Time resolved record of arc voltage provides an instantaneous diagnostic of the way by which a plasma spray torch is working. Slight changes in characteristic features of the voltage signal generally give informations concerning the actual state of torch electrodes, but at this time, no procedure was set up to predict the evolution of electrode erosion from in line measurements. It is nevertheless believed that usefull information is hidden in different transient events, such as acoustic pressure variations, light emission or arc voltage, occuring within various time scales and with frequency components spread over a large spectral range. The presented paper is devoted to the research of hidden structures in non-stationnary time series of arc voltage. The collected data are processed by using the wavelet transform which seems to be an ideal tool for this purpose.