George D. Hadden

Application Challenges: System Health Management for Complex Systems

System Health Management (SHM) is an example of the types of challenging applications facing embedding high-performance computing environments. SHM systems monitor real-time sensors to determine system health and performance. Performance, economics, and safety are all at stake in SHM, and the emphasis on health management technology is motivated by all these considerations. This paper describes a project focusing on condition-based maintenance (CBM) for naval ships. Condition-based maintenance refers to the identification of maintenance needs based on current operational conditions. In this project, system architectures and diagnostic and prognostic algorithms are being developed that can efficiently undertake real-time data analysis from appropriately instrumented machinery aboard naval ships and, based on the analysis, provide feedback to human users regarding the state of the machinery - such as its expected time to failure, the criticality of the equipment for current operation.

Condition-Based Maintenance: Algorithms and Applications for Embedded High Performance Computing

Condition based maintenance (CBM) seeks to generate a design for a new ship wide CMB system that performs diagnoses and failure prediction on Navy shipboard machinery. Eventually, a variety of shipboard systems will be instrumented with embedded high-performance processors to monitor equipment performance, diagnosis failures, and predict anticipated failures. To illustrate the general principles of our design, we are currently building a research prototype of the CBM system that applies to the shipboard chilled water system.

A general architecture for factory-based diagnosis of electronics

This paper describes an architecture, IFADS (Integrated Factory-based Automatic Diagnostic System), that is well suited for performing diagnosis of electronic assemblies that fail factory-based quality assurance testing. Diagnostics in a manufacturing setting differ in several ways from that in an end-user setting. These differences are noted and the motivation for the IFADS design is shown. The purpose of modern Automatic Test Equipment (ATE) is to indicate whether a piece of equipment is operating within specifications. “Diagnosis”, on the other hand goes beyond this to identify the cause of the problem as well. ATE usually contains most of the functionality required to extend this to diagnosis, e.g., the ability to stimulate the inputs in a variety of ways and to measure the outputs and testpoints. Lacking is the ability to analyze the data and determine the probable cause of the fault or the ability to control the tests applied based on the results of previous tests. One of the most interesting characteristics of the IFADS architecture is that it fuses several reasoning techniques to arrive at a diagnosis. Some of these are topological methods, including backtracing from “bad” testpoints and backtracing from “good” Testpoints. A new topological technique is presented which is able to assemble a first approximation of an ordered list of the devices components such that components earlier in the list are more likely to be the cause of the problem than components later in the list. Other techniques use causal information to explain abnormal testpoint data. Still other methods are heuristically based. (Our goal is to minimize the amount of heuristic knowledge required and, for generality, evolve our system toward one which can reason from circuit information alone.) Each of these techniques is fully discussed. In order to make full use of these reasoning methods, the IFADS architecture utilizes several types of domain knowledge. For a particular piece of equipment the following information is represented: the physical hierarchy (how the equipment is assembled), the functional hierarchy (how the equipment is broken down according to function — usually not the same as the physical hierarchy), a mapping between the physical and functional hierarchies, function names and cause-effect relations, and statistical failure probabilities. We built a working prototype of the IFADS architecture to diagnose a torpedo sonar receiver.