Automated Analysis of Automotive Read-Out Data for Better Decision Making

Abstract: The modern automobile is a complex electromechanical system controlled by control systems which consist of several interdependent electronic control units (ECUs). Analysis of the data generated by these modules is very important in order to observe the interesting patterns among data. At Volvo Cars Corporation today, diagnostic read-out data is retrieved from client machines installed at workshops in different countries around the world. The problem with this data is that it does not show a clear picture as what is causing what i.e. tracking the problem. Diagnostic engineers at Volvo Cars Corporation perform routine based statistical analysis of diagnostic read-out data manually, which is time consuming and tedious work. Moreover, this analysis is restricted to basic level mainly statistical analysis of diagnostic readout data. We present an approach based on statistical analysis and cluster analysis. Our approach focused on analysing the data from a pure statistical stand-point to isolate the problem in diagnostic read-out data, thereby helping to visualize and analyse the nature of the problem at hand. Different general statistical formulae were applied to get meaningful information from large amount of DRO data. Cluster analysis was carried out to get clusters consisting of similar trouble codes. Different methods and techniques were considered for the purpose of cluster analysis. Hierarchical and non-hierarchical clusters were extracted by applying appropriate algorithms. The results obtained from the thesis work show that the diagnostic read-out data consist of independent and interdependent fault codes. Groups were generated which consist of similar trouble codes. Furthermore, corresponding factors from freeze frame data which shows significant variation for these groups were also extracted. These faults, groups of faults and factors were later interpreted and validated by diagnostic engineers.