With an increasing number of influencing features, and a rising complexity in the industrial manufacturing world the identification of important relationships requires algorithmic support. Algorithms for causal inference use conditional independence (CI) tests to receive information about underlying relationships. These inferred causal relationships can be used to derive insights into the behaviour of complex production processes, and as basis for new applications that allow to monitor, predict, or optimize the complex processes in the Internet of Things (IoT).
In this project, we apply causal inference to car production processes. The car production process is split into several work groups each with a distinct responsibility, e.g., assembly of a part, paintwork or finishing. Within each workgroup, robots and sensors produce a constant stream of data. While robots in production provide status and error messages, sensors measure influences from the surrounding environment. The knowledge about the causal relationships in this complex setting allows to improve the detection and prediction of failures within the car production process.
The project will be a joint effort of HPI and Porsche AG/MHP, represented by Porsche Digital Lab Berlin (PDLB).