Impact Force Identification on Steel Pipelines Using Deep Learning and Bayesian Inference with Minimal Sensing Elements

Abstract

Pipelines are crucial for various industries but are vulnerable to impact damage, making it essential to identify impact forces for maintaining structural integrity and preventing accidents. Traditional inverse methods for impact force identification can be expensive, time-consuming, and require pipeline shutdowns. To address these challenges, this study proposes a novel approach using deep learning and Bayesian inference techniques with minimal sensing. Four accelerometers were placed on a 6- meter-long steel pipeline, and data was collected, pre-processed, and split into training and testing sets. A 5-layered ANN model was trained, achieving 87.2% accuracy, which was then used as a surrogate model for the Bayesian inference. The proposed approach, based on Approximate Bayesian Computing with Subspace Simulation (ABC-SS), demonstrated a reliable and robust solution for identifying impact forces on pipelines.