A Normalized Weighted-Coverage Model for Reference Vehicle Selection in Automotive Test-Fleet Software Updates

Modern automotive test fleets contain up to several hundred electronic control units (ECUs), each with its own combination of hardware variants, software versions and calibration datasets. Fleet software updates proceed in two waves: a small reference set is validated manually in an Initial Validation Phase, after which the remaining vehicles reuse the validated configurations automatically. This paper proposes a normalized weighted hardware coverage model in which the weight of a hardware variant follows its fleet frequency and compensates for logical duplication across ECU and diagnostic-address combinations. Reference vehicle selection is formulated as weighted maximum coverage under a cardinality constraint, and the objective is proved monotone and submodular, so that the greedy algorithm carries the classical (1 − 1/e) approximation guarantee. The recurrence is realized inside a business-intelligence engine that retains no state between evaluations, which required materializing the selected set explicitly. On anonymized industrial fleet data, ten reference vehicles cover 704 hardware variants in a homogeneous cluster and 697 in a heterogeneous one, corresponding to 95.1% and 79.7% of the respective weighted cluster objectives. Both selections attain 98.8% of the exact integer-programming optimum and exceed the mean coverage of 1000 random selections by 43% and 53%. The model applies to any fleet whose configuration coverage must be maximized under a budget constraint.