Reliable and solid transportation depend on the safety and efficiency of a high-speed train traction system. This paper offers distinctive idea to solve this problem by introducing an intelligent framework for traction system fault identifications in place. Since, this framework is based on machine learning techniques particularly multi-layer perceptron (MLP) models that can process various datasets such as drive signals, fan signals and concatenated drive-fan signals. Finally, this framework's architecture is tailored towards the complexities of traction system dynamics. hence demonstrating its suitability for classification tasks. Systematically pre-processing the data focuses on signal concatenation being important in understanding how systems function holistically. In particular, we extensively investigate these proposed models using drive dataset integrated with fan dataset and also combined datasets from both fan and driver. It was observed that this model achieved high accuracy, recall and precision levels while still being able to adjust to other datasets accordingly. The potentiality of smart mobility systems will be evident through the success of fault detection in traction systems which is available for real world implementation. Furthermore, this work fills gaps in literature concerning fault detection and became the first step into future researches regarding for fault detection with new point of view from previous studies. Index Terms—Traction system, Fault detection mechanism, Machine learning, Multi-Layer Perceptron, Intelligent transportation systems.