篇名 | Process-Monitoring Using Part Shape-Scales with Neural Networks: A Circular-Component Case |
---|---|
卷期 | 1:1 |
作者 | Kuang-Han Hsieh 、 C. Alec Chang |
頁次 | 030-044 |
關鍵字 | process monitoring 、 Fourier descriptor 、 feedforward neural networks 、 error pattern identification 、 Scopus |
出刊日期 | 200303 |
Existing approaches for conducting a control task for components machining generally include three methods: dimensional measurement, tolerance verification and equipment monitoring. However, spatial parameters from direct measurement present limited information about geometric features. Thus, many process monitoring systems must rely on acoustic information, torque and force sensors, vibration sensors, or an analysis of collected chips from a production process. These sensors can only detect problems caused by abnormal contact conditions between process tools and workpieces but not geometric deformations of industrial components produced in normal machining.
Frequency parameters that directly utilize coordinate data from an object can identify more detailed geometric features for the purpose of industrial process monitoring. Accordingly, many more process anomalies about manufacturing facilities can be revealed using neural networks to map geometric anomalies. This paper develops shape-scales extracted from Fourier descriptors from incoming part scans. By using these shape-scales, a pattern vector can be formed and fed into a feedforward neural network to identify error patterns. Also, using control charts, the identification of a process resetting point with shape-scales can be monitored. Since roundness is a recurring geometric form of industrial components, the monitoring of circular shape is used as an implementation example to illustrate the proposed system. This proposed system offers an alternative to obtain more process information through processed products in addition to hardware sensors.