Passengers' Comfort Modeling Inside Aircraft

Abstract

aircraft cabin, through probabilistic neural networks and general regression neural networks. The two alternative models are constructed using the physical environmental parameters as input–output patterns, reproducing real flight conditions (inputs) and the environmental comfort index (output), resulting from statistical analysis of passengers’ judgments during test campaigns. The values for the spread parameters characterizing the Gaussian neuronactivationfunctionsinthetwomodelsarechoseninordertoperformthebettergeneralizationofnewdatain the cases of discrete and continuous data, respectively. Results show tight correlation between estimated and actual comfort in both cases. The constructed networks being able to model human judgments about aircraft comfort starting from environmental parameters represents good tools for near-future aircraft design. I. Introduction N OWADAYS, evaluation of passengers’ and crew’s comfort inside the cabin is of key importance for modern aircraft designers;thistaskisnotstraightforwardasitmayseemat firstsight, because human perception of comfort is subjective and depends not only on environmental conditions such as temperature, humidity, noise level, vibration, pollutants, and so on, but also, and for a not negligible part, on personal conditions such as health, physiology, and psychological attitude. In general, the definition of comfort is itself odd, because what is actually perceived is the discomfort and it can only be evaluated by the person himself. Modeling the passenger’s evaluation of comfort through a mathematical tool is essentialforobtainingasoftwaretooltobeusedbyaircraftdesigners to optimize, together with the aircraft performances, the perceived comfort. Approaches for modeling passengers’ judgments about comfort should take into account both environmental and personal conditions and also the highly nonlinear nature of the human judgment process. The aim of the reported investigation is to design a mathematical modelreproducingresultsofhumanjudgmentsaboutenvironmental comfort inside aircraft cabins and its implementation by means of a computational tool based on artificial neural networks (ANNs) to be then used to reckon comfort levels for new possible environmental conditions under study. In the 1980s, following the exploit of the new regional turboprop aircraft, when internal noise comfort in aircraft was consistently addressed for the first time after a long period in which few general works could be found [1], the attention of designers was primarily addressed to vibration as a source of noise. The high internal noise levelsweresupposedtobeessentiallyaneffectofthepropellersthat, excitingthevibrationsofthemainstructure,radiateinsidethecabins. On these assumptions, several control systems, both active and passive,weredevelopedinordertoreducevibrationlevelsandhence improve the comfort. Soon, however, the vibration and noise fields were discovered to be not strictly related to each other, in the sense that controlling the vibrations was not always a guarantee of noise attenuation [2]. This was essentially due to the different capabilities of radiating noise by vibrations at different frequencies. The focus of the scientists was translated to noise reduction by considering noise parameters, and results were obtained in the automotive field for interior car noise reduction [3]: to increase the comfort in noisy interiors, it is necessary to consider the subjective response of passengers. The metrics related to comfort are not unique; a lot of psychoacoustic parameters [4] can be dealt with if the subjective response is desired to be taken into account: for instance, loudness, whichcanbeassociatedsomehowwithlow-band-frequencycontent; sharpness, which can be associated instead with high-bandfrequencycontent; tonality,whichtakesinto accountthepresence of response peaks in narrow bands; fluctuation strength, which considers variations of the spectral response during the time; roughness, which considers instead the pure broadband characteristics; and so on. The five mentioned parameters are, however, the most recurrent in the classical description [4]. The psychoacoustic parameters are estimated from both objective and subjective perspectives:estimationsareusuallymadethroughelaborationofthe physical output (acoustic response, hence objective) and elaboration of human answers (hence subjective). The two responses are correlated,butitisdifficulttoextractauniqueanddirectappreciation of the comfort impression. Another relevant problem is concerned with the necessity of performing very large experimental campaigns for collecting a limited amountofdata:ahugenumberofpeoplehavetobeinvolved in order to acquire just sufficient reference data that must be statisticallyanalyzed;theresponseofpeoplehastobeextractedfrom