A Machine Learning Approach to predicting properties of Drilling Fluids

Drilling fluid, also known as drilling mud, is a crucial component in the drilling operation of onshore and offshore oil and gas wells. It is a complex mixture of various chemicals additives, include clays, polymers, and surfactants, to form drilling mud in either water-based or synthetic-based. Drilling fluid serves several major drilling functions, including cooling and lubricating the drill bit, carrying rock cuttings to the surface, and maintaining the formation pressure in the wellbore to prevent blowouts. To enable a safe and effective well construction process, the parameters of the drilling fluid, such as its viscosity, density, and pH, must be strictly controlled and modified during drilling operations. The demand for better drilling fluid technology especially when including the application of data analytics has grown increasingly critical in the business as the demand for oil and gas rises.

Nowadays, the application of machine learning in drilling fluid is becoming increasingly important in the oil and gas industry. Machine learning can be used to evaluate vast amounts of data from drilling operations, including data collected on the properties of the drilling fluid, such as its viscosity, mud weight, emulsion stability and pH, as well as data on the geology of the well-bore. By using this data, machine learning algorithms can predict the behavior of drilling fluid under different conditions, and optimise the characteristics of the the fluid to improve drilling performance, and save cost and time.

Machine Learning Approach in Determining Viscosity

Let's use viscosity and gel strength as an example. In order to maintain good drilling mud viscosity, it is a critical properties to evaluate the mud's capacity to suspend drilled cuttings and the solid content in the mud based on its yield point (YP) and plastic viscosity (PV), as Bingham-plastic properties is commonly applied to explain its mud fluid behaviour. Placing a sample of the drilling fluid in the viscometer and measuring its resistance to flow (e.g., 600, 300, 200, 100, 60, and 30 dial readings) while being subjected to shear stress are the typical steps in the technique for establishing the yield point and plastic viscosity using a viscometer. The viscometer measures the fluid's shear rate and shear stress after applying a shear stress to it, typically by rotating a spindle or bob. The flow curve, which describes the relationship between shear stress and shear rate, can be used to calculate the yield point and fluid plastic viscosity.

To modify the viscosity of the drilling fluid, the specific amount of viscosifier will be added to a drilling fluid to achieve its desired viscosity. Viscosifiers are chemicals that are added to the drilling fluid to increase its viscosity and improve its ability to transport rock cuttings out of the wellbore. Bentonite is the common viscosifier that been used in the drilling formulation. From here, the relationship between the amount of viscosifer (Primary or secondary) added and the dial reading, yield point and plastic viscosity value is confirmed. The viscosity of the fluid increases together with the amount of viscosifier added into it during the formulation. As a result of the viscosifier molecules' interactions with the fluid, which result in the formation of continuous chains, the viscosity is increased along with the flow resistance.

So, without the need for a viscometer or testing, machine learning can be used to create the best regression model that predict the viscosity of drilling mud depending on the quantity of bentonite, or other chemical additives that might also affecting the viscosity, injected to the mud. A sizable dataset of measurements of the drilling fluid's viscosity, type of drilling mud, mud density, and the appropriate amounts of bentonite or other chemical additives that might alter the viscosity added must be gathered in order to prevent undertrained condition, and create a suitable machine learning regression model with sufficient data. A machine learning algorithm can be trained using the dataset to identify obvious or hidden patterns in the data and generate predictions based on the patterns and their relationships. Once trained, the model may be used by drilling fluid engineers to forecast viscosity with high accuracy by simply feeding it the chemical formulation amount.This can significantly enhance drilling efficiency, lower the danger of drilling issues like hole collapse, and spare engineers the time-consuming task of determining the viscosity value through repetitive and lengthy process of lab testing methods.

Machine Learning Approach in Determining Gel-Strength

Other than that, gel-strength properties can be also predicted by using Regression Model approach. Gel strength in drilling fluid is a measure of the strength of the fluid's gel-like structure. The capacity of the drilling fluid to suspend and transport rock shavings out of the wellbore as well as the minimum horsepower required by the drill string to initiate the rotation force necessary to ignore fluid resistance during drilling operations are both impacted by this fundamental rheological property. To minimise the additional horsepower required for drilling, the desirable mud will typically have stable constant gel-strength value with non-progressive qualities. In lab testing, 3 different duration of gel-strength tests (10 seconds, 10 minutes, and 30 minutes) will be conducted. The reason to carry out 3 different duration test is to evaluate the fluid’s behaviour via gel-strength properties during real drilling operations, especially when changing connection pump (10s), short trips (10m) and running casing process which include casing installation and cementing (30m). When a large amount of mud needs to be tested to establish the gel strength, conducting these experiments in the lab can be tedious and repetitive as each mud test might take at least 40 minutes and 10 seconds!

Therefore, a machine learning algorithm could be trained using data from past drilling operations to identify the relationship between mud properties, drilling parameters, and gel-strength. The algorithm may then be applied to assess data from ongoing drilling operations in real-time and forecast the drilling fluid's gel strength depending on the input data. The drilling fluid's gel strength can be also predicted by refering patterns and trends that can be found using this model. This could entail looking at data from numerous wells to determine common causes of gel-strength variations or looking at data from various drilling conditions to determine causes of gel-strength in various geological formations.

Conclusion

The accuracy and effectiveness of the drilling process might be greatly increased by using machine learning to the determination of drilling fluid parameters. It is possible to find patterns and links between drilling parameters, mud qualities, and the characteristics of not only the gel strength, viscosity, and yield point of drilling fluid, but also pH, emulsion stability value, and even predicting the amount of additives required to reach desired mud properties, by applying machine learning algorithms to evaluate huge volumes of data from numerous sources. However, proper training of the model must be carried out to ensure an effective model created with high accuracy. The data source must be low in correlation error to make sure the patterns and trends forecasted is accurate. The future of the drilling business and the way drilling operations are carried out show enormous promise for the use of machine learning in identifying drilling fluid parameters.