2002 AAAI Spring Symposium on Information Refinement and Revision for Decision Making: Modeling for Diagnostics, Prognostics, and Prediction Decision Making in M&D Background • In action
Home Description  Organizing Committee  Symposium Dates, times, and places  Why a symposium now?  Who should attend?  Format  List of Accepted Papers  Paper Submission  About AAAI  Decision Making in M&D Artificial Intelligence Challenge problems  Software and Data Existing software  Useful datasets  References FAQ  Web sites  Bibliography  Contact Comments  Subscriptions	Background Monitoring and Diagnosis have a long history in artificial intelligence. On the forefront was medical diagnosis with the historical MYCIN which pioneered the field of expert systems. Since then, almost any part of AI research has been applied to monitoring tasks. This includes Bayesian networks, neural nets, fuzzy logic, genetic algorithms, etc.  Examples for these techniques are enumerated on the software and data page. However, most diagnostic and prognostic systems provide only a suggestion on what the source of a fault or failure is. For true decision making, the notion of optimality needs to be incorporated. In particular, a utility needs to be found that allows the trade off between multiple criteria, including cost of equipment, maintenance, schedule changes (including downtime or delays), etc. Today, very few systems exist that provide the capability to perform this complex task. Below we will maintain an open list of existing example applications (that may not necessarily fulfil the requirements for the advanced decision making capabilities as defined here). However, these applications will give an appreciation for the state of the art. Contact the symposium organizers if you wish to suggest one or more such applications. Join our announcement list if you wish to be notified when examples are posted.  In action IDS The Integrated Diagnostic Systems (IDS) uses advanced information technologies and artificial intelligence techniques. It enhances time critical troubleshooting and prognostic capabilities in the context of the maintenance of complex machinery and equipment. The inputs and building blocks will be real-time and historical data, encoded expertise and data interpretation algorithms. The system provides a framework for the addition of new diagnostic techniques, knowledge, and technologies. In addition, it interacts with other client decision support software systems and functions.  As currently envisioned, IDS supports four main functions:       Provide Accurate Diagnosis       Advise Optimal Repair Strategy       Assess Equipment Health/Predict Incipient Failures       Establish Maintenance/Overhaul Workscope"  FLIDS FLIDS (or IDS, no relation to the IDS above) is initiated with human expert knowledge about a specific application system and connected to that system’s measurement data, FLIDS draws optimal diagnostic conclusions based on the information available. It is design for large complex application systems, and can be used for both real-time and non-real time situations. FLIDS consists of inference engines and a knowledge base. Three types of inference paradigms are used: a fuzzy logic engine using fuzzified maximum flow algorithm, a rule conversion engine, and a modular neural network engine. The user can enter knowledge in the form of connection relationships, data fuzzy membership functions, and linguistic rules. Also the system can take historical data for neural network training. The outputs of inference engines are optimized. Online learning can be performed using neural networks and reinforcement learning. ECANSE E C A N S E, Environment for Computer Aided Neural Software Engineering  realizes a concept of soft-computing on a high-abstraction level, where the software systems are designed visually by integrating a wide selection of predefined functional components or modules on a graphical interface. The modules in ECANSE range from data interfaces, signal generators,  mathematical and statistical functions, script language, and graphical displays to new soft-computing technologies including Neural Networks, Fuzzy Logic, Genetic Algorithms, Chaostheoretical Methods and Hybrid Combinations thereof. ECANSE provides a framework for evolutionary and rapid prototyping. By monitoring the system performance through various tools (statistical and visual) during simulation, the user may interact at any time and the results will be seen immediately.  Engine Health Monitoring Engine Health Monitoring (EHM) program is a tool which integrates vibration, performance, and component life monitoring for detecting and classifying developing engine faults necessary to reduce engine operating costs while optimizing the life of critical engine components. The modular EHM system has been developed for the USAF and is capable of sensor validation, vibration and performance diagnostics, "virtual" sensing, and real-time life consumption. Engine data currently sensed and recorded for post flight processing is analyzed in a continuous, real-time mode. The measured data is validated/trended and then passed through redundant anomaly detection routines for analyzing both performance and vibration related faults. These routines are based on extensive knowledge of how a healthy engine operates over the entire flight envelope, and any deviation from these "normal" patterns of engine operation will be detected and further analyzed. Once an engine anomaly is detected, a complete fault diagnostic analysis is performed in a real-time mode utilizing advanced fault pattern recognition schemes and fuzzy-logic decision analysis. The EHM system has been developed for the Rolls Royce F405 engine (Adour) which is used on the Navy's T-45 trainer aircraft. The system can be economically customized for other engine applications.   SmartSignal SmartSignal’s approach establishes an empirical model of normal operation for individualized machines. Using real-time signals and the model created from the previous step produces estimated signals. These estimated signals indicate the value of each sensor when a system is running normally. The final step is a comparison of the estimated signals to real-time signals. Very fine differences between the two indicate when a system or sensor is operating abnormally. The data reveal which signal or signals are at fault, in turn pointing out the parts of the process or equipment where problems are occurring or are about to occur.
Version 1.0 Updated 9/19/01	Background • In action