The paper proposes a new methodology for revamping design and optimization of a process piping system. Starting from ASME B31.3 Process Piping prescriptions for stress analysis, a nonlinear model is built to express the relationship between stress distribution generated by expansion and sustained loads (pressure, weight) and the geometry and routing of the pipeline, focusing on geometric parameters of expansion loops. The number of design variables affecting stress distribution over the pipe, together with the constraints to be respected, would make it hard to formulate an optimization procedure based on deterministic methods. This problem is overcome by applying a Feed Forward Neural Network, backpropagation trained, which makes it possible to interpolate a non-linear and multidimensional relation over a domain enclosed within the boundaries of a training set. Prediction of code stresses is obtained through the fitting of an artificial neural network for each examined loadcases. Network parameters are tuned offline, starting from a set of data obtained by finite element numerical simulation. As a result, an optimal geometry for expansion loops is found, allowing to revamp pipe routing by halving loops number and keeping code stress within the allowable limits.

Revamping Optimization of a Pressure Piping System Using Artificial Neural Networks

Caponetto, Riccardo;Fargione, Giovanna;Giudice, Fabio
;
2022-01-01

Abstract

The paper proposes a new methodology for revamping design and optimization of a process piping system. Starting from ASME B31.3 Process Piping prescriptions for stress analysis, a nonlinear model is built to express the relationship between stress distribution generated by expansion and sustained loads (pressure, weight) and the geometry and routing of the pipeline, focusing on geometric parameters of expansion loops. The number of design variables affecting stress distribution over the pipe, together with the constraints to be respected, would make it hard to formulate an optimization procedure based on deterministic methods. This problem is overcome by applying a Feed Forward Neural Network, backpropagation trained, which makes it possible to interpolate a non-linear and multidimensional relation over a domain enclosed within the boundaries of a training set. Prediction of code stresses is obtained through the fitting of an artificial neural network for each examined loadcases. Network parameters are tuned offline, starting from a set of data obtained by finite element numerical simulation. As a result, an optimal geometry for expansion loops is found, allowing to revamp pipe routing by halving loops number and keeping code stress within the allowable limits.
ASME B31.3; expansion loop; neural network; piping design; stress analysis
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.11769/545642
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 0
  • ???jsp.display-item.citation.isi??? ND
social impact