From statistics to strategy

Traditional methods for predicting lifespan often only provide point forecasts without indicating uncertainty. This makes decisions about maintenance, recalls, or spare parts inventory risky. Additionally, many methods require deep expert knowledge and are not scalable.

IAV takes a different approach: with probabilistic models, not just a single value is estimated, but an entire probability distribution. This results in a much more robust picture of the future, providing a solid basis for decision-making.

The method: Probabilistic Lifetime Prediction (PLP)

IAV has developed the Hirundo service for lifespan prediction (Probabilistic Lifetime Prediction), which combines statistics and machine learning. This patented method has been validated by a leading research institute and is now used there as a standard for warranty forecasts.

The PLP Hirundo is available as Software-as-a-Service – fully automated, scalable, and easy to use. All that is needed are common fleet data such as aging index and timestamps. The software takes care of the rest: from data cleaning to model selection to result presentation.

Two application examples from automotive and non-automotive practice

1. Predicting component failures

In the first example, failure statistics and fleet movements were combined. An individual mileage distribution is modeled for each vehicle. The failure probability is calculated using a weighted Weibull distribution – including uncertainty assessment. The result: a precise prediction of the number of future failures over defined periods.

 2. Predicting the lifespan of smart household appliances

In the second example, the method is applied to smart washing machines. The devices continuously transmit data to the manufacturer via the cloud – including error events such as motor wear or pump defects as well as aging indicators, such as the number of wash or rinse cycles. An individual usage distribution is modeled for each washing machine based on this data. The failure probability of individual components can thus be calculated using a weighted probability distribution – including explicit uncertainty assessment. The result: a precise forecast of the number of expected failures in the entire device inventory over defined periods.

The benefit: More security, lower costs

• With the PLP Hirundo, OEMs receive a tool that goes far beyond traditional forecasting methods:
• Reliable risk assessment: Uncertainties are explicitly considered.
• Optimized spare parts logistics: Demand-oriented inventory instead of overproduction.
• Efficient recall planning: Early identification of critical components.
• Cost reduction: Through preventive measures and targeted maintenance.

The future is predictable

Probabilistic models like PLP show that modern data analysis can do more than just recognize trends. It creates trust in predictions – and thus in decisions. For OEMs, this means: fewer surprises, more control, and a real competitive advantage.