This paper presents an LSTM-autoencoder deep learning model for anomaly detection in electric motors. The model combines LSTM layers with an autoencoder to effectively handle large amounts of temporal data. The study focuses on detecting anomalies in axial, radial, and tangential vibrations of an electric motor, which are indicative of potential faults or failures. The model is trained on a dataset collected from sensors installed on a machinery fault simulator. The results show that the LSTM-autoencoder model outperforms a regular autoencoder in terms of loss values and MSE anomalies, although it has a longer training time due to the complexity of the LSTM layers. The study also compares the performance of the LSTM-autoencoder and regular autoencoder models on three key aspects: training time, loss function, and MSE anomalies. The analysis demonstrates that the LSTM-autoencoder model has significantly smaller loss values and MSE anomalies compared to the regular autoencoder, while the regular autoencoder performs better in terms of training time. The study concludes that the LSTM-autoencoder model is more effective in detecting anomalies in electric motors, despite its slower training time. The results highlight the importance of using deep learning techniques for anomaly detection in industrial applications, particularly in predictive maintenance. The study also discusses the potential for future research in integrating real-time monitoring and feedback mechanisms, as well as comparing the LSTM-autoencoder model with other methods such as generative models, variational autoencoders, and temporal logic-based learning.This paper presents an LSTM-autoencoder deep learning model for anomaly detection in electric motors. The model combines LSTM layers with an autoencoder to effectively handle large amounts of temporal data. The study focuses on detecting anomalies in axial, radial, and tangential vibrations of an electric motor, which are indicative of potential faults or failures. The model is trained on a dataset collected from sensors installed on a machinery fault simulator. The results show that the LSTM-autoencoder model outperforms a regular autoencoder in terms of loss values and MSE anomalies, although it has a longer training time due to the complexity of the LSTM layers. The study also compares the performance of the LSTM-autoencoder and regular autoencoder models on three key aspects: training time, loss function, and MSE anomalies. The analysis demonstrates that the LSTM-autoencoder model has significantly smaller loss values and MSE anomalies compared to the regular autoencoder, while the regular autoencoder performs better in terms of training time. The study concludes that the LSTM-autoencoder model is more effective in detecting anomalies in electric motors, despite its slower training time. The results highlight the importance of using deep learning techniques for anomaly detection in industrial applications, particularly in predictive maintenance. The study also discusses the potential for future research in integrating real-time monitoring and feedback mechanisms, as well as comparing the LSTM-autoencoder model with other methods such as generative models, variational autoencoders, and temporal logic-based learning.