Abstract:Accurate prediction of remaining useful life (RUL) plays a significant role in ensuring the safe flight of aircraft. With the recent rapid development of deep learning, there has been a growing trend towards more precise RUL prediction. However, while many current deep learning methods are capable of extracting spatial features—those along the sensor dimension—through convolutional kernels or fully connected layers, their extraction capacity is often limited due to the small scale of kernels and the high uncertainty associated with linear weights. Graph neural networks (GNNs), emerging as effective approaches for processing graph-structured data, explicitly consider the relationships between sensors. This is akin to imposing a constraint on the training process, thereby allowing the learned results to better approximate real-world situations. In order to address the challenge of GNNs in extracting temporal features, we augment our proposed framework for RUL prediction with a Transformer encoder, resulting in the adaptive graph convolutional transformer encoder (AGCTE). A case study using the C-MAPSS dataset is conducted to validate the effectiveness of our proposed model.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the accurate prediction of the remaining useful life (RUL) of aero - engines. Specifically, the paper points out that aero - engines work in harsh environments and have numerous components, which makes their reliability limited and prone to failure. However, if the occurrence time of the fault can be predicted in advance, that is, the remaining useful life (RUL) of the current component, unreasonable waste of time and cost can be avoided, thereby ensuring flight safety. Therefore, how to accurately predict the RUL of aero - engines is an important issue that needs to be studied.
### Background and Challenges
1. **RUL Prediction Methods**:
- **Model - based Methods**: Predict RUL by analyzing degradation processes such as cracks, fatigue, and deformation, but this method requires sufficient research on the degradation mechanism and is difficult to achieve for complex aero - engine systems.
- **Data - based Methods**: With the rapid development of sensor technology, a large amount of data on aero - engines can be obtained, making data - based methods possible in industrial applications. These methods are further divided into traditional machine learning methods and deep learning methods.
2. **Limitations of Existing Methods**:
- **Convolutional Neural Network (CNN) and Auto - encoder (AE)**: Although they can extract spatial features, they lack the ability to model time - series changes.
- **Recurrent Neural Network (RNN) and Transformer**: They can model time - series changes, but are insufficient in extracting spatial features. In particular, when using fully - connected layers, the linear weights may not reflect the true relationships between sensors.
### Solution
To overcome the limitations of the above methods, the paper proposes a new method - Adaptive Graph Convolutional Transformer Encoder (AGCTE). This method combines the Graph Convolutional Network (GCN) and the Transformer Encoder, aiming to extract spatial and temporal features simultaneously.
### Method Overview
1. **Adaptive GCN Layer**:
- Generate a learnable adjacency matrix by learning the embedding vectors of each node, so as to better capture the relationships between sensors.
- Use the GCN layer to process the signal, with the expression \( Y=\sigma[AXW] \), where \( A \) is the normalized adjacency matrix, \( X \) is the original signal, \( W \) is the weight matrix, and \( \sigma \) is the activation function.
2. **Transformer Encoder**:
- Used to model long - term sequence changes, including input embedding, position embedding, multi - head attention, residual connection, layer normalization, and feed - forward part.
- The input embedding converts the input dimension to a dimension suitable for multi - head attention calculation through a fully - connected layer.
- The position embedding is used to maintain the order information of the input data.
- The multi - head attention module captures different aspects of the input signal through multiple parallel self - attention modules.
- The feed - forward part contains two linear mappings and uses the GeLU activation function.
### Experimental Verification
The paper uses the C - MAPSS data set to verify the effectiveness of the proposed AGCTE model. The experimental results show that AGCTE performs well on multiple sub - data sets, especially on the FD002 and FD004 sub - data sets, and its average performance is better than other methods.
### Conclusion
By combining GCN and Transformer Encoder, AGCTE can more accurately predict the remaining useful life of aero - engines, showing its potential in practical applications.