LSTM based fault detecti̇on wi̇th edge computi̇ng

The Internet of Things (IoT), which has emerged under the framework of Industry 4.0, plays a significant role by providing communication and control capabilities in various systems. It is estimated that there were approximately 20.4 billion IoT devices in 2020. However, the diversity and ubiquity of IoT devices also bring forth certain challenges. The first challenge is the difficulty in transmitting and processing data generated by IoT devices. Currently, cloud-based solutions are commonly preferred, and cloud computing approaches are employed in the design of IoT system architectures. However, allocating a separate cloud resource for each system can lead to latency issues and increased costs in real-time IoT systems. To overcome this disadvantage, designing systems based on edge computing approach can be more efficient. In this case, data is collected and processed at local points, reducing the computational load on the cloud, decreasing latency, and enabling resource savings. Another challenge is the detection of errors in IoT systems. Failure to detect problems in a timely manner, lack of spare parts inventory, and long lead times for replacement parts can disrupt system continuity. Literature studies indicate that artificial intelligence models are continuously trained with data obtained from IoT devices in the cloud layer and used for fault detection. LSTM (Long Short-Term Memory), a type of artificial intelligence model, has a recurrent neural network (RNN) architecture that can learn sequential dependencies in time series data. LSTM is a highly successful model for making predictions based on time series data. Considering that IoT data is often time-dependent, LSTM has an advantage over other models in this field. However, running such an artificial intelligence model in the cloud layer can pose challenges such as storage capacity, computational load, network congestion, and server costs. This study introduces the Edge Computing with LSTM-Based Fault Detection (UBLTAT) system, which combines cloud computing and edge computing infrastructure for processing IoT device data and detecting faults. UBLTAT transfers the data obtained from IoT devices to the cloud layer to train the LSTM model. Once the desired training accuracy is achieved, the weights of the model are transferred to UBLTAT, which operates at the edge layer, and fault detection is performed using the LSTM model in the edge layer. By running artificial intelligence models with UBLTAT at the edge, data generated by devices can be processed and faults can be detected locally, while also reducing the costs associated with the cloud.