An automobile manufacturer wishes to predict the failure probability of its cars at a sub-system level. Accurate failure probability prediction would allow the manufacturer to offer appropriate warranty periods to its customers and would also allow it to maintain its warranty reserves at optimal levels. Using a combination of parametric and non-parametric survival models, an automated rules engine is built to predict the failure probability of the different sub-systems and causal car parts. In addition, examination of the failure rates before & after engineering changes allows the manufacturer to measure the impact of such engineering changes on failure rates.
An automobile manufacturer wants to predict the failure probability of the cars produced by it during the cars’ warranty periods. The manufacturer wants these failure probability predictions at a car sub-system level (i.e., at a car part level). Typical car sub-systems are engines, brakes, etc. In addition, for the engine sub-system, primary causal parts are ignition, transmission, etc.
Accurate prediction of the failure probability of different car sub-systems is critical in order to estimate likely car recall rates. Vehicle recalls are sensitive for an automobile manufacturer, as these lead to additional costs that cannot be passed on to consumers. In addition, instances of substantial recall volumes are detrimental to the manufacturers brand perception and are a source of competitive disadvantage.
Manufacturers that are able to efficiently predict failure probabilities can offer appropriate and optimal warranty periods to their customers. In addition, accurate failure probability prediction allows manufacturers to optimize their warranty reserves.
Robust mechanisms to predict instances of car sub-system failure are also necessary to evaluate the effectiveness of engineering changes. As automobile manufacturers strive for continuous improvement and engage in continual modification of engineering designs & processes, it becomes imperative for manufacturers to measure the effect of such engineering changes on the failure rates of the different car sub-systems.
Both parametric as well as non-parametric techniques are used to estimate survival rates, failure rates and hazard rates of the various car sub-systems. Parametric survival models include those that make use of the Weibull, Pareto, Gamma and Exponential distributions. Non-parametric models utilize Kaplan Meier techniques. The survival modelling is conducted for each of the different sub-systems, such as engines, brakes, etc. For the engine sub-system, further drill-downs are employed for causal car parts, such as ignition, transmission, etc.
For each sub-system and causal car part, the selection between a parametric and a non-parametric survival model depends on the goodness of fit of the model’s output with the observed data.
In all instances where the car manufacturer has effected engineering changes, the survival-failure data is carefully segregated to isolate the “before change” and “after change” stages, so that the effect of the engineering changes can be examined.
The end result is a consolidated automated rules engine that is implemented as a front-end tool for business users.
Armed with an automated rules engine that predicts the failure probability of the different car sub-systems, the car manufacturer is able to accurately predict failure rates. This allows the manufacturer to offer appropriate warranty periods as well as maintain its warranty reserves at optimal levels. An additional benefit is the ability to measure the impact of engineering changes on failure rates for the various sub-systems.
