Prediction of Process Parameters by Using an Neural Network Model for Die Casting Alloys––Practice Ai of Die Casting Industry

View Statistics Email Alert RSS Feed

Information

Details

Project title

Code/計畫編號

MOST108-2221-E019-051

Translated Name/計畫中文名

使用神經網路模型預測壓鑄合金的製程參數－實踐壓鑄人工智慧化

Project Coordinator/計畫主持人

Shuei-Wan Juang

Funding Organization/主管機關

National Science and Technology Council

Department/Unit

Department of Mechanical and Mechatronic Engineering

Website

https://www.grb.gov.tw/search/planDetail?id=13121650

Year

2019

Start date/計畫起

01-08-2019

Expected Completion/計畫迄

31-07-2020

Bugetid/研究經費

960千元

ResearchField/研究領域

機械工程

壓鑄為鑄造業中自動化極高且適合大量生產的方式，相當適合用來開發智慧壓鑄技術。由於少子化加上鑄造業環境髒亂，越來越少年輕人投入鑄造業，導致壓鑄產業人才嚴重斷層，許多新進壓鑄技術員以無經驗之前輩帶領及提供經驗，及為減少鑄造廠以傳統試誤法提高鑄件品質所消耗之生產成本。本研究旨在結合智慧壓鑄技術，提供調整製程參數之依據。本研究利用機器學習分別建立兩個模型，其中模型一可得知機台設定參數對機台反應參數的影響，模型二可依機台反應參數預測壓鑄件品質，結合模型一和模型二後可得知若需提高壓鑄件品質應如何調整機台參數，然因數據量不足，導致模型一建立不完全，無法直接以設定參數預測壓鑄件品質。本研究僅為初步的機器學習模型，尚無法達到自我學習。研究方法係以多項式回歸分析建立模型一，找出機台設定參數與機台反應參數間的關係；以支持向量分類演算法建立模型二，找出機台反應參數與壓鑄件品質間的關係，以預測壓鑄件品質。實驗結果顯示，模型一之參數配適度皆達0.5以上，其表示設定參數與反應參數有一定的相關性；模型二利用SVC演算法並結合交叉驗證，其模型穩定且準確率可達74%，以數據量少的情況下來說，已達可應用的標準 Die casting is a highly automated method suitable for mass production in the foundry industry, and is quite suitable for the development of intelligent die casting technology. Due to the declining birthrate and the messy environment in the foundry industry, fewer and fewer young people are investing in the foundry industry, resulting in a serious gap in the talents of the die-casting industry. Many new die-casting technicians lead and provide experience with inexperienced predecessors. The production cost consumed by wrong method to improve the quality of castings. This research aims to provide a basis for adjusting process parameters in combination with smart die casting technology. This research uses machine learning to establish two models respectively. Model 1 can know the influence of machine setting parameters on machine response parameters, and model two can predict the quality of die castings based on machine response parameters. Combine model 1 and model 2. Knowing how to adjust the machine parameters to improve the quality of die-casting parts, but due to insufficient data, the model is not completely established, and it is impossible to directly predict the quality of die-casting parts by setting parameters. This research is only a preliminary machine learning model, and it has not yet achieved self-learning. The research method is to establish model one by polynomial regression analysis to find out the relationship between machine setting parameters and machine reaction parameters; to establish model two by support vector classification algorithm to find out the relationship between machine reaction parameters and die casting quality. To predict the quality of die castings. The experimental results show that the parameter matching of model 1 is above 0.5, which means that the setting parameters and the reaction parameters have a certain correlation; the model 2 uses SVC algorithm and combined with cross-validation, the accuracy rate of the model is 74%. Although the accuracy rate is not more than 90%, but with a small amount of training data, the prediction model is stable and acceptable, this model has been applied to the die casting industry.

Keyword(s)

資料分析
神經網路
深度學習
人工智慧
壓鑄智慧化
data analysis
neural network
deep learning
artificial intelligence (AI)
practice of die casting

DSpace CRIS

Prediction of Process Parameters by Using an Neural Network Model for Die Casting Alloys––Practice Ai of Die Casting Industry

Details

Description