Hoists and cranes exist in many contexts around the world, often carrying veryheavy loads. The safety for the user and bystanders is of utmost importance. Thisthesis investigates whether it is possible to perform fault detection on a systemlevel, measuring the inputs and outputs of the system without introducing newsensors. The possibility of detecting dangerous faults while letting safe faultspass is also examined.A mathematical greybox model is developed and the unknown parametersare estimated using data from a labscale test crane. Validation is then performedwith other datasets to check the accuracy of the model. A linear observer of thesystem states is created using the model. Simulated fault injections are made,and different fault detection methods are applied to the residuals created withthe observer. The results show that dangerous faults in the system or the sensorsthemselves are detectable, while safe faults are disregarded in many cases.The idea of performing model-based fault detection from a system point ofview shows potential, and continued investigation is recommended.

This dissertation addresses fault detection and isolation (FDI) for nonlinear systems based on models using two different approaches. The first approach detects and isolates single and multiple faults, particularly when there are restrictions in measuring process variables. The FDI model-based method is based on nonlinear state estimators, in which the estimates are calculated under high filtering, and a high fidelity residuals model, obtained from the difference between measurements and estimates. In the second approach, a robust fault detection and isolation (RFDI) system, that handles both parameter estimation and parameters with uncertainties, is proposed in which complex models can be simplified with nonlinear functions so that they can be formulated as differential algebraic equations (DAE). In utilizing this framework, faults are identified by performing a statistical analysis. Finally, comparisons with existing data-driven approaches show that the proposed model-based methods are capable of distinguishing a fault from the diverse array of possible faults, a common occurrence in complex processes. / text