Hwarng, H. Brian; Hubele, Norma F. (1991, ASQC) Arizona State University, Tempe, AZ
This abstract is an edited version of the author's original.
One of the difficulties encountered in implementing Statistical Process Control (SPC) in today's computer integrated manufacturing environment is the need for automatic identification of special disturbances in the process. This paper presents the results of research to develop a pattern-recognition methodology, which is suitable for real-time applications and which addresses special disturbances.
The major deficiencies of conventional control chart schemes for monitoring the manufacturing process include: (1) only the information about the process contained in the last plotted point is used, and (2) no information about the special disturbances in the process is explicitly indicated.This paper proposes a control chart, pattern-recognition methodology using a neural network approach. The major components of the proposed methodology include a pattern generator, a data preprocessor, and a network trainer. The methodology consists of two steps: training and testing. The training process, which uses the back-propagation algorithm, trains the network to incorporate six unnatural patterns: (1) trends, (2) cycles, (3) system variables, (4) stratification, (5) mixture, and (6) sudden shifts. Next, the testing procedure measures the performance of the trained network. Measurement criteria include: (1) the classification/misclassification rate, (2) the rate of first detection (ROFD), and (3) the average pattern detected run length (APDRL).
Simulation results indicate that the performance of the proposed pattern-recognition neural network (PRNN) is quite promising.
Control charts,Manufacturing,Neural networks,Pattern recognition,Statistical process control (SPC),Statistics,X-bar control charts,Computers