Acoustic emissions: diagnosing tribological phenomenon in artificial joint materials

Olorunlambe, Khadijat Abiola (2022). Acoustic emissions: diagnosing tribological phenomenon in artificial joint materials. University of Birmingham. Ph.D.

Preview

Olorunlambe2022PhD.pdf
Text - Accepted Version
Available under License All rights reserved.
Download (5MB) | Preview

Abstract

Studies have shown that many reported causes of failure of artificial joints such as hip, knee and spine are wear and friction related. Current modes of diagnosing failed artificial joints involve the use of imaging techniques like X-rays and CT scans, which although effective, are costly, time-consuming and harmful to patient health due to frequent exposure to radiation. There is the added limitation of the delay experienced before signs of failure become visible, causing further discomfort to the patient and, at times, health complications resulting from possible migration of wear debris into blood tissues. These complications have necessitated the need for a simpler and more dynamic system for identifying and diagnosing failed artificial joints, which is where the acoustic emission (AE) testing has shown promise.
AE testing is a non-destructive test method used to detect the onset and progression of mechanical flaws that has proven advantageous in the analysis and understanding of tribological interactions in mechanical systems. In recent times, it has been increasingly used in the study of the tribology of artificial and natural human joints thereby showing potential as a tool for the identification and diagnosis of failed artificial joints. Thus, this research aimed to use AE to monitor the tribological characteristics of artificial joint materials as a first step toward using AE to diagnose artificial and natural joint pathologies.
To gain an initial understanding of how AE features can be related to tribological mechanisms such as friction, in particular, a bio-tribo-acoustic tests system was developed. This enabled the acquisition of AE signals during biotribological testing of artificial joint materials. This proof-of-concept study showed that time-dependent (TDD) AE features can be used to predict the friction profile of a simulated polymer-metal artificial joint articulation. The prediction was carried out using a Non-linear Auto Regression with Exogeneous inputs (NARX) model. During testing of the trained model, predicted data had R2 values of 94% in tests on PEEK reciprocating at 2 Hz test and 98.6% for UHMWPE at 2 Hz. These regression results support the hypothesis that AE TDD features can be used to predict the friction profile which can then be related to the wear behaviour of the simulated joint articulation.
Having proved the potential of AE as a biotribological diagnostic tool, the next step is to be able to use the acquired AE signals to identify the perceived damage mode prompting the need for a method by which AE signals can be differentiated according to different wear mechanisms. To this end, AE signals from adhesive and abrasive wear, simulated under controlled joint conditions, were classified using supervised learning. Principal component analysis was used to derive uncorrelated AE features and then classified using three methods – logistic regression, k-nearest neighbours and back propagation (BP) neural network. The BP network emerged as the best performing network with a classification accuracy of 98%.
One of the limitations of traditional artificial neural networks (ANN) such as the BP network is the complex feature engineering required to obtain a model with high accuracy and high sensitivity. To mitigate this, deep transfer learning, with GoogLeNet as the base convolutional neural network (CNN) model, was used to classify AE signals from simulated damage mechanisms observed in retrieved polyethylene inserts of failed knee implants - burnishing and scratching wear. It was found that using CNN to extract features to be trained with an SVM model obtained a higher classification accuracy (99.3%) than just training with CNN model (96.5%).
The work presented in this thesis has shown that AE testing can be used to monitor the tribological properties of simulated articulating joint surfaces. With machine learning and deep transfer learning techniques, models with high accuracy and high sensitivity can be built to classify the acquired AE signals based on simulated real-life artificial joint damage modes. This confirms the initial hypothesis that with AE testing, a more dynamic, highly specific and highly sensitive process of identifying and diagnosing artificial joint pathologies can be developed, thereby reducing patient discomfort and NHS expenditure.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):