This　paper　is　to　generalize　the　research　on　chaotic　resonance　(CR)　towards　fractional-order　chaotic　systems　and　then　develop　a　new　technique　for　detecting　weak　signals　embedded　in　strong　background　noise.　For　illustration,　a　fractional-order　Duffing　oscillator　system　is　evaluated　by　means　of　bifurcation　analysis,　revealing　the　phenomenon　of　chaotic　resonance,　with　the　optimal　driving　amplitude　falling　within　a　chaotic　interval.　It　is　found　that　the　weak　signal　can　be　amplified　by　the　intrinsic　fluctuations　in　the　chaotic　system　instead　of　stochastic　noise.　Based　on　this　investigation,　a　novel　weak　signal　detection　method　is　developed　and　successfully　applied　to　mechanical　fault　diagnosis　without　the　need　of　signal　preprocessing.　Extensive　numerical　results　show　that　the　signal-to-noise　ratio　of　the　incipient　fault　signal　of　machinery　can　be　greatly　improved.　©　2020　Elsevier　Ltd