Spatiotemporal dual dimensional disturbance perception and Bayesian quantitative assessment method for flexible job shop scheduling

Flexible job shops are highly vulnerable to stochastic disturbances, including machine breakdowns, rework, processing time variations, and rush orders, while traditional robust scheduling often leads to conservatism or instability. Existing models struggle to quantify how local anomalies propagate nonlinearly across the system. This paper presents a system-level disturbance perception approach combining a digital twin with a two-layer Bayesian network. At the process level, the network captures single-operation delay probabilities; at the system level, it maps operations and machine constraints to simulate cascade propagation. Large-sample data from digital twin simulations are Discretised to objectively assign conditional probability tables. A spatiotemporal disturbance index evaluates both critical operation delays and the number of affected downstream operations, condensing complex disturbances into a unified metric [0, 1]. Multi-scenario simulations show that the method enables comprehensive system-level perception and provides dynamic thresholds for rescheduling in intelligent manufacturing.