The wavelet transform and its basic properties are described. The principle and method based on signal break detection of wavelet transform modulus maximum varying along the scale are also discussed. All sorts of the frequency are analyzed and processed on various scales to amplify the odd and break points of signal and improve the resolving power and SNR . The application of wavelet theory in signal processing in mechanical engineering field is also discussed and a practical technique for signal singularity detection is proposed. The initial fault locations of roll bearing are effectively detected using the method of the multi－scale wavelet transform modulus maximum. It delivers of exploring beneficial to online fault diagnosis.