Modeling heavy-duty vehicle fuel economy based on cycle properties

This study, by researchers at the West Virginia University Center for Alternative Fuels, Engines and Emissions, developed a methodology for predicting heavy‐duty vehicle fuel economy during operation over “unseen” activity based on fuel economy data gathered from operation measured from vehicles exercised over chassis dynamometer cycles and properties of those cycles.

 

Heavy‐duty chassis dynamometer data for over‐ the‐road trucks and 40‐foot transit buses were gathered from the CRC E55/59 program and the WMATA emission testing program, respectively. A linear model, a black box neural network model, and a commercial software model (PSAT) were used to predict either fuel economy in a distance traveled per volume of fuel consumed basis (miles per gallon) or fuel consumption inferred from CO₂ emissions mass rate (grams per second) basis. Most of the resources of this project were dedicated to the linear model. The methodology allowed for the prediction of fuel economy from vehicles operating on a number of different chassis dynamometer cycles based on relatively few experimental measurements.

 

The results of the application of the linear model to a set of 56 heavy heavy‐duty trucks operating over five different cycles showed that the use of average velocity and average positive acceleration as metrics produced the lowest average percentage error (less than 5%). The results of the application of the linear model to a set of five buses operating over 16 or 17 different cycles showed again that average velocity and average positive acceleration were suitable metrics to predict fuel economy with reasonable accuracy (less than 10% average percentage error). It was also found that baseline cycles must include idle cycle, along with a relatively slow transient cycle and a relatively high speed cycle, preferably with an average velocity at or above the average velocity of the unseen cycle. Based on the results obtained with both data sets, it was recommended that the prediction be made in terms of CO₂ mass rate (g/s) and then convert to fuel economy (mpg).

 

The results of the application of the black box neural network model and the commercial software model produced average percentage errors of the order of 10% and 4%, respectively. The main disadvantages of these alternative approaches with respect to the linear model were their inherent complexity (application difficulty) and the need to use continuous (second‐by‐ second) data.