Abstarct: This thesis presents a novel approach for the classification of load unbalance faults in induction motors, caused by external resistance, through the analysis of image-baxsed data representations. The methodology leverages a combination of Convolutional Neural Networks (CNNs) and Transformer architectures to achieve robust fault diagnosis. The core challenge addressed is the conversion of one-dimensional time-series signal data, which constitutes the primary dataset, into two-dimensional textured images suitable for image-baxsed analysis. To this end, the Recurrence Plot (RP) technique was employed to transform the 1D signals into 2D images that encapsulate the phase space characteristics of the original signal. The developed deep learning model utilizes a CNN for feature extraction, dimensionality reduction, and conversion of the generated images into array formats compatible with a Transformer network. The Transformer architecture then serves as the primary classification engine. The study systematically investigates the impact of sampling frequency on classification accuracy. Initially, to ensure the model's generalizability and avoid prior knowledge bias, experiments were conducted using low-resolution images generated from limited signal Sampeling Rate (4 Hz, 8 Hz, and subsequently 16 Hz). Experimental results demonstrate a direct correlation between sampling frequency and classification accuracy. The findings confirm the initial hypothesis that increasing the sampling rate, particularly around the motor's nominal frequency, significantly enhances the model's ability to accurately classify load unbalance faults. This research establishes the viability of using image analysis techniques, specifically Recurrence Plots combined with a CNN-Transformer frxamework, for effective fault diagnosis in electric motors from time-series data.