Abstract
<title>Abstract</title> <p>To address the insufficient accuracy of traditional methods in industrial robot bearing fault diagnosis, this study proposes a novel artificial intelligence–based diagnostic method that optimizes a Transformer model using an improved Black Kite Algorithm (IBKA). First, to overcome the drawbacks of the original Black Kite Algorithm—namely unstable convergence, susceptibility to local optima, and low efficiency in handling mixed parameter spaces—the algorithm is improved from three aspects: an adaptive nonlinear convergence factor is introduced to balance exploration and exploitation; a dynamic step-size adjustment mechanism is designed to enhance convergence stability; and a hybrid encoding strategy is proposed to efficiently handle mixed continuous–discrete hyperparameter spaces. Next, the IBKA is integrated with the Transformer model to construct an end-to-end fault diagnosis framework. Multi-dimensional feature engineering is employed to extract vibration signal features, while IBKA is used to simultaneously optimize Transformer hyperparameters and perform feature selection. The optimized Transformer then completes the fault classification task. Finally, experimental validation is conducted on a self-constructed industrial robot bearing fault dataset covering multiple fault types and operating conditions. The experimental results demonstrate that the proposed method achieves a diagnostic accuracy of 99.86%, significantly outperforming comparative methods such as ACGAN, SVM, and CNN. Ablation experiments further confirm the effectiveness of the proposed improvement strategies.</p>