AWAM-BRT: a hybrid deep learning architecture for intelligent fault diagnosis of drilling pumps

Precise fault diagnosis is imperative for ensuring the secure and dependable operation of drilling pumps. The main challenge lies in effectively extracting discriminative fault features from signals with strong nonstationary noise, while also accurately modeling their long-range temporal evolution. This paper introduces an intelligent fault diagnosis method (namely, Adaptive Wavelet Attention and Block-Recurrent Transformer Integrated Network (AWAM-BRT)) for drilling pumps. This method is based on the WaveletKernelNet (WKN), incorporates the convolutional block attention module (CBAM), and utilizes the Block-Recurrent Transformer (BRT) as its foundation, replacing manual and experience-based fault awareness patterns. To optimize the performance of the AWAM-BRT network for specific tasks, Bayesian optimization techniques are applied to fine-tune the hyperparameters. The proposed model consists of three main parts: (1) The WKN for robustly extracting transient fault features from noisy signals. (2) The CBAM attention mechanism for refining the feature representation by focusing on critical diagnostic information. (3) The BRT module for modeling the long-range dependencies of the entire fault evolution process. The proposed method is evaluated using experimental data from a five-cylinder drilling pump test platform. Through ablation experiments, the effectiveness of WKN, CBAM, and BRT was validated. The proposed approach exhibits excellent fault diagnosis capabilities in comparison experiments.