Overview

This thesis investigates statistical hierarchical modelling for collaborative prognosis and anomaly detection in industrial asset fleets. With the increasing reliance on machine learning for predictive maintenance, modeling challenges arise due to diverse operating conditions and data sparsity in some assets. Hierarchical models offer a structured approach to address these challenges by leveraging shared information across similar assets.

Aims:

• Develop a systematic approach for collaborative prognosis using hierarchical modeling.
• Improve predictive maintenance by addressing bias-variance trade-offs in fleet-wide and independent models.
• Demonstrate the effectiveness of hierarchical models for anomaly detection in industrial asset data.

Methodology

• Formulate hierarchical models where lower-level asset-specific parameters are sampled from higher-level distributions.
• Enable knowledge transfer across assets by incorporating similarities due to age, maintenance, and manufacturing processes.
• Compare hierarchical models with independent and fleet-wide models in terms of accuracy and variance.
• Validate findings using a case study on a fleet of long-haul trucks.

PhD Student

Maharshi Dhada

Supervisor

Prof. Ajith Parlikad

Sponsor

Siemens Turbo-machinery UK