Modeling of Fluid Content in Vibratory Screening Residual Solids Using Artificial Neural Networks

Abstract

The proper management of drilling waste, particularly the fluid content in residual solids from shale shakers, remains a critical challenge in oil and gas operations. Traditional methods relying on laboratory analysis introduce significant delays, hindering real-time process optimization. This study proposes an artificial neural network (ANN)-based virtual sensor to predict fluid content in vibratory screening residual solids in real time. Experimental data were collected from an industrial shale shaker system under varied operational parameters, including motor speed, feed flow rate, and screen inclination. A multilayer perceptron model was developed using TensorFlow, featuring input normalization, dropout regularization, and optimized training with stochastic gradient descent. The ANN architecture achieved a mean absolute error of 0.03 and a loss of 0.002, demonstrating robust convergence without overfitting. Statistical validation via t-tests confirmed no significant difference between predicted and experimental values (p-values of 0.67 for test data and 0.85 for the full dataset). The model’s accuracy under stable operating conditions enables continuous monitoring without additional hardware, addressing the industry’s reliance on delayed laboratory measurements. Key implications include real-time operational adjustments, reduced waste management costs, and a scalable solution for existing systems. This work bridges a critical gap in Artificial Intelligence (AI) applications for solids control, offering a practical framework for enhancing separation efficiency and sustainability in drilling operations.