Term 252

AI MUD PROPERTY PREDICTOR & DOSE RECOMMENDOR

Project Type: Self-Initiated

Project Description

explainable machine learning-based decision support system for the formulation of water-based drilling muds. The formulation of drilling muds has traditionally been a trial-and-error process. A set of ingredients is mixed according to a recipe, and laboratory testing is performed on the mixture. The recipe is then revised, and the mixture is re-made and re-tested, and so on, until the rheological and fluid loss properties are within the desired range. The new system will predict the plastic viscosity, yield point, gel strengths at 10 seconds and 10 minutes, and API fluid loss of the drilling mud, all as a function of the recipe ingredients. The system will then suggest bounded doses to aid the engineer in finding the target range. The final design of the system will be a web-based system with two modes: Evaluate Recipe, which will have prediction and explanation, and Hit Targets, which will have a dose recommender. The system design will incorporate the following: a dataset with data validation gates, CPU-friendly machine learning inference, chemical and operation constraints, audit trails, and user interface. The system will be validated according to a set of specifications, including prediction accuracy, system latency, target hit ratio, and polymer reduction.