**4. Obtained Results**

The proposed technique efficacy was validated through 100 different trials under each treated motor condition. A holdout-type data set was employed, where the first 70 experiments were employed for teaching the network and the remaining ones were used for checking the proposed methodology efficacy. Table 1 shows the performance results, through a confusion matrix, of the introduced technique during the identification and classification of the IM operational condition as HLT, 1BRB, 2BRB, BRN, and UNB. The right-most column in Table 1 displays the average success rate of the suggested procedure.

The results obtained experimentally are depicted in Table 1 and they show that the introduced methodology reached 100% of effectiveness on identifying and discriminating among a healthy motor (HLT), a motor with one broken rotor bar (1BRB), a motor with two broken rotor bars (2BRB), a motor with outer-race bearing damage (BRN), and a motor with an unbalanced mechanical load (UNB). The proposed technique was executed in a 2.20-GHz Intel Core i7-8750 processor, making use of the software MATLAB 2020a.


**Table 1.** Confusion matrix and overall effectiveness of using homogeneity and kurtosis features for fault detection and classification in SCIM.

#### *Discussion*

The proposed methodology is compared against previous approaches used for detecting SCIM faults in Table 2. The obtained results from real experimentation demonstrated the usefulness of homogeneity and kurtosis as indexes for IM diagnosis and their high reliability as indicators for multiple fault identification and classification [25,26]. The proposed

approach can recognize and classify the operational condition of an induction motor as in a good state (HLT), a motor with one split rotor bar (1BRB), a motor with two damaged rotor bars (2BRB), a motor with outer race damage in the bearing (BRN), and a motor with an unbalanced mechanical load (UNB) with high certainty, attaining up to 100% of effectiveness, utilizing just two features of a single phase from the three-phase electrical current supply to the SCIM, as inputs to a multilayer perceptron ANN. This is different from other approaches reported in the reviewed literature that even require the signal transformation from the time domain into the frequency domain and back to the time domain to carry out the signal processing in order to extract up to 29 features of the current signals from the three phases and the multi-axis vibration signals, in conjunction, in order to be capable of performing the fault detection. In the proposed procedure, the electrical current signal from the SCIM startup transient was analyzed just in time domain without any pre-treatment, which is an evident convenience compared to the previous works in Table 2, which require the combination of two or more processing techniques to carry out a qualitative diagnosis or to perform it in a quantitative style by analyzing current and vibration signals in time, frequency, and even time-frequency domains. Therefore, the proposed methodology is a reliable tool that ensures high certainty during different fault detections and classifications in induction motors through the analysis of just one phase from startup electric current supply, outperforming previous approaches in the state of the art.

**Table 2.** Comparison chart of the proposed methodology against the state of the art in related literature for fault detection in IM.

