Since　multi-way　kernel　principal　component　analysis　(MKPCA)　is　usually　inadequate　in　monitoring　nonlinear　and　multimodal　faults　of　batch　production　processes,　a　new　method　based　on　physical　information　entropy　was　proposed　for　fault　monitoring　(named　multiple　sub-stage　multi-way　kernel　entropy　component　analysis　(MSMKECA)).　The　data　was　first　mapped　from　low-dimensional　space　to　high-dimensional　space　via　kernel　mapping.　Different　steady　and　transitional　stages　of　batch　processes　were　then　divided　by　calculating　the　similarity　index　of　data　matrices　according　to　the　structure　information　entropy　in　the　high-dimensional　feature　space.　Moreover,　fixed　covariance　was　replaced　by　time-varying　covariance　in　transitional　stages.　Finally,　models　were　built　in　different　stages　for　batch　process　monitoring　to　resolve　dynamic,　non-linear　and　multi-stage　characteristics　of　batch　processes.　The　proposed　algorithm　was　applied　in　a　penicillin　fermentation　simulation　system　for　on-line　monitoring　and　the　effectiveness　of　this　method　was　verified.　©,　2015,　Zhejiang　University.　All　right　reserved.