Advanced Fault Detection And Diagnosis Methods

Project Title:

Advanced fault detection and diagnosis methods and applications

Project Description:

Loose particles within aerospace power supply and aerospace relay, such as tin granular, copper shot, iron filling and wire skin, can cause catastrophic failures. For example, their moving into bared bond wires could short out the parts for a brief time during usage. Therefore, aerospace power supplies and relays must be screened for loose particles.
Currently, Particle Impact Noise Detection (PIND) test is used to detect the presence of the loose particles. Its typical structure is shown in Figure 1. Its process is as follows: (1) the component under test is attached on an acoustic sensor mounted on the top of a vibration shaker. (2) The shaker induces a series of mechanical shocks and vibrations to free loose particles clinging to the interior of the component. (3) The acoustic wave is generated from the resulting impacts between the freed particles and the component interior. (4) The above wave is converted into electrical signal by the acoustic sensor. (5) The electrical signal is processed by the electronic circuit into two kinds of signals, one kind is an audio signal to drive a speaker and the other kind is a visible waveform shown in an oscilloscope. (6) The operator can determine the presence of the loose particles by both the audio and visual signals.

Typical%20structure%20of%20PIND%20test%20system.png

Figure 1: Typical structure of PIND test system
PIND test is a semi-automatic method due to the fact the operator draws the experimental conclusion either by watching the waveform or by listening to the audio signal, where the operator's fatigue, neglect and lack of experience could easily lead to wrong conclusions. Although the PIND test could detect the existence of large particles effectively, the drawbacks of this method are its low accuracy for the detection of tiny particles and its inability to provide information about the particle materials. The information is of great importance to manufacturers for determining sources of loose particles.
The key point for particle detection is how to extract useful particle signal from noise precisely, especially in the case of weak particle signal with heavy noise, generated by tiny particles. In addition, it is difficult for material identification due to the complexity that even the same material presents different time and frequency domains characteristics with respect to different shapes, sizes and weights. To address these issues, building models using wavelet neural networks for the particle signal and noise are effective to distinguish noise and partial signal generated by different material.
For the project of loose particle detection, we have cooperated with Prof. Shujuan Wang and Prof. Guofu Zhai and they have been working loose particle detection for over eight years at Harbin Institute of Technology in China. This program aims to develop new detection methods based on wavelet networks to improve the detection and identification accuracies.