Enhancing efficiency in CO 2 scroll compressors: A data-driven optimization approach incorporating clearance tolerance

Traditional optimization of scroll compressors often focuses on geometric adjustments and overlooks the coupled effects of real-gas behavior and dynamic leakage, especially under the high-pressure conditions of CO 2 transcritical cycles. To address the computational challenges of high-dimensional collaborative optimization, this paper proposes an integrated optimization framework that combines physical modeling with data-driven approaches. The framework first constructs a thermodynamic model that couples transient leakage mechanisms with real-gas effects. On this basis, a neural network surrogate model is established to reduce computational load, and Sobol global sensitivity analysis based on response surfaces is employed to identify key design variables, thereby systematically overcoming the computational efficiency bottleneck in multi-parameter collaborative optimization. Under the high-pressure conditions of the CO 2 transcritical cycle, minute variations in clearances significantly influence leakage flow and overall efficiency. Consequently, this study treats radial and tangential clearances as active optimization variables and incorporates manufacturing tolerance perturbations into the optimization objectives. A multi-objective collaborative optimization of volumetric efficiency ( η v ), isentropic efficiency ( η s ), and robustness (R) is performed, where robustness is defined as performance degradation not exceeding 2% under ±5% manufacturing tolerance variations. The optimization results are ultimately synthesized into a piecewise functional design criterion. This criterion links manufacturing and assembly-controllable parameters (radial clearance, scroll wrap height, and rotational speed) to provide design rules for compressors operating across different speed ranges. The optimization results demonstrate that, while ensuring robustness greater than 0.95, the framework achieves average improvements of 9.56% in η v and 5.12% in η s .