The invention discloses an on-line diagnosing method for faults of a metering production facility. The method comprises the following steps that firstly, centralized monitoring data are collected; secondly, a diagnosing expert base (training sample) is established; thirdly, a fault prediction model based on an artificial neural network is established; fourthly, on-line diagnosis is carried out on the faults of the metering production facility. All of hardware production units of the metering production facility are provided with alarming equipment, however, when the faults actually occur, production is usually already influenced, and the alarming equipment does not send alarms aiming at production anomalies and some equipment faults. An on-line diagnosing system for the metering production facility constantly analyzes 'four-line-one-storeroom' system running data, a current system running state is judged, the faults are predicted, when the production anomalies or the equipment faults are predicted, staff can solve the problems in time, threats which might influence the production are eliminated, and the reliability of the metering production facility is greatly improved.