The coefficient of friction (COF), defined as the maximum of the adhesion coefficient for a given contact condition, fluctuates rapidly due to environmental and operational factors. This paper introduces a torque modulation-based method for COF estimation. A simplified analytical model of the Manchester benchmark bogie operating under dry adhesion conditions is used to evaluate this method. The study presents an analytical equation that confirms earlier simulation-based findings showing a phase difference between applied torque modulation and resulting motor angular velocity. This phase relationship is shown to reflect the shape of the adhesion-slip curve. Notably, when the phase difference approaches 90°, the locomotive operates near the point of maximum adhesion, corresponding to the COF. Furthermore, the sensitivity of this approach to key system parameters, including normal load, wheel rolling radius, and modulation frequency, is examined. The findings provide valuable insights into the robustness and applicability of torque modulation-based COF estimation techniques in real-time traction control systems. The estimated COF can be further leveraged for adhesion management, driver advisory systems, and autonomous train operation.

The swift advancement of electric vehicle technology has led to increased requirements for ensuring the safety of batteries. Various models for predicting battery life and aging have been introduced to facilitate the appropriate utilization of batteries. Timely prediction and alert systems for identifying potential battery failure due to mechanical abuse are of utmost importance. The ongoing progress in machine learning (ML) algorithms and the evolution of extensive cloud-based models offer viable solutions for predicting and issuing early warnings for battery failure. This study focuses on a crucial aspect of EV safety: the timely prediction and prevention of battery failure caused by mechanical abuse. It introduces a cloud-based framework designed for the prediction and early detection of battery failure. The framework comprises three components, with the first being a model for recognizing failure modes resulting from mechanical abuse of batteries. To achieve this aim, a self-organizing map-back propagation (SOM-BP) model is employed, which integrates both supervised and unsupervised learning capabilities to identify three distinct failure conditions: bending, compression, and indentation. The second part involves the implementation of a prediction and pre-short-circuit warning. This is achieved through the utilization of whale optimization algorithm-support vector regression (WOA-SVR) and tuna swarm optimization-support vector regression (TSO-SVR) models to forecast the remaining duration until mechanical failure and short-circuit occurrence. Additionally, these models facilitate the prediction of voltage and temperature levels at the subsequent sampling time. The third part deals with the implementation of battery post-short-circuit prediction using WOA-SVR, TSO-SVR, and random forest models. This involves sampling the temperature, subsequent current, and voltage under various SOCs and then comparing the characteristics of the three models. The findings indicate that the ML models are capable of accurately identifying, predicting, and providing early warnings for failure modes. This work proposes a scalable and potentially efficient solution by leveraging cloud computing for data storage, processing, and model training. The collaboration between the cloud model and vehicle-side information can effectively ensure the safety of passengers.