Abstract:Accurately estimating the remaining useful life (RUL) of industrial machinery is beneficial in many real-world applications. Estimation techniques have mainly utilized linear models or neural network based approaches with a focus on short term time dependencies. This paper, introduces a system model that incorporates temporal convolutions with both long term and short term time dependencies. The proposed network learns salient features and complex temporal variations in sensor values, and predicts the RUL. A data augmentation method is used for increased accuracy. The proposed method is compared with several state-of-the-art algorithms on publicly available datasets. It demonstrates promising results, with superior results for datasets obtained from complex environments.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the Remaining Useful Life (RUL) estimation problem of industrial machinery. Accurately estimating the RUL of industrial machinery is of great significance in practical applications, such as predicting failures to make maintenance decisions in advance, thereby preventing permanent damage and significantly reducing operation and maintenance costs. ### Problem Background 1. **Limitations of Existing Methods**: - Existing RUL estimation methods mainly rely on linear models or neural - network - based methods, but these methods usually only focus on short - term time - dependencies. - The sliding - window method only considers the relationships within the sliding window, so it can only capture short - term dependencies. - Although the Hidden Markov Model (HMM) can model long - term dependencies, it has high computational complexity and storage requirements. - Although the Recurrent Neural Network (RNN) can learn time - dependencies, it is prone to the vanishing - gradient problem when dealing with long - term dependencies. 2. **Research Motivation**: - The RUL estimation of industrial machinery is crucial for equipment health management and preventive maintenance. - Data - driven methods have more advantages than physical - model - based methods because it is difficult to establish an accurate physical failure model for the latter. ### Solutions Proposed in the Paper This paper proposes a deep - learning architecture that combines Temporal Convolutional Layers and Long - Short - Term Memory (LSTM) networks, aiming to capture both short - and long - term time - dependencies in sensor data simultaneously. Specifically: - **Temporal Convolutional Layers**: Extract the time features of sensor data through 1D convolution operations, avoiding the noise that may be introduced by 2D convolution. - **LSTM Layers**: Learn long - term time - dependencies by stacking LSTM layers. - **Data Augmentation**: Use a new data - augmentation algorithm to generate partial training trajectories to improve the generalization ability and accuracy of the model. ### Main Contributions 1. **Improved Feature Extraction**: Combining temporal convolution and LSTM can more effectively extract and learn the complex time - variations in sensor data. 2. **Data Augmentation**: The proposed data - augmentation algorithm generates partial training trajectories, making the model perform better in the test phase. 3. **Superior Performance**: On complex environmental data sets, especially on the FD002 and FD004 sub - data sets, this method performs excellently and outperforms the existing advanced algorithms. ### Summary This paper provides a novel and effective RUL estimation method by combining temporal convolution and LSTM networks and introducing data - augmentation techniques, which is especially suitable for industrial machinery and equipment in complex environments.

Temporal Convolutional Memory Networks for Remaining Useful Life Estimation of Industrial Machinery

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

A Novel Deep Learning-Based Encoder-Decoder Model for Remaining Useful Life Prediction

TCRSCANet: Harnessing Temporal Convolutions and Recurrent Skip Component for Enhanced RUL Estimation in Mechanical Systems

Remaining Useful Life Estimation Via Transformer Encoder Enhanced by a Gated Convolutional Unit

Long Short-Term Memory Neural Network with Transfer Learning and Ensemble Learning for Remaining Useful Life Prediction

Multi-Dimensional Self Attention based Approach for Remaining Useful Life Estimation

A recurrent neural network approach for remaining useful life prediction utilizing a novel trend features construction method

A novel spatio-temporal hybrid neural network for remaining useful life prediction

Trend-augmented and temporal-featured Transformer network with multi-sensor signals for remaining useful life prediction

Dilated Convolution Neural Network for Remaining Useful Life Prediction

Memory-enhanced hybrid deep learning networks for remaining useful life prognostics of mechanical equipment

An Improved Method for Predicting the Remaining Useful Life Using a Spatial–Temporal Feature Extraction Network With Attention Mechanism

Cloud–Edge-Based Lightweight Temporal Convolutional Networks for Remaining Useful Life Prediction in IIoT

Convolutional AutoEncoder and Bidirectional Long Short-Term Memory to Estimate Remaining Useful Life for Condition Based Maintenance

Temporal and Heterogeneous Graph Neural Network for Remaining Useful Life Prediction

An Adaptive and Dynamical Neural Network for Machine Remaining Useful Life Prediction

Attention Sequence to Sequence Model for Machine Remaining Useful Life Prediction

Automatic Remaining Useful Life Estimation Framework with Embedded Convolutional LSTM as the Backbone

Joint model for residual life estimation based on Long-Short Term Memory network

Remaining Useful Life Prediction Based on a Double-Convolutional Neural Network Architecture