Similarity Based Methods for Faulty Pattern Detection in Predictive Maintenance
Somayeh Bakhtiari Ramezani, Logan Cummins, Brad Killen, Richard Carley, Shahram Rahimi, Maria A. Seale · 2021 International Conference on Computational Science and Computational Intelligence (CSCI) · 2021
Detecting similarity between instances in large datasets is key to accurately predicting faulty patterns in Predictive Maintenance (PM). Most of the existing methods use the whole data to train a clustering or classification model, resulting in lower accuracy, especially in noisy data. Similarity-based methods (SBM) aim to increase the performance of data-driven algorithms by selecting the most similar training instances as the prototypes. This study evaluates the state-of-the-art SBMs used in the PM and summarizes the main challenges yet to be considered. Based on our observations, Dynamic Time Warping and Longest Common Sub Sequence are the most promising methods in PM.