MultiSenseNet: Multi-Modal Deep Learning for Machine Failure Risk Prediction

Predictive maintenance plays a key role in industrial manufacturing by identifying machine failure risks early. This prevents downtime and improves efficiency, contributing towards industry innovation. This paper presents a method for predicting equipment failures using multi-modal sensor data. Their approach combines advanced techniques, including convolutional neural networks (CNNs) for feature extraction, long short-term memory networks (LSTMs) for temporal patterns, transformer-based attention mechanisms for critical feature identification, and graph neural networks (GNNs) for modeling sensor-machine relationships. The model was tested on two real-world datasets: the Machine Failure Classification Dataset and the Machine Failure Prediction Dataset. MultiSenseNet outperformed existing benchmarks, achieving accuracy rates of 95.4% and 95.2%, respectively. It also excelled in precision, recall, F1-score, and AUC metrics compared to traditional machine learning models and recent deep learning architectures. The model performed consistently across different hyperparameter settings, reaching peak accuracy of 94.7% on the classification dataset and 95.6% on the prediction dataset after 40 training epochs. Training and validation curves showed effective convergence with minimal overfitting.