This　article　proposes　an　improved　multi-scale　principal　component　analysis　approach　to　recover　the　structural　dynamic　response　in　the　time　domain　from　data　measured　by　performance-deteriorated　sensors.　The　proposed　approach　builds　on　discrete　wavelet　transform,　principal　component　analysis,　and　inverse　discrete　wavelet　transform.　The　main　approach　is　based　on　a　cross-correlation　calculation　to　better　preserve　the　true　structural　vibration　behavior　as　well　as　to　recover　the　structural　response　from　data　measured　by　deteriorated　sensors.　Three　sensor　performance　degradation　models　are　considered　in　this　study.　Numerical　studies　are　conducted　first　to　demonstrate　the　effectiveness　and　accuracy　of　the　proposed　approach　to　recover　the　structural　vibration　response　of　an　example　spatial　truss　model　considering　sensor　performance　degradation　under　different　loading　scenarios.　Experimental　data　on　an　in-field　benchmark　bridge　are　also　analyzed　to　further　validate　and　demonstrate　the　performance　of　the　proposed　approach.