IAV Hirundo - Probabilistic Lifetime Prediction

**Our solution**

• Probabilistic approach with inherent accuracy assessment. This allows to quantify the uncertainty of the forecast.
• Low budget risk. Our Software as a service scales with your needs.
• Widely applicable: Only commonly available data like age index and date from fleet and failure parts are required.
• Full automation from data cleansing to model selection.

**Background**

Warranty costs represent a significant expense, which can reach billions in the automotive industry. A significant fraction of these costs arise from the storage of parts that are either never needed or stored in insufficient quantities, leading to costly late-stage reproduction. A data-driven lifespan analysis that incorporates early data points into the durability forecast of key parts is crucial for reducing these costs.

Traditionally, lifespan analyses are conducted by fitting a single Weibull distribution to failure data using regression methods. Commonly, the estimates are point estimations without considering their uncertainty. As a result, the quality of the forecast is highly dependent on the procedure and the individual experience of the user.

Our novel approach models multiple failure mechanisms by estimating a weighted superposition of Weibull distributions, including their expected uncertainties. The optimal number of components and their training are determined with a probability-based method (EM Algorithm). The expected uncertainties are estimated with bootstrapping.

<b><i>Notice Regarding Use in the United States</i></b><br>

<i>Please note that our standard License Terms do not permit the use of this software within the United States of America. If your company is based in the U.S. or you intend to use the software in the U.S., please contact IAV directly before concluding any agreement. We will work with you to establish the appropriate conditions for compliant usage.</i>