Detection-oriented Identification of Nonlinear Systems Using The NARMAX Model in Neural Networks

本提案旨在加强PI的研究/教育活动，PI因家庭原因从密歇根理工大学（MTU）转到西南路易斯安那大学（USL）。 主要的教育活动将是开发跨学科课程。 为此已经迈出了第一步：PI在MTU开发并教授了一门研究生水平的神经网络跨学科课程。 下一个计划是：（1）开发本课程的本科版本，（2）为本课程创建基于网络的教程，（3）与全国的学生和教师分享教程和课程大纲以及其他材料。 神经网络本科课程将为广泛的学生（所有工程和计算机科学专业）开发。在本课程的初步实验阶段结束后，PI有一个长期计划，为它编写一本合适的教科书。 技术研究将涉及一名研究生和一名本科生，重点是使用神经网络实现NARMAX模型进行非线性系统识别。 神经网络已经被其他研究人员用于建模和系统辨识，但通常是以控制而不是故障检测为目标。 所提出的研究的主要目标是使用递归神经网络（与连续的细化程度）进行系统识别的特定目的产生的残差，可用于设计有效的方案检测，隔离和故障诊断。

This proposal is for the enhancement of research/education al activities of the PI who is moving from Michigan Technological University (MTU) to the University of Southwestern Louisiana (USL) due to family reasons. The major educational activity will be the development of interdisciplinary courses. The first step toward this has already been taken: the PI developed and taught a graduate-level interdisciplinary course on Neural Networks at MTU. The next plan is to: (1) develop an undergraduate version of this course, (2) create a web-based tutorial for this course, (3) share the tutorial and the course outline and other materials with students and faculty nationwide. The undergraduate course on Neural Networks would be developed for a broad spectrum of students (all engineering and computer science majors). After the initial experimental phase with this course is over, the PI has the long-term plan to write an appropriate textbook for it. The technical research will involve a graduate student and an undergraduate student, and focus on the use of neural networks for the implementation of NARMAX a models for nonlinear system identification. Neural networks have been used for modeling and system identification by other researchers, but usually with the goal of control rather than fault detection. The main goal of the proposed research is to use recurrent neural networks (with successive degrees of refinement) to carry out system identification of the specific purpose of generating residuals that can be used to design efficient schemes for detection, isolation, and diagnosis of faults.