HYBRID DIGITAL TWIN AND MONTE CARLO SIMULATION FOR RELIABILITY OF ELECTRIFIED MANUFACTURING LINES WITH HIGH POWER ELECTRONICS

Abstract

Reliability assessment of electrified manufacturing lines with high power electronics had remained difficult because disruption behavior had been driven by electro-thermal stress, operating-regime variability, restoration logistics, and line-level propagation effects that had produced heavy-tailed downtime and throughput-loss distributions. This study had developed and evaluated a hybrid digital twin integrated with Monte Carlo simulation and discrete-event line modeling to estimate distribution-based reliability and production-impact outcomes for a converter-driven manufacturing line. The merged operational timeline had covered 12 months, had included 18 converter-driven stations, and had captured 86,400 scheduled production minutes. A total of 3,462 disruption events had been reconciled, comprising 2,691 protection trips, 412 hard stops, and 359 reduced-capacity events. Line-level cumulative downtime had totaled 8,964 minutes, corresponding to an operational availability of 89.6%, and downtime per event had been right-skewed with a mean of 2.59 minutes and a median of 1.10 minutes. Downtime concentration had been evident, as the top three stations had contributed 41.2% of total downtime. Correlation analysis had shown that the thermal stress proxy had been strongly associated with downtime totals (r = 0.68) and throughput loss (r = 0.57), while utilization intensity had been most aligned with stop counts (r = 0.62). Regression modeling based on 6,480 station-days had indicated that thermal stress increased stop incidence (IRR = 1.24, 95% CI 1.16–1.33, p < 0.001), and high-load regimes increased stop rates (IRR = 1.19, 95% CI 1.10–1.28, p < 0.001). Repair-duration modeling using 412 repairs had shown that maintenance delays inflated restoration time by 18% (p < 0.001) and hard-stop class produced longer repairs (ratio = 1.74, p < 0.001). Monte Carlo–discrete-event simulation had matched weekly downtime distributions with a Kolmogorov–Smirnov distance of 0.09 and improved tail accuracy, reducing ninety-fifth-percentile downtime absolute error from 210 minutes (independence) to 172 minutes (dependency), an 18.1% reduction.