Particle impact noise detection in sealed relays based on neural network

Remainder particles failure and mechanical failure are the two main failure modes of sealed relays. In the production, packaging and usage processes of sealed relays, all kinds of metal and nonmetal particles remain or are produced inside the enclosure. These particles might result in a remainder particles failure. The moving parts of the relay may result in a mechanical failure due to looseness and friction. The traditional failure diagnostic method depended on human recognition of the vibration signature. It was difficult to distinguish these two types of failure. In this paper, based on an analysis of the characteristics of the remainder particles signal and the moving parts signal, their probability classification was realized by a neutral network. According to this method, the unknown signals from the remainder particles or from the moving parts could be successfully assigned an efficient estimation. By real testing on an experimental system, the feasibility of the method was confirmed. At the same time, it was proved that the failure identification rate was increased by this method. It would be a benefit for decreasing failures and increasing the reliability of sealed relays.