Statistical evaluation of the performance of a mechatronic human leg prosthesis

Abstract

Modern devices constituting a supply for lost extremities experienced a boost in product development in the last decades. The work at hand focuses on a performance analysis of one of the latest models of lower limb prosthesis. The herein observed knee joint (OKJ) facilitates almost natural walking behaviour and several additional functions such as running or cycling. The embedded software allows for individual customization and closed-loop control of the product. Moreover, automated data acquisition is implemented in the device to satisfy the mandatory documentation regulation of the Food and Drug Administration (FDA). Therefore e.g. customization and service procedures are documented, delivering datasets which are the basis of a fundamental statistical device assessment. In this work the performance of the OKJ through analysis of field data acquired in FDA reports was evaluated. The hereby obtained data was used for an assessment of future perspectives for service engineering. First, promising data sources were elected and preconditioning including subset linking and a reliability check was performed. Thereafter several multivariate statistical approaches were used to analyse data acquired during the first service process containing usage information of devices with early service demand. Results of an exploratory attempt of the Principal Component Analysis (PCA), verified through the Partial Least Squares - Discriminant Analysis (PLS-DA) showed the influences of usage features on the frequency of part exchange. Herein a -special- usage leading to e.g. a long duration of device usage, extraordinary maximal device temperature values or the selection of a certain assembly was shown to be related to an alteration of the exchange frequency of specific parts. A multiple linear regression approach was performed to obtain a history-based data driven prediction model. The service demand date of devices produced and returned in 2014 was estimated with a standard deviation of approx. 40 days. In addition, strategic inputs on the procedure for evolving from reactive to proactive service were delivered. Besides a variety of suggestions for future data management, the obtained results of this study establish an opportunity for fundamental product and service value optimization.