Predictive maintenance sensors are the vibration and temperature sensors that a vendor fits to machines that record nothing themselves, so that failures can be predicted from the sensor's own readings. Many Indian factories run motors and pumps that are twenty or thirty years old. These machines have no data link at all. The vendor adds one. This is called retrofitting, which means adding a new feature to an old machine.
Here is a typical case. A textile mill in Surat has a pump driven by an electric motor. A technician fixes a small sensor to the motor casing with a magnet or a screw. The sensor is about the size of a matchbox and runs on a battery. It measures vibration and temperature at fixed intervals and sends them over the mobile network to the vendor's cloud. Some other sensors send their readings to a small gateway box in the mill, which then sends them to the cloud. Nothing has to be connected to the pump's wiring or to the mill's control room.
The software does not know at first what is normal. It first spends a few weeks learning the pattern of this motor, and of similar motors the vendor already watches. After that, a change in the pattern is treated as a warning. A worn bearing, for instance, makes the motor shake in a particular way, and the sensor can pick up this change weeks before the bearing fails.
The vendor does not need the plant's own control system. This is the difference from software that reads data a plant already collects. Fitting sensors is slower, but it reaches machines that a process historian never recorded. Vendors usually charge a yearly fee for each sensor, so a mill can start with its ten most important machines.
The sensors are usually fitted to rotating machines such as electric motors, bearings and gearboxes and pumps. The vendor's alerts go to the maintenance team, and the readings can also feed a plant's own asset performance management (APM) software. The sensors send their data through the same kind of link as any industrial IoT device.