<link rel="stylesheet" href="styles.cf390c6bfbe7555f.css">

Engineering evaluation of an embedded computer-vision system for static dimensional screening of cartridge-case overall length

Loading...
Thumbnail Image

Date

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

This thesis evaluated CaseMetrology, an embedded computer-vision system developed for static dimensional measurement of cartridge-case overall length (OAL). The formal study was limited to overall length and used a frozen cohort of 60 specimens, an independent manual reference process, and three independently randomized computer-vision placements per specimen for each prospective system configuration. The original V1 configuration used a principal component analysis (PCA) based estimator and established the prospective baseline. It achieved 100 percent measurement availability but met only one of six quantitative performance criteria. Its ninety-fifth percentile pairwise repositioning difference was 0.004831 inch, indicating substantial presentation-dependent variability. Subsequent engineering work investigated acquisition stability, fixture behavior, estimator design, and physical presentation. The final V3 configuration combined controlled specimen presentation with a frozen Endface-Hybrid estimator and was prospectively evaluated after configuration freeze. V3 again achieved 100 percent availability. At the specimen level, mean bias was negative 0.000435 inch, mean absolute error was 0.001517 inch, root mean squared error was 0.001794 inch, the nearest-rank ninety-fifth percentile absolute error was 0.003170 inch, and the maximum absolute error was 0.003913 inch. The ninety-fifth percentile pairwise repositioning difference was 0.000633 inch. This approximately 86.9 percent V1-to-V3 reduction reflects changes in both the estimator and physical presentation, so it cannot be attributed to presentation alone. V3 met five of the six retained quantitative criteria. The only unmet criterion was the ninety-fifth percentile absolute error target of 0.0030 inch. Under the unguarded specification rule, V3 produced one false pass and three false fails. The results demonstrate that repeated presentation stability and absolute agreement with the manual reference were distinct properties of the measurement system. The large improvement in repositioning did not eliminate residual specimen-level disagreement near the specification boundaries. A post-validation guard-band analysis showed that guarded decision logic could reduce observed false pass behavior within the study cohort, but those findings were retrospective and were not validated as a production decision rule. The completed research supports further development of CaseMetrology as a controlled static dimensional screening architecture. It does not establish the prototype as a replacement for precision reference metrology, a production-ready inline inspection system, or an unconditional final-acceptance device across the full specification interval.

Description

Keywords

Computer Vision, Non-Contact Measurement

Graduation Month

December

Degree

Master of Science

Department

College of Technology and Aviation

Major Professor

Michael J. Pritchard

Date

Type

Thesis

Citation

Endorsement

Review

Supplemented By

Referenced By