When components fail on complex machinery, the costs to the customer in terms of replacement parts and downtime soon start to add up. If parts fail when they are still under warranty, the producer bears these costs, so it is essential to calculate if this could happen. Estimating such factors can be a major challenge with parts made in small quantities, or with new models, and with machines that come in different varieties. The Stuttgart-based Steinbeis Transfer Center for Applied System Analysis (STASA) is specialized in the statistical analysis of complex technical systems and it now has a solution to this problem.
Schnell Motoren AG, an engine maker based in the city of Amtzell in Germany’s Allgäu region, has been selling modern cogeneration units for biogas plants since 1992. It develops its system in house and the components used vary dramatically in terms of wear and tear, mainly because the conditions on farms (where the equipment is typically used) can also vary dramatically, plus there are few standard setups. If parts fail while they are still under warranty, it can be an expensive process just trying to work out the causes. The experts at STASA have been helping and advising the company with the identification and prediction of component failures on its biogas plants, also in an attempt to predict the expected warranty costs in the future.
The project team’s first aim was to look at the engine components that account for most of the warranty costs and develop a technique to determine and predict their running time and service life. Based on this, the Steinbeis experts developed a statistical model that was individually tailored to the challenges faced by Schnell. This could predict the number of failure incidences and the resulting warranty costs for a period of 36 months. The advantage of using such a model is that scenarios can be set up using different assumptions for subsequent developments and this makes it possible to estimate the number and rate of failures that can be expected as well as the associated costs. This would make “guesstimating” warranty costs based on gut feelings a thing of the past.
Even if there are perfect customer service records, or running times have been logged – theoretically making it possible to calculate the known service life of an exchanged part – it is still difficult to capture and evaluate all failure statistics. Conventional statistical methods such as determining Weibull distributions based on component failures can only be used under certain circumstances. Consideration has to be given to the small number of instances, as well as the service life of certain components over time. For example, supplier quality problems can be enough to cause data blips and shorten service lives. It is also quite normal for components to have different average service lives on different types of units, simply because these cause differences in wear and tear.
The solution that was identified for Schnell showed that even a small number of failure incidences can be enough to make reliable forecasts of failure rates. What is important is the choice of method for the particular application and this has to be tailored properly to specific needs. The team at STASA analyzed the failure rates for individual components at Schnell. Based on this they developed a statistical model for forecasting future numbers. To do this, they took the statistics for part replacements and examined how these developed over time over the previous two years. They also looked at the running times of components still in active use. This has now made it possible to predict the failure rates of individual components in relation to the units that a company is expected to use in the future. By looking at replacement costs, it is also possible to calculate and predict anticipated warranty claims. This also takes into account that the warranty period is reset each time a component is replaced. Calculations can even be made for more complex warranty arrangements, such as the pro-rata warranties offered by Schnell Engines. With this model, clients only have to meet a portion of the costs after expiration of the statutory warranty period, irrespective of the actual running time of the defective part.
It is already possible to validate the results of forecasting. After just four months, the actual warranty costs are within the best-caseworst- case forecast made by STASA. Failure rates over time will be checked regularly by the experts at STASA so that adjustments can be made to warranty forecasts if necessary. The aim is to keep accruals needed to cover warranty costs within a sensible range. At the same time, regular checks are a useful early warning system that there may be a quality issue with suppliers. Summarizing the positive results of the project, Viktor Gaspar, Chief Operating Officer at Schnell, says, “The outcome of the analysis carried out by STASA made a decisive contribution to the accuracy of our strategy planning and it helped us by providing a good illustration of a difficult situation. We were extremely impressed not just with how quickly they got their minds around the complexity of the information but also how professional and amicable the collaboration was.”