Bearings are critical yet vulnerable components in mechanical equipment, with potential failures that can significantly impact system performance. As stochastic resonance methods effectively convert noise energy into fault characteristic energy within bearing vibration signals, they remain a research focus in bearing fault diagnosis. This study proposes a coupled neuron model based on biological stochastic resonance effects for processing bearing vibration signals. To enhance parameter optimization, we develop an improved deep reinforcement learning algorithm that incorporates a prioritized experience replay buffer into the network architecture. Using the SNR as the evaluation metric, the algorithm performs data screening on the replay buffer parameters before training the deep network for predicting coupled neuron model performance. In terms of experimental content, the study performed data processing on simulated signals and vibration signals of gearbox bearing faults collected in the laboratory environment. By comparing the coupled neuron model optimized with a reinforcement learning algorithm, particle swarm algorithm, and quantum particle swarm algorithm, the experimental results show that the coupled neuron model optimized with a deep reinforcement learning algorithm has the optimal signal-to-noise ratio of the output signal and recognition rate of the bearing faults, which are −13.0407 dB and 100%, respectively. The method shows significant performance advantages in realizing the energy enhancement of the bearing fault eigenfrequency and provides a more efficient and accurate solution for bearing fault diagnosis, which has important engineering application value.

As a crucial role in industrial production, fault diagnosis has limitations as traditional noise-reduction-based methods often lose critical information, compromising weak fault feature identification. In biological neuron networks, stochastic resonance leverages the noise of neurons to detect minute features. However, the lack of biological interpretability of traditional neuron models often leads to the decline of model performance. Accordingly, a Gaussian neuron discharge network (GNDN) is developed, enhancing model diversity and fault feature detection capability through incorporation of coupled discharge mechanisms. The effects of six key parameters in the neuron network cluster were systematically investigated, with model optimization achieved through a quantum particle swarm parameter alternating (QPSPA) algorithm. Experiments on feature signal enhancement were conducted on the vibration signals of generator bearings from a drivetrain diagnostics simulator and a wind turbine. When compared with traditional four algorithms, the QPSPA algorithm demonstrated superior performance in optimizing six parameters of the GNDN model with average fitness, mean fitness difference, signal to noise ratio (SNR) and spectral amplitude. Simultaneously, in contrast to traditional four models, the GNDN model has higher SNR, spectrum amplitude, and recognition rate. Results indicate that the proposed approach, validated on both drivetrain diagnostics simulator and wind power generators, can substantially improve the capability of mechanical fault diagnosis and demonstrates significant potential for application within the energy industry.

Through the analysis of multidimensional vibration signals of machinery, existing faults in mechanical equipment can be timely identified to ensure normal equipment operation. In biological nervous systems, noise influences the generation, transmission, and response of neural signals, causing these signals to exhibit diverse dynamic behaviors. Inspired by this mechanism, to improve fault diagnosis accuracy, this study investigates the advantages of stochastic resonance effects in discharge neural networks and proposes an adaptive cluster discharge resonance neuron (CDRN) model. This model enhances characteristic frequency extraction capability for bearing fault vibration signals. In addition, by embedding a multi-dimensional neuron model into the CDRN framework, the limitation of one-dimensional signals being vulnerable to complex noise interference and false detection is overcome. Using the output signal-to-noise ratio and neural network classifier recognition rate as evaluation metrics, the detection performance of the CDRN method is compared with the single one-dimensional stochastic resonance (SOSR) method and the single hyperbolic tangent neuron (SHTN) method for early-stage mechanical fault detection in wind turbine bearing inner and outer rings. In the two experiments, the fault detection rate of the CDRN method reaches 100%. Experimental results demonstrate that the CDRN method outperforms both SOSR and SHTN in mechanical fault feature recognition and significantly improves early fault detection accuracy.