Traditional torque converter modeling approaches rely on look-up tables derived from steady-state data, which fail to capture the transient behaviors. This paper employs a fully connected neural network to predict the fluid torque, which incorporates both angular velocities and angular accelerations as inputs, enhancing the prediction accuracy of pump and turbine torques in transient conditions. Through extensive training and validation, the proposed method demonstrates superior performance compared to previous models that only considered angular velocities, achieving a mean squared error less than 1. This approach not only mitigates the uncertainties associated with parameter measurement but also provides a robust framework for understanding torque converter dynamics under varying operational conditions.