In industrial process automation, sensors (pressure, temperature, etc.), controllers, and actuators (solenoid valves, electro-mechanical relays, circuit breakers, motors, etc.) make sure that production lines are working under the pre-defined conditions. When these systems malfunction or sometimes completely fail, alerts have to be generated in real-time to make sure not only production quality is not compromised but also safety of humans and equipment is assured. In this work, we describe the construction of a smart and real-time edge-based electronic product called PreMa, which is basically a sensor for monitoring the health of a Solenoid Valve (SV). PreMa is compact, low power, easy to install, and cost effective. It has data fidelity and measurement accuracy comparable to signals captured using high end equipment. The smart solenoid sensor runs TinyML, a compact version of TensorFlow (a.k.a. TFLite) machine learning framework. While fault detection inferencing is in-situ, model training uses mobile phones to accomplish the `on-device' training. Our product evaluation shows that the sensor is able to differentiate between the distinct types of faults. These faults include: (a) Spool stuck (b) Spring failure and (c) Under voltage. Furthermore, the product provides maintenance personnel, the remaining useful life (RUL) of the SV. The RUL provides assistance to decide valve replacement or otherwise. We perform an extensive evaluation on optimizing metrics related to performance of the entire system (i.e. embedded platform and the neural network model). The proposed implementation is such that, given any electro-mechanical actuator with similar transient response to that of the SV, the system is capable of condition monitoring, hence presenting a first of its kind generic infrastructure.