Bosch Research has teamed up with the UK AI (artificial intelligence) specialist Fetch.ai to test a technology that will gather data from multiple machine users to build up a bigger picture of potential failures. The aim is to overcome the problem of relying on data from a single machine user where problems do not occur often.
The tests will use Cambridge-based Fetch.ai’s “collective learning” network technology and shared data sets to predict potential failures in Bosch machinery, while maintaining the data privacy of the individual machine users.
Bosch will use the collective learning technology to reduce the risk of equipment failures by helping multiple users train a machine-learning model and allowing each of them to vote on whether a model proposed by one of the participants has improved the performance on their local dataset. Bosch hopes this will help it to understand and adopt machine-learning algorithms relevant to manufacturing using publicly available datasets.
“Using machine learning to identify equipment failures is a difficult problem to solve as these events occur very infrequently,” explains Fetch.ai’s chief technology officer, Jonathan Ward. “The collective learning system enables the different manufacturers that use Bosch’s equipment to share information with each other without sharing the raw data, thereby greatly improving their ability to detect failures, and thus improve the efficiency of their operations.”
Bosch and Fetch.ai say that the decentralised nature of using collective learning techniques based on DLT (distributed ledger technology) to gain insights from the machine users’ data, without compromising their data security or privacy, will be the first implementation of its kind.
The project fits in with Bosch’s strategic direction of becoming a leading AIoT (artificial intelligence of things) company.
“Secure and trustworthy computation across several participants, while keeping the raw data and possibly even the learned model private is key to unlock the true value of distributed data,” says Dr Alexander Poddey, a Bosch AI researcher involved in an EoT (economy of things) advanced engineering project at Bosch, which is conducting the trials of the machine-learning technology. “In our view, collective learning is a key enabler to leading digital socio-economy to efficiency.”
Bosch Research fills the innovation pipeline for Bosch businesses and focuses on research and implementation of new technologies. It first entered into a collaboration with Fetch.ai in 2019 and deployed a node on a Fetch test network in early 2021. The launch of a collective learning proof-of-concept for Bosch’s manufacturing operations is an extension of that partnership that will evaluate the feasibility and effectiveness of this technique applied to predictive maintenance.