What question did this study set out to answer?

The research aims to enhance the prediction of remaining useful life in industrial equipment using a novel algorithm.

April 8, 2026Open Access

Equipment remaining useful life prediction algorithm based on hyperparameter optimization and LSTM

Key Points

The research aims to enhance the prediction of remaining useful life in industrial equipment using a novel algorithm.
Developed a model integrating improved particle swarm optimization and LSTM with an attention mechanism.
Utilized a standardized data preprocessing and feature engineering pipeline based on the C-MAPSS dataset.
Implemented bidirectional LSTM to analyze time-series data and focus on key degradation features.
Conducted comparative experiments on four sub-datasets to evaluate model performance against traditional algorithms.
The proposed model significantly reduces root mean squared error compared to traditional LSTM and other algorithms.
In complex operating conditions, the model's RMSE is 47.86% of SVR's, demonstrating superior accuracy.
Performance evaluation scores show the new model's adaptability and stability in varying scenarios.

Abstract

Aiming at the key challenges in industrial equipment’s remaining useful life (RUL) prediction, such as difficulty in capturing long-term dependencies in time-series data, low efficiency of hyperparameter optimization, and easy trapping in local optima of traditional algorithms, a prediction model integrating improved Particle Swarm Optimization (PSO) and a LSTM attention mechanism is proposed. Taking aero-engines as the research object, a standardized data preprocessing and feature engineering pipeline is constructed based on the C-MAPSS benchmark dataset, and the quality of input data is improved through data cleaning, redundant feature elimination, and time series sequence reconstruction. In terms of model design, bidirectional LSTM is used to mine the internal correlation of time-series data, and the self-attention mechanism is combined to accurately focus on the key features of life degradation; meanwhile, an improved PSO algorithm is proposed, which introduces a nonlinear decreasing inertia weight and an adaptive Gaussian mutation operator to realize global optimization of the core hyperparameters of LSTM. Comparative experiments on four sub-datasets (FD001–FD004) of the C-MAPSS dataset show that the proposed model significantly outperforms baseline algorithms such as traditional LSTM, Gated Recurrent Unit (GRU), Support Vector Regression (SVR), and Random Forest in terms of two core indicators: Root Mean Squared Error (RMSE) and Performance Evaluation Score (Score) and also outperforms variant models without an attention mechanism or improved PSO optimization. In the complex scenario of multiple operating conditions and compound faults (FD004), the RMSE of the model is only 47.86% of that of SVR, and the score is only 62.10% of that of GRU, showing excellent prediction accuracy, stability, and adaptability to complex scenarios. The research verifies the synergistic effect of the attention mechanism and improved PSO hyperparameter optimization and provides reliable RUL prediction technical support for industrial equipment predictive maintenance.

Ask AI

Helpful

Bookmark

View Full Paper