Contact Us
CNN-Based Fault Diagnosis for Metal Mine Crushers
Crushers are critical equipment in metal mines. Over 30% of unplanned downtime is caused by crusher failures. Traditional diagnosis methods have high false alarm rates. This study uses a Convolutional Neural Network (CNN) to realize high-precision intelligent fault diagnosis and predictive maintenance for crushers.
1. Typical Crusher Fault Types & Features
| Fault Type | Mechanism | Typical Performance |
|---|---|---|
| Bearing Failure | Poor lubrication, overload, foreign matter | High-frequency vibration, local overheating |
| Liner Wear / Detachment | Impact, loose bolts | Periodic shock vibration, abnormal discharge size |
| Rotor Imbalance | Uneven wear, spindle deformation | 1× rotational frequency amplitude surge |
| Transmission Fault | Gear broken, belt slip, misalignment | Torque fluctuation, modulation sidebands |
2. Data Acquisition System
- Sensors: 3-axis vibration, infrared thermal imager (±1℃), acoustic emission (50–400 kHz)
- Frequency range: 0.5–10 kHz; sampling rate: 25.6 kHz
- Samples: 200 groups per fault; total 800 groups
- Set division: training set 70%, validation set 30%
3. CNN Model & Hyperparameters
- Network: CNN (convolution + pooling + full connection)
- Activation function: tanh
- Batch size: 16; epochs: 200
- Input size: 96×96; kernel: 5×5; filters: 64
- Learning rate: 0.001
4. Diagnosis Accuracy Results
| Fault Type | CNN Accuracy |
|---|---|
| Bearing Failure | 95.0% |
| Liner Wear / Detachment | 93.3% |
| Rotor Imbalance | 96.7% |
| Transmission System Fault | 95.0% |
| Overall Average | 95.0% |
5. Model Comparison
- CNN overall accuracy: 95.0%
- SVM overall accuracy: 90.8%
- Decision Tree overall accuracy: 89.2%
- CNN reduces false alarm rate to below 5%
6. Advantages & Conclusions
- CNN automatically extracts fault features without manual dependence
- Strong ability to capture nonlinear and coupling features
- Suitable for strong impact, high dust, unsteady mine conditions
- Supports predictive maintenance and reduces downtime loss
Summary
The CNN-based intelligent fault diagnosis model realizes high-precision recognition of four major crusher faults with an overall accuracy of 95.0%. It outperforms SVM and decision tree models, providing a reliable solution for intelligent operation and maintenance of metal mine crushing systems.