Comparison of neural networks based on accuracy and robustness in identifying impact location for structural health monitoring applications

Abstract

Structural health monitoring systems must provide accuracy and robustness in predicting the structure’s health using the minimum intervention to ensure commercial viability. Characterization of impact is useful in assessing its severity, deciding if detailed damage analysis is necessary, and re-evaluating the present health of the structure under monitoring with better confidence. In this characterization process, the impact location is significant since some positions within a structure are more sensitive to damage. The inherent noise and uncertainties present in the sensor response pose a substantial hurdle to estimating the external impact correctly. This paper quantitatively compares three of the widely used neural networks, namely, Artificial Neural Network (ANN), Convolutional Neural Network (CNN), and Long Short-Term Memory network (LSTM), to estimate impact location from the lead zirconate titanate (PZT) sensor response. For this purpose, a square aluminum plate of 500 × 500 mm was equipped with four PZT sensors; each placed 100 mm away in both the plate directions from a corner and impact loads were given on a grid covering the whole plate. The PZT responses were used to train the three neural networks under study here, and their estimations were compared based on the Mean Absolute Error (MAE). In addition, increasing Gaussian noise was added to the PZT responses, and the robustness of the three neural networks was monitored. It was found that the ANN gives better accuracy with a Mean Absolute Error of 22 mm compared to Convolutional Neural Network (MAE = 31 mm) and Long Short-Term Memory (MAE = 25 mm). However, CNN is more robust when encountering noise with a 2% reduction in accuracy, while LSTM and ANN lost 7% and 11% accuracy, respectively.

A shallow 2D-CNN network for crack detection in concrete structures

“…The safety of the structure and building can also be considered among these items. Many researchers believe that the accuracy and precision of structural health monitoring can be provided with techniques based on deep learning (Abbas et al ., 2023; Balasubramanian et al ., 2023; Nevarez et al ., 2023; Pan et al ., 2023; Sabato et al ., 2023). But as shown in the Related work, less research has emphasized the aspects of execution time and reducing the number of calculations.…”

An Impact Localization Solution Using Embedded Intelligence—Methodology and Experimental Verification via a Resource-Constrained IoT Device

“…For the same experimental setup, Balasubramanian et al [ 14 ] developed and validated a convolutional neural network (CNN), a long short-term memory (LSTM) network, and an artificial neural network (ANN), with respect to mean absolute error (MAE) metric. Although, the accuracy of these networks is reduced compared to the SNN (when being shallower and wider), are found to be more robust in cases where noise is present.…”

“…Azimi et al [ 26 ], provide an extensive list of those publications that do share their datasets (with vibration and mostly vision-based data) that have been recently used in deep learning-based SHM. However, for impact detection and localization in plate structures with PZT sensors, which is also a very well studied problem in the literature (for example, [ 13 , 14 , 15 , 27 , 28 ]), there are no openly available datasets.…”

Piezoelectric-array-based MISO diagnostic system for CNN-condition monitoring of bearing/gearbox instruments

“…The regression approach is based on the minimization of MSE (mean square error) loss of target coordinates positioned by the model. 22 The classification method is based on minimizing the cross entropy loss of the output. Both 1D CNN 23 and 2D CNN 24 were used for defect detection and localization.…”