Kalman-enhanced artificial intelligence in machining: A systematic review

Hybrid methods combining Kalman filters (KFs) and Artificial Intelligence (AI) integrate state estimation with nonlinear learning in machining. This systematic review develops a taxonomy of Kalman–AI coupling patterns, examines evaluation practice, and derives deployment implications for machining applications. The review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement and the PRISMA extension for search reporting (PRISMA-S). Searches in Scopus and the Web of Science (WoS) Core Collection (peer-reviewed English journal/full conference papers, 2022–2026) were completed in November–December 2025, with a May 2026 Scopus/WoS update and a supplementary sensitivity search. From 7176 original and 1325 update records, 4419 database records were screened; with supplementary screening, 21 studies (2022–2026) were included. Four coupling patterns were identified: (A) Kalman-based preprocessing and feature conditioning; (B) learned system or observation components within Kalman loops; (C) Kalman-updated neural or polynomial models; and (D) cascaded or ensemble observer–learning schemes. Patterns A (9/21) and B (7/21) dominated. Reported comparisons usually favoured hybrids. Applications covered milling, turning, drilling, spindle thermal behaviour, and robotic machining, targeting tool wear and remaining useful life (RUL), force/deformation, fault detection, thermal error, and path correction. Evidence remains fragmented because datasets are narrow, tightly matched baselines and controlled ablations isolating the Kalman and AI contributions are limited, explicit uncertainty quantification is uncommon, and shared benchmarks are scarce. Shared datasets, controlled ablation protocols, physics-informed hybrids, and shop-floor demonstrators are priorities.