Original Article Prediction of the maximum air velocities created by metro trains using an artificial neural network approach Gencer Koc 1 , Cuneyt Sert 2 and Kahraman Albayrak 2 Abstract The maximum air velocity created by a moving train inside a tunnel is obtained using an artificial neural network approach. A neural network model is developed to represent a single train travelling in a single tunnel. A set of non- dimensional groups, which are known to influence the induced flow characteristics, is used for the training of the neural network. Various test runs are compared with the results of the authoritative software, Subway Environmental Simulation. The presence of ventilation shafts within a tunnel is included in the model by defining an aerodynamically equivalent single tunnel using major head loss characteristics of different parts of the system. This approach eliminated the requirement to train the neural network for a large number of possible tunnel/shaft configurations. Keywords Piston effect, vehicle-induced flow, artificial neural network, underground transportation Date received: 6 December 2012; accepted: 25 March 2013 Introduction Underground metro systems are one of the most effective means of transportation in terms of energy efficiency. Underground structures such as tunnels and stations should be designed so as to meet comfort and safety requirements. Vehicles travelling inside tunnels induce air flow, a phenomenon known as the ‘piston effect’. It significantly influences passenger comfort levels and the amount of ventilation required in tunnels and stations. In this regard, the maximum air velocity induced by a metro train has always been the most important design parameter. For a given set of dynamic and geometric parameters, such as the frontal cross-sectional area, speed and the drag coef- ficient of the train, vehicle-induced air velocities can be determined using numerical and approximate ana- lytical methods. One of the most commonly used methods to deter- mine time-dependent fluid velocity in closed conduits is the method of characteristics (MOC). MOC is a numerical method used to reduce mass and momen- tum conservation equations to ordinary differential equations. 1,2 The Tunnel Research Group at Dundee University used it to obtain velocity and pres- sure distributions for cases such as a train passing by cross passages, ventilation shafts or another train. 3 Aradag˘ 4 used MOC to simulate vehicle-induced air flow in tunnel systems. However, her model cannot predict the amount of air flowing through vent open- ings; therefore it cannot be used for practical applica- tions. MOC can also be used to generate appropriate boundary conditions to allow moving boundary fea- tures to be considered by computational fluid dynam- ics (CFD) software that lack this capability. For example, Ke et al. 5 used MOC-based Subway Environmental Simulation (SES) software together with the PHOENICS code to optimize the subway environmental control system for the Hsin Chuan Route of the Taipei Rapid Transit System. They obtained the temperature change in a station for dif- ferent train speeds by considering the effect of venti- lation shaft length, area and geometry. They also obtained pressure values on platform screen doors and compared their results with empirical data. In a similar study, Galindo et al. 6 utilized a combination 1 TARU Engineering Inc., Ankara, Turkey 2 Department of Mechanical Engineering, Middle East Technical University, Ankara, Turkey Corresponding author: Gencer Koc ¸, TARU Mu ¨hendislik A.S¸., ODTU ¨ Teknokent, Silikon Blok, No. 12, Ankara, Turkey. Email: gencerkoc@taru.com.tr Proc IMechE Part F: J Rail and Rapid Transit 2014, Vol. 228(7) 759–767 ! IMechE 2013 Reprints and permissions: sagepub.co.uk/journalsPermissions.nav DOI: 10.1177/0954409713488100 pif.sagepub.com