System models combine tens or hundreds of failures modes. Although system models may be represented by lognormal or even binomial distributions, the Weibull distribution is used most often. The combination can be done by Monte Carlo simulation or by analytical methods. If the data cannot be segregated into individual failure modes or if early data is missing, the Crow-AMSAA or the Kaplan-Meier models may be applied to provide trending and failure forecasting.

System models are useful for predicting spare parts usage, availability, module returns to depot and maintainability support costs. System models are frequently updated with the latest Weibull probability plots. Past predictions may be compared with actual results to estimate the model uncertainties and fine tune the model.

For complex systems, early failure modes are likely to “cover” later failure modes. This means that unless early failure modes are eliminated, later failure modes are never identified. For this reason, complex systems that involve safety are exposed to accelerating testing well beyond their design life to uncover and eliminate any later failure modes that may be catastrophic. Because all problems are never found or solved, there are always unknown failure modes that will occur in the future.