Method, system and device for predictive recognition of errors in a manufacturing system, said comprising: online monitoring of data of the manufacturing system; and online comparing said data with archived error patterns to predict errors supposed to occur in the manufacturing system; whereby the archived error patterns are built and recognized by statistical methods and data mining methods. The present invention can be used for discrete or continuous or batch processes.