Directed flow contrast residual descriptor for schedule level fault diagnosis in complex production systems

Fault diagnosis in automotive production lines is challenging because disturbances propagate through routing-coupled stages, manifesting not only in local equipment signals but also in structured deviations between released schedules and realized execution. Although digital-twin and execution-monitoring platforms make such plan-execution residuals increasingly observable, existing diagnosis methods provide limited support for encoding their routing-dependent propagation patterns, even though such patterns are characteristic of complex routed production environments. This paper addresses this representation gap by proposing a directed flow contrast residual descriptor for fault diagnosis from schedule residuals in routing-coupled production lines. Planned and observed operations are aggregated into machine-window residuals and temporally weighted to emphasize the most informative intervals. The weighted residual map is then decomposed into a plant-wide common timing drift component and a topology-dependent contrast component. Based on the routing graph, low-order propagation, centered edge variation, graph-energy statistics, and forward and reverse diffusion summaries are integrated to characterize abnormal flow patterns in a compact and interpretable manner. Experiments on a reproducible multi-scale digital-twin benchmark providing simulation infrastructure for automotive-scale schedule residuals show that the descriptor is particularly effective for lightweight classifiers suited for auditable industrial analytics, and the improvement remains consistent under stress conditions, concurrent faults, and limited route flexibility. These results indicate that plan-execution residuals can serve as an effective complementary diagnostic signal for monitoring routing-coupled automotive production systems.