Neural Networks for Statistical Modeling
Neural Networks for Statistical Modeling
| Hi-index | 0.00 |
This paper proposes an analytical tool that supports the design process of a disc-spring valve system used in hydraulic dampers. Such a valve is representative of a broader family of valve systems used in many industrial applications. The tool will speed up the valve-system design process.The paper considers a valve system that combines a number of circular metal plates, referred to further in the paper as 'a stack of plates'. The proposed analytical tool obtains a key design characteristic of a valve, the flow rate, and the corresponding maximum stress level in the stack of plates as a function of pressure load. The stress level enables the ranking of valve settings in terms of durability. The calculation process is based on the response of a nonlinear, data-driven model that approximates the a priori simulated cases to cover the complete range of input design parameters, namely the number of plates, their thickness and their diameter. The cases are produced by a first-principle model using a Finite Element (FE) approach. The model was calibrated based on experimental results to provide accurate results in the entire range of input parameters. The advantage of the analytical tool is its ability to immediately provide the pressure-stress-flow characteristic of a valve instead of repeating time-consuming calculations for each new setting of input parameters. The objectives of the paper are as follows: to (a) adapt the model to a three-plate stack by applying large strain theory, (b) select and rank the Artificial Neural Networks used in order to approximate simulation data, and (c) demonstrate validation and application results obtained with the Approximation Tool.