Modern industrial processors, engineering systems and structures, have grown significantly in complexity and in scale during the recent years. Therefore, there is an increase in the demand for automatic processors, to avoid faults and severe break downs, through predictive maintenance. In this context, the research into nonlinear systems analysis has attained much interest in recent years as linear models cannot be used to represent some of these systems. In the field of control systems, the analysis of such systems is conducted in the frequency domain using methods of Frequency Response Analysis. Generalised Frequency Response Functions (GFRFs) and the Nonlinear Output Frequency Response Functions (NOFRFs) are Frequency Response Analysis techniques used for the analysis of nonlinear dynamical behaviour in the frequency domain. The problem of Condition Monitoring and Fault Diagnosis has been investigated in the perspective of modelling, signal processing and multivariate statistical analysis, data-driven methods such as neural networks have gained significant popularity. This is because possible faulty conditions related to complex systems are often difficult to interpret. In such a background, recently, a new data-driven approach based on a systems perspective has been proposed. This approach uses a controls systems analysis method of System Identification and Frequency Response Analysis and has been shown before as a potential technique. However, this approach has certain practical concerns regarding real-world applications. Motivated by these concerns in this thesis, the following contributions are put forward:
1. The method of evaluating NOFRFs, using input-output data of a nonlinear system may experience numerical errors. This is a major concern, hence the development of a method to overcome these numerical issues effectively.
2. Frequency Response Analysis cannot be used in its current state for nonlinear systems that exhibit severe nonlinear behaviour. Although theoretically, it has been argued that this is possible, even though, it has been impossible in a practical point of view. Therefore, the possibility and the manner in which Frequency Response Analysis can be conducted for these types of systems is presented.
3. Development of a System Identification methodology to overcome the issues of inadequately exciting inputs and appropriately capturing system dynamics under general circumstances of Condition Monitoring and Fault Diagnosis.
In addition to the above, the novel implementation of a control systems analysis approach is implemented in characterising corrosion, crack depth and crack length on metal samples. The approach is applied to the data collected, using a newly proposed non-invasive Structural Health Monitoring method called RFID (Radio Frequency IDentification) wireless eddy current probing. The control systems analysis approach along with the RFID wireless eddy current probing method shows the clear potential of being a new technology in non-invasive Structural Health Monitoring systems.
