Fault　detection　in　multiple　phase　processes　is　a　complicated　problem,　because　it　is　needed　in　both　the　steady　phase　and　the　transition　from　phase　to　phase.　To　overcome　the　hard-partition　and　misclassification　problems,　and　also　to　monitor　batch　processes　more　accurately　and　efficiently,　we　propose　a　novel　strategy　for　fault　monitoring　and　diagnosing　in　batch　processes　based　on　the　kernel　principal　component　analysis-principal　component　analysis　(KPCA-PCA).　In　this　work,　a　phase　division　algorithm　is　designed　based　on　the　similarity　index　between　different　time-slice　data　matrices　of　batch　processes,　following　by　a　fuzzy　membership　grade　transition　identification　step.　The　steady　phase　ranges　and　the　transition　ranges　are　then　modeled　by　PCA　with　time-varying　covariance　structures　and　KPCA　separately.　Results　of　simulation　study　and　industrial　application　to　penicillin　fermentation　process　clearly　demonstrate　the　effectiveness　and　feasibility　of　the　proposed　method,　which　detects　various　faults　more　promptly　with　desirable　reliability.