Change detection in rotational speed of industrial machinery using Bag-of-Words based feature extraction from vibration signals

Detection of early changes in rotational speed is highly required in on-line process monitoring of industrial manufacturing and numerical controlled machining. This paper proposes a Bag-of-Words (BoW) based feature extraction method that uses vibration signal with such a motivation. Initially, the BoW model is adopted to cluster a prior collection of vibration signals. Then, for a new vibration signal, two strategies: histogram based encoding (HBE) and embedding based encoding (EBE), are investigated respectively and comprehensively to encode it based on the BoW model, in order to extract its dynamic characteristic. The entropy is subsequently computed, supporting continuous analysis of dynamic machine status over time. Distance metric is finally adopted to make decision by hypothesis testing. The method is validated with both simulated and real-engineering signals. Results reveal excellent performance by using EBE, coupled with Kolmogorov distance, in the proposed method. Comparison with state-of-theart competitors demonstrate the priority and robustness of the method. (C) 2019 Elsevier Ltd. All rights reserved.