Sucker rod pumping systems are the most widely used artificial lift methods in the oil and gas industry. These pumps require regular maintenance to ensure that they are operating at optimal levels, and failure to do so can result in costly downtime and high maintenance costs. The operating circumstances are reflected in the shape of dynamometer cards, and various conditions can be identified by the card’s typical features. In this paper, sensor-collected data from dynamometer cards is utilized for feature extraction and dimensionality reduction using Elliptical Fourier Descriptors, Principle Component Analysis (PCA), and Convolutional Autoencoders. Machine learning models like Convolutional Variational Autoencoders (CNN-VAE), Long Short-Term Memory (LSTM) autoencoders and Support Vector Machine (SVM) models are used to generate some synthetic data and train on a combination of real and synthetic data. This approach can save significant costs associated with downtime and repairs while improving overall system performance.