مشخصات پژوهش

خانه /Fault Diagnosis of Unknown ...
عنوان
Fault Diagnosis of Unknown Systems: Developing a Novel Active-Learning Technique
نوع پژوهش پایان‌نامه
کلیدواژه‌ها
Discrete event systems, partially-known systems, active-learning, complex systems, systems identification, automata theory, manufacturing systems, automotive industries.
چکیده
The growing demand for autonomy in newly engineered systems has generated a substantial increase in technological complexity. With this increase of autonomous systems comes an imperative need for an improvement of their reliability. Unfortunately, in spite of all efforts to contain or prevent system deficiencies, faulty behavior cannot be practically eliminated. The occurrence of these faults that typically lead to errant behavior are often times unpredictable and can occur anywhere in the system; creating a massive challenge for the autonomous system community. Therefore, proper techniques must be employed to diagnose (detect and identify) these faults in a timely and systematic fashion. In many cases, our understanding of an autonomous system may be limited due to proprietary reasons, extreme complexity, or lack of access into the entire system leaving us with partial or no knowledge of the system. This dissertation addresses two main parts regarding the aforementioned challenges. First, a novel systematic active-learning technique for constructing a fault diagnosis tool for completely unknown finite-state Discrete Event System (DES) is proposed. The developed tool, termed diagnoser, detects and identifies occurred faults by methodically querying chosen observable behaviors of a plant. The proposed algorithm utilizes an active-learning mechanism to incrementally complete the information about the system. This is achieved by completing a series of observation tables in a systematic way, leading to the construction of the diagnoser. Secondly, a novel systematic active-learning method for realizing a partially-known DES is proposed which takes the available information about the system into account by tabularly capturing the known data from the system. Then, the algorithm discovers the unknown part of the system via an active-learning procedure. In order to accomplish this, a series of tables will be constructed to first infer the information about the system from th
پژوهشگران ایرا وندل بیتز (دانشجو)، علی کریم الدینی (استاد راهنما)، محمد کریم الدینی (استاد راهنما)