Machines are a key part of the infrastructure that powers modern life. From the pumps that gives us fresh water in the tap, to transformers that distributes electricity, and the ventilation systems that keeps our buildings airy and cool. Millions of such machines exist, and they should all operate consistently without faults. Monitoring the health of the machines continiously can be done using sensors, detecting anomalies and triggering alarms - preferably before downtime occurs. Soundsensing has developed sound and vibration sensors that use Machine Learning to perform this task. Edge processing enables this to be done while maintaining privacy, and keeping the amount of data transfer and storage down. In this talk we will share some of the problems we are working on, challenges we face and solutions that we came up with.