The　reliability　of　chiller　is　very　important　for　the　safe　operation　of　refrigeration　system.　In　order　to　solve　the　problem　that　the　traditional　linear　discriminant　analysis　(LDA)　based　on　L2　norm　is　sensitive　to　outliers,　this　paper　introduced　a　novel　dimensionality　reduction　algorithm　for　chiller　fault　data　set　-　RSLDA.　Firstly.,　L2,　1　norm　is　used　to　extract　the　most　discriminant　features　adaptively　and　eliminate　the　redundant　features　instead　of　L2　norm.　Secondly,　an　orthogonal　matrix　and　a　sparse　matrix　are　introduced　to　ensure　the　extracted　features　contain　the　main　energy　of　the　raw　features.　In　addition.,　the　recognition　rate　of　the　nearest　classifier　is　defined　as　the　performance　criteria　to　evaluate　the　effectiveness　of　dimensionality　reduction.　Finally.,　the　reliability　of　algorithm　was　verified　by　experiences　compared　with　other　algorithms.　Experimental　results　revealed　that　RSLDA　not　only　improves　robustness　but　also　has　a　good　performance　in　the　Small　Sample　Size　problem　(SSS)　of　fault　classification.　©　2019　IEEE.