Artificial　boundary　method　is　widely　used　in　the　numerical　modeling　of　unbounded　wave　problem.　However,　the　accurate　modeling　of　truncated　infinite　domain　with　general　geometry　and　heterogeneous　materials　is　still　a　challenging　task,　especially　for　direct　time-domain　analysis　in　three　dimensions　(3D).　In　this　paper,　a　novel　3D　time-domain　artificial　boundary　method,　called　Scaled　Boundary　Perfectly　Matched　Layer　(SBPML),　is　proposed.　This　method　is　a　generalization　of　the　Perfectly　Matched　Layer　(PML)　based　on　a　scaled　boundary　coordinates　transformation　inspired　by　the　Scaled　Boundary　Finite　Element　Method　(SBFEM),　which　is　capable　of　using　artificial　boundary　of　general　geometry　(not　necessarily　convex)　and　considering　plane　physical　surfaces　and　interfaces　extending　to　infinity　in　the　truncated　infinite　domain.　Local　scaled　boundary　coordinates　are　firstly　introduced　on　the　element-level　into　the　truncated　infinite　domain　to　describe　general　geometry　properties　of　the　infinite　domain.　Then,　a　complex　stretching　function　from　PML　is　applied　to　radial　direction　of　the　local　scaled　boundary　coordinates　to　map　the　physical　space　onto　the　complex　space,　resulting　in　a　SBPML　domain.　The　spatial　discretization　of　the　SBPML　domain　produces　semi-discrete　mixed　displacement-stress　unsplit-field　formulations　of　third　orders　in　time.　The　order　of　the　obtained　formulation　can　be　reduced　by　one,　enabling　a　seamless　coupling　with　the　standard　displacementbased　finite　element　formulation　of　the　interior　domain.　The　coupled　system　can　be　solved　by　an　explicit　time　integration　algorithm　efficiently.　The　validation　of　the　SBPML　is　demonstrated　through　several　benchmark　tests,　including　wave　problems　in　unbounded　domains　with　general　geometries　and　heterogeneous　material　properties.　Furthermore,　the　application　of　the　SBPML　in　dynamic　soil-structure　interaction　(SSI)　is　demonstrated　using　two　engineering　problems,　including　an　impact　analysis　of　soft　rock-nuclear　island　system　and　a　vibration　analysis　of　soil-lined　tunnel　system.　(c)　2022　Elsevier　B.V.　All　rights　reserved.