Abstract

Advanced computing brings opportunities for innovation in a broad gamma of applications. Traditional practices based on visual and manual methods tend to be replaced by cyber-physical systems to automate processes. The present work introduces an example of this, a machine vision system research based on deep learning to classify bridge load, to give support to an optical scanning system for structural health monitoring tasks. The optical scanning system monitors the health of structures, such as buildings, warehouses, water dams, etc. by the measurement of their coordinates to identify if a coordinate displacement befalls that could indicate an anomaly in the structure that can be related to structural damage. The use of this optical scanning system to monitor the structural health of bridges is a little more complicated due to the vehicle’s transit over the bridge that causes a vehicle-bridge interaction which manifests as a bridge oscillation. Under this scheme, the bridge oscillation corresponds to their coordinate’s displacement due to the vehicle-bridge interaction, but not necessarily due to bridge damage. So, a bridge load classifier is required to correlate the bridge coordinates measurements behavior with the bridge oscillation due to vehicle-bridge interaction to discriminate the normal behavior of the structure to abnormal behavior or identify tendencies that could indicate bridge deformation or discover if the bridge behavior due to loads is changing through the time.

中文翻译：

---

基于深度学习的桥梁位移分类器用于结构位移关联

摘要

先进的计算为广泛的应用带来了创新的机会。基于视觉和手动方法的传统实践往往被网络物理系统所取代，以实现流程自动化。本工作介绍了一个示例，这是一个基于深度学习的机器视觉系统研究，以对桥梁载荷进行分类，从而为用于结构健康监测任务的光学扫描系统提供支持。光学扫描系统通过测量其坐标来监视建筑物，建筑物，仓库，水坝等结构的健康状况，以识别坐标位移是否下降，这可能表明结构异常可能与结构损坏有关。使用光学扫描系统来监视桥梁的结构健康情况会稍微复杂一些，这是因为车辆在桥梁上的行驶会导致车桥相互作用，表现为桥梁振动。在这种方案下，由于车桥相互作用，桥梁振动对应于其坐标的位移，但不一定由于桥梁损坏。因此，桥梁载荷分类器需要将桥梁坐标测量行为与因车桥相互作用而引起的桥梁振动相关联，以将结构的正常行为与异常行为区别开，或者识别出可能指示桥梁变形或发现桥梁行为的趋势。由于负载随时间而变化。