In the field of mining, open-pit mining occupies a large proportion. Electric shovels, as important mining equipment, are widely used. The bucket teeth of electric shovels are key components that directly contact with ores and are subject to severe working conditions. Due to the large reaction force, bucket teeth often suffer from wear, fracture, and shedding, which can cause serious consequences such as damage to crushers and production line failures, resulting in significant economic losses.
1.2 Research Significance
The detection of bucket teeth faults is crucial for ensuring the normal operation of mining equipment, reducing maintenance costs, and improving production efficiency. It can also protect the safety of workers and equipment.
2. Related Research
2.1 Bucket Teeth Fault Analysis
Wear: The surface of bucket teeth may become rough due to plastic deformation and brittle cracks, accelerating wear.
Fracture: High impact loads and stress concentrations can cause bucket teeth to fracture, especially at the junction of the tooth tip and the tooth seat.
Shedding: The connection between bucket teeth and the tooth seat may become loose, resulting in tooth shedding.
2.2 Existing Detection Methods
Laser Ranging Method: Luo et al. proposed a method to reconstruct the 3D shape of bucket teeth using a laser rangefinder and detect faults by template matching.
Image Processing Method: Li et al. proposed an object detection algorithm based on image processing techniques, using the position relationship between bucket teeth and adding structural feature constraints to improve detection accuracy.
3. Detection System Design
3.1 Overall Scheme
The detection system consists of an image acquisition system, an image processing system, and a display system. The image acquisition system uses a camera to capture images of the bucket teeth, the image processing system uses deep learning and image processing algorithms to detect faults, and the display system shows the detection results.
3.2 Hardware System
Camera: A low-temperature infrared camera is used to ensure imaging in harsh environments.
Computer: A high-performance graphics workstation with a high-end graphics card is used for image processing.
3.3 Software System
The software system includes system initialization, bucket teeth fault detection, and result information display modules. It can achieve functions such as bucket image acquisition, fault detection, result visualization, and fault alarm.
4. Dataset Preparation
4.1 Model Bucket Data
Model bucket data is obtained by shooting 3D printed model buckets and collecting data from image sharing websites. The data includes normal buckets, buckets with one faulty tooth, and buckets with two faulty teeth.
4.2 Real Bucket Data
Real bucket data is collected from search engines and image databases.
4.3 Data Augmentation
Data augmentation methods such as flipping, rotation, and noise addition are used to increase the amount of data and improve the robustness of the model.
5. Bucket Teeth Target Detection
5.1 Convolutional Neural Network Overview
The convolutional neural network consists of an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. The convolutional layer extracts features, the pooling layer reduces dimensions, and the fully connected layer classifies the features.
5.2 Faster R – CNN Algorithm
The Faster R – CNN algorithm is used for bucket teeth target detection. It consists of a feature extraction network, a region proposal network, an interested region pooling layer, and a classification and regression loss function.
5.3 Model Training and Evaluation
Anchor Initialization: The anchor initial values are adjusted according to the characteristics of the bucket teeth dataset.
Model Training: The model is trained using a fine-tuning strategy on the MSTOD and RSTOD datasets.
Model Evaluation: The performance of the model is evaluated using the PASCAL VOC evaluation standard, including metrics such as AP and mAP.
6. Bucket Teeth Fault Detection Based on Image Segmentation
6.1 Full Convolutional Neural Network for Image Segmentation
The full convolutional neural network is used to segment the bucket teeth images. It can output pixel-level classification results and is more suitable for image segmentation tasks.
6.2 Fourier Descriptor for Contour Feature Extraction
The Fourier descriptor is used to extract the contour features of the bucket teeth. It has good invariance to rotation, translation, and scaling, and can be used to distinguish between end and middle bucket teeth.
6.3 Fault Location and Wear Degree Calculation
Fault Location: The position of the faulty bucket teeth is determined by combining the position relationship between the bucket teeth and the contour features.
Wear Degree Calculation: The wear degree of the bucket teeth is calculated by comparing the length of the faulty bucket teeth with that of the normal bucket teeth.
7. System Implementation and Verification
7.1 System Design
The system design includes the design of the system initialization function, bucket teeth fault detection function, fault alarm function, and result visualization function.
7.2 Graphical User Interface Design
The graphical user interface is designed using PyQt4 in Python. It includes an input image area, a target detection area, an image segmentation area, and a detection information area.
7.3 System Experiment and Result Analysis
Experiment Setup: Three model buckets are used for testing, including normal buckets, buckets with one faulty tooth, and buckets with two faulty teeth.
Result Analysis: The detection accuracy of the target detection program is high, but the performance of the fault detection program is affected by the image segmentation performance and the pose of the bucket.
8. Summary and Outlook
8.1 Research Summary
System Design: A bucket teeth fault detection system is designed, including the overall design of the system, the implementation of the target detection algorithm, the implementation of the fault detection algorithm, and the design of the user interface.
Algorithm Innovation: The Faster R – CNN algorithm is used for target detection, and the full convolutional neural network and Fourier descriptor are used for fault detection.
8.2 Research Outlook
Dataset Optimization: The dataset needs to be further optimized to improve the recognition effect.
Model Improvement: The model needs to be improved to reduce parameters, increase calculation speed, and ensure recognition performance.
Detection Precision Improvement: The detection precision needs to be improved by enhancing the segmentation performance of the image segmentation model and using better contour features.
Table 1: Bucket Teeth Fault Types and Causes
Fault Type
Cause
Wear
Plastic deformation, brittle cracks, material abrasion
Fracture
High impact loads, stress concentrations
Shedding
Loose connections
Table 2: Comparison of Existing Detection Methods
Method
Advantage
Disadvantage
Laser Ranging Method
High accuracy, insensitive to environmental changes
Complex equipment, high cost
Image Processing Method
Simple equipment, low cost
Sensitive to environmental changes, lower accuracy
Table 3: Detection System Components
Component
Function
Image Acquisition System
Capture images of bucket teeth
Image Processing System
Detect faults using algorithms
Display System
Show detection results
Table 4: Hardware System Specifications
Device
Specification
Camera
Low – temperature infrared camera, high resolution
Computer
High – performance graphics workstation, high – end graphics card
Table 5: Software System Modules and Functions
Module
Function
System Initialization
Initialize system parameters, set storage paths
Bucket Teeth Fault Detection
Detect faults in bucket teeth
Result Information Display
Display detection results
Table 6: Dataset Sources and Characteristics
Dataset
Source
Characteristic
Model Bucket Data
3D printed model buckets, image sharing websites
Includes normal and faulty buckets
Real Bucket Data
Search engines, image databases
Real – world data
Table 7: Data Augmentation Methods
Method
Description
Flipping
Flip the image horizontally or vertically
Rotation
Rotate the image by a certain angle
Noise Addition
Add noise to the image
Table 8: Convolutional Neural Network Layers and Functions
Layer
Function
Input Layer
Receive input images
Convolutional Layer
Extract features
Pooling Layer
Reduce dimensions
Fully Connected Layer
Classify features
Output Layer
Output detection results
Table 9: Faster R – CNN Algorithm Components
Component
Function
Feature Extraction Network
Extract features from images
Region Proposal Network
Generate candidate regions
Interested Region Pooling Layer
Map regions to feature maps
Classification and Regression Loss Function
Calculate losses for classification and regression
Table 10: Model Training and Evaluation Metrics
Metric
Description
Anchor Initialization
Adjust anchor values according to dataset
Model Training
Use fine – tuning strategy on datasets
Model Evaluation
Use PASCAL VOC standard, calculate AP and mAP
Table 11: Full Convolutional Neural Network for Image Segmentation Advantages
Advantage
Description
Pixel – level Classification
Output detailed segmentation results
Adaptability
Suitable for various image segmentation tasks
Table 12: Fourier Descriptor Properties
Property
Description
Invariance
Invariant to rotation, translation, and scaling
Feature Extraction
Extract contour features of bucket teeth
Table 13: Fault Location and Wear Degree Calculation Methods
Method
Description
Fault Location
Determine position by combining position and contour features
Wear Degree Calculation
Calculate by comparing lengths of faulty and normal teeth
Table 14: System Design Functions
Function
Description
System Initialization
Set parameters and paths
Bucket Teeth Fault Detection
Detect faults in bucket teeth
Fault Alarm
Alert when faults are detected
Result Visualization
Display detection results
Table 15: Graphical User Interface Areas and Functions
Area
Function
Input Image Area
Display input images
Target Detection Area
Show target detection results
Image Segmentation Area
Display image segmentation results
Detection Information Area
Show detection information
Table 16: System Experiment Results
Result
Description
Detection Accuracy
High for target detection, affected by image segmentation for fault detection
Fault Location Accuracy
Depends on image segmentation performance and bucket pose
Table 17: Research Summary Highlights
Aspect
Highlights
System Design
Overall design, algorithm implementation, user interface design
Algorithm Innovation
Faster R – CNN for target detection, full convolutional neural network and Fourier descriptor for fault detection
