An IoT architecture based on the control of Bio Inspired manufacturing system for the detection of anomalies with vibration sensors

Abstract

This work presents an IoT architecture for the detection of anomalies in motors with vibration sensors using a real-time autoen-coder, based on a new bio-inspired control architecture for recently proposed manufacturing systems. Unlike other approaches, this work analyzes the behavior of the anomaly detection system in real time, seeking to cover the new requirements for real-time processing and scalability in control systems. A neural network is implemented to control anomalies based on a bio-inspired architecture, achieving the detection of anomalies in the time domain based on the evaluation of different models based on recurrent neural networks. Similarly, an evaluation is shown regarding the latency of each component of the system, thus finding possible bottlenecks in real-time operation. The system was implemented on a prototype conveyor belt with low-cost accelerometers, commercial-use microcontrollers, and free software.