In the bustling world of subway infrastructure, ensuring the smooth and safe operation of vehicles is paramount. A recent breakthrough in identifying wheel state degradation levels could revolutionize how we maintain and monitor subway systems, with significant implications for the energy sector. Xichun Luo, a researcher from Yunnan Jingjian Rail Transit Investment Construction Co., Ltd., has developed a novel data-driven method that promises to enhance the accuracy and efficiency of subway vehicle inspections.
The challenge of detecting periodic harmonic irregularities on subway wheel surfaces has long plagued engineers. Traditional methods often struggle with nonlinear interferences and low signal-to-noise ratios in vibration signals collected during operation. Luo’s innovative approach addresses these issues head-on by introducing a three-stream feature fusion deep learning framework. This method avoids the pitfalls of manual signal processing and feature engineering, offering a more robust and automated solution.
The framework processes vibration signals through three parallel channels: a 1D convolutional neural network (1D-CNN) for raw time-domain signals, a 2D convolutional neural network (2D-CNN) for continuous wavelet transform (CWT) time-frequency representations, and a bidirectional gated recurrent unit (BiGRU) network for fast Fourier transform (FFT) spectra. The extracted features from these branches are then fused, and a multi-head self-attention mechanism is embedded at the tail of the network to enhance classification performance.
“The proposed method achieves an accuracy of 97.62% in recognizing wheel state degradation levels,” Luo explained. “This level of precision is a game-changer for the industry, as it allows for more proactive maintenance and reduces the risk of costly and dangerous failures.”
To meet the multi-parameter diagnostics required in engineering, Luo introduced a multi-task learning (MTL) framework based on hard parameter sharing. This framework establishes task-specific branches at the back end of the aforementioned model, maintaining an average accuracy of 97.25%. “The MTL framework is comparable to or exceeds that of single-task models, making it a versatile tool for various diagnostic needs,” Luo added.
The implications of this research extend beyond the subway industry. In the energy sector, where the reliability of machinery is crucial, such advanced diagnostic tools can lead to significant cost savings and improved safety. By identifying potential issues before they escalate, companies can minimize downtime and avoid expensive repairs.
Published in the journal *Advances in Mechanical Engineering* (translated from Chinese as “机械工程进展”), this research represents a significant step forward in the field of mechanical diagnostics. As subway systems and other forms of public transportation continue to expand, the need for efficient and accurate maintenance solutions will only grow. Luo’s work provides a promising path forward, offering a glimpse into the future of mechanical diagnostics and maintenance.
“This research is not just about improving subway operations; it’s about setting a new standard for mechanical diagnostics across industries,” Luo concluded. “The potential applications are vast, and I am excited to see how this technology will shape the future of maintenance and safety.”