Digital Twin Technology in Mechanical Engineering: AI Integration for Simulation and Lifecycle Management

Digital twin technology-the maintenance of a continuously synchronised virtual replica of a physical mechanical system throughout its design, manufacturing, and operational lifecycle-has evolved from a computationally prohibitive concept into an operationally viable engineering capability specifically because artificial intelligence has resolved the fundamental tension between simulation fidelity and computational speed that constrained earlier digital twin implementations. This paper proposes an AI-Integrated Digital Twin Framework (AIDTF) combining a physics-informed reduced-order modelling and real-time synchronisation engine, a predictive simulation and multi-physics coupling module, and a lifecycle management and maintenance optimisation gateway, evaluated across 34 manufacturing and engineering organisations (aerospace$\mathbf{n}=\mathbf{1 2}$, automotive$\mathbf{n}=\mathbf{1 0}$, industrial equipment$\mathbf{n}=\mathbf{8}$, energy systems$\mathbf{n}=\mathbf{4})$processing 1,384,000 asset telemetry and simulation records (968,800 training / 415,200 test, 70/30 stratified split) and 58,600 monitored mechanical assets across 6 engineering domains (rotating machinery, structural components, thermal systems, fluid systems, precision manufacturing equipment, and power transmission systems) over a 36-month evaluation horizon. The AIDTF achieves a composite digital twin effectiveness score of$\mathbf{r}=\mathbf{0. 9 2 7}$against expert-validated engineering benchmark assessments, outperforming ML-only simulation platforms$(\mathbf{r}=\mathbf{0. 7 5 6}; +\mathbf{2 2. 6 \%}$[(0.927-0.756)/0.756]), traditional CAE-based digital twins without AI acceleration$(\mathbf{r}=\mathbf{0. 5 7 3}; +\mathbf{6 1. 8} \%$[(0.927-0.573)/0.573]), and manual engineering assessment without digital twin infrastructure$(\mathrm{r}=0.437$; +112.1% [(0.927-0.437)/0.437]). Deployment outcomes: +46.2% simulation accuracy improvement, +38.8% predictive maintenance precision, -42.6% unplanned downtime reduction, +34.2% design iteration efficiency, 89.4% model validation confidence, and +24.8% asset lifecycle extension. Sub-domain performance: Digital Twin Fidelity and Synchronisation$(\mathbf{r}=\mathbf{0. 9 6 9})$, Predictive Simulation Accuracy (0.917), Lifecycle Optimisation (0.912), and Real-Time Data Synchronisation (0.883). Model-reality divergence in novel operating regimes degrades AIDTF accuracy to 0.816 (-12.0% [(0.927-0.816)/0.927]), and multi-physics coupling complexity to 0.803 (-13.4% [(0.927-0.803)/0.927]), with deployed mitigation protocols for both conditions.

Hali tarjima qilinmagan