Particle Swarm Optimization-Based Adaptive Wavelet Threshold Denoising for Vibrating-Screen Bolt Vibration Signals in Combine Harvesters

Bolted connections in combine-harvester vibrating screens are affected by reciprocating screen motion, frame vibration, and intermittent impact, making weak local vibration components difficult to distinguish from background fluctuations. This study used particle swarm optimization to adaptively select the wavelet basis, decomposition level, and thresholding rule for vibration signal denoising. Triaxial acceleration signals were collected under tightened- and loosened-bolt conditions. Gaussian white noise with input signal-to-noise ratios from 0 to 14 dB was added to evaluate parameter selection under controlled noise levels. The selected wavelet basis and decomposition level varied with the input signal-to-noise ratio. Three or four decomposition levels were selected under stronger noise, whereas two levels were generally sufficient at higher signal-to-noise ratios. Soft thresholding was selected in all tested cases. The optimized method reduced random fluctuations while retaining the main waveform trend in simulated noisy signals. For measured loosened-bolt signals, it reduced background fluctuation and retained local waveform changes.