Based　on　the　issue　of　torsional　vibration,　this　article　analyzes　the　structure　of　the　drive　part　of　coaxial　drive　printing　system,　constructs　its　mathematic　model,　and　explores　the　mechanism　of　generation　of　torsional　vibration　of　press-that　is　the　coupling　of　inherent　frequency　between　input　signal　and　machine　drive　system　which　leads　to　torsional　vibration.　In　view　of　this,　it　adopts　the　closed　loop　input-shaping　based　on　chaotic　particle　swarm　to　restrain　the　torsional　vibration.　In　the　algorithm,　it　firstly　finds　out　the　optimal　control　structure　by　the　transfer　function　based　on　the　structure　of　three　kinds　of　closed-loop　input-shaping　and　secondly　adopts　the　approach　of　chaotic　particle　swarm　optimization　to　make　off-line　optimization　of　parameter　of　input-shaping.　Such　kind　of　approach　avoids　the　prematurity　phenomenon　existing　in　traditional　particle　swarm　optimization　to　some　extent.　Besides,　it　enlarges　the　scope　of　parameter　selection.　Finally,　it　achieves　the　onetime　collection　of　vibration　signal　and　off-line　optimization　by　the　transfer　function.　This　approach　can　not　only　avoid　system　oscillation　caused　by　online　optimization　but　also　solve　the　problem　of　low　precision　of　pure　offline　modeling　optimization.　Meanwhile,　it　builds　multi-mass　experimental　platforms　as　well　as　makes　a　contrast　with　site　collected　signal.　The　data　shows　that　experimental　platform　proves　the　system　oscillation　caused　by　low　frequency　vibration　point　of　the　drive　system　of　press.　It　carries　out　experimental　verification　by　simulation　platform　and　it　proves　that　chaotic　particle　swarm　optimization　has　fewer　iterations　and　quick　convergence　compared　with　traditional　particle　swarm　optimization　in　the　first　place.　Secondly,　after　adding　chaotic　particle　swarm　optimization,　compared　with　traditional　PID　control,　the　closed-loop　input-shaping　has　simple　parameter　adjustment　and　obvious　effect　of　vibration　restraint.　After　adopting　ITSE　evaluation　function　to　evaluate　the　inhibitory　effect,　it　is　found　that　the　vibration　damps　by　54.4%　compared　with　the　fact　without　adding　input-shaping.　©　2014　Academy　Publisher.