Related Experiment Video
Updated: Jan 7, 2026
05:47
Simulation of a Scaled Assembly Process with Collaboration of a Robotic Arm and Monitoring through a Vision System for Quality Control
Published on: August 29, 2025
362
An AM-CNN-BiGRU network with spatiotemporal feature fusion for industrial robot predictive maintenance
Zihao Zang1,2, Yiwen Zhang1,2, Shouxin Ruan3
1School of Mechanical and Electrical Engineering, Changchun University of Science and Technology, Changchun, 130022, China.
Scientific Reports
|December 30, 2025
Summary
This study introduces an advanced predictive maintenance (PdM) method for industrial robots using an attention-based convolutional neural network-BiGRU (AM-CNN-BiGRU) model. The novel approach effectively predicts robot failures by integrating multi-source sensor data for enhanced reliability.
Area of Science:
- Robotics
- Artificial Intelligence
- Machine Learning
Background:
- Industrial robots face frequent failures due to aging, heavy operational loads, and harsh environments.
- Predictive maintenance (PdM) is crucial for mitigating downtime and ensuring operational continuity.
- Existing PdM methods may not fully capture complex failure patterns in industrial robot systems.
Purpose of the Study:
- To develop and validate a novel PdM method for industrial robots.
- To enhance the accuracy and reliability of failure prediction in industrial robot systems.
- To integrate multi-source sensor data effectively for robust fault detection.
Main Methods:
- A hybrid deep learning model, AM-CNN-BiGRU, was proposed, combining Convolutional Neural Network (CNN) for spatial feature extraction and Bidirectional Gated Recurrent Unit (BiGRU) for temporal dependency learning.
- Multi-source sensor data (vibration, current, torque) were fused as distinct channels for the CNN input.
- An Attention Mechanism (AM) was incorporated to adaptively weight different data sources, improving the identification of critical fault signatures.
- The Lion optimizer was utilized for training the deep learning models.
Main Results:
- The AM-CNN-BiGRU model demonstrated superior performance in predicting industrial robot failures.
- Comparative experiments showed that AM-CNN-BiGRU outperformed other state-of-the-art time series forecasting algorithms.
- The model achieved optimal results across key evaluation metrics: R², Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Squared Error (RMSE).
Conclusions:
- The proposed AM-CNN-BiGRU model is effective and reliable for predictive maintenance of industrial robots.
- The integration of CNN, BiGRU, and an attention mechanism enables robust feature extraction and data fusion for accurate fault prediction.
- This research contributes a significant advancement to the field of industrial robot health monitoring and maintenance.
