<p>Operation costs and safety are vital in most engineering industries. An effective machinery fault diagnosis approach is required to minimise operation costs and ensure safety. An algorithm known as extreme learning machine (ELM) constitutes a machine learning paradigm that effectively resolves several issues associated with traditional machine learning approaches. However, the efficacy of the ELM may exhibit instability and imprecision attributable to its parameters, which encompass hidden neurons number, bias, and input weights. Hence, this article introduced an integration of ELM and the greater cane rat algorithm (GCRA), also called the ELM-GCRA method, to select parameter values for ELM. First, an optimizer known as GCRA is used to select ELM parameter values. Second, the selected value is then fed into the ELM-GCRA for fault classification. In this study, statistical features pertinent to the time domain are employed to derive significant insights from the vibration signals, which encompass data from online bearing datasets, experimental datasets pertaining to bearings, and datasets associated with wind turbine gearboxes. The findings indicate that the ELM-GCRA approach enhances both the robustness and efficacy of the ELM technique. Additionally, it is able to surpass and compete with recent proposed fault diagnostic approaches.</p>