| Summary: | 碩士 === 國立成功大學 === 製造資訊與系統研究所 === 106 === With the advancement of manufacturing technologies, assuring product quality becomes an important issue for the manufacturing industry. Because the automatic virtual metrology (AVM) can achieve real-time and on-line total inspection on workpieces at less cost than traditional inspection methods, it has been applied to several manufacturing industries, such as semiconductor, solar cell, and precision machining, for monitoring workpieces. Electrical discharge machining (EDM) is a manufacturing process where a workpiece is transformed into a desired shape by removing its materials using electrical discharges. EDM can be used to machine hard metals or those difficult to machine using traditional techniques and is commonly used for die making, mold making, and small hall drilling in the CNC industry. Due to the characteristics of the EDM process, it is required to install sensors (e.g., voltage, current, and vibration sensors) with a high sampling rate to acquire machining data, leading to a high data generation rate, up to 130 GB per machined hole. Thus, applying AVM to the EDM process encounters a big data processing issue in terms of data preprocessing for computing machining features. Aimed at resolving the big data processing issue of EDM, this thesis proposes a novel efficient big EDM data processing scheme (i.e., BEDPS) based on Hadoop and Spark. First, BEDPS detects the machining waves using the proposed concept of gaps and saves each machining wave into a file with no internode communications in Hadoop. Then, BEDPS computes machining features by pre-loading the machining-wave files in memory to reduce the amount of data access. Finally, testing results of applying BEDPS to the EDM process in a case study show that the proposed BEDPS can effectively detect machining waves from big raw data and efficiently compute the key features of machining data for the EDM process. Compared to the existing sequential data processing scheme, the proposed BEDPS is a promising efficient parallel data processing approach for the EDM process.
|
|---|
