The project has been conducted in three stages, i.e. the initial technology scoping and design stage, the sub-scale testing and initial model development stage and the final full-scale testing and model validation stage. During the first stage, the design of the smart bearings inculding the smart system such as sensors, wireless communication systems, energy harvesters have been designed and preliminarily tested in laboratories before they were implimented on sub-scale smart bearings for testing. All devices are selected to meet the smart bearing in UHPE ground demonstrator requirements such as temperature, speed, load and noise conditions. The following sensors were selected for the smart bearing systems:
- Temperature: Thermocouple
- Load: Resistive strain gauge and fibre bragg grating sensor
- Vibration: Piezoelectric charge mode accelerometer
- Shaft speed: Eddy current probe
- Cage Speed: Inductive sensor
To enable onboard power supply to the sensing system in the smart bearing, thermoelectric generator (TEG) technologies have been explored. Together with ultracapacitors, power management board and microcontrollers, TEGs have been designed and tested in laboratory before it was implemented in the smart bearing systems on the sub-scale test rig.
To enable wireless data transmission and control, a wireless communication system that can be used in the UHPE environment (high temperature and surrounded by metallic housing parts) was designed and tested as part of the smart bearing solutions.
To evaluate the smart bearing design, a sub-scale bearing test rig was designed, developed and manufactured in the second stage of this project. Sub-scale smart bearing testing started in April 2019 and over 50 tests were conducted under a range of bearing conditions. Sensor data collected from the smart bearings have been used to develop bearing fault detection tools and ML-based models for automated health monitoring.
Three data processing and bearing health monitoring software platforms have been established, including
- A bespoke onboard data collection and display platform, which provides real time trending, detection and warning of rig/bearing faults relating to operation, e.g. failure of lubricant supply, and instantaneous sensor response information.
- An automated bearing fault detection toolbox (called “AtoB toolbox”), which enables bearing fault detection and diagnosis by tracking bearing characteristic frequencies through statistical analysis of the ‘big data’.
- A range of artificial intelligence (AI) and machine learning (ML) methods, such as deep neural networks and convolution neural networks have shown a potential to further improve the detection and diagnosis.
During the final stage of this project, the wireless communication system, including high temperature amplifiers that are designed for the UHPE environment, has been tested on full-scale rig.
A small number of full-scale bearing tests were conducted under UHPE flight cycle conditions. Bearing defects were seeded to experiment and initiate spalls and propogate them to failure for ML algorithm development. The results show that, based on the sub-scale and full-scale test data as well as bearing test data from literature, a novel hybrid signal processing method developed in this study can significantly improve the fault detection and diagnosis. By using the hybrid method, a generalised ML model has been developed for generic bearing fault detection and diagnosis. Details of the results have been captured by peer reviewed publications.