Effective alerting for bridge monitoring via a machine learning-based anomaly detection method

Structural Health Monitoring, Год журнала: 2024, Номер unknown

Опубликована: Авг. 24, 2024

To alert upcoming structural failure is a critical task for health monitoring of bridges. Traditional methods mainly rely on thresholds, which are often fixed values and may cause missing or too sensitive reports. Identifying abnormal data, locating the source anomalies delivering proportional alerts require new, dynamic, robust algorithms running massively streaming data. This article proposes new machine learning-based anomaly detection method historical data mining as well real-time alerting. The transforms one-dimensional time series into two-dimensional tensors, enabling encoder-like model to simultaneously learn changes in multiple sensors within between temporal cycles space. Training validation proposed presented with from bridge system service, comparisons against traditional threshold-based alerting made. can accurately identify abnormalities beyond thresholds effectively detect deviations sensors, thus constituting promising module systems

Язык: Английский

---

From Binary to Multi-Class: Neural Networks for Structural Damage Classification in Bridge Monitoring Under Static and Dynamic Loading

Dynamics, Год журнала: 2024, Номер 4(4), С. 786 - 803

Опубликована: Окт. 25, 2024

Structural Health Monitoring (SHM) plays a vital role in ensuring the health status of wide range structures, such as bridges, buildings, and large infrastructure general. The advantages this process can be further enhanced by incorporating more numerical statistical approaches into traditional methods, finite element analysis Machine Learning. In study, truss bridge structure is examined, neural networks are trained with data derived from analyses under static loads dynamic excitations. contributions work based on comparing analyses, well deriving important insights key parameters that impact their performance SHM. Initially, binary classification problem addressed, where numerically classifiers tasked identifying whether healthy state or not. This category divided two subcategories, depending extent damage present structure. Subsequently, multi-class defined, three different classes same considered, network required to distinguish between them. Although training all was highly satisfactory, prediction results varied, success rates ranging 55% 90%. Finally, conclusions drawn study regarding model error influence, size, types used.

Язык: Английский

---

Bolt Loosening and Preload Loss Detection Technology Based on Machine Vision

Buildings, Год журнала: 2024, Номер 14(12), С. 3897 - 3897

Опубликована: Дек. 5, 2024

Steel bridges often experience bolt loosening and even fatigue fracture due to load, forced vibration, other factors during operation, affecting structural safety. This study proposes a high-precision key point positioning recognition method based on deep learning address the high cost, low efficiency, poor safety of current identification methods. Additionally, angle is proposed digital image processing technology. Using data, angle-preload curve revised. The established correlation between pretension for commonly utilized high-strength bolts provides benchmark identifying angles. finding lays theoretical foundation defining effective detection intervals in future systems. Consequently, it enhances system’s ability deliver timely warnings, facilitating swift manual inspections repairs.

Язык: Английский

---

Conditional generative adversarial networks for the data generation and seismic analysis of above and underground infrastructures

Tunnelling and Underground Space Technology, Год журнала: 2024, Номер 157, С. 106285 - 106285

Опубликована: Дек. 13, 2024

Язык: Английский

---

Iterative-Based Impact Force Identification on a Bridge Concrete Deck

Sensors, Год журнала: 2023, Номер 23(22), С. 9257 - 9257

Опубликована: Ноя. 18, 2023

Steel-reinforced concrete decks are prominently utilized in various civil structures such as bridges and railways, where they susceptible to unforeseen impact forces during their operational lifespan. The precise identification of the events holds a pivotal role robust health monitoring these structures. However, direct measurement is not usually possible due structural limitations that restrict arbitrary sensor placement. To address this challenge, inverse emerges plausible solution, albeit afflicted by issue ill-posedness. In tackling ill-conditioned challenges, iterative regularization technique known Landweber method proves valuable. This leads more reliable accurate solution compared with traditional methods it is, additionally, suitable for large-scale problems alleviated computation burden. paper employs perform comprehensive force encompassing localization time–history reconstruction. incorporation low-pass filter within Landweber-based procedure proposed augment reconstruction process. Moreover, standardized error metric presented, offering effective means accuracy assessment. A detailed discussion on placement optimal number iterations presented. automatedly localize force, Gaussian profile proposed, against which reconstructed compared. efficacy techniques illustrated utilizing experimental data acquired from bridge deck reinforced steel beam.

Язык: Английский

---