Abstract:Accurate remaining useful life (RUL) predictions are critical to the safe operation of aero-engines. Currently, the RUL prediction task is mainly a regression paradigm with only mean square error as the loss function and lacks research on feature space structure, the latter of which has shown excellent performance in a large number of studies. This paper develops a multi-granularity supervised contrastive (MGSC) framework from plain intuition that samples with the same RUL label should be aligned in the feature space, and address the problems of too large minibatch size and unbalanced samples in the implementation. The RUL prediction with MGSC is implemented on using the proposed multi-phase training strategy. This paper also demonstrates a simple and scalable basic network structure and validates the proposed MGSC strategy on the CMPASS dataset using a convolutional long short-term memory network as a baseline, which effectively improves the accuracy of RUL prediction.

What problem does this paper attempt to address?

This paper attempts to solve several key problems in the prediction of the remaining useful life (RUL) of aero - engines: 1. **Insufficient feature space structuring**: Current RUL prediction methods mainly adopt the regression paradigm and only use the mean - squared error (MSE) as the loss function, lacking research on the structure of the feature space. This results in the model being unable to effectively distinguish and align samples in the feature space. 2. **Sample imbalance and batch - size problems**: In supervised contrastive learning, directly using RUL labels will lead to an imbalance between positive and negative samples, and a large batch size is required to balance these samples, which places high demands on hardware resources. To solve these problems, the paper proposes a multi - granularity supervised contrastive framework (MGSC), which specifically includes the following aspects: - **Multi - granularity supervised contrastive strategy**: - **Coarse - grained contrast**: Use the health state (HS) label as the class label to conduct supervised contrastive learning on a large classification scale, thereby initially regularizing the embedding with a smaller batch size. - **Fine - grained contrast**: Use the RUL label for contrastive learning within each HS class to ensure that samples within the same HS class are aligned according to the RUL label. - **Multi - stage training strategy**: By gradually achieving contrastive learning tasks and regression prediction tasks at different granularities, ensure that the sample embeddings in the feature space are gradually aligned and ultimately improve the accuracy of RUL prediction. - **Base network structure**: Design a simple and extensible base network structure, including an encoder, a projector, and a regression layer, to verify the effectiveness of the MGSC strategy. Through the above methods, the experimental results of the paper on the CMAPSS dataset show that the MGSC framework significantly improves the accuracy of RUL prediction, especially when dealing with complex working conditions and failure modes.

A Multi-Granularity Supervised Contrastive Framework for Remaining Useful Life Prediction of Aero-engines

Aero-Engine Remaining Useful Life Prediction Via Physics-Informed Self-Attention Encoder

Remaining Useful Life Prediction of the Aviation Engine Based on the Long-Short-Term-Memory Network with a Feature Selection Mechanism.

A Multi-Head Neural Network with Unsymmetrical Constraints for Remaining Useful Life Prediction

Structural-Guided Interaction and Attention-Enhancing Compensation Network for Machine Remaining Useful Life Prediction

A dual path hybrid neural network framework for remaining useful life prediction of aero‐engine

Enhancing aircraft engine remaining useful life prediction via multiscale deep transfer learning with limited data

Aeroengine Remaining Life Prediction Based on Advanced Health Index Construction and Grey Similarity Multi-Scale Matching

Multiscale BLS-Based Lightweight Prediction Model for Remaining Useful Life of Aero-Engine

Spatial-Temporal Feature Extraction Network for Online Aero-engines Remaining Useful Life Prediction

Multi-Head Self-Attention-Based Fully Convolutional Network for RUL Prediction of Turbofan Engines

Remaining Useful Life Prediction for Multi-Sensor Systems Using a Novel End-to-end Deep-Learning Method

A Dual-Dimension Convolutional-Attention Module for Remaining Useful Life Prediction of Aeroengines

Remaining Useful Life Prediction Method Based on Dual-Path Interaction Network with Multiscale Feature Fusion and Dynamic Weight Adaptation

Adaptive Feature Utilization with Separate Gating Mechanism and Global Temporal Convolutional Network for Remaining Useful Life Prediction

Aero-engine remaining useful life prediction method with self-adaptive multimodal data fusion and cluster-ensemble transfer regression

Machine Remaining Useful Life Prediction Via an Attention-Based Deep Learning Approach

A Multi-channel Long-term External Attention Network for Aeroengine Remaining Useful Life Prediction

Multiscale Feature Extension Enhanced Deep Global-Local Attention Network for Remaining Useful Life Prediction

A Hybrid Prognostics Deep Learning Model for Remaining Useful Life Prediction

A Continuous Remaining Useful Life Prediction Method With Multistage Attention Convolutional Neural Network and Knowledge Weight Constraint