In-process pokayoke development in multiple automatic manufacturing processes

Abstract

<p>In this dissertation, three in-process pokayoke systems were developed to prevent defects from occurring, so as to ensure product quality for three automated manufacturing processes.;The first pokayoke development resulted in an in-process, gap-caused flash monitoring (IGFM) system for injection-molding machines. An accelerometer sensor was integrated in the proposed system to detect the difference of the vibration signals between flash and non-flash products. By sub-grouping every two consecutive molded parts with the vibration signal, the online statistical process control (OLSPC) was able to monitor 100% of the molded products. The threshold of this system established by the SPC approach can determine if flash occurred when the machine was in process. The testing results indicated that the accuracy of this IGFM system was 94.7% when flash is caused by a mold-closing gap.;The second pokayoke development led to an in-process surface roughness adaptive control (ISRAC) system for CNC end milling operations. A multiple linear regression algorithm was successfully employed to generate the models for predicting surface roughness and adaptive feed rate change in real time. Not only were the machining parameters included in the ISRAC pokayoke system, but also the cutting force signals collected by a dynamometer sensor. The testing results showed this proposed ISRAC system was able to predict surface roughness in real time with an accuracy of 91.5%, and could successfully implement adaptive control 100% of the time during milling operations.;The third pokayoke development brought an in-process surface roughness adaptive control (ISRAC) system in CNC turning operations. This system employed a back-propagation (BP) neural network algorithm to train the models for in-process surface roughness prediction and adaptive parameter control. In addition to the machining parameters, vibration signals in the Z direction used as an input variable to the neural network system were included for training. The test runs showed this pokayoke system was able to predict surface roughness in real time with an accuracy of 92.5%. The 100% success rate for adaptive control proved that this proposed system could be implemented to adaptively control surface roughness during turning operations.</p>