Bayesian Estimation of Defect Inspection Cycle Time in TFT-LCD Module Assembly Process

The defect inspection station is one of the major processes in module assembly stage of TFT-LCD panel manufacturing. Because this process is usually examined manually through human vision, its cycle time estimation is more uncontrollable and therefore could easily affect the customer response time. Hence, this study would like to apply Bayesian networks approach to establish a reliable cycle time prediction model for this key procedure. Our initial model includes work-in-process, throughput, yield, and number of product mixes as the possible explanatory drivers of defect inspection cycle time. To validate the applicability of proposed model, structural and parameter learning is further performed through the data of a TFT-LCD manufacturing plant. Our findings not only demonstrate the feasibility of Bayesian approach in cycle time estimation but also prove its prediction quality by comparing to the results from discriminant analysis.