Design of a quality risk early warning system for discrete manufacturing based on stepped neural network

The complexity of mechanical part manufacturing process calls for quality control from the production line level. Regarding the fuzzy mechanism, low efficiency and high cost of error diagnosis and resolution, AI methods are used to carry out prognostics that avoid the failure occurs. This paper proposes an error prediction model based on a new form of neural network constructed by serially connected branch networks, in which each branch network corresponds to a manufacturing stage. Then an early warning system for the production line concentrating on discrete manufacturing industry is developed based on the proposed model. Application research is carried out in a mechanical parts manufacturing factory where the prototype system is testified to be effective and meaningful. The study contributes to reduce the production shutdown and improve quality control capability for discrete manufacturing factories.