Visual Fingerprints of Vibration Signals Using Time Delay Embeddings

Most machines generate vibrations during operation, but effectively visualizing these vibrations is often a challenge, due to large and high-resolution data. Line charts suffer from overplotting, while frequency-domain analysis requires specialized knowledge in signal processing. We introduce a method that bridges the gap between time-domain and frequency-domain analysis: a visual fingerprint computed through the time delay embedding of the vibration data. This fingerprint helps identify segments exhibiting periodic behavior and can be used to cluster similar segments within a vibration signal. Additionally, we demonstrate its practical application in predictive maintenance, showcasing its potential for real-world industrial use.