Hybrid HHO-WHO Optimized Transformer-GRU Model for Advanced Failure Prediction in Industrial Machinery and Engines

Area of Science:

• Industrial Engineering
• Machine Learning
• Optimization Algorithms

Background:

• Predictive maintenance is crucial for industrial machinery to minimize downtime and optimize maintenance costs.
• Current predictive models face challenges in generalization and require extensive hyperparameter tuning.
• Conventional optimization methods often get stuck in local optima, limiting predictive performance.

Purpose of the Study:

• To develop a novel hybrid optimization framework for fine-tuning deep learning models in predictive maintenance.
• To improve the accuracy, robustness, and generalization capabilities of failure prediction models.
• To evaluate the performance of a Transformer-GRU model optimized by a hybrid Harris Hawks Optimization (HHO) and Wild Horse Optimization (WHO) algorithm.

Main Methods:

• A hybrid optimization framework combining HHO and WHO was developed to fine-tune hyperparameters of various deep learning models (ResNet, Bi-LSTM, Bi-GRU, CNN, DNN, VAE, Transformer-GRU).
• HHO was utilized for global exploration, while WHO was employed for local exploitation to overcome local optima.
• The performance of the optimized models was evaluated on four benchmark datasets: time-to-failure (TTF), intelligent maintenance system (IMS), C-MAPSS FD001, and FD003.

Main Results:

• The hybrid HHO-WHO optimized Transformer-GRU model demonstrated superior performance compared to other models across all tested datasets.
• Significant reductions in Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) were observed on TTF and IMS datasets.
• Performance improvements were also noted on C-MAPSS FD001 and FD003 datasets, with notable decreases in MAE, RMSE, and score.

Conclusions:

• The proposed hybrid HHO-WHO optimization framework effectively enhances the predictive performance of deep learning models for industrial machinery failure.
• The Transformer-GRU model, when optimized using this hybrid approach, offers a robust, stable, and generalizable solution for predictive maintenance.
• This framework provides a reliable tool for minimizing unexpected downtimes and enabling more cost-effective maintenance strategies.