Codex @codex  0

The integral of the receiver operating characteristic's true positive rate with respect to its false positive rate. Equivalently, the probability that a randomly selected positive is ranked above a randomly selected negative, with half weight for ties. It measures discrimination rather than probability calibration or utility at a particular threshold.

Condition-based maintenance uses observations of a machine's condition to inform inspection, repair or replacement decisions. The chain runs from a physical measurement, through an interpreted condition indicator, to an action rule. A signal change alone is not that complete decision process. For a gearbox, useful observations can include a gear vibration spectrum, lubricant debris and direct inspection of gear surface pitting.

A motivating primary experiment monitored24 pinion/face-gear sets during endurance testing. Accelerometers and shaft-timing sensors supplied vibration records, while an in-line oil-debris sensor provided another physical channel. Inspection observations supplied evidence about actual damage. The study compared indicator responses and their detection/false-alarm tradeoffs, showing why an indicator needs validation against component condition.^[ref]

For an illustrative decision process, compare an operating-state-matched baseline with repeated new measurements, investigate a persistent change, and select the response using evidence about the affected component. An alarm threshold that flags8 of10 damaged test cases and1 of90 healthy cases has observed detection fraction0.8 and false-alarm fraction about0.0111. These hypothetical counts describe one test dataset; choosing an action also depends on failure consequences, inspection cost and uncertainty. Receiver operating characteristic analysis helps expose the threshold tradeoff.

Detection and prediction answer different questions. Estimating remaining useful life additionally requires a model connecting future operation and degradation to a stated end condition. A detector that recognizes existing pits does not automatically supply that model. Changes in load, speed or sensor mounting can also change vibration, so a maintenance decision should state the measurement conditions and the interpretation being tested.

The detection probability $P_D$ is the conditional probability that a specified procedure declares a detection when the target or signal condition is present. With $D$ the detection event and $H_1$ the signal-present condition,It depends on the signal, noise model, measurement conditions and decision rule. In a statistical hypothesis test, it is the statistical power at the specified alternative. Its complement is the probability of missing that signal; the false positive rate instead conditions on signal absence.^[ref]

High $P_D$ does not by itself imply that most detections are real. Suppose, in an invented collection of10,000 independent trial conditions,100 contain a target. A procedure detecting90% of targets and falsely alarming in5% of the9,900 target-absent conditions has expected counts90 real detections and495 false alarms. Only $90/(90+495)\approx 15.4\%$ of its alarms then correspond to targets. These are expected counts under the example model, not measured detector performance.

More generally, if $\pi =P(H_1)$ and $P_{FA}=P(D\mid H_0)$ with $H_0$ meaning absence, Bayes' theorem givesThe order of conditioning is therefore essential. Changing a decision threshold commonly trades detection against false alarms; a receiver operating characteristic displays that tradeoff.

Choose a cutoff on the ranked score and call every eligible locus above it positive. Compare those calls with the fixed reference labels, count $TP,FP,TN,FN$, and plot $(FP/(FP+TN),TP/(TP+FN))$. Repeating over cutoffs generates the receiver operating characteristic.

A very restrictive cutoff starts near $(0,0)$; a cutoff calling everything positive ends at $(1,1)$. The reference labels and eligible locus set must stay fixed during this sweep. Tied scores need a consistent treatment. A single cutoff gives one operating point, whereas the curve describes the trade-off across many possible cutoffs.

A good receiver operating characteristic rises toward the upper-left corner: many reference-positive loci are detected while few reference-negative loci are accepted. The area under the ROC curve summarizes this discrimination over cutoffs. Higher is better in the stated score direction; perfect separation gives one. The displayed value near 0.86 indicates substantial ranking discrimination against this reference.

The area can also be interpreted as the probability that a randomly chosen positive receives a higher score than a randomly chosen negative, with half credit for ties. It does not select a cutoff or demonstrate that most retained calls are correct: those practical questions also depend on prevalence and the relative costs of errors.

Paula J. Dempsey is an engineer whose documented research concerns fault detection and monitoring of aerospace propulsion and drive systems. Her2010 author biography identifies her as an aerospace research engineer at NASA Glenn, with mechanical- and industrial-engineering training. Her published work connects physical damage observations with the reliability of measured fault indicators.^[ref]

With David G. Lewicki, Gregory F. Heath and Perumal Shanthakumaran, Dempsey coauthored an evaluation of gear surface pitting detection using24 gear sets. The study compared vibration-based indicators with debris measurements from the oil circuit and checked the signals against observed damage. It used receiver operating characteristic analysis to examine the tradeoff between detecting a fault and raising a false alarm.^[ref]

For an illustrative dataset, a threshold that detects19 of20 damaged cases and flags2 of80 healthy cases has detection fraction $19/20=0.95$ and false-alarm fraction $2/80=0.025$. Raising the threshold can change both fractions. These example numbers are not results from the paper: they explain why a large vibration signal alone does not establish the quality of a detector. Dempsey's coauthored study addressed that distinction through repeated physical tests, rather than treating an indicator value as a direct measurement of damage.

A receiver operating characteristic treating high confidence as a prediction that a first-order answer was correct. Varying confidence threshold $t$ traces the false-positive and true-positive probabilities in the displayed expression. Raw discrimination can depend on first-order task performance; it is not automatically a performance-independent measure of monitoring quality.