Digital service engineering: Remote diagnostics and prediction of product field failures for healthcare and electronic equipment based on event logs and manufacturing measurements

Service or field performance phase of product life cycle is becoming increasingly important to reduce cost and improve quality of service provided to the customer. Current quality improvement efforts for design and manufacturing phases may not be sufficient to guarantee product's satisfactory performance in field.;This thesis focuses on integrated analysis of product life cycle for (i) Improving quality and robustness of product in field: This is based on warranty information for mass produced products such as cell-phones which contains limited field information in the form of nature of customer complaint but substantial stored manufacturing measurements; (ii) Field failure prediction and diagnosis for optimum service interventions: This is performed using the in-situ field performance monitoring for critical and expensive equipment such as medical instruments (MRI, CT etc.) which have extensive field information in the form of event logs.;Warranty analysis focuses on Fault Region Localization (FRL) based on generalized rough sets approach to determine key-manufacturing measurements related to the failure region in the tolerance. Based on the FRL analysis, we propose a FC-Space based process monitoring scheme in manufacturing to prevent warranty failures. Current advances in data acquisition and storage technology have enabled monitoring of products field performance in the form of event logs wherein all events occurring during system operations are recorded which can be used for failure diagnosis and prediction. The methodology uses historical event logs to obtain fault patterns associated with various failure modes to construct a fault library. A dynamic programming based inexact matching approach is proposed to provide optimum alignment scores of the patterns with the event log. Further the statistical significance of the optimal alignment scores is calculated to determine the most probable pattern match for diagnosis. The optimum failure prediction point is determined by considering the expected value of future events and the deterioration of the system condition as more events are observed. With respect to all the aforementioned research topics, corresponding case studies are provided to demonstrate the presented methodologies.