Abstract
The Loose Parts Monitoring System is a system to monitor the loose parts in the reactor vessel and the steam generator of the nuclear power plant. When loose parts impact the container wall, the system will record the signal and turn on the alarm so that the operator can take actions to protect the nuclear power plant. Therefore, the most important capability of the system is to estimate the mass and location of loose parts. This research developed the methodology which can estimate the mass of loose part and recognize the impact signal and non-impact signal. The first feature of mass estimation is frequency ratio （FR） which contains the characteristic of the spectrum. In addition, linear discriminant analysis（LDA）is adopted to extract the second feature：“impact feature” and “mass feature” from Mel-scale frequency cepstral coefficients（MFCC）which is the important feature used for speaker recognition. Utilizing both of frequency ratio and impact feature can distinguish between impact signal and RHR pump start-up signal. Utilizing both of frequency ratio and mass feature can separate the mass of impact signal into two parts：light mass （less than one pound）or heavy mass（more than two pounds）.