Towards Collaborative Predictive Maintenance Leveraging Private Cross-Company Data

Abstract

The accuracy of a predictive maintenance model is largely determined by the available training data. This puts such machine learning systems out of reach for small and medium-sized production engineering companies, as they are often unable to provide training data in sufficient quality and quantity. Building a collaborative model by pooling training data across many companies would solve this issue, but this data cannot simply be consolidated in a central location while at the same time preserving data integrity and security. This paper enables a collaborative model for predictive maintenance on cross-company data without exposing participants' business information by connecting two recent methodologies: blockchain and federated learning.