Recent Publications
Mechatronics traditionally integrates mechanical systems, electronics, sensing and computer control through pre-specified models, rules and feedback structures. Artificial intelligence (AI) extends this paradigm by enabling representation learning, probabilistic inference, adaptive optimisation and data-driven decision-making; however, published evidence remains fragmented across robotics, maintenance, manufacturing, digital twins and cyber-physical systems. This critical evidence review synthesises a verified evidence core of 30 publications spanning 1997–2025 and defines a reproducible protocol for its future systematic extension. Records were classified by mechatronic function, AI method, application and study design; descriptive distributions and an evidence-density map were generated in Python without imputing missing performance data. The critical synthesis shows that the decisive architectural change is not the replacement of feedback control, but the addition of a learning and supervisory intelligence layer linking multimodal sensing, state estimation, prediction, planning and lifecycle optimisation. The strongest evidence concerns fault diagnosis, predictive maintenance, machine perception, data-enabled manufacturing and digital twins. AI-integrated systems offer context sensitivity, adaptation and decision support, whereas conventional systems retain advantages in determinism, certification, latency assurance and explainability. Quantitative cross-study effect pooling is presently unjustified because metrics, baselines, datasets and operating conditions are heterogeneous. Critical gaps concern real-time validation, uncertainty calibration, safety cases, cybersecurity, concept drift, interoperable data, energy-aware edge deployment and human authority allocation. The review proposes a function-based AI-Mechatronics Maturity Matrix that separates perception, estimation, prediction, decision, action and governance across five integration levels. The evidence supports hybrid, safety-bounded intelligence rather than unrestricted end-to-end autonomy as the most defensible near-term trajectory for industrial mechatronics and provides testable criteria for future comparative studies.