Predicting automobile seat comfort using a neural network

Abstract

This paper demonstrates the viability of employing a neural network for the purpose of predicting subjective perceptions of automobile seat comfort. The inputs included eight seat interface pressure measures, three anthropometric and demographic variables, and a subjective rating of the seat's aesthetic quality. The output was an overall comfort index derived from occupant responses to a survey with proven levels of reliability and validity. The neural network was developed and validated using data collected from 12 subjects, representing a broad range of anthropometry. The subjects evaluated five different front driver bucket seats in a repeated measures fashion. The neural network performance statistics were as follows: r 2=0.83, average error=1.19, and cross-validated- r(15)=0.85, p=0. The resulting knowledge (in terms of the relative importance of inputs) and design guidelines (i.e., human criteria for seat interface pressure measures) should reduce the cost and time associated with the current automobile seat comfort development process, which relies mainly on jury evaluations and is, therefore, executed in a trial-and-error fashion. Relevance to industry This study suggests that subjective perceptions of automobile seat comfort can be predicted using a neural network. The resulting design and product validation implications are pertinent to the current product development processes employed by the automotive seating industry.