A universal, automated approach to condition-based maintenance of drilling rig mud pumps is developed using acoustic emission sensors and deep learning models for early detection of pump failures to help mitigate and reduce costs and nonproductive time generally associated with catastrophic pump failures.