Paul Schreier touches on the many uses of scientific software in the automotive industry
Thoughts of scientific software for automobiles naturally turn towards CFD (computational fluid dynamics) for aerodynamic analysis.
Meanwhile, though, software has started to make major contributions to the auto industry in countless other areas, many of them aimed at improving fuel economy and reducing emissions.
Here is just a sampling of the myriad areas where scientific software has become a key element in auto design and development.
Virtual wind tunnels
Let’s begin, though, by looking at advances in CFD to study aerodynamics. In years past, the limitations of computer memory and the speed of solvers made it practical to study only portions of a car. Meanwhile, software and hardware have improved tremendously, as evidenced by the use of Fluent by the BMW Sauber F1 Team, who chose to invest in a custom-built supercomputer, the Albert2, rather than in a second wind tunnel. Powered by 512 Intel Xeon 5160 dual-core processors, Albert2 can make 12.28812 calculations/sec.
Further, the team can run models with more than 1 billion cells. Thanks to improved chassis construction and improved aerodynamics achieved with the help of this software, the BMW Sauber F1 Team improved its point total in the Constructors’ Championship from 34 points in 2006 (for a fifth place finish) to 101 points in 2007 (finishing second to Ferrari).
It’s possible to do a full-car simulation with far fewer than one billion cells depending on the desired accuracy and speed of simulation; in fact, with that many cells it’s possible to simulate two or more cars in a racing scenario concurrently and see the effects of ‘dirty’ air – airflow hitting one car and then impacting following cars. Rather than running huge simulations faster, F1 teams are more likely to perform more medium-sized solution in the same time, allowing them to better scrutinise key components. Typical areas where the mesh is most dense, giving more accurate results, include the wheels and front/rear wings where the car’s geometry is the most complex and the flow physics is the most complicated. Wind tunnels still have an important role to play, but the big difference with CFD compared to wind tunnels, says Mario Theissen, BMW motorsport director, is that you not only get results, but also get an understanding of what goes on.
A CFD analysis on an F1 racing car using the Fluent package shows just one slice of the mesh, with a finer mesh near the car itself and tetrahedrals in the far field to save computational time. Note the vortex shedding on the side of the car.
FEA for electromagnetic value actuators
FEA has applications far beyond CFD and aerodynamics, as exemplified by an application from TRW Automotive Engine Components in Barsinghausen, Germany. The firm uses Comsol Multiphysics in the development of new types of valve-train systems. Operating engine valves without a camshaft and instead with an electromagnetic valve train (EMVT) allows the adjustment of valve opening and closing times for any engine speed and load.
Dethrottling an engine with an EMVT system leads to potential fuel savings of 18 per cent compared to camshaft-driven engines. Further, the EMVT can achieve the low-end torque typically known only in diesel engines. In these actuators, two electromagnets hold the valve and armature in the closed and open positions against a spring force. The transition time from open to closed depends solely on the resonance frequency of the spring-mass-oscillator, and that time must be short enough to allow valve timing and control strategies even at high engine speeds – which means a lightweight design including the magnets. However, energising the magnets generates alternating magnetic fields and electric voltages, and these induced voltages can cause undesirable eddy currents that in turn creates losses and deteriorate system dynamics. A reduction in eddy currents can be achieved only with high electrical resistivity in the soft magnetic material and with design features such as narrow slots that disturb the eddy currents’ paths.
An electromagnetic field forces the valve actuator to move, and those movements change the EM field. Thus, a transient simulation in Comsol Multiphysics calculates the eddy currents as well as accounts for the equation of motion and the coupling between them.
Using the software’s equation-based modelling capabilities, engineers enter the equations of motion. Further, they can represent nonlinear materials using tables and functions. With the simulation of a complete transition cycle it is possible to calculate the electrical power transfer into the actuator. During geometry optimisation, engineers can examine various mechanical configurations for the distributions in copper losses, eddy current losses and mechanical losses. It is also easy to analyse the impact of material parameters such as permeability, electrical conductivity and saturation flux density for various soft magnetic materials.
Comsol Multiphysics plot showing distribution of eddy currents in the armature plate of an EMVT 2msec after switchoff of the upper electromagnet, shortly before being caught by the energized lower electromagnet.
LMS Virtual.Lab enabled Woco engineers to predict the effect of changes in the configuration of a powertrain suspension. On this graph of dB vs rpm, the solid red line shows the effect of lowering mount stiffness by half.
Designing for the Chinese market
Finite-element analysis is also helping auto manufacturers penetrate new markets. General Motors, for instance, has reengineered a version of its Buick LaCrosse mid-sized sedan for the Chinese market, specifically for the country’s road conditions, fuel requirements, and customer expectations. In developing the powertrain suspension for that car, GM came to Woco Industrietechnik (Bad Soden-Salmünster, Germany) for meeting some important NVH (noise, vibration, harshness) responsibilities: evaluating competitive vehicles, developing the powertrain suspension system, and ensuring that the engine noise transmitted through the chassis did not exceed specified levels in the passenger compartment.
For this assignment, Woco relied heavily on technology from LMS International. They collected data with LMS Scadas III hardware integrated with LMS Test.Lab software, which operated in conjunction with LMS Virtual.Lab software for noise and vibration simulation and virtual prototyping. For FE calculations, Virtual.Lab uses common solvers such as Nastran, Ansys or Abaqus, but users drive the solver from the Virtual.Lab environment and perform all pre/postprocessing inside that software.
This is the case for structural-analysis calculations or vibration assessments, but the package includes proprietary solvers for fatiguelife prediction, acoustic simulation and dynamic multibody simulation.
The design team began by studying various suspension systems, and they determined that a four-point suspension – consisting of an antiroll torque restrictor strut mount, front/rear engine mounts, and a transmission mount – was more effective, lighter and less expensive than their initial concept. Then the team moved to detailed design of the powertrain suspension in which they sized and positioned mounts and brackets. Here, individual mounts as well as the suspension cradle assembly were evaluated at various frequencies to determine the noise transfer function. With Virtual.Lab they were able to model the complete powertrain suspension system and determine the combined vibration at selected locations in the interior.
Then they could optimise the behavior of the assembly by modifying the size and position of the mounts in a virtual environment. In one case they found a ‘booming’ at 3700rpm, and with this system they were quickly able to determine which parts of the vehicle were responsible for the highest structural contribution in a certain problem zone.
In the analysis of noise in a chain drive, Ford engineers took advantage of Maple’s ability to combine a worksheet, report and model in one file.
A speedy alternative to FEA
Setting up and running an FE analysis can take quite a bit of time: it requires geometry creation or import, setting up boundary conditions and material properties, meshing, running the solvers, and then postprocessing. In some cases there are faster methods to arrive at the same results. This was the experience of Jack Liu of Ford Motor Company, who was studying chain drives widely used for power transmission in automotive systems. The chain link and sprocket tooth impact during meshing have been identified as the most significant noise source. A severe 1800-1900Hz chain noise was detected in a prototype transmission, and sound pressure levels were 10 to 15dB over nominal values. The causes for this noise were initially unknown, but engineers worked on the assumption that chain resonances can amplify the radiated chain-meshing noise.
A colleague first developed an FE model of the chain drive system using Abaqus software, but Liu wanted to work more directly with the fundamental equations. He derived a system of second order ODEs (ordinary differential equations) to model the system and then programmed them in Maple from Maplesoft.
He also created a predictive design tool, so other engineers could easily analyse chaindrive dynamics using on-screen components. The Ford team was able to accurately determine the exact locations of the noise source and the problematic noise peak. The engineers then went to the chain manufacturer and had them randomise the link and tooth profile to cancel the meshing resonance.
Both the analytical and FEA models predicted the longitudinal chain resonance as observed in test data, but setting up and solving the Maple model took far less time, says Liu. In addition, with Maple he gets more insight into the underlying equations, whereas he feels that engineers sometimes employ FEA tools ‘blindly’ with little knowledge of the underlying mathematics.
Jack additionally likes the integration of the worksheet, report and model. Based on his experience, Liu is expanding the Maple model to include chain link/tooth impulse predictions and study the dynamics of chain spin.
Using LabView for control and pre/postprocessing, this powertrain test stand at Argonne National Laboratory consists of a 4-wheel drive dynamometer, emissions bench, a cart for measuring fuel during testing, and miscellaneous instruments such as a water analyser or hydrogen analyser.
ECU for retrofitting to a hybrid
Another set of design software, this time from The MathWorks, helped Eaton Corp retrofit delivery vehicles from a major freight carrier. The new design – which combines a diesel engine with an electric motor in a hybrid powertrain – reduced emissions by 90 per cent and improved fuel economy by 50 per cent.
The team set out to create their first singleshaft parallel hybrid construction, and the system would need to work with any OEM’s chassis, engine or clutch. One key element was building the engine control system for the lowemission powertrain. For this the team designed and simulated a model of the controlscheme architecture in Simulink that included energy management, subsystems of the engine, clutch, battery, motor and other components.
They then created a plant model to simulate all the inputs and outputs that the hybrid control module would use to communicate with the real vehicle components. With the RealTime Workshop they generated C code, which they downloaded onto xPC Target for hardware-inthe- loop (HiL) simulation. At this point they essentially had a working vehicle on the PC that allowed for the easy examination of adjustments or modifications.
After validating the PC results against actual I/O by running these HiL simulations, the engineers developed a prototype controller. They again turned to the RealTime Workshop to generate C code, which they downloaded to a Motorola microcontroller for realtime operation in the freight carrier’s engine control unit (ECU). Hybrids using this controller are now in use across the US in real-world testing.
This 3D display from Lotus’ Sharc suspension software shows a full-vehicle model with a double wishbone front suspension and an H frame type rear suspension. Three example graph results show camber angle, toe angle and castor angle. Graph displays are for bump/rebound displacement, but users can also display roll, steer and combined motions.
What to expect from alternate fuels?
What are realistic fuel consumption and efficiency goals for alternate fuel and hybrid vehicles? Finding out is one task for the Advanced Powertrain Research Facility (APRF) at Argonne National Laboratory, which uses a test stand based on National Instruments’ LabView software. That group is testing advanced vehicles such as hybrids and hydrogen cars to establish benchmarks of where the technology is today to better guide and direct future developments and set realistic efficiency goals for individual components.
They also look at the impact of alternate fuels on powertrain efficiency and how they impact emissions. For instance, because a hybrid more often turns on a cold engine, it might be more fuel-efficient but with higher emission levels.
The test system uses one PC to collect facility data such as vehicle speed and emissions, a second to collect vehicle data such as engine speed and temperatures or signals from the vehicle’s CAN bus, while a third LabView program provides an on-screen speed profile for a test driver to follow. These predefined profiles allow a vehicle manufacturer to establish its fuel consumption during typical conditions such as city or highway driving. The test driver follows the changing speed on the indicator within a small window for speed variances, accelerating and braking as necessary.
This data is collected on one PC in its raw form, and the program must account for the fact that the sensors and subsystems are running at different rates from 2Hz to 100Hz, doing all up and down sampling as necessary to synchronise the data. In addition, some emission data has a lag of as much as 20s as the gases move through the exhaust system, and the software must correct for this time.
A LabView program performs post-processing, such as determining how close the test driver adheres to the profile, or evaluating the modal (realtime) and phased (average) fuel economy and fuel consumption based on an examination of the exhaust gases. Engineers can also view displays that graph all the events with appropriate scaling, shift data, and extract data as necessary. In addition, sometimes the researchers want to look for vehicle-specific data, and for this they can write plug-in VIs (virtual instrument software modules) that they can drop into the LabView dynamically.
Proprietary software goes commercial
The auto industry has long been known for its use of specialised, proprietary codes. However, some of this software is becoming commercially available. For example, Lotus Engineering (Norfolk, UK) has released its Shark LSA (Lotus Engineering Suspension Analysis) package that it has been developing internally for 15 years. In addition, three other packages are now commercially available: Lotus Engine Simulation to predict the performance of two or four-stroke engines; Lotus Vehicle Simulation allows users to build a virtual vehicle and analyse fuel consumption, exhaust emissions and fuel consumption over a range of fixed-speed runs to accelerations; while Lotus Concept Valve Train offers kinematic performance analysis of value-actuation mechanisms.
Responsible for the vehicle dynamics of many of the world’s class-leading passenger cars, Lotus’ ride and handling team was integral to the development of LSA and continues to use the package. Intended for the design and analysis of suspension geometry, it provides templates that define how the suspension is connected to the vehicle and in which you supply dimensions and loads. More than 20 standard templates describe common suspension layouts including double wishbones, Macpherson struts, trailing/semi-trailing arms, pus rod as well as and H-frame configurations, while a user-definable template enables designers to construct unique configurations. An optional module permits the addition of compliant bushes, the rate, position, and orientation of which can easily be modified to allow effects such as lateral force steer due to compliance to be included in the analysis. Suspension characteristics include camber, castor, toe, kingpin angles as well as roll center positions, swing axle lengths, damper ratios, anti-dive, anti-squat and ackermann.
Results can be displayed either graphically or numerically over specified bump, rebound, roll and steer articulations. External forces can be applied to the suspension as well as suspension spring loads, anti-roll bar forces and the influence of steering rack coupling and compliance.
Calculations and results are updated instantaneously after each design change. Result curves show how the suspension geometry changes with the variation of a particular element. An optional module allows users to add stiffness properties of compliant bushes to calculate compliant displacements and bush forces.