Ball　mill　is　a　key　heavy　energy　consuming　equipment　in　the　grinding　process.　It　is　based　on　the　closed　rotation　operation　mode　and　the　tens　of　thousands　of　steel　balls　loaded　inside　it　to　realize　the　crushing　effect　on　ore.　The　vibration　signals　generated　by　the　impact　of　steel　balls　with　different　size　distributions　on　the　mill　shell　with　different　amplitudes　and　frequencies　contain　rich　information　related　to　the　mill　load　parameters.　The　inherent　non-stationary　and　non-linear　characteristics　of　shell　vibration　signals　make　it　difficult　to　extract　valuable　features,　and　the　physical　meaning　of　different　characteristic　frequency　bands　is　even　more　difficult　to　explain.　Although　the　empirical　mode　decomposition　(EMD)　algorithm　and　its　improved　version　can　adaptively　decompose　the　vibration　signals　of　the　mill　shell,　there　are　problems　such　as　mode　aliasing.　This　paper　presents　a　method　for　analyzing　vibration　signals　of　mill　shell　based　on　variational　mode　decomposition　(VMD).　Firstly,　the　number　of　layers　K　of　VMD　is　determined　according　to　the　existing　research　results.　Then,　time　and　frequency　domain　analysis　are　performed　on　the　decomposed　IMFs.　Finally,　the　contribution　of　different　IMFs　is　analyzed　based　on　the　predictive　performance　of　simple　linear　models.　The　effectiveness　is　shown　by　the　vibration　signal　of　the　experimental　mill.　©　2020　Technical　Committee　on　Control　Theory,　Chinese　Association　of　Automation.