With the advancement of Artificial Intelligence (AI)-based methods and the establishment of diverse databases, significant research has been conducted on the application of AI in railway track monitoring, inspection and maintenance. Although several review studies exist in this field, each has been confined to a limited scope, focusing on specific data types, data collection methods, or AI-based techniques. To date, no comprehensive review has been published that encompasses all data types, data collection methods, and AI-based approaches to assess prior research holistically. This study aims to address this critical gap by covering both passenger and freight railway transport systems. Firstly, the available databases used for AI applications in railway track inspection and maintenance were categorized and reviewed, distinguishing between peer-reviewed and non-peer-reviewed sources. Secondly, this review introduces a novel three-level classification framework, based on data acquisition method (including track response methods, on-board data approaches, and remote data methods), target railway track component or feature, and input data type, to systematically organize and analyze 567 studies the field published between 2005 and 2025. The findings reveal that the majority of research in this field (88 %) is concentrated on on-board data methods. Approximately 90 % of these studies focus on railway track components, specifically their identification or damage detection. Among the track components, rails and fastening systems, being both critical and vulnerable, have been the primary focus of most research efforts. Image data emerges as the most prevalent and widely utilized data type in on-board data approaches for all railway track components. An in-depth gap analysis was conducted on the literature to identify the limitations of previous studies and outline a roadmap for future research and open directions from multiple perspectives. A comprehensive review of the literature indicates a pressing need for the development of AI-based methods capable of processing multiple data types simultaneously to identify both internal and external damages across all railway track components. The limited number of studies addressing the integration of multiple data types underscores the significant research opportunities in this area. This review not only synthesizes AI-based methods for railway track monitoring and maintenance but also highlights their role in advancing industrial information integration by enabling scalable and intelligent fusion of multi-source data for real-time decision-making.

中文翻译：

---

人工智能在铁轨监测、检查和维护中的应用综述

随着基于人工智能（AI）的方法的进步和多样化数据库的建立，人们对人工智能在铁路轨道监测、检查和维护中的应用进行了大量研究。尽管该领域存在多项综述研究，但每项研究都局限于有限的范围，重点关注特定的数据类型、数据收集方法或基于人工智能的技术。迄今为止，尚未发表涵盖所有数据类型、数据收集方法和基于人工智能的方法的全面综述，以全面评估先前的研究。本研究旨在通过涵盖客运和货运铁路运输系统来弥补这一关键差距。首先，对用于铁路轨道检查和维护中人工智能应用的可用数据库进行了分类和审查，区分了同行评审的来源和非同行评审的来源。其次，本文介绍了一种新的三级分类框架，基于数据采集方法（包括轨道响应方法、车载数据方法和远程数据方法）、目标铁路轨道组件或特征以及输入数据类型，系统地组织和分析了该领域在 2005 年至 2025 年间发表的 567 项研究。研究结果表明，该领域的大部分研究 （88%） 集中在车载数据方法上。这些研究中约 90% 集中在铁路轨道部件上，特别是它们的识别或损坏检测。在轨道部件中，铁轨和紧固系统既关键又脆弱，一直是大多数研究工作的主要焦点。图像数据成为所有铁路轨道部件的车载数据方法中最普遍和最广泛使用的数据类型。 对文献进行了深入的差距分析，以识别先前研究的局限性，并从多个角度勾勒出未来研究的路线图和开放方向。对文献的全面回顾表明，迫切需要开发基于人工智能的方法，能够同时处理多种数据类型，以识别所有铁路轨道组件的内部和外部损坏。涉及多种数据类型集成的研究数量有限，凸显了该领域的重大研究机会。这篇综述不仅综合了基于人工智能的铁路轨道监测和维护方法，还强调了它们通过实现多源数据的可扩展和智能融合以进行实时决策，在推进工业信息集成方面的作用。