Improving Process Traceability Using Deep-Learning Based Unsupervised Feature Extraction

Computer Vision and Deep learning are making visible inroads in the manufactur-ing industry. One of the application areas in manufacturing includes process trace-ability. Process Traceability refers to the process of tracing parts back to the source components (e.g., raw material, machine, tools) that were used to produce the part. In Bosch transmission technologies, one component of a continuously variable transmis-sion belt called as the elements are traced back to their source using manual methods based on measuring the part characteristics. The objective of this thesis is to auto-mate this process. In particular, an investigation is conducted with deep learning al-gorithms to extract features in images of elements automatically, which are then used to achieve traceability. Experiments were performed to identify the best performing model in terms of traceability performance. Results indicate that the parts could be traced with only moderate success using a deep learning based automated method. However, there are definitive hints from the results suggesting that the performance could be improved with a better design on data collection.