HybriDX - Hybrid fault diagnosis of unknown faults in industrial systems

Purpose and goal

The HybriDX project aims to develop an AI-based decision support for monitoring and fault diagnosis in complex industrial systems. The goal is to enable condition-based maintenance instead of time-controlled service intervals, to reduce unplanned downtime, costs and environmental impact. The project focuses on combining physical insights with data-driven models and machine learning to estimate degradation and isolate faults in a user-friendly interface for human–AI collaboration.

Expected effects and result

This project will develop methods for designing machine learning models that can support an operator to reason about fault scenarios that the models have not seen before. The methods in this project will streamline the development of new diagnostic systems for industrial applications, which will increase the competitiveness of Swedish industry.

Planned approach and implementation

The work is divided into a number of work packages focusing on: case study and data collection, design of fault detectors and health indicators, data analysis and decision support, and development and evaluation of a diagnosis-based decision support. Each work package contributes different parts to the final prototype that will be evaluated on realistic case studies together with engineers at Siemens Energy.