Machine Learning for Engineering Knowledge Capture, MALEKC

Important results from the project

The goal was: “To create conditions to effectively identify and reuse knowledge by utilising machine-learning algorithms on engineering change reports (ECR)”. We have utilised Markov chains to identify patterns and NLP to create an efficient search to identify and cross reference knowledge in engineering change reports ECR. Even though the responses from the industrial test was better than expected there are hurdles to cross. E.g. the knowledge asked for was quite general, whereas spearhead knowledge may be more difficult to ask questions about and may be easily misunderstood.

Expected long term effects

The goal was to identify new ways of automatically identifying new knowledge through machine learning algorithms and then following up this work with validation with the help of smart assistants. This work has largely succeeded to the technical level aimed for in the proposal. However, work remains to make a cohesive chain work. Searching for knowledge works in itself, and we have learned a lot about how to ask questions to get useful answers, but the work is far from done.

Approach and implementation

The project has been carried out in four different themes, the first two concerned the development of advanced search in ECR databases. The two latter themes dealt with methodology development for knowledge reuse and experimentation on knowledge validation. The main applicants in the project were Chalmers and the Wingquist laboratory, the other participants being Fraunhofer Chalmers Centre (FCC), AB Volvo and Rejmes Transportfordon AB. 6 publications, 4 master´s theses and 2 doctoral dissertations have been written connected to the project.