Top

Rubber Seal Deformation Measurement with a Causality-Enhanced S4D-Mixer Prediction Model

Authors

Files

PDF

Abstract

Rubber seal deformation measurement is required for assembly quality inspection because seal compression, edge displacement and recovery behavior determine pressure-bearing capacity and long-term sealing performance. This paper introduces a causality-enhanced S4D-Mixer prediction model for deformation measurement of rubber sealing rings based on visual and mechanical data. Seal images, edge-offset features, compression displacement, load response, fixture position and image-quality indicators are aligned into a deformation window. The S4D branch models ordered loading, holding and recovery changes, while the Mixer branch fuses image and mechanical variables without relying on a single local texture cue. A causal regularisation module is introduced to distinguish real deformation evidence from unstable associations caused by lighting variation, fixture displacement or reflective rubber surfaces. The simulated experiments include normal compression, under-compression, over-compression, eccentric compression and edge-folded states, and the production-style dataset contains 5,280 measurement windows. The proposed model achieves a mean absolute deformation error of 0.043 mm and 91.8% high-risk deformation recall, while reducing false alarms caused by reflective surfaces compared with convolutional and recurrent baselines. The output includes predicted deformation, confidence score, dominant evidence and review priority. This structure supports high-speed seal screening, measurement traceability and maintenance inspection for automated rubber-component assembly lines.

PDF

Details