Aero-Engine RUL Prediction Using Differential Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural networks for predicting the remaining useful life (RUL) of aero-engines process data uniformly, failing to effectively mine different levels of feature information, leading to poor prediction accuracy.
Innovation Solution
A method utilizing an Automatic Differential Learning Deep Neural Network (ADLDNN) with a multibranch convolutional neural network (MBCNN) and multicellular bidirectional long short-term memory (MCBLSTM) model for feature extraction and prediction, incorporating level and trend division units, and differential learning to classify and update health indexes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing neural networks process data in a uniform mode, then the processing is simple, but the prediction accuracy is poor
Solution Approach 1:
The patent segments the data processing into multiple branches with different processing modes. The MBCNN model divides feature extraction into spatial and temporal branches, while the MCBLSTM model further segments processing into short-term and long-term memory branches. This segmentation allows different levels of feature information to be mined through various feature extraction modes, resolving the contradiction between simple processing and accurate prediction.
Solution Approach 2:
The patent introduces differential learning as an additional dimension to the traditional neural network processing. By incorporating differential operators into the learning process, the model can capture both instantaneous and cumulative degradation information, transforming the processing from a single-dimension uniform mode to a multi-dimensional approach that enhances prediction accuracy.
2Loss of information
If existing neural networks use uniform processing mode, then the model is simple, but different levels of feature information cannot be mined
Solution Approach 1:
The model segments feature extraction into multiple parallel branches: MBCNN provides spatial feature extraction with multiple convolutional branches, while MCBLSTM provides temporal feature extraction with short-term and long-term memory branches. Each branch processes different levels of feature information independently, preventing information loss while maintaining a structured approach to complexity.
Solution Approach 2:
The patent creates a multi-functional model where the same network architecture serves multiple purposes: extracting spatial features, temporal features, short-term dependencies, and long-term dependencies. This universality allows comprehensive feature information extraction without requiring entirely separate models for each function, thus managing complexity efficiently.
3Measurement precision
If traditional neural networks are used, then the implementation is straightforward, but the RUL prediction accuracy is poor
Solution Approach 1:
The patent implements automatic differential learning where the model automatically learns the appropriate differential operators and processing parameters from the data itself, rather than requiring manual configuration. The MBCNN and MCBLSTM structures automatically adapt to capture different degradation patterns, reducing the need for complex manual processing while improving prediction accuracy.
Solution Approach 2:
The model dynamically changes processing parameters through differential learning, adapting the order and type of differential operations based on the degradation stage. This allows the system to automatically adjust processing complexity to match the information content in the data, improving accuracy without requiring fixed complex processing for all cases.
Data Source
AI summary
The present disclosure provides a method for predicting a remaining useful life (RUL) of an aero-engine, specifically including: acquiring multidimensional degradation parameters of an aero-engine to be predicted to obtain acquired data; segmenting the acquired data by a sliding window (SW) to obtain preprocessed data; constructing a RUL prediction model of the aero-engine including a multibranch convolutional neural network (MBCNN) model, a multicellular bidirectional long short-term memory (MCBLSTM) model, a fully connected (FC) layer FC1, and a regression layer; taking the preprocessed data as input data of the MBCNN model, extracting an output of the MBCNN model, taking the output of the MBCNN model and recursive data as input data of the MCBLSTM model, and extracting an output of the MCBLSTM model; obtaining an output of the FC layer FC1, and inputting the output of the FC layer FC1 to the regression layer to predict a RUL.