Liquid-Liquid Coaxial Swirl Injector Performance Prediction Using General Regression Neural Network Kaveh Ghorbanian,* Mohammad R. Soltani,* Mehdi Ashjaee,** Mohammad R. Morad*** (Received: 6 May 2007; in revised form: 20 December 2008; accepted: 20 December 2008) DOI: 10.1002/ppsc.200701104 1 Introduction Liquid fuel/oxidizer atomization is used extensively in rocket engines for exploiting high mixing efficiency within a given length of combustion chamber. Charac- teristics of the resulting spray field after an atomization process have a significant impact on the combustion stability and the propulsion efficiency. Hence, the inher- ently complex process of continuous liquid fragmenta- tion into smaller fluid elements is of critical perfor- mance. Historically, pressure-swirl injectors have been designed based on experience and empirical design tools derived from subscale databases. The disadvantages are that, in general, empirical design methodologies are limited to the test conditions and range of variables for which the tests are performed. Design improvements can only be achieved through extensive sub- and full-scale cold and hot test programs. Furthermore, due to the lack of an optimization scheme for pressure-swirl injectors, a large number of studies must be performed for a better im- pact assessment of the geometrical parameters, operat- ing conditions, and propellant properties on the spray and the resulting flow characteristics. As a result, acquir- ing data from test programs for the purpose of injector design, as well as for performance optimization, make the design procedure expensive and time consuming. Over the past few decades a series of theoretical and experimental studies have been launched in industry and academia towards finding a better fundamental un- derstanding of liquid atomization phenomena. Studies have been carried out to predict the spray characteristics of coaxial swirl atomizers; determining the velocity and the drop size distribution and its influence on the eva- poration and mixing of fuel and oxidizer spray [1–5]. Eroglu et al. [3] and Hardalupas et al. [4] employed Phase Doppler Anemometry (PDA) to investigate the local spray characteristics of single coaxial injectors. In the latter article, spray characteristics were examined on the basis of the exit Weber number and the gas-to-liquid velocity ratio. In another investigation by Hardalupas et al. [5], spray and merging characteristics of three identical coaxial air-blast atomizers, placed in a triangu- lar arrangement, were investigated. Ramamurthi et al. 454 Part. Part. Syst. Charact. 25 (2008) 454–464 * Assoc. Prof. K. Ghorbanian* (corresponding author), Prof. M. R. Soltani, Department of Aerospace Engineering, Sharif University of Technology, Tehran (Iran). E-mail: ghorbanian@sharif.ir ** Prof. M. Ashjaee, Department of Mechanical Engineering, University of Tehran, Tehran (Iran). *** Assist. Prof. M. R. Morad, Department of Aerospace Engi- neering, Science & Research Branch, Islamic Azad Univer- sity, Tehran (Iran). Abstract A general regression neural network technique was ap- plied to design optimization of a liquid-liquid coaxial swirl injector. Phase Doppler Anemometry measure- ments were used to train the neural network. A general regression neural network was employed to predict dro- plet velocity and Sauter mean diameter at any axial or radial position for the operating range of a liquid-liquid coaxial swirl injector. The results predicted by neural network agreed satisfactorily with the experimental data. A general performance map of the liquid-liquid coaxial swirl (LLCS) injector was generated by convert- ing the predicted result to actual fuel/oxidizer ratios. Keywords: coaxial injector, liquid-liquid injector, neural network, swirl atomizer http://www.ppsc-journal.com © 2008 WILEY-VCH Verlag GmbH & Co. KGaA,Weinheim