Diagnostics of machines and structures: dynamic identification and damage detection

This research work deals with damage detection of engineering machines and structures. This topic, developed in particular for bearing diagnostics in the first part of the work, is strictly related to dynamic identification when structures are considered. Thus, subspace-based methods are investigated in the second part of the work, with particular attention to nonlinear system identification. Changes in operational and environmental conditions for structures (such as air temperature, temperature gradients, humidity, wind, etc.) or machines (such as oil temperature, loads, rotating regimes, etc.) are known to have considerable effects on signal features and, consequently, on the reliability of diagnostics. Useful tools for eliminating this influence are provided by a Principal Component Analysis (PCA)-based method for damage detection. The same way as many published works applied PCA-based diagnostics of structures, in this research work a bearing diagnostic application is considered. After a detailed description of the test rig, the huge amount of acquired data, on several different damaged bearings, is investigated. Results are useful for giving an overview on how the PCA-based method for damage detection can be applied on a complicated real-life machine. In general cases of real structures, the application of efficient identification techniques is crucial for correctly exploiting the capabilities of the PCA-based method for damage detection. Moreover, in many cases damage causes a structure that initially behaves in a predominantly linear manner to exhibit nonlinear response: the application of nonlinear system identification methods to the feature-extraction process can also be used as a direct detection of damage. For these reasons, a detailed study of the nonlinear subspace-based identification methods is presented in the second part of this work. Since the classical data-driven subspace method can in some cases be affected by memory limitation problems, two alternative techniques are developed and demonstrated on numerical and experimental applications. Moreover, a modal counterpart of the nonlinear subspace identification method is introduced, to extend its relevance also to realistic large engineering structures. In a conclusive application, two of the main sources of non-stationary dynamics, namely the time-variability and the presence of nonlinearity, are analysed through the analytical and experimental study of a time-varying inertia pendulum, having a nonlinear equation of motion due to its large swinging amplitudes.