A new approach to DWT design for real time de-noising of vibration signatures related to the induction machine defects

Extracting only useful information for vibration monitoring from those sensed by an accelerometer decreases the amount of data to be transmitted and greatly reduces the size of required monitoring and diagnosis applications. This paper presents a suitable architecture for real time FPGA implementation of the discrete wavelet transform for extracting useful signatures of electrical machine defects. Based on wavelet bases selected in order to meet the specificities of these signatures, the implemented algorithm only uses a single pair of LUT-based filters controlled by an FSM. Design complexity and used resources are greatly reduced, while maintaining a very satisfactory execution rate.