The classical approach to machinery fault detection is one where a machinery’s condition is constantly compared to an established baseline with deviations indicating the occurrence of a fault. With the absence of a well-established baseline, fault detection for variable duty machinery requires the use of complex machine learning and signal processing tools. These tools require extensive data collection and expert knowledge which limits their use for industrial applications.

The thesis at hand investigates the problem of fault detection for a specific class of variable duty machinery;​ parallel machines with simultaneously loaded subsystems. As an industrial case study, the parallel drive stations of a novel material haulage system have been instrumented to confirm the mechanical response similarity between simultaneously loaded machines. Using a table-top fault simulator, a preliminary statistical algorithm was then developed for fault detection in bearings under non-stationary operation. Unlike other state of the art fault detection techniques used in monitoring variable duty machinery, the proposed algorithm avoided the need for complex machine learning tools and required no previous training.

The limitations of the initial experimental setup necessitated the development of a new machinery fault simulator to expand the investigation to include transmission systems. The design, manufacturing and setup of the various subsystems within the new simulator are covered in this manuscript including the mechanical, hydraulic and control subsystems. To ensure that the new simulator has successfully met its design objectives, extensive data collection and analysis has been completed and is presented in this thesis.

The results confirmed that the developed machine truly represents the operation of a simultaneously loaded machine and as such would serve as a research tool for investigating the application of classical fault detection techniques to parallel machines in non-stationary operation.