OhmNet: Advanced neural network-based viscosity prediction of sauces for efficient Ohmic heating processing

Industrial food processes such as Ohmic Heating (OH) are gaining popularity due to their lower carbon emissions and improved energy efficiency. The effectiveness of OH largely depends on the electrical conductivity, physical properties, and rheological characteristics of the food product, with dynamic viscosity directly influencing the fluid flow, residence time, and heating rate in a Continuous Flow Ohmic Heating (CFOH) system. Therefore, accurate prediction of viscosity during CFOH processing is crucial for optimising heating efficiency and maintaining the desired output temperature, ultimately reducing energy consumption and operational costs. To address this challenge, this study introduces OhmNet - an advanced Neural Network (NN)-based predictive model designed to accurately estimate the dynamic viscosity of tikka sauce during OH, offering a robust solution for viscosity prediction in CFOH applications. The predictive model has been developed using real-time data obtained from heating experiments, where viscosity measurements were recorded using a rheometer at varying target temperatures. To achieve the optimal configuration of OhmNet, three different approaches were explored: separate network development for each target temperature, a transfer learning-based neural network, and a one-hot encoding-based unified neural network model. These approaches were systematically evaluated through a grid search for hyperparameter tuning to identify the most accurate and robust dynamic viscosity predictive model during Continuous Flow Ohmic Heating. The resulting OhmNet model demonstrates high performance and reliability, achieving a Mean Squared Error (MSE) of 0.002, a Mean Absolute Error (MAE) of 0.025, and a coefficient of determination (R²) equal to 0.99. This optimal configuration of OhmNet offers a powerful tool for enhancing process efficiency and control in industrial food processing applications. In the future, the model can be seamlessly integrated with advanced process controllers for precise temperature control and power consumption optimisation, driving sustainable and energy-efficient food processing applications.