dindoruk
|
Utkarsh Sinha
|
The chatbot transforms the review paper into an interactive knowledge assistant. Instead of manually navigating the paper, users can ask technical questions and receive context-aware answers grounded directly in the source material. This enables faster exploration of complex topics related to physics-informed machine learning in subsurface engineering.
The system is built using a Retrieval-Augmented Generation (RAG) framework combined with FAISS vector search and Group Relative Policy Optimization (GRPO). When a user submits a question, FAISS retrieves the most relevant text sections from the paper, which are then supplied to the language model through RAG to ensure responses remain grounded in the literature. GRPO further improves answer quality through context-aware reinforcement learning and prompt optimization, enhancing clarity, relevance, and factual consistency.
Physics-informed machine learning integrates governing physical laws with data-driven algorithms to improve predictive reliability in complex engineering systems. In subsurface engineering applications, this hybrid framework helps overcome limitations of purely physics-based simulations and purely data-driven models by combining the strengths of both approaches. As a result, PIML methods are increasingly being applied to reservoir characterization, flow modeling, and energy-transition technologies.