| Résumé | As aircraft fleets are required to expand their roles and usage, the accurate estimation of component loads in a helicopter is an important capability for safety and security reasons as well as for life cycle management and life extension efforts. Although dynamic component loads can be measured and monitored directly, these measurement methods are not reliable and are costly and difficult to maintain. Computational intelligence techniques have been successfully used for estimating helicopter dynamic loads and their resulting fatigue life using flight system and control parameters. However, other approaches work on low dimensional spaces with the advantage of smaller number of features and noise reduction due to information fusion. Nonlinear transformations have been used for this purpose, but their computation via implicit methods becomes more complex, time consuming and impractical with data growth. Moreover, the relationships between the features of the original and the target spaces are more difficult to uncover. Extreme Learning Machines (ELM) are used as an explicit functional representation for implicit methods, in particular for the t-SNE mapping. It was found that ELMs provided a good approximation to the implicit mapping, which preserves the appropriateness of the load prediction and damage estimation of critical helicopter components. In addition, the ELM model can be used for processing incoming streams of data, overcoming the limitation of the computation of the low dimensional mapping inherent to the use of implicit methods. |
|---|
