Investigation of Time Series Data Augmentation Methods for Improving Deep Learning Models in Conveyor Belt Load Classification

This study explores time series data augmentation techniques to enhance the classification accuracy of deep learning models for conveyor belt tension signals. The study examines conveyor belt load data for weights of 0.5 kg, 1 kg, 2 kg, 3 kg, and 5 kg. Various time series data augmentations, including magnitude warping, Laplace noise, Gaussian noise, and uniform noise, are applied and evaluated. Additionally, the TimeVAE model is employed to synthesize new tension signals for the conveyor belt load datasets. The research investigates the impact of time series augmentation methods on the accuracy of deep learning models. A CNN-LSTM deep learning model is developed for classifying signal data from conveyor belt loads. Results indicate that the addition of Laplace noise yields the most significant improvement in classification accuracy, boosting the baseline accuracy (without augmentation) of 0.6 second signal data by 1.85% and of 1.2 second signal data by 0.59% to achieve an accuracy of 94.38%.