I must admit I had some trouble with the earliest sensorless drives. Sometimes auto tune did not work and the drives did not follow the speed reference exactly. We were then forced to call the factory and request assistance in manually tuning the drive. Confidence in the sensorless drive was reduced after a few failures like this. It is rare for auto tune to fail today.
Only a very brave person would go sensorless for a line pacer, unwind or windup drive. Pay for the sensor here. The sensor will probably be required for diameter calculations.
Sensorless drives may be used for some web handling applications. I would consider sensorless for an accumulator screw drive, a turret index drive, or roll handling drives. I would also consider sensorless drives for some intermediate web handling drives, particularly if these drives are repeated several times. It may be wise to put an encoder on one of these repeated drives simply to improve confidence (verify) that all these drives are operating as expected.
Sensorless drives operate by creating a model of the speed of the motor and using this model instead of encoder feedback in regulating the speed. In modern state-controllers this is referred to as an estimator or observer. Because of the volume of drives sold every year, the models have become very good. A large number of models have been used inside sensorless drives.
Some signs indicating the encoderless drive is not working correctly is that the speed is not correct at zero or very low speed (<2%). I have seen examples of roll handling carts stalling when used with sensorless drives. I have seen cases of turret index drives stalling with sensorless drives. I have seen cases of conveyors used for threading web failing to hold accurate speed. Of course with no encoder it is difficult to demonstrate that the speed is incorrect. With effort, all of these problems have been successfully resolved. When present, the cost (in time) of solving the problems exceeded the savings of going sensorless.