Wen L., Sherman P.

Engine cooling system is the main noise source in a tractor. In an engine cooling system, where a fan operates together with a radiator/shroud in a compact engine compartment, the radiated noise is complex containing both harmonics of the blade passing frequency (BPF) and broadband noise. Existing experimental methods used to estimate the overall SPL are insufficient to characterize the complex noise structure. The goal of this work was to characterize the stochastic structure of the noise from a tractor engine cooling system using advanced statistical signal processing tools and to decide whether a given change in the engine shape has a significant effect on the noise components by hypothesis test. A standard octave band analysis indicated the most important bands. Families of common spectral estimators were used to detect tones and estimate the tonal power at multiples of the BPF of the measured noise process. A comparison is made between the confidence interval (C.I.) of the spectral estimator from our C.I. formula and from commonly used spectral analysis software Matlab. Our formula considers the mixed spectral nature of the process, while Matlab assumes purely continuous spectrum. The comparison showed that Matlab overestimates the C.I. at the tone’s frequency by a factor proportional to the window size. This overestimation may fail to identify an actual significant effect of a specific design on the noise components. The C.I. of the 1/3 octave band level (OBL) was estimated with our formula and used to decide whether the presence of engine block significantly influences the noise OBLs. In the bands without tones, the process was modeled as a Gauss-Markov one. A hypothesis test on the GM bandwidth (BW) parameter was used to determine whether the engine block significantly influences the BW parameters, which is shown to be more robust than the non-parametric test.