Structural Health Monitoring - deep learning approach
Rapport, 2021

The development of the sensor on Structural Health Monitoring (SHM) provides some useful data that indicates the condition of the structures. A data-driven approach as an option for achieving the goal of SHM in predicting structural conditions and obtaining useful information from the numerical data. This thesis introduces deep learning methods to perform supervised learning on damage conditions in SHM.

Deep learning methods such as one-dimensional Convolutional Neural Networks (1D-CNN) and Long Short-term Memory (LSTM) are applied to predicting crack position, crack width and deflection of the concrete beams. A Linear Regression (LR) is also investigated to compare with the deep learning models.

Given multidimensional time-series strain data that simulated from finite element methods and the labeled crack positions, 1D-CNN and LSTM models are proposed to handle the binary classification problem. The result shows that an LSTM model is a more promising model than a 1D-CNN model on crack position prediction while handling multidimensional input and output and time-series classification. LSTM model could be a potential solution to achieve automatic monitoring on structural health with only using strain data obtained from DOFS.

In predicting crack width and deflection, a predictive model as LR is also a promising method for solving the regression problem. While exploring different sets of input variables for the LR model, such as strain and geometry variables as inputs, only training with strain data results in a better performance on prediction. 1D-CNN and LSTM models are also implemented and evaluated for comparison with the LR model, which achieved good performance results.


Weng Hang Wong

Linköpings universitet

SensIT – Sensorstyrd molnbaserad förvaltningsstrategi av infrastruktur

Thomas Concrete Group, 2018-07-01 -- 2020-08-31.

WSP Sverige, 2018-07-01 -- 2020-08-31.

Trafikverket (2018/27871), 2018-07-01 -- 2020-08-31.

NCC AB, 2018-07-01 -- 2020-08-31.

Microsoft Research, 2018-07-01 -- 2020-08-31.


Hållbar utveckling


Building Futures (2010-2018)



Sannolikhetsteori och statistik

Mer information