The Birth of the Virtual Car:
How Hardware-in-the-Loop is Revolutionizing Product Design
at General Motors and Ford Motor Company
by
Edward C. Jennings
Concurrent Computer Corporation, USA
Presented at:
International Symposium on Automotive Technology and Automation
32nd Isata – June 14-17, 1999
Vienna, Austria
Abstract
Hardware-in-the-loop and man-in-the-loop simulations are revolutionizing the methods by which automobile manufacturers design and test controllers. Design costs and time-to-market are drastically reduced, and the end products are safer and more reliable. As a result of the increasing power and affordability of computer systems and the increasing sophistication of modeling software, the virtual car is on the brink of becoming a reality. In this paper, I will discuss the use of simulation and the evolution of the virtual car and their effects on two global automobile manufacturers, the Ford Motor Company and General Motors.
The birth of the virtual car has been an evolutionary process, but each step in its actualization is bringing about a revolution in automobile design. As Dr. Alan Kay, one of the developers of the modern workstation, aptly stated, "The best way to predict the future is to invent it." While I'm speaking, the real-time computer systems and the modeling software that will enable production of the car of the future are being invented. The systems already in existence are being refined and are leading inevitably towards the virtual car, and this virtual car will become a tool that can be used by automobile manufacturers with 100% confidence in its reliability.
Computers already control many of our cars' most important systems, from fuel injection through brakes and air bag sensors. There are now approximately 18 computers in a car, with more predicted for the future. The computerization of the car is instrumental in the development of the next frontier in design – the intelligent car. By that I mean a car that can actually make safety decisions and compensate for driver reactions to road and weather conditions.
But before I discuss the future, let's take a brief look at the history of design in the automotive industry. In the past, automobile companies developed and tested actual physical controller systems with what I like to call the waterfall method of development. Just as the drops of water go over a waterfall one by one, the controller was developed step by painstaking step. First, the hardware was designed, and an actual controller was constructed. Next, the software was developed, and the complete unit was tested. Inevitably, testing revealed problems, so the process began again and was repeated generally several times before the controller was finally ready for market.
As a result of this rather lengthy development process, it took American automobile manufacturers approximately two to three years to take a new product from conception to production. In some cases, the process took as long as five years. Now, Japanese automakers develop new products in about 24 months, and Ford and GM are working towards reducing the time to 18 months.
To reduce this time to market, automobile industry paradigms are changing. Hardware-in-the-loop, the marriage of deterministic real-time computer systems and modeling software, is now used to reduce prototypes in system design and testing. With hardware-in-the-loop simulation, time to market can be reduced by approximately 50% because components and processes can be completed in parallel. Now manufacturers can develop the hardware and the software at the same time. We use a model to simulate how the hardware will react to a variety of scenarios, and not only to many scenarios, but also to every possible scenario. When the hardware is actually produced, it's right the first time.
Let's take a closer look at how the hardware-in-loop actually works. If I'm developing a system in which the number of times an engine rotates is a critical factor, I can input different RPMs to simulate how the hardware will react at each of these different values. I can also develop a variety of inputs and outputs from the engine to determine the effects of different scenarios. From the results of these tests, I can verify and optimize performance without ever seeing the system produced.
For example, in designing a new fuel injector controller, I previously would have had to design the controller, develop a prototype, and modify the existing fuel injector. I would then have to test the prototype in a car and hope that it would work without too many modifications. Now I design the controller on my computer and test and fine-tune it from the laptop computer that I am using to control the engine. By the time the controller is actually built, it is right the first time.
Building the product right the first time is important to every manufacturer, and Continental Teves proved that it can be done in the automobile industry with a safety system it designed for Ford. Continental Teves ranks first in the world in the production of disk brakes and second in anti-locking brake systems. It was developing a braking controller system for icy conditions that would correct a skid before it occurs. This controller actually senses the direction in which a car is headed after a driver slams the brakes on ice. By varying the pressure on each tire, the safety system corrects the wheels to straighten the car.
The system was completely developed using hardware-in-the-loop simulation. When the controller was tested in a driving simulator, it worked 100% of the time. To test the reliability of the system even further, the old controller was also evaluated in the simulator with the expected outcome - the wheels locked and the car skidded on the virtual ice when the brakes were slammed. Without simulation, a system like this would require about 40 or 50 prototypes before the controller was ready for market.
Not only did hardware-in-the-loop save Continental Teves time and money and assist it in producing a superior product, but it had the same benefits for Ford, the company for which the controller was designed. Ford maintains test tracks in both the United States and Argentina so that it can provide summer or winter conditions for test drives as needed. To test brakes for icy conditions during summer in Detroit, Ford sends its engineers, test drivers and cars to Argentina and vice versa at considerable expense. If any problems surface during the test drives, the engineers are sent back to Detroit with the equipment to redesign the deficient components. The test drive itself can be potentially hazardous to both the driver and the equipment.
With these test tracks, the most probable scenarios can be evaluated. With hardware-in-the-loop, as I previously mentioned, every conceivable scenario can be evaluated. I personally am glad that they can test for even the most improbable occurrences. When I step on my brakes, I want them to work predictably each and every time. I want assurance that there are no defects in my car's controllers, regardless of which system they govern. I can obtain that assurance from components designed by hardware-in-the-loop because this method does not merely reduce defects, it eliminates them.
When a Japanese automobile manufacturer was asked over a decade ago to account for his company's success, he said, "The difference between Japanese manufacturers and American manufacturers is that the Americans test for problems, and we don't create them." Today, Ford and General Motors have adopted this philosophy and are using hardware-in-the-loop and man-in-the-loop simulations to avert potential problems before they occur.
With today's modeling tools, it has become easier to test even the most improbable scenario. As recently as five or ten years ago, there were few software modeling packages available to design engineers without knowledge of high-level programming languages. Today, we have algorithms to design systems, and these algorithms are incorporated into the icons that the engineer selects. Packages such as ISI's MATRIXx, Boeing's EASY5, the Mathworks' Simulink, and other similar modeling software have drastically reduced engineering time, which is important because it translates into reduced time to market.
Hardware-in-the-loop has been utilized in the aerospace industry since the 1960s when it was first used in the Apollo space program. Boeing Aircraft has been using hardware-in-the-loop for the past 14 years. Boeing has reached a high enough level of confidence to develop all avionics without constructing prototypes. When automotive engineers demonstrate that virtual controllers designed and tested through hardware-in-the-loop are equal to the production controllers 100% of the time, then the automobile industry, too, will be able to dispense with prototypes. Some controller systems developed at Boeing, Ford, and General Electric Transportation Systems through hardware-in-the-loop are illustrated here:
General Motors has also adopted that philosophy. One of GM's major goals is to reduce time and costs by moving towards a 100 percent math-based product development process1. For a global company such as GM, math-based development processes serve many purposes – reduction in design time and costs, greater reliability, standardization of processes and development of common denominators between its international divisions and products. Math-based design, algorithms, hardware-in-the-loop – whatever it is called, it allows GM's brands to be tailored easily to meet the needs and tastes of specific markets.
Math-based design engineering played a pivotal role in several automotive system advances at GM, including its stability control system, electronically enhanced steering, continuously variable suspensions, and more precise engine control. Sensors and computers are at the heart of StabiliTrak, a chassis control safety system that monitors the car's speed and performance. Stabilitrak uses the resulting data to guide the car's responsiveness to the driver's steering. If necessary, the car can correct the direction through the application of different braking pressures on each wheel.
Mathematics-based systems are widely used at GM for a variety of other purposes. The new concept cars were designed or fabricated using math-based technology, and math-based systems underlie GM's three-dimensional modeling simulations of vehicle structures and safety restraints. They are also used in the design and analysis of engine combustion systems, transmissions, and vehicle aerodynamics. In the future, GM plans to develop computer-controlled suspension systems which can be tuned, crash-avoidance systems, and other "intelligent" vehicle features.2
As you can see, safety is one of the pivotal concerns of the automobile industry today. Manufacturers want to reduce time to market and costs, but they also want to build a safer and more reliable car. To do this, they must be confident that every time the driver steps on the brakes or the accelerator, these components will respond predictably. Think about the cost of screening 500,000 controllers or finding out after selling the 2 millionth car that there is a problem serious enough to warrant a recall. With hardware-in-the-loop design and testing, the chance of a design defect is almost completely eliminated.
Just as hardware-in-the-loop has become not only a viable but also a preferred method of controller design, optimization, and testing, the preferred method of road testing is becoming man-in-the-loop simulation. Used for a long time to train pilots for military and civilian aircraft, man-in-the-loop facilitates the evaluation of the complex relationships between the car or controller being tested, road conditions, weather conditions, traffic, and driver behavior.
Ford has its own driving simulator that can accommodate the front of all Ford models. If a controller for a Continental is being tested, drivers feel as if they actually are in a Continental. If a van is being tested, drivers are in a van. In each case, drivers experience the same conditions as they would in a particular model vehicle.
The state-of-the-art driving simulator at the University of Iowa, powered by an 8-processor Concurrent Night Hawk™, features a six-degree-of-freedom hexapod motion base. From the interchangeable cabs – a Ford Taurus, a GM Saturn, and a high-mobility, multipurpose wheeled vehicle – the driver can select a panoramic or front view. The software allows researchers to control over 40 semi-autonomous vehicles, develop a variety of traffic scenarios, and control traffic devices, lighting, and weather conditions. Distributed interactive simulation (DIS) protocols allow the simulator to interact in real-time with other DIS simulators.
An even more sophisticated research driving simulator, the National Advanced Driving Simulator (NADS), shown here, is under development.
Funded by the United States National Highway Traffic Safety Administration, it will be completed in the year 2000. This high-fidelity, real-time simulator will provide the driver with realistic representations of the driving environment through its six modules:
- A visual system with three-dimensional front, rear, and side views
- A motion system which combines translational and angular motion with six degrees of freedom
- Control feel systems for steering, brakes, clutch, transmissions and throttle that respond to driver inputs
- An auditory system that provides motion-correlated three-dimensional realistic sounds
- Vehicle dynamics systems that represent vehicle motions and control feel conditions in response to driver actions, road surface conditions, and aerodynamic disturbances
- A vehicle cab system which contains an actual vehicle cab and a full range of instrumentation interfaces
The virtual controller tested on the virtual test track is the methodology of the present. The virtual car tested on the virtual test track will be the methodology of the near future. The inexorable move towards the virtual car is being propelled by economic and technical considerations. Originally hardware-in-the-loop and man-in-the-loop were the almost exclusive domains of the worldwide aerospace industry and military contractors. It was simple economics. Their products sold for billions of dollars, so they could afford large, expensive computer systems.
Remember, ten years ago a computer that cost $1 million had only a small fraction of the power of today's $20,000 or $30,000 workstation. Now it is economically feasible for companies, such as automobile manufacturers that sell products for as little as $20,000 or $30,000, to purchase
them. As prices of computer systems decline further and as software capabilities continue to increase, companies selling products for as little as $5,000 or $6,000 will be able to afford hardware-in-the-loop.
Although economics is one of the primary driving forces behind the virtual car, these economic benefits are realizable only because of the technological advances in computer systems. General Motors benefits from cost savings by combining a dual-processor Concurrent Power Hawk™ system with its own custom-designed testing software for an automatic dynamometer test cell. Engineers can use the test cell to design their own transmission tests and custom operator displays without the assistance of software programmers. In addition, the dual-processor computer system has the flexibility to expand from running the test cell to conducting the simulated transmission test.
Among the more important technological developments has been the production of closely-coupled symmetric multiprocessing systems. These systems allow one processor to execute the operating system while the other processors can be running the model, collecting data, or completing other tasks that are entirely separate from the operating system overhead.
Another advance that I consider very important is Concurrent's development of a Real-time Clock Interrupt Module, which we call the RCIM. The RCIM enables us to synchronize different models running across different processors. For example, in a virtual car, separate models might control the cruise control, antilock brake system, the instrument controller, the chassis controller, and so on. Typically, these models are too large for only one processor, so they are assigned to separate ones. RCIM enables us to synchronize these systems to replicate the functioning of an actual car where even millisecond delays can be critical.
Just how critical a millisecond can be is illustrated by a glitch that occurred during the operation of a 1,000-Hertz model at Ford. This model was running at 1,000 Hertz, but every once in a while it would overrun its frame and complete one cycle in 1.2 milliseconds. However, to run the model in real-time just as it would execute in the car, each and every cycle had to be exactly 1 millisecond. With the help of Concurrent's NightTrace™, a tool that correlates activities between multiple applications and the kernel, the problem was traced to an interrupt. The cause was found to be a bit of rare and obscure code which had never previously been executed and which contained an instruction to "divide by zero." That particular instruction will never be used again at Ford.
Summary
In summary, less expensive, more powerful computer systems combined with more sophisticated yet easier-to-use modeling software will dramatically decrease time to market for automobile manufacturers. The time will soon be here when the virtual car operating in its virtual world will be the only precursor of the real car operating in the real world.
Footnotes
1 Baker, Ken R, GM Vice President, General Motors Global R & D Operations.
"Leading the Globalization of the Product Development Process – Vision,
Challenges, Call to Action." Speech to the Global Automotive Management
Council. July 28, 1997.
2 General Motors Corporation. GM Research and Development Center.
References
Baker, Ken R. GM Vice President, General Motors Global R & D Operations. "Teaming on Innovation: A Brave New World for General Motors. Presented at Tribute 1998. 1997 Supplier of the Year Awards. Mexico City. April 25, 1998.
"GM Product Safety – A Team Commitment." Presented to the Crash Injury Research & Engineering Network Conference. University of Michigan. Ann Arbor, Michigan. October 20, 1997.
" Leading the Globalization of the Product Development Process – Vision, Challenges, Call to Action." Presented to the Global Automotive Management Council. Erice, Italy. July 28, 1997.
Ford Motor Company. World Wide Web Site www.ford.com
General Motors. World Wide Web Site. www.gm.com
Green, Ron and Ken Jackson. "The Design Drive: Advanced Simulation Systems for Automotive Controllers." Modern Simulation and Training. 2/1997.
Smith, John F., Jr. Chairman and Chief Executive Officer. General Motors Corporation. "A New General Motors." Presentation. August 5, 1998.