Customer Usage Space Classification and Representative Duty Cycle Development Using K-Means Clustering

Understanding customer usage space and its impact on engine, after treatment, and vehicle duty cycles poses challenges in terms of data noise, data variability and complex interrelations. Moreover, humans are only able to concurrently visualize at most 2 to 3 dimensions, limiting the number of engine parameters that can be considered. Previous studies in this field have been limited to understanding trends in data based on single duty cycle, comparatively short application period and time domain segmented clustering analysis. These techniques have been used to determine representative cycles for specific applications. In this paper, K-Means Clustering is used to classify customer usage space based on tens of dimensions, for multiple duty cycles, and over years of operation. The clusters are evaluated based on system, sub-system, and component-based metrics on a day based unsegmented engine parameter values. Some particular applications of this methodology are discussed in the paper including - Generalized Clustering and Post Processing Visualization Techniques, Critical Customer Identification and Customer Representative Nominal Short-Cycle Generation - a moving window approach. Case studies with real customer data for various On-Highway commercial diesel applications are discussed for different sub systems to demonstrate the applications and power of the tool.