To　enhance　the　generalization　capability　of　rotating　machinery　fault　diagnosis,　a　novel　generalized　fault　diagnosis　framework　is　proposed.　Phase　entropy　is　introduced　as　a　new　method　for　measuring　mechanical　signal　complexity.　Furthermore,　it　is　extended　to　refined　time-shift　multi-scale　phase　entropy.　The　extended　method　effectively　captures　dynamic　characteristic　information　across　multiple　scales,　providing　a　comprehensive　reflection　of　the　equipment＇s　state.　Based　on　signal　amplitude,　multiple　time-shift　multi-scale　decomposition　subsignals　are　constructed,　and　a　scatter　diagram　is　generated　for　each　sub-signal.　Subsequently,　the　diagram　is　partitioned　into　several　regions,　and　the　distribution　probability　of　each　region　is　calculated,　enabling　the　extraction　of　stable　and　easily　distinguishable　features　through　the　refined　operation.　Next,　the　one-versus-onebased　twin　support　vector　machine　classifier　is　employed　to　achieve　high-accuracy　fault　identification.　Case　analyses　of　a　wind　turbine,　an　aero-engine,　a　train　transmission　system,　and　an　aero-bearing　demonstrate　that　the　accuracy,　precision,　recall,　and　F1　score　of　the　proposed　framework　are　over　99.51　%,　99.52　%,　99.51　%,　and　99.51　%,　respectively,　using　only　five　training　samples　per　state.　The　proposed　framework　achieves　higher　accuracy　compared　to　nine　existing　models　via　deep　learning　or　machine　learning.　The　aforementioned　analysis　results　validate　the　accuracy　and　generalizability　of　the　proposed　framework.