David W. Tong

Diagnosing an analog feedback system using model-based reasoning

The approach developed by J. de Kleer and B.C. Williams (1987) has been extended to diagnose steady-state faults in an analog feedback system. A prototype system called AMFI (automated model-based fault isolation) is described which automatically diagnoses a target system given only information on the connectivity and component transfer functions. The prototype detects inconsistencies among measurements, calculates single- and multiple-fault hypotheses and their probabilities, and suggests the best next measurement.<<ETX>>

Multi-branched diagnostic trees

The authors describe the application of quantitative model-based reasoning to the automatic generation of multi-branched diagnostic trees using only a system model description containing connectivity and functional information. The technique is demonstrated using two examples, diagnosing a simple adder-multiplier circuit and a more complex analog feedback control system. Quantitative measures are defined for the performance of the generated trees, and data show that both diagnostic accuracy and efficiency increase with larger branching factors. This technique is believed to hold significant potential for increasing the productivity of developing fault isolation test programs.<<ETX>>

Diagnostic tree design with model-based reasoning

A reasoning procedure using quantitative models of connectivity and function has been developed to generate automatically multibranched diagnostic trees which can isolate faults within feedback loops and in the presence of multiple faults. The authors describe how the model-based reasoning system is used to generate automatically diagnostic trees that can have variable degrees of branching, from binary to ternary (nodes with high, OK, and low branches) to n-ary trees. With branching degrees at or above ternary, these trees are capable of fault isolating within loops and can in fact isolate multiple faults. The trees can utilize much of the information content in quantitative measurements to make efficient and accurate diagnoses not possible with the binary tree. Both efficiency and accuracy of diagnosis increase with the branching factor of the tree. Automated tree generation provides effective automated diagnostics to applications requiring low-cost hardware and fast response time.<<ETX>>

Space charge probing with electron beam

A diffused monoenergetic electron beam of variable energy is used to measure space charge profiles in thin polyethylene terephthalate films. The method and a quantitative model used for the profile calculations are presented. Results are given of charge profiles set up both by electron beam deposition and by voltage poling. The method is shown to have a spatial resolution of about 0.5 pm and a minimum detectable charge density of about 10−4 C/cm3. Inherent experimental problems and potential uses of the method are discussed.

Automated diagnosis of analog circuits

While a number of different approaches have been proposed to automatically troubleshoot electronic systems given schematic information, few are sufficiently powerful to tackle the complexity of analog circuits at the resistor/transistor level. This paper describes work which applies quantitative model-based reasoning techniques to this problem. The circuit schematic is converted into a constraint diagram to which the combination of constraint propagation and dependency tracking are applied to search for inconsistencies and identify the implicated components. Instead of resorting to the propagation of symbols, the technique of aggregate models is used to enhance deductive power but with manageable computation. These ideas have been implemented in the program FIX which diagnoses a given circuit by recognizing inconsistencies among measurements, identifies the set of fault candidates and their posterior probabilities, and suggests the best next measurement. Modeling and inference issues are discussed, and diagnosis of various faults in an example circuit by FIX is described.