一種用於放電加工的新式虛擬量測方法

文化大學機構典藏 CCUR > 理工學院 > 工學院 > 資訊工程學系 > 博碩士論文 > Item 987654321/51857

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: https://irlib.pccu.edu.tw/handle/987654321/51857

题名:	一種用於放電加工的新式虛擬量測方法 A Novel Approach to Virtual Metrology for Electrical Discharge Machining
作者:	林禮傑(DENATA RIZKY ALIMADJI)
贡献者:	資訊工程學系
关键词:	Virtual Metrology Big Data Processing Electrical Discharge Machining
日期:	2022
上传时间:	2023-03-24 15:00:49 (UTC+8)
摘要:	EDM (Electrical Discharge Machining) is a process to remove metal from conductive materials using electrical sparks. To monitor the EDM process using virtual metrology (VM), we need to obtain the electrode’s voltage and current signals of a machine tool. Due to the nature of EDM, the sensors installed on the machine tool acquire the signals at a high sampling rate and generate a vast amount of data in a short time, thereby raising the big-data processing issue. In previous work, the Big EDM Data Processing Scheme (BEDPS) research has proposed an efficient EDM big data processing scheme based on Hadoop to process the EDM big data. This paper presents a novel big data processing approach to feature extraction for EDM by using container technology (i.e., Docker and Kubernetes). This paper re-implements some Spark algorithms of BEDPS in Python (originally in Scala) and then runs the refined BEDPS in containers in a Kubernetes cluster. Testing results show that the refined BEDPS developed in this study can reduce the execution time by almost half, compared to the original Scala version (9.6577 minutes vs. 19.2735 minutes). The adoption of Python in Spark has a similar performance to Scala. However, there are some cases where Python performance falls short, for example, parallel processing using Python parallel processing library. The results also show that the Kubernetes cluster promises to be an alternative way, other than the Hadoop, to process big data. At the same time, it can bring some advantages to the big data processing applications, such as easy deployment, robustly running, load balance, self-healing, failover, and horizontal auto-scaling for containerized applications. This paper also proposes Virtual Metrology (VM) on the EDM data by implementing CNN, Multi-Input CNN, Multi-Input MLP, and the adoption of autoencoder to predict four metrology items, specifically diameter, roundness, surface roughness (Ra), and upper diameter. The result shows that Multi-Input MLP with autoencoder can predict the metrology items quite well compared to other machine learning models such as CNN and Multi-Input CNN. In predicting diameter and upper diameter metrology items, Multi-Input MLP with Autoencoder can achieve MAPE lower than 1%. However, due to the limited amount of data and the variability of the output parameter in the datasets predicting roundness and Ra is challenging as both MAPE values are above 10%.
显示于类别:	[資訊工程學系] 博碩士論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	120	检视/开启

检视Licence

在CCUR中所有的数据项都受到原著作权保护.

数据加载中.....