ABSTRACT
In practice, reliability analyses in the automotive industry are done for various purposes. First of all field data analysis is used to evaluate the potential of an emerging failure mode, to estimate the number of expected failures within a defined timespan or to get a risk estimate for single vehicles, to concentrate campaigns to the group of vehicles under the highest risk. Further manufacturers (OEMs) are often interested in information about the general failure behavior of a component, e.g. its Weibull shape parameter. This information helps, in combination with further reliability analyses of testing results, to evaluate the reliability of components during the development. The performance of all these applications highly depends on the failure model’s ability to explain the actual damage process that leads to a failure. In many years of working in the automotive industry, we noticed in a lot of cases that this damage process is more complex than common univariate reliability models are able to explain. This observation is reflected in false failure prediction and significantly changing results of reliability analyses over time. The main reason for this circumstance is, that one single variable rarely includes all the necessary information, to fully describe the complexity of a failure mode. In common univariate reliability analyses, these single variables are mostly either mileage or time in service. The true stress that a component is experiencing is the result of
to increase. Some of the most relevant aspects are given by these three following categories:
• The drivers behavior Sportive driving Fuel efficient driving
• The driving conditions Quality of the roads Traffic conditions, e.g. frequent traffic jams Climate
• Main purpose of the vehicle Private usage Commuter traffic Professional usage, e.g. Taxi vehicles
Thus in general driving profiles are complex there are two reasons why only reduced profiles are often used to work with in reliability models. First of all most automotive OEM’s possibilities are limited by the available data. In the case of customer data analysis, they are facing the situation of very big samples but only limited information about each vehicle. Further an obvious assumption is that only a part of the information is relevant regarding a component’s failure behavior. Some of the possible variables can be excluded from the start on, due to no expected influence on the failure mode.
