In　view　of　the　fact　that　it　is　difficult　to　predict　the　location　and　size　of　the　failure　of　rotating　machinery,　the　sensors　sometimes　cannot　collect　the　fault　information.　When　the　fault　location　differs　from　the　sensor　direction　by　180　degrees,　the　fault　information　in　the　signal　is　weak　and　may　be　masked　by　noise.　Multi-channel　signal　acquisition　is　one　of　the　methods　to　solve　the　problem.　Although　existing　algorithms　can　extract　fault　information　from　a　single-channel　signal,　they　are　inefficient　and　sometimes　fail　to　extract　weak　fault　information.　Therefore,　the　information　in　each　channel　needs　to　be　associated　and　combined　in　advance.　This　paper　proposes　a　multi-channel　signal　processing　method　called　quaternion　empirical　wavelet　transform　(QEWT).　QEWT　based　on　the　quaternion　Fourier　transform　can　process　multiple　groups　of　vibration　signals　at　the　same　time,　which　realizes　synchronous　spectral　division,　fusion　and　modal　decomposition.　The　modal　separation　method　based　on　spectral　trend　avoids　the　problems　of　modal　aliasing　and　low　efficiency　of　scale-space　representation.　The　proposed　method　combines　spectral　negentropy　with　the　identification　of　the　fault　characteristics　of　the　outer　ring　and　the　inner　ring,　and　successfully　realizes　the　fault　diagnosis　of　the　rolling　bearing.