Multi-way　Independent　Component　Analysis　can　obtain　higher　order　statistics　of　the　signal,　which　has　gotten　great　progress　for　fault　detection　of　batch　processes.　FastICA　algorithm　easily　affects　by　the　initial　point　when　solving　non-Gaussian　independent　ingredients,　which　cannot　convergence　to　the　minimum　point　and　has　no　idea　for　the　principal　independent　component　number　before　running　it.　To　solve　the　above　mentioned　problems,　a　particle　swarm　optimization　based　on　MICA　algorithm　is　proposed.　Also,　support　vector　data　description　method　is　introduced　to　determine　the　control　limit　of　monitoring　statistics,　avoiding　the　＇dimension　disaster＇　problem　caused　by　kernel　density　estimation.　Design　of　experiments　has　performed　by　penicillin　fermentation　simulation　platform.　The　result　shows　that　the　proposed　method　is　superior　to　traditional　MICA,　which　can　maximize　the　non-Gaussian　characteristic　of　the　extracted　independent　components,　and　make　fault　detection　more　timely　and　effectively.　©,　2015,　Science　Press.　All　right　reserved.