<p>Recent technological advancements and the availability of monitoring equipment have facilitated the collection of time-series data in the mining industry, specifically for monitoring the dynamic loads on roof bolts. These data are important for ensuring miner safety and predicting bolt load variations over time. However, due to the challenging environment in coal mines, data quality can be compromised, highlighting the need for robust real-time anomaly detection. In this study, we proposed two deep-learning models, including the convolutional autoencoder and long short-term memory (LSTM) autoencoder, to detect anomalies in dynamic roof bolt load monitoring data. The convolutional autoencoder model’s encoder consists of three convolutional layers and three pooling layers, with a decoder composed of three transposed convolutional layers. In contrast, the LSTM autoencoder model leverages two LSTM layers in its encoder to capture sequential dependencies in a compact form, while the decoder uses two additional LSTM layers to expand the representation back to the original sequence length. Results indicate that these models effectively detect anomalous patterns, learning and simulating normal operational conditions with high accuracy. The convolutional autoencoder achieved a reconstruction error (MSE) of 0.004, while the LSTM autoencoder yielded an MSE of 0.14. An anomaly detection threshold was set using the sum of the mean reconstruction error and the standard deviation of the reconstruction error on the validation set. Both models detected multiple anomalies across bolts, with two bolts out of the studied bolts exhibiting a notably higher count. The LSTM autoencoder, optimized for sequential data, performed better in identifying time-dependent anomalies compared to the convolutional autoencoder. These findings demonstrate the potential of these deep-learning models for real-time anomaly detection, contributing to enhanced safety monitoring and predictive maintenance in underground mining environments.</p>