Steady State Hydraulic Valve Fluid Field Estimator Based on Non-Dimensional Artificial Neural Network (NDANN)

Abstract

An automatic transmission (AT) hydraulic control system includes many spool-type valves that have highly asymmetric flow geometry. A simplified flow field model based on a lumped geometry is computationally efficient. However, it often fails to account for asymmetric flow characteristics, leading to an inaccurate analysis. An accurate analysis of their flow fields typically requires using the computational fluid dynamics (CFD) technique, which is numerically inefficient and time consuming. In this paper, a new hydraulic valve fluid field model is developed based on non-dimensional artificial neural networks (NDANNs) to provide an accurate and numerically efficient tool in AT control system design applications. A grow-and-trim procedure is proposed to identify critical non-dimensional inputs and optimize the network architecture. A hydraulic valve testing bench is designed and built to provide data for neural network model development. NDANN-based fluid force and flow rate estimators are established based on the experimental data. The NDANN models provide more accurate predictions of flow force and flow rates under broad operating conditions (such as different pressure drops and valve openings) compared with conventional lumped flow field models. Because of its non-dimensional characteristic, the NDANN fluid field estimator also exhibits good input-output scalability, which allows the NDANN model to estimate the fluid force and flow rate even when the operating condition parameter or design geometry parameters are outside the range of the training data. That is, although the operating/geometry parameter values are outside the range of the training sets, the non-dimensional values of the specific operating/geometry parameters are still within the training range. This feature makes the new model a potential candidate as a system design tool.