Identification and interpretation of manufacturing process patterns through neural networks

To produce products with consistent quality, manufacturing processes need to be closely monitored for any deviations in the process. Proper analysis of control charts that are used to determine the state of the process not only requires a thorough knowledge and understanding of the underlying distribution theories associated with control charts, but also the experience of an expert in decision making. The present work proposes a modified backpropagation neural network methodology to identify and interpret various patterns of variations that can occur in a manufacturing process. The neural network methodology is developed utilizing the delta-bar-delta learning rule and hyperbolic tangent activation function. The methodology adopted is designed with the objective to recognize both small and large magnitude deviations and also to identify the nature of process change, so that proper corrective action may be taken to remedy the problem. The network can identify patterns of variation such as shift patterns and trend patterns, as well as normal patterns with a fewer number of subgroups when compared with control charts. This can be utilized to signal an out-of-control condition well in advance to facilitate the prevention of nonconforming products from being produced.