How an algorithm revolutionizes the road test in Volkswagen´s production
In all car manufacturing plants, every produced vehicle passes different driving and testing stations before being delivered to its new owner. The production processes are becoming more and more reliable, stable and the success rate of quality checkpoints is increasing constantly. In other words, plants are spending more and more time on testing cars without failures. This potential was addressed at Volkswagen Autoeuropa in Palmela. The Road Test Predictor application is a tool for quality inspectors to help them decide which vehicles should be tested and which can bypass the road test inspection process on the assembly line. This can result in significant time and cost savings, as well as an increase in productivity and carbon neutrality as fewer vehicles go through the road test and emissions are avoided. The Road Test Predictor application analyzes production data across more than 10,000 attributes per car and is also highly scalable to other brands and plants in the automotive industry. Read this success story about how the Palmela plant of Volkswagen Autoeuropa makes use of big data to predict production processes – a first insight into the future of manufacturing.
The statistical analysis of data we perform with the Road Test Predictor application shows us what is negatively influencing the road test outcome. This is an example of how data helps us to better understand and optimize our processes
Challenge – the hidden knowledge of production data vs. mandatory road tests
Safety is a top priority at Volkswagen. To achieve this, a car undergoes several tests at the end of production, including water or road tests in which each car is exposed to the respective situation under harsh conditions for several minutes. During the tests, quality experts are checking possible failures. To make sure that conditions are as close to reality as possible and to ensure the accuracy of the results, the test facilities have been adapted with comprehensive equipment to simulate the dynamic testing and generate data about the vehicles. Further data on e.g. torques, status, possible defects are required and stored for compliance reasons. Thanks to facility optimization and transparency in production processes, the results of the tests are increasingly reliable and fewer cars have to be reworked after for example the road test. Nevertheless, all cars are still tested. The unlocked potential of production data is a challenge to be discovered not only at Volkswagen Autoeuropa in Palmela, but in all Volkswagen Group production plants around the world as well as in the entire automotive industry.
Solution – leveraging knowledge from production data for precise road test predictions
An interdisciplinary team from Volkswagen Autoeuropa, Amazon Web Services (AWS), and the Industrial Cloud community partner Amorph Systems has taken advantage of the ideal conditions for collecting, using, and evaluating data from the road test in the assembly line in Palmela. The result is a tool that analyzes the generated production data and provides the probable outcome of the dynamic test – the Road Test Predictor application. With the expertise and integration platform SMARTUNIFIER of Amorph Systems, the production equipment of the plant is connected to the cloud. The system of the Road Test Predictor application is based on a lean AWS architecture that enables the data flow of each vehicle from the shopfloor to the prediction algorithm.
The algorithm behind the application uses data from three different databases and thus includes more than 10,000 attributes per vehicle, for example the time spent inside the paint shop or a single torque of a fastener in the front axle, into its calculations. The prediction then comes from the cloud back to the shopfloor in two to three seconds per operation. When the respective car arrives at the dynamic testing station, the prediction is available and visible in the application for the quality experts. The accuracy of the prediction was already above 97% at the beginning of 2021 in a test environment where the prediction is compared to the actual Road Test result in order to calculate the overall accuracy. The reliability of the results will increase in the future due to more data that will be included in the calculations. In the future, the result will be visualized in a traffic light, which provides the quality experts in the assembly line with a precise indication of whether a vehicle should go through the road test or not. On this basis, the expert can then decide and, in many cases, skip the road test and seamlessly route the vehicle to the next station. This saves several minutes per vehicle in production. All in all, the Road Test Predictor is an application that is easy and intuitive to use and provides precise predictions about the vehicle in production.
Key benefits – using what we have, it often does not take much to create new paradigms for production
The great part about the application is that most of the framework for using the technology is already in place: the production facilities are well equipped and take precise measurements, which only need to be consolidated and analyzed by the Road Test Predictor. No additional hardware is needed to implement the use case in other Volkswagen plants, and existing databases can be further added to the application with the result of increasing the number of attributes per car and therefore allowing an even higher precision of the predictions. Therefore, the Road Test Predictor is highly scalable to other plants of automotive manufacturers. Furthermore, the algorithm behind the application is generic, so the road test is just the beginning – other testing stations in production, such water testing, are already the focus of the development teams. The potential benefits are cost reduction and process optimization – much leaner processes are the first big steps in using big data for predictive analysis applied to car manufacturing processes.