IGWO- and LSTM-Based Fault Diagnosis for Gear Transmission Systems in Mining Equipment

This is the authors' abstract. We don't add key points for this paper.

Large-scale mining excavators often operate under complex and variable working conditions, making the transmission system susceptible to non-stationary vibration signals, substantial environmental noise, and weak early-fault features. To address these challenges, this study proposes a fault diagnosis method that integrates vibration-signal processing, improved Grey Wolf Optimization, long short-term memory (LSTM) networks. First, the raw vibration signals are decomposed using the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) method. A hybrid screening strategy based on Pearson correlation coefficients and Shannon entropy is then employed to identify informative intrinsic mode functions and reconstruct the vibration signals, thereby suppressing noise and enhancing fault-related features. Subsequently, an improved Grey Wolf Optimization (IGWO) algorithm is developed by introducing an adaptive convergence-factor adjustment mechanism and a dynamic weighting strategy for position updating, which improves the algorithm’s global exploration capability and local exploitation accuracy. The IGWO algorithm is further used to optimize the key parameters of an LSTM network, resulting in a CEEMDAN–IGWO–LSTM fault-diagnosis model for identifying different operating and fault conditions of the transmission system. The experimental results show that the proposed denoising method outperforms moving-average filtering and the least-mean-square algorithm in preserving and enhancing fault-related vibration features. The optimized LSTM model achieves average diagnostic accuracies of 96.87%, 97.86%, 96.79%, and 96.67% for the four test datasets, respectively, representing improvements of 10.17, 5.26, 6.99, and 9.07 percentage points over the unoptimized one-dimensional LSTM model. Moreover, compared with the deep-learning-based VMD–DBN method, the proposed approach achieves a further 2.21% improvement in diagnostic accuracy. The results demonstrate that the proposed method can effectively improve the accuracy and robustness of fault diagnosis and provide a reliable technical basis for condition monitoring and predictive maintenance of transmission systems in large-scale mining excavators.