Before vehicles leave the factory, automobile manufacturers have to ensure their safety and reliability. To do this they conduct tests on test benches. There are massive amounts of different car configurations. These configurations result in different test durations on the test benches. In the automotive industry, test benches are critical bottleneck resources where simultaneous assignments and idling must be avoided at all costs.
The aim of the project was therefore to develop a model that precisely predicts the test time a vehicle will spend on the test bench depending on its configuration, so that scheduling of test bench allocations can be optimized.
The training dataset included 4,200 car configurations and the corresponding test times as well as 400 features.
The test dataset contained new 4,200 configurations and the same features, too. However, test times were missing.
Challenges & Solutions
Possessing a big amount of features and proportionally few data points marked the first challenge that could be solved with Dimension Reduction. Compared with other data analysis techniques, only the Multiple Correspondence Analysis (MCA) could reduce the provided 400 features to a total of 14 without any loss of information. That is why MCA served as basis for the developed Machine Learning Model.
The second challenge was the selection of the most appropriate regression algorithm as the different applicable algorithms are similar in principle. The training dataset was split and Cross Validation was applied. Fine-tuning the parameters of the chosen algorithm allowed for a prediction of the original test dataset.
Lacking of data confounding external factors could not be considered in this Machine Learning Model.
The Machine Learning Model that has made it possible to predict test bench times for all vehicle configurations. This allows test cycles to be run more efficiently in the future to better optimize both allocation and timing.
Through the use of the previously mentioned dimensionality reduction, it has also been possible to identify the most essential features influencing test time duration of a vehicle.