A Machine Learning Model for the Development of a Digital Twin for a Control Valve for Oil and Gas Pipelines

Control valves are integral to the piping infrastructure in most process industries including nuclear, oil and gas, and chemical. They serve as pressure, temperature, and flow rate regulators. Hence detailed information about these parameters within the complex valve shape is particularly useful for design, optimisation, and efficient operation of control valves. For better reliability and operation under safety-critical installations, extensive research is being carried out to develop multi-layer and knowledge-intensive digital twins for such valves. This paper reports a computational investigation to determine the local valve flow coefficient (Cv) at various valve opening positions (VOP)/flow rate conditions and develop an artificial neural network (ANN) model to predict the local Cv distribution within the valve-pipe system. This is expected to provide sufficient information to embed extensive knowledge base in the valve’s digital twin. Ten internal locations (at L/D=1 to 10) were chosen for data extraction to give a detailed Cv profile from the pipe inlet through the valve to the pipe outlet. Three VOPs (20, 60, and 100%) were considered and for each, three inlet velocities of 1, 2 and 3 m/s were simulated. This combination resulted in a database of 90 data-points which were used to train the ANN using the Levenberg-Marquardt algorithm. MATLAB’s Neural Network Toolbox was used with 70%, 15%, and 15% of the data used for training, validation, and testing, respectively. Model performance was measured using the mean-square-error and square-of-residuals (R2) metrics. Sensitivity analysis of hidden layer neuron number was conducted, and the results show that 10 neurons produced a robust prediction of the CV profile. An overall average R2 of 0.92 was obtained. Hence, it may be efficiently used for the valve’s internal (rather than global) CV prediction for design and other purposes – including implementation within digital twin architectures for remote/condition monitoring in difficult to reach areas such as offshore and within complex industrial sites.