Many industries rely on the continual and unimpeded operation of turbines, pumps, pulleys, fans, motors, gearboxes, and other associated fixed and mobile plant equipment. As such, reliable and remote condition monitoring and fault localisation can improve safety, prevent unnecessary downtime and reduce maintenance costs. Current methods of condition monitoring such as acoustic emissions (AE) testing can prove difficult to automate and require careful analysis by a trained analyst. This research investigates the use of adaptive beamforming for source localisation and signal extraction in conjunction with a convolutional neural network classification system based on spectrogram plots. Furthermore, it tests the effects of reducing the number of microphones in the microphone array on the deep network classification accuracy. This technology has been investigated with the use of 12-volt computer fans as an analogue for rotating machinery, with the primary challenge of reliably separating and classifying the unique spectral signal of each fan. The outcome of this research from over 450 test samples demonstrates damage detection accuracy consistently above 97% based on available data when adequate beamforming resolution and array gain are achieved. This technology shows promise for use in an automated monitoring system for industrial applications, with available scope for further refinements.