Uncertainty of diagnostic features measured by laser vibrometry: the case of optically non-cooperative surfaces

This paper discusses the uncertainty in the measurement of characteristic features by laser Doppler vibrometry useful to industrial diagnostics when measuring on polished, highly reflective, low diffusive surfaces, such as the enamelled metal sheet of the cabinet of electrical household appliances. This case is relevant to on-line quality control applications, where it is not possible to adopt any surface treatment to improve optical scattering properties. The paper illustrates in particular the effect of dropout noise on the measured vibration signal and develops a joint analysis of drop-out noise due to poor optical properties and its effect on the diagnostic process, presented in statistical terms. A nondimensional quantity is introduced to describe the amplitude of the Doppler signal and the presence of drop-out noise is shown to be correlated to its amplitude. Starting from the consideration that drop-out noise is impulsive, with a pseudo-random occurrence, this paper presents an experimental assessment of uncertainty in the measurement of some spectral features used for the diagnosis of electrical appliances on the production line. It can be seen that the effect of drop-out leads to an increase in scatter and to a systematic shift in the distribution of the features examined; this effect is relatively larger for features with low amplitude. The Monte Carlo simulation of measurement uncertainty propagation confirms the same trend and allows statistical distributions to be obtained for the features, thereby enabling us to draw some conclusions as regards diagnostic errors. This study shows that in the presence of pseudo-random drop-out noise a diagnosis based on spectral features with low amplitude has poor reliability and false-positives are highly probable. An analysis of this occurrence is made for cases of production exhibiting features with different statistical distributions and possible actions to limit such problem are highlighted.