Risk monitoring of additive manufacturing as a regime-based multi-state system with sensor fusion

Monitoring strategies for Additive Manufacturing processes are predominantly based on binary anomaly detection. However, these processes rarely operate in purely nominal or failed states. This work proposes a data-driven framework for regime-aware risk monitoring of Wire Arc Additive Manufacturing (WAAM) processes based on a multi-state system (MSS) perspective, in which operating regimes are inferred directly from high-frequency multisensor data. The main contribution is a two-step procedure for regime-aware operational risk monitoring in WAAM. First, we introduce an unsupervised tree-based partitioning rule that identifies operating regimes by describing each deposition segment through its distance and direction of deviation from a global medoid representing the normal-operation signature. Second, based on the identified regimes, we define a regime-conditioned risk indicator that combines the distance from the nominal process signature with the severity assigned to the corresponding operating regime. The resulting indicator provides both a quantitative risk level and an interpretable description of the current WAAM condition, beyond conventional normal/anomalous detection. The methodology is validated on an industrial WAAM setup, and the results show that the proposed approach achieves an F1-score above 90%, while benchmark methods consistently remain below 70%, with a substantial reduction in false alarms. Beyond detection performance, the proposed framework provides interpretable information on the current operating regime and the associated risk level of non-conforming production, supporting decision-making beyond binary alarms. The proposed methodology provides a strong alternative for industrial applications, where datasets are typically highly imbalanced across operating states and prior knowledge of process regimes is often limited or unavailable.