A Pattern Recognition Technique forSystem Error Analysis

The error analysis problem istheresolving ofalimited improbable, theoccurrence ofseveral ofthese errors is setofmeasurements interms ofalarge setofpossible but significantly lesslikely thantheoccurrence ofalesser improbable physical errors. Therelation between themeasurements numberoferrors. Thisconcept isformalized inSection II. andtheerrors ismodeled inpartasasetoflinear undetermined equations 6v=Abr,where 6visavector ofthemeasurements andbr Tepurposecof thiserks toilustrathe pern isavector oferror parameters, andinpart byspecifiication oftherecognition technique thatseeksa solution totheerror relations between theparameters 6ri andthephysical errors. An analysis problem. The solution isbothnumerically approximate solution tothemodel equations isdeemed physically reasonable inthat itadequately fits themeasurements and reasonable ifitreflects oneoronly afewofthephysical errors. To physically reasonable inthatitreflects onlyoneorjust a evaluate acandidate solution consisting of6randits interpretation fewphysical errors. Thetechnique ispresented inSection asphysical errors, weintroduce acriterion function y=yo+ yI;yo isameasure ofllSv -AsrIl and'yl isameasure ofthelikelihood of IVandanexample illustrates formulation oftheproblem thecomposite physical error associated with 6r.With this criterion, andapplication ofthetechniques. thecommonleast-squares (pseudoinverse) solution ofthemodel equations isshowntobeinadequate (itminimizes yobutnoty). A pattern recognition technique ispresented andshowntoyield