Automotive engineers face a constant challenge to produce designs that are both high in quality and cost-efficient. These designs must work under the extreme conditions experienced by an automobile, and there is pressure to reduce the design time. These requirements can be met by working smarter using virtual prototyping and simulation.
Early in the design process, an engineer may want to determine the relationship between key performance measures and component choices in the design. Using the example of an automotive electrical power system, there will likely be a number of requirements such as the output voltage range for various input voltages and loads, the drift in the output voltage with temperature, and the behavior of the supply when the output terminals are shorted. Each of these requirements would correspond to a key performance measure; developing an understanding of how these measures are influenced by the components can help the engineer adjust the design to meet the specifications. Some of these relationships can be obtained by building a prototype and swapping in and out different components and observing the change in the behavior. This is both a tedious and time-consuming process, and some changes may not be practical. For example, adjusting the electrical current/torque relationship for an electric motor load can only be practically accomplished by testing a collection of motors to find one with the desired characteristics. Moreover, testing under various temperatures requires an expensive temperature chamber.
These issues are overcome if this testing is done using a virtual prototype. In this case, changing a component value or the ambient temperature is done by specifying a new value in a text file or software dialog box. More importantly, specialized analyses such as sensitivity analysis allow an engineer to work smarter so that they can quickly and automatically identify the most important components in a design. Sensitivity analysis generates a bar chart that ranks the components of the design in order of importance for a particular performance measure. This information is invaluable when making design changes to improve the value of a performance measure. Further information about the relationship between a performance measure and a component value can be obtained with a parametric sweep analysis. This information, while quite useful, is for a single implementation of a design. The next step is to consider the impact of component variations on volume manufacturing.
Component and manufacturing variations will lead to measurable (and potentially disastrous) differences in the behavior of a design when it is manufactured in volume. These variations must be investigated to see how they impact the performance measures. This can be done experimentally by building a number of prototypes and measuring the corresponding performance measures of the design. But this can be expensive and time-consuming so typically only a few prototypes will be built—not enough to ensure the design will work when built in quantities. A more accurate and efficient approach is to use a virtual prototype with specified tolerances on the component values. In this case, a simulation such as Monte Carlo or worst-case analysis can be run to simulate an entire production run. Each key performance measure can be tested using the simulation results, and an estimate of the yield for the given assumed component tolerances can be computed. The time to run 100 or more Monte Carlo simulations will typically be shorter than the time to build and measure a single real prototype experimentally. This process provides an assessment of the design for a given set of tolerances but does not tell the engineer how to reduce the variability or the cost of the design.
Once the variation in a design’s performance has been determined for a given set of component tolerances, the next step is to specify more appropriate tolerances to the components to either reduce the performance variability or the design cost. The simplest solution, if the performance variability is too high, is to tighten the tolerance on all of the components, but this will add unneeded cost to the design. With virtual prototyping, several specialized analyses exist that can quickly and automatically identify the key components that are causing the variation in the design. Tolerance-based sensitivity analysis and statistical sensitivity analysis both produce a ranked bar chart showing the components that are causing the variability in the design. These analyses can be run at various temperatures or under different environmental or input conditions. Cost trade-offs can be performed on the identified components to see which ones should be tightened so that the design will meet its requirements when manufactured in quantity. Additionally, these analyses identify components that may have been over-specified and are adding unneeded cost to the design; increasing the allowed tolerances on these components can lower costs.
David Bedrosian of Mentor Graphics Corp. wrote this article for Automotive Engineering.