Statistical Distribution
Fourth-order statistical signal processing tool mapping the non-Gaussian nature of an oscillating signal across discrete frequency bands identifies hidden transient impulses within operating machinery. Spectral kurtosis reveals repetitive mechanical shock events that remain completely masked by ordinary background noise in vibrating paper mill components. The technique applies directly to condition monitoring of roll bearings and drive gearboxes across continuous paper and board production machines.
Transient Detection
Mechanical impacts from early-stage bearing fatigue spalls generate brief shock waves that excite natural structural resonances. Conventional frequency spectrums miss these micro-impacts because their energy distributes across broad noise floors created by pulp agitation and rolling nip friction. Spectral kurtosis computes the normalized kurtosis for every individual frequency bin, producing sharp statistical peaks wherever transient, non-stationary impulses dominate.
Vibration specialists use these sharp peaks to design optimal band-pass filters that isolate the resonance band containing the defect signal for envelope demodulation. Filtering raw signals around these kurtosis maxima reveals bearing defect frequencies with high signal clarity.
Machine Preservation
Paper mills avoid unplanned machine shutdowns by deploying this diagnostic method on high-speed press sections and calender bearings. Spectral kurtosis detects bearing outer race fractures weeks before thermal degradation or high overall vibration amplitudes show up on standard maintenance gauges. Identifying faults early permits scheduled replacement during planned wire and felt changes, preventing catastrophic paper web breaks.
Continuous automated monitoring relies on this statistical algorithm to protect capital equipment.