The vibrational signals of several motors for jet pumps and circulation pumps have been collected and used in the experiment in the research. The data are collected using one of the two processes. The first process stores the signals in the embedded system, and the signals are then manual moved to the database in batches. The second process uses wireless sensors to collect and transmit the signals directly to the database. Additionally, a dimension reduction method has been proposed by quantifing the degree of sawtooth of the MSE curves. Using too many attritubes incurs much overhead on the embedded system. By replacing the MSE attributes with the sawtooth attributes, fewer attributes are needed when building the classifier for abnormal detection.