The development of an in situ efficiency estimation technique is a challenging task where the lowest level of intrusion and the highest possible accuracy are required. In this paper, a new algorithm is discussed for the in situ efficiency estimation of induction machines under unbalanced power supplies. Prior work in the literature has concentrated on balanced supplies. In addition, to have a nonintrusive speed measurement, a specific adaptive nonlinear algorithm is applied for the extraction of the speed-dependent current harmonics from the measured current signal. A similar algorithm is used to extract the symmetrical components from the current and voltage signals to handle the unbalanced supply conditions. Experimental results with two different machines are used to prove the effectiveness and generality of the proposed method. Measurement error analysis, as well as repeatability tests, has been done to determine the credibility of the proposed method.Index Terms-Error analysis, evolutionary algorithm, induction motor, in situ efficiency estimation, nonlinear adaptive filter, repeatability test, unbalanced supply.

International efficiency testing standards such as the IEEE 112-B and IEC 34-2-1 can determine an induction machine's efficiency accurately at the cost of hindering the machine's productivity. Alternatively, various methods used to determine a machine's efficiency in-situ do so at the cost of accuracy. This paper proposes a method that determines an induction machine's efficiency over a range of load conditions from tests conducted and centered around one thermally stable load point in the least intrusive manner possible. The results are then compared to those of the IEEE 112-B and IEC 34-2-1 motor testing standards using segregated loss analysis. It was found that despite the proposed algorithm being unable to accurately determine core losses, the efficiency of a machine can be estimated to within 0.5-2.1% and 1.1-1.7% error when compared to the IEEE 112-B and IEC 34-2-1 standards respectively over a 25-150% machine load profile (dependent upon the proposed method's implementations).

No abstract