Poster: Lambda architecture for robust condition based maintenance with simulated failure modes

Condition based maintenance (CBM) is increasingly seen as a promising approach for addressing downtime issues which are a common occurrence in the manufacturing industry and are a major cause of lost productivity. However, it has been a challenge to develop a generic CBM solution that works for all assets since each asset has unique sources of noise. This mandates use of manual diagnostics to custom tailor a solution for each asset for accurate failure mode identification (FMI). This problem is further compounded by the scarcity of failure data. In this paper, we propose a lambda architecture for FMI of industrial assets that achieves low initial deployment cost while securing a reasonable classification accuracy. The lambda architecture consists of a light-compute edge node, such as Raspberry Pi, that processes high-speed vibration data in real-time to extract useful features and applies a deep-learning (DL) engine which is trained in a cloud platform, such as AWS. In addition, we also incorporate a failure modes' feature simulator so that DL models can adapt to different industrial assets without costly failure data collection. Finally, experimental results are provided using the bearings' failures dataset validating the proposed cost-effective CBM architecture with high accuracy and scalability.