Adaptive Transient Event Detection for Industrial Applications

Foreign elements may destroy sensitive parts of mobile industrial machines such as harvesters. When such undesired transient events shall be detected, varying machine noise scenarios demand adaptive algorithms that allow for a robust detection performance. Moreover, a detection system that is too responsive reduces the machine's speed of operation. We have evaluated three algorithms that are capable of detecting transient events, and that allow for timely precautionary measures. Two of the methods apply a fixed and adaptive threshold to the short-time energies of the high-pass filtered sensor signal, respectively, while a new method employs linear prediction-based filtering and an adaptive frame-energy threshold, and incorporates the variance of the high-frequency frame content enabling the distinction between events resulting from foreign elements and events originated by the machine. The algorithms were applied to four types of transient events that were combined with a set of machine noise recordings at different signal-to-noise-ratio (SNR) levels. Our results show that the new method provides 95 % correct detections down to a SNR of -1 dB, and that all methods provide a very low rate of misdetected events.