Abstract: The key supporting bearings of heavy-duty mining transportation equipment are constantly exposed to heavy loads, impacts, dust pollution, and variable load conditions. Their lubrication status directly affects the reliability of equipment operation, production continuity, and maintenance economy. To address the issues of existing lubrication status monitoring methods relying on a single signal source, insufficient feature characterization capabilities, and low recognition accuracy and robustness under variable conditions, a lubrication status diagnosis method based on vibration electrostatic heterogeneous spatial temporal deep fusion with ResNet-BiLSTM (VEH-STDF-ResNet-BiLSTM) for the bearings of mining heavy-duty transportation equipment is proposed. A multi-physics field coupling simulation model of bearing fluid-solid-electric is established to reveal the intrinsic correlation mechanism between electrostatic response characteristics and lubrication status; a vibration-electrostatic synchronous monitoring experimental system is built to simulate typical lubrication states and multiple rotational speed conditions, and a heterogeneous fusion dataset is constructed. To address the strong noise and non-stationary characteristics of multi-source signals, a multi-level adaptive hybrid denoising method for vibration signals and a combined denoising method based on variational mode decomposition (VMD) and adaptive wavelet threshold denoising (AWTD) for electrostatic signals were respectively designed, and Gramian angular field (GAF) encoding is adopted to achieve two-dimensional representation of time-series signals. An improved ResNet-CBAM network is utilized to extract spatial features, while a bidirectional long short-term memory (BiLSTM) network was used to extract temporal features. Combined with feature concatenation and an adaptive weighted double-layer fusion strategy, collaborative diagnosis of heterogeneous information is realized. Experimental results show that the recognition accuracy reaches 98.7% under fixed rotational speed conditions, and the average recognition accuracy exceeds 95% under variable rotational speed conditions, with overall performance superior to single-signal-source diagnostic models. This method can effectively enhance the accuracy and adaptability of bearing lubrication status diagnosis under complex mining conditions, and provides technical support for intelligent monitoring and predictive maintenance of key components in mining conveyance systems.