STP1472

    Wear Scar Prediction Based on Wear Simulator Input Data—A Preliminary Artificial Neural Network Approach

    Published: Jan 2006


      Format Pages Price  
    PDF (524K) 8 $25   ADD TO CART
    Complete Source PDF (13M) 8 $55   ADD TO CART


    Abstract

    A significant difference in wear scar formation between tested and retrieved knee implants of the same type has been reported. In this study, an Artificial Neural Network (ANN) model has been designed with the aim to gain knowledge of relationships between simulator input parameters and generated wear scars. One hundred twenty-four short-term tests were conducted with four implants of a single design using a four-station knee simulator in load control mode. Data points of the wear scar boundaries were transferred into bitmap images for computer analysis. Eighty percent of these discretized wear scars formed the output training set for a back-propagation neural network. The input training set was selected from the related simulator input motion and load parameters. The remainder of the testing matrix was used for network cross-validation and testing. Training resulted in 82.9% accuracy of the input-to-output relationship and 69.3% predictive capability. The predictive capabilities of the network may be further enhanced by utilizing a modification of the learning algorithm.

    Keywords:

    wear, total knee prosthesis, wear simulation, polyethylene, UHMWPE, simulator tuning, backpropagation network


    Author Information:

    Orozco, D
    Graduate Student, Laboratory Manager, and Assistant Professor and Director, Rush University Medical Center, IL

    Schwenke, T
    Graduate Student, Laboratory Manager, and Assistant Professor and Director, Rush University Medical Center, IL

    Wimmer, MA
    Graduate Student, Laboratory Manager, and Assistant Professor and Director, Rush University Medical Center, IL


    Paper ID: STP40881S

    Committee/Subcommittee: F04.93

    DOI: 10.1520/STP40881S


    CrossRef ASTM International is a member of CrossRef.