In the automotive industry, multiple prototypes are used for vehicle development purposes. These prototypes are typically put through rigorous testing, both under accelerated and real world conditions, to ensure that all the problems related to design, manufacturing, process etc. are identified and solved before it reaches the hands of the customer. One of the challenges faced in testing, is the low repeatability of the real world tests. This may be predominantly due to changes in the test conditions over a period of time like road, traffic, climate etc. Estimating the repeatability of a real world test has been difficult due to the complex and multiple parameters that are usually involved in a vehicle level test and the time correlation between different runs of a real world test does not exist. In such a scenario, the popular and the well-known univariate correlation methods do not yield the best results. The current work deals with the development of a new repeatability analysis approach using multivariate analysis.
The technique is developed with a non-parametric multivariate method called Mantel test which brings down all the complex parameters of the analysis to one number for checking the repeatability and take corrective measures accordingly. The Mantel test compares two matrices, i.e. two multivariate distributions in this case to assess the correlation between the two matrices. This method is discussed in detail for real world driving conditions. The analysis is carried out by collecting the data from a passenger vehicle driven for 10,000 kilometers with 34 different drivers. The paper further discusses about a repeatability analysis procedure. This procedure compares a base test and the data collected from the 34 drivers for repeatability. Implementation of Data Depth plot, a multivariate method for such applications, is also reported.