SpikeFusionNet: A Hybrid Approach to Robotic Fault Diagnosis Using Spiking Neural Dynamics

In the domain of robotics, rapid and accurate fault diagnosis is crucial for maintaining the reliability and efficiency of systems. Traditional convolutional neural networks (CNNs), though adept at fault detection, often falter in handling the temporal intricacies inherent in multivariate time-series data, further compounded by their considerable computational requirements. To address these challenges, we propose SpikeFusionNet, a fusion framework that marries the temporal acuity of spiking neural networks (SNNs) with the potent feature extraction capabilities of CNNs, thereby effectively overcoming the shortcomings of conventional methodologies. This new method uses the neural temporal-intensity coding (NTIC) technique to better record changes in time and an improved Leaky Integrate-and-Fire (LIF) neuron model to make the neurons more responsive to input stimuli. Our experiments showed that this fusion framework significantly boosts recognition accuracy by 6.36% and processing speed by 33.2% over spike-CNNs in complex time series analysis. The SpikeFusionNet not only ensures minimal resource utilization but also significantly augments the fault diagnosis prowess of traditional networks.