CAREER: Neural Network Strategy for Machining When Data is Sparse

职业：数据稀疏时的神经网络加工策略

• 批准号: 9733747
• 负责人: Janet Twomey
• 金额: $ 25.5万
• 依托单位: Wichita State University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 1998
• 资助国家: 美国
• 起止时间: 1998-09-01 至 2005-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9733747&HistoricalAwards=false
• 关键词: CAREER; Neural; Network; Strategy; Machining

DMI-9733747 Twomey Artificial Neural Networks (ANN) require large numbers of observations to ensure good generalization, and a large number of independent observations (or data) to evaluate the networks generalization performance. ANN are potentially one of the most important data processing technologies in a truly intelligent system for the monitoring and control of manufacturing processes. However, manufacturers are often reluctant, if not resistant, to use an ANN approach because of the associated costs of user expertise, long development times, and the need for large amounts of training data. Very little research has been performed on evaluation methods, and the problems of training and evaluation concurrently, when data is sparse, have been overlooked. The research objective of this CAREER project is to develop an ANN training and evaluation strategy for manufacturing situations where data is sparse. The approach in the project seeks the simultaneous use of sparse data for both network training and evaluation. The results of this research will be applied to two manufacturing processes: drilling and electrochemical machining. The utilization of ANN and other information processing technologies will be developed into courses for the undergraduate curriculum. The research will increase the utility of ANN through the development of an ANN training and validation strategy in cases where data is sparse. Together with the development of curriculum in information processing, the project will provide a means for attracting talented students to careers in research, and will embody practices that encourage women and ethnic minorities to see themselves as tomorrow's engineers.

电话：[phone] 人工神经网络（ANN）需要大量的观测值来保证良好的泛化能力，同时需要大量的独立观测值（或数据）来评估网络的泛化性能。 人工神经网络（ANN）是一种潜在的最重要的数据处理技术，在一个真正的智能系统，用于监测和控制的制造过程。 然而，制造商通常不愿意（如果不是抵制的话）使用ANN方法，因为用户专业知识的相关成本，开发时间长，并且需要大量的训练数据。很少有研究已经进行评估方法，和训练和评估的问题，同时，当数据稀疏，被忽视。 这个CAREER项目的研究目标是开发一个人工神经网络的训练和评估策略的制造情况下，数据是稀疏的。 该项目中的方法寻求同时使用稀疏数据进行网络训练和评估。本研究的结果将应用于两个制造过程：钻孔和电化学加工。 人工神经网络和其他信息处理技术的利用将被开发成本科课程的课程。 这项研究将增加人工神经网络的效用，通过开发一个人工神经网络的训练和验证策略的情况下，数据是稀疏的。 该项目将与信息处理课程的开发一起，为吸引有才华的学生从事研究职业提供一种手段，并将体现鼓励妇女和少数族裔将自己视为明天的工程师的做法。