HIL and virtual test reduce test times and save money

Perhaps the most active test topic at this year's SAE World Congress was hardware-in-the-loop (HIL) test. The topic was well covered on the show floor, where several companies were exhibiting products for HIL test, as well as in the technical sessions, where many papers covered the topic. In addition, the related topics of simulation and virtual test were also the subject of several papers.

An example of the papers presented on HIL test is "Development of Transmission Hardware-in-the-Loop Test System" (paper 2003-01-1027) written by two engineers from the Southwest Research Institute (SwRI, San Antonio, TX; http://www.swri.org) and two engineers from General Motors. The authors note that developers have traditionally relied on vehicle and engine-driven testing to evaluate new transmissions, but this approach has many drawbacks: Using prototype vehicles to test new transmissions means waiting months for prototype vehicles to be built; vehicle tests are notoriously unrepeatable; weather conditions vary from one test to another; and human drivers don't always drive in a repeatable way.

GM and SWRI engineers developed this HIL test rig to test transmission in the laboratory. Courtesy of Southwest Research Institute.

To simulate the engine, the test system uses a low-inertia AC motor and a high-speed control system. The system can simulate engine firing pulses at frequencies up to 250 Hz. This corresponds to an engine speed of 7500 rpm for a four-cylinder engine, 5000 rpm for a six-cylinder engine, and 3750 rpm for an eight-cylinder engine.

To provide road load and vehicle simulation, the system has a motoring dynamometer connected to the transmission output. Other hardware features of the test system include:

• a customer-supplied transmission controller,
• an automatic transmission fluid (ATF) temperature control system,
• a hinged test bed to vary the tilt angle,
• can insulated chamber to control the ambient temperature when necessary, and
• a data-acquisition system to measure the transmission parameters.

While all of this hardware is necessary, the keys to performing a realistic test in the laboratory are the road, engine, and driver models that the engineers use to control the test system.

The road model consists of values for the grade angle and road surface coefficient for a typical test track. Values for drag coefficient, vehicle weight, frontal area, and tire radius are put in by the operator; the system uses these values, along with the vehicle speed from the driver model, to calculate the road load torque. The system also calculates the vehicle acceleration and the amount of torque needed to simulate this inertia. From these values, the system determines the set point for the motoring dynamometer.

The system uses the engine model to calculate the rpm of the low-inertia AC motor. Some of the inputs that the system uses to determine the motor speed are throttle position, road grade, and inputs from the transmission controller. For example, when the road grade is negative, the motor will be coasting and actually absorbing power from the transmission to simulate braking effects that would happen on the road.

The system uses the driver model to provide the closed-loop control function that a test driver would normally perform. Inputs to the model include the test schedule and transmission output speed; the model's output is a throttle position. If the test schedule calls for a particular speed, the system uses the driver model to perform closed-loop control, varying the throttle position.

After building the system, the engineers ran several tests simulating engines as small as 2.2 liters and as large as 6.0 liters, using a single-output dynamometer to simulate different vehicles and different road conditions. The test schedule they used was a 0 to 80 mph partial-throttle acceleration. After running the tests, the engineers compared the data collected from the test system to data collected during actual vehicle tests and found that they agreed very closely. They concluded that they could use this system to test new transmissions in the lab.

Another paper, "CAE Virtual Door Slam Test for Plastic Trim Components" (paper 2003-01-1209) by Hong Su, Chuck Dunn, and Alex Krajcirovic of Visteon Corp. (Dearborn, MI; http://www.visteon.com), described how to evaluate the durability of door trim components. The engineers' objective was to reduce the costs of key life tests (KLTs) required by OEM specifications and speed up the development process. For plastic trim components, a KLT usually calls for a high number of door slams, typically in the hundreds of thousands.

The engineers developed a finite element model (FEM) of an entire side door assembly, including the sheet metal frame, the glass, the mirrors, and the plastic trim subassembly. The models included the preset deflections and preloaded stresses from assembly processes modeled as steady-state loads.

The authors used MSC/Nastran software to simulate the door slam. They used the direct transient method for assessing how the structure responded to the event, calculating velocities, acceleration, and element stresses.

To virtually evaluate component durability, the engineers considered: dynamic slam stress time histories, simulated assembly preload stresses as mean stress corrections; material fatigue properties; and geometrical and surface conditions for each component.

To estimate the amount of fatigue and damage a plastic part might incur, they developed a model using S-N data gathered for the material using test methods described in ASTM D671, "Flexural Fatigue of Plastics by Constant-Amplitude-of-Force."

For this study, the authors identified three components with a high probability of failure: The rear trailing edge of the arm rest, the rear middle hook, and the rear top hook. Using their models, they predicted that the arm rest would have a life of 67,300 slam cycles, the rear middle hook a life of only 10,300 cycles, and the rear top hook a life of 54,000 cycles.

Next, they tried to correlate simulation results with physical test measurements. They placed accelerometers at the high stress points and measured the acceleration over a number of different slam cycles. They then compared these values with the values predicted by the simulations. As shown in the table, the error was less than 2% on channel #10; other locations showed similar errors.

Finally, they performed a series of slam tests using 12 side door assemblies. After 67,300 cycles, 10 of the 12 armrests had failed; after 10,300 cycles, eight of the 12 rear middle hooks had failed; and after 54,000 cycles, three of the rear top hooks had failed.

From this data, the authors concluded that while they needed to refine their models, performing these virtual tests yielded useful information. The virtual tests identified potential durability failures, and even in their less than optimal condition, would help designers focus their efforts.

As this paper points out, the success of virtual tests depends greatly on the models used to run the tests. After developing virtual tests, you need to refine your models and correlate simulation data with real-world results. It's a long, tedious process, but it's a necessary one if you want to reap the benefits of successful tests.