Early and timely detection of surface damages is important for maintaining the functionality, reliability, and safety of concrete bridges. Recent advancement in convolution neural network has enabled the development of deep learning‐based visual inspection techniques for detecting multiple structural damages. However, most deep learning‐based techniques are built on two‐stage, proposal‐driven detectors using less complex image data, which could be restricted for practical applications and possible integration within intelligent autonomous inspection systems. In this study, a faster, simpler single‐stage detector is proposed based on a real‐time object detection technique, You Only Look Once (YOLOv3), for detecting multiple concrete bridge damages. A field inspection images dataset labeled with four types of concrete damages (crack, pop‐out, spalling, and exposed rebar) is used for training and testing of YOLOv3. To enhance the detection accuracy, the original YOLOv3 is further improved by introducing a novel transfer learning method with fully pretrained weights from a geometrically similar dataset. Batch renormalization and focal loss are also incorporated to increase the accuracy. Testing results show that the improved YOLOv3 has a detection accuracy of up to 80% and 47% at the Intersection‐over‐Union (IoU) metrics of 0.5 and 0.75, respectively. It outperforms the original YOLOv3 and the two‐stage detector Faster Region‐based Convolutional Neural Network (Faster R‐CNN) with ResNet‐101, especially for the IoU metric of 0.75.

Wavelet transform (WT) is a mathematical tool that can decompose a temporal signal into a summation of time-domain basis functions of various frequency resolutions. This simultaneous time-frequency decomposition gives the WT a special advantage over the traditional Fourier transform in analyzing nonstationary signals. One drawback of the WT is that its resolution is rather poor in the high-frequency region. Since structural damage is typically a local phenomenon captured most likely by high frequency modes, this potential drawback can affect the application of the wavelet-based damage assessment techniques. The wavelet packet transform (WPT) adopts redundant basis functions and hence can provide an arbitrary time-frequency resolution. In this study, a WPT-based method is proposed for the damage assessment of structures. Dynamic signals measured from a structure are first decomposed into wavelet packet components. Component energies are then calculated and used as inputs into neural network models for damage assessment. Numerical simulations are performed on a three-span continuous bridge under impact excitation. The results show that the WPT-based component energies are good candidate indices that are sensitive to structural damage. These component energies can be used for various levels of damage assessment including identifying damage occurrence, location, and severity.

The dynamic behavior of a magnetorheological (MR) damper is well portrayed using a Bouc–Wen hysteresis model. This model estimates damper forces based on the inputs of displacement, velocity, and voltage. In some control applications, it is necessary to command the damper so that it produces desirable control forces calculated based on some optimal control algorithms. In such cases, it is beneficial to develop an inverse dynamic model that estimates the required voltage to be input to the damper so that a desirable damper force can be produced. In this study, we explore such a possibility via the neural network (NN) technique. Recurrent NN models will be constructed to emulate the inverse dynamics of the MR damper. To illustrate the use of these NN models, two control applications will be studied: one is the optimal prediction control of a single-degree-of-freedom system and the other is the linear quadratic regulator control of a multiple-degree-of-freedom system. Numerical results indicate that, using the recurrent NN models, the MR damper force can be commanded to follow closely the desirable optimal control force.