A Multimodal IoT Sensing System with Hybrid LSTM-CNN Model for Predictive Fault Diagnosis in Elevators

This paper presents a multimodal Internet of Things (IoT)-enabled sensing system integrated with a hybrid deep learning framework for predictive fault diagnosis in elevator systems. The proposed system incorporates a compact sensor node that combines triaxial accelerometers, gyroscopes, temperature, humidity, and microwave radar modules, all managed by an STM32 microcontroller and ESP8266 wireless unit. These modules enable real-time acquisition and cloud-based transmission of operational and environmental data from elevators. To enable accurate fault classification, a dual-branch neural network is developed, combining Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) architectures. The LSTM module captures temporal dependencies in time-series data, while the CNN module extracts local features from frequency-domain representations. Experimental validation under three operational scenarios—normal operation, guide shoe wear, and sensor-induced startup anomalies—demonstrates that the hybrid model achieves a classification accuracy of 99.88%, outperforming traditional machine learning methods and single-model baselines. This work offers a scalable and efficient framework for real-time monitoring and predictive maintenance of vertical transportation systems, with broader applicability to industrial equipment health monitoring.