108 C. Q. G. Mun˜oz et al. Fig. 3 Location of different bumpers on the plate to generate abnormal vibrations. Fig. 4 Video image showing the movement of the place through pseud color representation. • Video Generation for Each Case: Using the interpolated data, a video was generated for each case. This visual representation allowed for a clearer understanding of the plate’s dynamics under different conditions. • Trimming video duration to 20 seconds: It was determined that beyond this time, the videos did not provide additional relevant information. By removing segments that did not contribute significantly, processing was streamlined, and unnecessary computational resource consumption was avoided. • Balancing the number of videos per frequency and class: The dataset was unified to have a homogeneous distribution, obtaining one video per frequency for each class (bumper). This allowed for better generalization during training and avoided biases towards classes with more data.
RkJQdWJsaXNoZXIy MTMzNzEzMQ==