Automated Distributed Neuromorphic Pipelines for Wind Turbine Fault Analytics at Scale

Abstract

The growth in the use of wind farms of large scale has escalated the need to use fault analytics systems that can be used at low latency, high reliability and minimum manual intervention. Traditional centralised diagnostics pipelines are not scalable to increasing amounts of data and non-homogeneous conditions of turbine operations. This paper shows an automated distributed neuromorphic pipeline system of wind turbine fault analytics, which is able to work effectively with geographically dispersed assets. The suggested solution combines automated machine learning workflow orchestration, neuromorphic feature-to-spike conversion, as well as distributed spiking inference nodes to apply fault detection in scalable and low-energy fashion. In contrast to solutions that are digital twin based, or based on causal inference, the framework is based on end-to-end pipeline automation and distributed neuromorphic execution. A workflow engine with modular architecture is used to coordinate the data ingestion process, normalization of features, encoding spikes, scheduling inferences and aggregation of results without human intervention. Spiking neural network inference is performed at distributed neuromorphic nodes, which are located near data sources to minimize the overhead of communications, and respond faster. The experimental assessment through vibration-based and supervisory control information indicates that the suggested pipeline has lower-end to end latency, higher fault classification consistency, as well as better scalability than centralised deep learning pipelines. Findings also show that automated orchestration greatly simplifies system operations whilst keeping analytical accuracy within both high capacity and heavy system loads. The results provide distributed neuromorphic pipelines as a viable and scalable framework on the next generation wind turbine fault analytics.