Masters Theses

Peter Kim

Date of Award

12-2001

Degree Type

Thesis

Degree Name

Master of Science

Major

Nuclear Engineering

Major Professor

Wesley Hines

Committee Members

Belle Upadhyaya, Andrei Gribok

Abstract

Accelerometers may be used to measure vibration in many areas such as Loose Part Monitoring Systems (LPMS), Anthropomorphic Dummy Response Studies (ADRS) and Machinery Vibration Monitoring (MVM).

The purpose of this study is to identify the failure and degradation of accelerometers in nuclear power plants. The monitoring of systems with accelerometers provides an indication of a problem within the system such as impacting loose parts, sliding and rolling action, and cavitation. However, the ability to identify and locate this problem is compromised when accelerometers are not within calibration or have degraded. It is thus very important to make sure that accelerometers are not malfunctioning in the system.

Since we cannot manipulate accelerometers at a nuclear power plant due to safety, expenses, and so on, we installed eight accelerometers on a SpectraQuest Machinery Fault Simulator. The simulation system was used to analyze accelerometers. An outside force was introduced using a Magnetic Break Torque (MET), which simulates impacting. The changes of the Power Spectral Densities (PSD) of accelerometers can be related to the impacting simulation.

We provide several different loadings (16 levels from 0 lb-inch MET to 10 lb-inch MET). If the PSDs of the various accelerometers are correlated, these correlations can be used to predict the PSD of a particular accelerometer using PSDs of the other related accelerometers in the system. Further, if the predicted power spectral densities can be compared to the PSDs obtained from the actual measurements, then differences between the actual and predicted PSDs can be used to indicate accelerometer degradation or failure.

Regression techniques are very useful tools to model relationships among variables, when these relationships are not explicitly known. Multiple linear regression and ridge regression were used to model relationships between the accelerometers. Multiple linear regression uses the method of least squares to regress given input variables onto a set of output variables. Ridge regression uses the same least squares method as Multiple Linear Regression, but adds a regularization parameter into the regression which assists in dealing with collinear, ill-conditioned data sets.

The results show that a data-based predictive system can be constructed to monitor accelerometer performance. The regularized ridge regression method out performed the simple linear regression system on this ill-conditioned data set.

Download
Files over 3MB may be slow to open. For best results, right-click and select "save as..."

Share

COinS