Deep learning approaches for multi-modal sensor data analysis and abnormality detection

Abstract

The information gathered within the structure health monitored (SHM) device would display a range of irregularities mainly a result of sensing defeat, noise disturbance, and different causes. This will greatly impair the structure's security evaluation. This research presents a multipurpose deeper neural network-based data-driven abnormality diagnostic system called SHM. The multipurpose deeper neural networks fuse single-dimensional as well as two-dimensional properties regarding the sensory signals to increase the detecting efficiency. Two separate Convolutional Neural Network, streams within the system are used for obtaining time-frequency characteristics from information collected by sensors (also referred to as two-dimensional-CNN medium) as well as unprocessed one-dimensional characteristics (also referred to as one-dimensional-CNN medium). Following the 2D as well as 1D streams' individual clustering and filtering processes using the sensing information, the two categories of recovered properties have been distorted through single-dimensional matrices that combined within the fusion level. The ideal framework shows the efficacy as well as potential of the suggested approach having a precision percentage of 95.10%. Considering an accurate AI-assisted electronic instrument for evaluating security in structured health management networks, the suggested approach has an exciting period ahead of it.

Keywords

• Deep learning
• Multi-modal sensor data
• Abnormality detection
• Sensor fusion
• Neural networks
• Feature extraction