A Laplacian Image-Guided Semantic Segmentation Framework for Disc Brake Vibration Displacement Measurement

Measuring the vibration displacement of disc brakes is essential since aberrant vibration has a direct impact on braking function and driving safety. In this research, we offer a semantic segmentation-based non-contact vision measurement framework that overcomes the accuracy and adaptability constraints of conventional sensors. Considering that the fine localization of disc brakes depends more on spatial details, we design a wavelet-based down-sampling approach (HWTD), a global-local feature extraction residual structure (GLFE), and a Laplacian image-guided feature fusion network (LIFFN) to achieve effective local information fusion, extraction, and retention. Experiments on video data captured by a high-speed camera under two operating conditions, namely smooth operation and actual braking, show that the proposed method respectively achieves an average <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">RMSE</i> of 12.559 μm and 18.803 μm for the vibration displacement data measured under the two conditions, and performs excellently in both the time and frequency domains compared to existing visual measurement methods.