An IMU-based anomaly detection mechanism for tower structures and model evaluation

Watching over the construction integrity of tall things like power sending sticks and talk-to-sky towers has problems, especially when it is needed to react right away, use little bit of power, and have not much to do with making plans. This study uses a mixed approach of dynamic threshold for triggering with deep learning classification model for classifying events and IMUs as the primary sensor device. The experiment has made a small model of experimental setup to imitate both normal and false operation, so as to obtain representative movement and vibration data. To assess the classification results, these three neural networks were used, CNN, LSTM and Transformer. Every model was correct and in agreement when identifying structural errors, making the light weight option a viable option for an embedded or edge based monitoring system.