Ingenuity Centre Innovation Park NG7 2TU
1 Introduction

Synchronous generators are widely used in power generation applications. Gen Sets (generators coupled to typically diesel or gas-powered engines) are common for electricity generation for remote off-grid locations. Wind turbines and other renewables which convert mechanical to electrical power also rely upon synchronous generators for this conversion process.

Further applications include power regeneration on hybrid trucks, recovering otherwise wasted power as trucks descend into quarries in order to electrically assist ascents.

Key design requirements vary between each of these applications. Gen Sets require the greatest possible efficiency to reduce fuel costs, and renewables also benefit from high efficiency, recovering as much of the mechanical energy as possible. Power density is important for mining trucks in order to minimize the volume and weight carried by the truck.

An effective thermal design is critical to push the boundaries of generator performance.

Stator vents have been used in through flow generators for almost 100 years to reduce operating temperature. As the airflow through the vents and their cooling behaviour is complex and difficult to predict, the number of stator vents to use and their exact positioning has historically required a trial and error approach.

This series of papers present a method to combine Computational Fluid Dynamics (CFD) simulations with Lumped Parameter Thermal Models (LPTM) to arrive at an optimal stator vent design with reduced simulation time. It provides an example of combining high-fidelity simulations with quick thermal modelling methods to arrive at an optimized solution with high confidence in the shortest time possible.

Part One of the paper series describes the CFD modelling approach taken to create a Reduced Order Model to allow the prediction of vent flow and HTC values of the stator vent.

Part Two uses these Reduced Order Models as inputs to a LPTM to determine the best choice of the vent position and amount of baffling in the housing-passage, not restricted to the variants simulated in CFD.

This approach can be applied to the optimization of many thermal management features on all types of electric machine.

2 Stator Vents in Through Flow Generators

Synchronous Generators

This paper deals specifically with through-flow air cooled synchronous generators. Air is pulled through the generator by means of a (normally) centrifugal fan. Air enters the opposite end of the machine to the fan, flows axially through the passages along the machine, and into the fan inlet where it is ejected radially outwards.

Thermal management of this relatively old technology has matured considerably over the last two decades. Employment of thermal simulation tools, such as those developed and used by ECS, allows a greater insight into the heat transfer mechanisms.

Cooling in large generators can be difficult, as the thermal resistance between heat generation source and cooled surface is proportional to the distance between them. As the stator diameter increases, the distance from the slots to the back iron, an important cooling surface, increases. Further, the length of the machine is usually of the same order of magnitude as the stator outer diameter. Stator end-windings are an important cooling source of through ventilated generators, and as core length increases the maximum distance between end winding and centre of machine increases. Therefore, the internal thermal resistance paths increase with the generator size.

Stator Vents

Stator vents are commonly employed in air-cooled synchronous generators to improve cooling and have been used since as early as 1924.

A region of stator laminations is removed and replaced by an insert which provides structural strength whilst allowing air to pass between the airgap and passage between stator back-iron and housing (housing-passage) e (see Figure 1). These provide additional cooling surfaces along the length of the generator, compensating for the increased thermal resistance to the back-iron which is incurred in larger machines.

As stator vents remove lamination steel, the stator winding region in this region becomes inactive, thus reducing the output of the machine, all other things held constant. Compensation for stator vents must be made by increasing the total core length (maintaining the same active length) or increasing the rotor current. In other words, the cooling provided by stator vents comes at a cost, and therefore should be used minimally. Being able to design them for an optimal cooling effect is necessary to reduce trade-off costs.

Determining the optimal number and exact positioning of stator vents is a difficult task. Flow through vents is difficult to predict, given it is dependent on a number off actors, such as the vent location, pressure difference between airgap and housing-passage, the ‘fan’ effect caused by the rotor, rotor width and the axial velocity through the airgap. These complexities mean that simple flow models, such as those commonly used along side lumped capacitance thermal models, are inadequate.

Additional measures are often taken to increase the stator vent flow. Baffling is a common example, where the housing-passage passage is partially closed to manipulate the static pressure. This can increase the pressure difference between airgap and housing passage passage, increasing the flow rate through the vent.

3 Thermal Simulations of Electric Machines
Predicting the operating temperature of an electric machine, such as the generator in this study, can be performed using several different numerical methods
Calculation speed, solution accuracy and solution resolution are three key factors in deciding on which method to use.

Lumped Parameter Thermal Models

At the early design stage, it is preferable to use a method which can provide temperature predictions very quickly in order to consider many design parameters. The speed of these calculations come at the cost of solution accuracy and detail. At this stage, trends in the temperature are as important as being able to predict temperature very precisely. Lumped Parameter Thermal Model (LPTM) approaches are widely popular to achieve this. 

A weakness of this method is in the approximation of the fluid side cooling performance, specifically the fluid flow rates and Heat Transfer Coefficients (HTCs) on different surfaces. Whilst empirical correlations have been devised that can be used as a first guess, these are notoriously inaccurate. Deviations from the machine size, operating speed and other influencing geometric factors from those used to create the correlations will decrease the accuracy.

Computational Fluid Dynamics

When the range of design decisions is reduced to a smaller field, it becomes more important to be able to predict the temperatures more accurately and in more detail. Computational Fluid Dynamics (CFD) provides accurate simulation, allowing a greater insight into the thermal performance and cooling characteristics. 

At the simplest level, fluid flow behaviour can be examined, from which the flow rate and Heat Transfer Coefficients can be determined. These values can then be used along with LPTMs, replacing the empirical correlations, providing much more accurate predictions. However, due to the nature of LPTMs, these models will always lack detail and cannot be used to optimise the finer details of the motor.

Conjugate heat transfer CFD models, which model the heat conduction through the solid components in addition to the fluid flow, are more insightful providing predictions of winding temperatures. However, these models take longer to set up and solve, which can lead to long computing time to solve many scenarios in an optimisation exercise. They are useful to run at intermittent points, validating the LPTM results.

Reduced Order Models

CFD can be used to produce Reduce Order Models (ROMs) for use alongside LPTMs. These ROMs provide more detailed and considerably more accurate flow characteristics compared to empirical correlations.

A series of CFD models are simulated, with changes in the design parameters to be optimised. From the flow and HTC values evaluated from these models, suitable numerical fitting techniques can be used to allow flow characteristics to be determined for all parameter values. Used alongside the LPTM, optimal design decisions can be arrived at quickly and with confidence.

The paper investigates the optimal position for a single stator vent, and the amount of baffling at the Non-Drive End (NDE) of the housing-passage. These factors are chosen as they are difficult to predict without the use of CFD.

4 Generator Description

The generator used for this analysis is a field wound synchronous generator.

Generators Dimensions

A 500mm outer diameter stator, with a 500mm core length is used as the example machine. The 4-pole field wound rotor has a diameter of 356mm, the airgap being 2mm.

Specific attributes of the machine are unimportant, as this paper serves to present a method that could be applied to any generator of any size and operating conditions.

Various views of the generator are shown between Figure 2 and Figure 3.

Vent and Baffling Configurations

CFD models were created for 9 vent and baffle configurations: 3 vent positions, at 25%, 50% and 75% of the way along the stator core, and 3 housing-passage baffle configurations: No baffling, 33% closed and 66% closed.

Figure 2: The generator analysed, shown with and without the frame. Here the stator vent is located 25% along the stator core

Figure 2: The axial air passages through the machine

Figure 3: Axial  slice through the generator. The barrel gap baffle (blue) is shown in the in set.

CFD Process

The CFD process used by ECS is based around the OpenFOAM solvers and snappyhHexMesh mesher. This process has been developed and refined to allow CFD simulations to be setup and solved at much reduced times compared to traditional CFD methods. The process has tailored for electrical machines, which accelerates the setup procedure for CFD.

Standard naming conventions along with a detailed and reliable mesh script enables the generation of the CFD mesh with minimum user input. User control of the meshing process is maintained, without the need to manipulate geometry and apply mesh settings manually from within a GUI.

 Scripts control the setup and solve procedure as well. Boundary conditions – such as inlet and outlet conditions, machine rotation speed and – can be modified easily in the script file. Necessary monitors for the solution are applied to ensure solution convergence.

All script files are generally reusable across electrical machines of all types and sizes, with small adjustments easily applied where needed.

All meshing and solving procedures were performed on a 96-core HPC cluster at ECS.

CFD Setup

The geometry for each case was meshed using the mesh scripts.

Simulations assumed atmospheric pressure at the inlet and outlet, all flow being driven by the fan and rotor rotation, which where simulated at 1500rpm – equivalent to 50Hz operation for a 4-pole generator.

Each mesh took approximately 1 hour to generate, with approximately 60 million cells for each case. The solver took a further 3 hours per case, totalling 27 hours.

Once the CFD solve process was completed the automated export process, developed by ECS for electrical machines, automatically exported the HTC coefficients on all key surfaces.

In total, CFD process took around 2.5 days to complete from geometry prep to data analysis.

Reduced Order Models

The CFD results provide information about the cooling performance of the machine under the 9 scenarios examined.

Thermal modelling techniques will allow us to predict the generator operating temperatures based on these cooling parameters, and operating losses. This would allow the best baffling and vent position of the 9 scenarios to be determined. However, it is unlikely this is the best setup, considering all other baffling and vent positions possible.

To allow the global range of baffling and vent positions to be explored, the 9 simulated scenarios can be represented by Reduced Order Models (ROMs), which represent the cooling parameters as a function of the two parameters we are aiming to optimise.

The majority of cooling parameters – air flow rates, and heat transfer coefficients – are sensitive to baffling and vent positions. Each of these can be simplified to an equation to predict their values for alternate vent setups.

Generator Volume Flow Rate

The air flow rate through the machine can be plotted for each CFD model. The flow rate does not very notably with the start vent position but is highly dependent on the amount of baffling in the barrelgap.

Further, the flow distribution can be viewed as a function of barrelgap baffling. Like the total flow rate, the barrelgap flow, shown as a percentage of the total flow, is independent of the stator vent position.

Both total flow and the percentage of flow through the barrelgap can be represented by a polynomial equation, fitted to the fit lines shown. These equations allow the flow characteristics to be predicted for any baffling level.

Figure 4: Air volume flow rate through the generatorfort different amounts of barelgap baffle.
Figure 5: Amount of air flow through the barrelgap for different baffle amounts, as a percentage of total flow.

Stator Vent Flow and HTC

Another example of ROM is presented in how to predict the flow rate through the stator vent, and the corresponding Heat Transfer Coefficients.

From the 9 data points generated, mathematical software can be used to generate a surface plots of vent flow rate as a function of the barrelgap baffling and stator vent position. Figure 6 shows the 9 data points, and the generated mesh grid which represents predicted values at all other points.

Figure 6: Surface plot of Stator Vent flow rate as a function of barrelgap baffle and stator vent position.

Once the volume flow rate is determined for a baffle and vent position, the Heat Transfer Coefficients can be determined. The stator vent HTC is evaluated and average over two regions. The tooth region between the stator slots has the highest cooling coefficient, due to the high air velocities. As the air dissipates in the yoke region between the stator slots and back iron the HTC is reduced.

The correlation between vent flow rate and HTC is approximately linear, and can be predicted for any value of flow rate through curve fitting of the collected data points, shown in Figure 7

Figure 7: The correlation between Stator Vent flow rate and Heat Transfer Coefficient in two main vent regions.

Backlron HTC

A final example is one for the backiron HTC. This is a particularly interesting example due to the cooling effects created by the baffle.

With an increased a baffle amount, the flow rate through the barrelgap decreases (Figure 5) which logically leads to decreased cooling. However, as the opening to the barrelgap decreases, the velocity of the air through the baffle increases. The result is a local increase in HTC at the NDE.

Figure 8 shows velocity contours through the baffle, causing locally enhanced HTCs. These are visible in Figure 9, where the normalised HTC values for three baffle scenarios (each with a vent at 25% along the core) varies. These plots show the initially high HTC on the backiron when a baffle is present, which decreases further along the core length.


Part one of this White Paper series has laid out the importance of employing an analytical approach to selecting the axial position of stator vents in through flow electrical machines.

The cooling impact of stator vents are impossible to predict using general empirical correlations.

CFD can be used to simulate stator vent flow accurately. However, to simulate all possible vent positions would be time consuming, especially if there are other parameters which can be changed that impact vent flow.

Figure 8: Air velocity entering the barrelgap through the baffle. Baffling increase the velocity near the backiron.
Figure 9: Normalised Heat Transfer Coefficients on the backiron surface, from NDE to DE.

In this study, the stator vent position and level of baffling in the barrelgap are considered variables to be optimised. Nine CFD simulations provided Reduced Order Models of all generator cooling parameters across the range of possible vent positions and baffling. In total, the CFD process took 2.5 days for all models from CAD setup of data analysis, using the CFD process ECS have developed around the OpenFOAM solvers.

ROMs were created to describe the cooling of the generator for different vent locations and barrelgap baffle amounts. These models can be used in a Lumped Parameter thermal model to calculate the generator temperatures and find the optimal vent position and amount of barrelgap baffling. This process is presented in Part 2 of this paper series.