NEW APPROACHES OF THE FORECAST OF THE AGEING OF PLASMA JET NOZZLE IN INDUSTRIAL SETTINGS OF THERMAL SPRAYING
Résumé
The optimization of the extension of the lifetime of the torches without hazard of drop of the product is an unceasing industrial difficulty. It needs a reliable criterion to characterize the wear, and thus indicate when the nozzle used should be switched by a new nozzle. These criteria are still quite subjective and hold the operator's experience. In 2003, David RIGOT, in collaboration with the Volvo Aero Corporation, has defined a number of criteria in his thesis dissertation [ ]. This study led to the realization of an electronic module alert detecting the hazardous time of process. The focus method was based on research and tracking characteristic frequencies in arc voltage fluctuations and in acoustic noise that is the image. It seems that a rough statistical study of signals is able to help define and visualize simple criteria ruthless variation is usable in objective phenomena wear alert. This is ad hoc measures for smooth power spectrum [ ] [ ] and its variability. Filtered signal is used to distinguish typically one or two epochs of slow wear followed by an epoch of fast wear. This difference is clear after a suitable choice of used frequency bands. Graphical representations are very expressive; they are supplemented by statistical tests. Signals & method: results, by David RIGOT on campaign made shots in September 2003 at the aircraft manufacturer Volvo Aero. Files led to collect a series of measures standardized on the time life of the nozzles. For one, whose use was particularly long, a few shots "vacuum" were prosecuted beyond of the usual threshold used in order to study the ageing. Sound torch and tension thus collected at various times of operation nozzles, with several frequencies of sampling, feed and no feed of powder. They allow first approach directly extract global statistical parameters which it represents dissimilarity based on time: moments, skewness, kurtosis, entropy [ ]. To each power spectra measurement, time is calculated from each signal multiple sampling. Each of the frequencies spectra is smooth inside of a convolution window by a kernel assessor [ ]. A simultaneous representation of the power of signals and their erraticism are used to visually choose the most discriminating frequency bands. Two sets of settings are then assessed: mean and variance of spectrum in the band selected; global parameters newly evaluated from filtered in the chosen band signals. Results: most selective statistics evolve in a similar manner and allows periods wear, while the overall results are most often efficient. For example templates to slope failure (piecewise linear models) are used. The spectra of tension frequencies and the spectra of sound frequencies give similar results. The sound is usable with the feed or without feed of powder. Sequential simple tests may be expected to assess the time of rapid wear.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...