M² BIST-SPNet: RUL prediction for railway signaling electromechanical devices Xiaoxi Hu, Lei Tan, Tao Tang Accepted: 26 March 2024 / Published online: 16 April 2024 © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024 Abstract Railway signaling electromechanical devices (RSEDs) play a crucial role in the railway industry. Normal wear and tear of these devices occur during day-and-night operation and can progressively develop into failures. Therefore, predicting the remaining useful life (RUL) is essential for reliable services. However, there are three existing challenges in addressing this issue. To overcome these challenges, we introduce M² BIST-SPNet for RSEDs RUL prediction. The model incorporates advanced spatiotemporal feature extraction mechanisms, including a spatial-temporal attention-based convolution network (STACN), multiple branches with multiscale multifrequency module (MBM), and communication module (CM). Extensive experiments on three typical object datasets (i.e., railway safety relay, EMU contactor, switch circuit controller) show that the proposed approach has demonstrated state-of-the-art performance in long-term multiparameter prediction and RUL prediction. Keywords: RUL · Long-term prediction · Railway signaling electromechanical device · Spatial-temporal attention-based convolution network · Multi-scale multifrequency module · Communication module ## 1 Introduction Railway systems play an important role in modern transportation infrastructure, serving as the backbone of both passenger and freight movements across vast geographic regions. Moreover, railway signaling system (RSS) is the core of ensuring the safe and efficient operation of trains. The RSS covers many electromechanical devices that are responsible for pivotal functions, such as relays in various control & indication circuits and switch circuit controllers (SCCs) particularly constituting the turnout control and indication circuit. Because of daily operational wear and tear and harsh service environment, these railway signaling electromechanical devices (RSEDs) are prone to degeneration and eventually failures. Hence, railway signaling maintainers have to take breakdown maintenance and usage-based and calendar/time-based preventive maintenance, although the currently-adopted strategies are very costly for maintenance teams and facility managers. Fortunately, numerous condition monitoring and intelligent prognosis have achieved rapid development due to the advancement of sensors, communication technologies, modeling techniques, and machine learning-related techniques. Moreover, parameter degradation prediction and remaining useful life (RUL) provide unlimited opportunities that effectively reduce operational disruptions and avoid unnecessary fatalities and economic loss in the railway industry. Regretfully, there are some serious challenges we have to face. Firstly, due to the over-maintenance and frequent repairs, there are rare data collections. Meanwhile, safety concerns that the installed condition monitoring system is not likely to hamper the function of the inspected object as well as the special requirement thatM² BIST-SPNet: RUL prediction for railway signaling electromechanical devices Xiaoxi Hu, Lei Tan, Tao Tang Accepted: 26 March 2024 / Published online: 16 April 2024 © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024 Abstract Railway signaling electromechanical devices (RSEDs) play a crucial role in the railway industry. Normal wear and tear of these devices occur during day-and-night operation and can progressively develop into failures. Therefore, predicting the remaining useful life (RUL) is essential for reliable services. However, there are three existing challenges in addressing this issue. To overcome these challenges, we introduce M² BIST-SPNet for RSEDs RUL prediction. The model incorporates advanced spatiotemporal feature extraction mechanisms, including a spatial-temporal attention-based convolution network (STACN), multiple branches with multiscale multifrequency module (MBM), and communication module (CM). Extensive experiments on three typical object datasets (i.e., railway safety relay, EMU contactor, switch circuit controller) show that the proposed approach has demonstrated state-of-the-art performance in long-term multiparameter prediction and RUL prediction. Keywords: RUL · Long-term prediction · Railway signaling electromechanical device · Spatial-temporal attention-based convolution network · Multi-scale multifrequency module · Communication module ## 1 Introduction Railway systems play an important role in modern transportation infrastructure, serving as the backbone of both passenger and freight movements across vast geographic regions. Moreover, railway signaling system (RSS) is the core of ensuring the safe and efficient operation of trains. The RSS covers many electromechanical devices that are responsible for pivotal functions, such as relays in various control & indication circuits and switch circuit controllers (SCCs) particularly constituting the turnout control and indication circuit. Because of daily operational wear and tear and harsh service environment, these railway signaling electromechanical devices (RSEDs) are prone to degeneration and eventually failures. Hence, railway signaling maintainers have to take breakdown maintenance and usage-based and calendar/time-based preventive maintenance, although the currently-adopted strategies are very costly for maintenance teams and facility managers. Fortunately, numerous condition monitoring and intelligent prognosis have achieved rapid development due to the advancement of sensors, communication technologies, modeling techniques, and machine learning-related techniques. Moreover, parameter degradation prediction and remaining useful life (RUL) provide unlimited opportunities that effectively reduce operational disruptions and avoid unnecessary fatalities and economic loss in the railway industry. Regretfully, there are some serious challenges we have to face. Firstly, due to the over-maintenance and frequent repairs, there are rare data collections. Meanwhile, safety concerns that the installed condition monitoring system is not likely to hamper the function of the inspected object as well as the special requirement that