Digital-Twin-Assisted Spatio-Temporal Visual Soft Sensing for Multi-Gauge Retrofit of Legacy Bearing-Roller Superfinishing Machines

Legacy bearing-roller superfinishing machines display operating settings on analog dials without digital interfaces, leaving their parameter histories unavailable to supervisory systems. We developed digital-twin-assisted spatio-temporal robust gauge recognition (DT-ASTR), which combines shared seven dial geometry, calibrated proposal consensus, and separately inspectable observation and history streams. Evaluation uses the original 181 timestamp manual record, a second 15.57 s constant setting recording, and 23 procedurally rendered videos, with 864 candidate-level episodes providing a separate estimator stress test. On five pressure channels excluding unresolved G3, DT-ASTR achieves a retrospective manual record discrepancy of 0.00433 MPa compared with 0.01123 MPa for the refitted original reader; on the second recording, its discrepancy is 0.01154 MPa at 93.75% coverage. Conventional reader comparisons and rendered video tests distinguish reading discrepancy, coverage, and temporal fidelity, supporting an auditable, low-intrusion route to offline equipment history reconstruction within the tested panel configuration.