1670155 - Implementation and Assessment of a Rail Surface Defect Detection System from Onboard Measurements.

Maintaining the railroad infrastructure in a state of good repair to prevent failure and service disruptions is a challenge of paramount importance to the safety and economy of operations.  Rail Surface Spot Irregularities (RSSI) are among the most common types of damage in railway networks that affect track operating conditions and may lead to rail breaks, track failure and derailments. Such irregularities are local defects on the rail surface arising from rolling contact fatigue, joints and welds, among others, and will grow as load accumulates. Relative to random irregularities, RSSI consists of shorter wavelengths that induce significant dynamic excitation on the track and track structures. Traditionally, RSSI detection and characterization is conducted manually by trained inspectors traversing the tracks seeking to visually identify the localized defects and classify the occurrences according to their experience. This approach is labor-intensive, inefficient, and does not produce a well-structured diagnostic since it is subjective and frequent inspections are not feasible. These methods are being gradually replaced by automated detection systems. Several studies involving automated detection of internal and/or surface defects have been reported, including ultrasonic inspection, eddy current inspection, magnetic flux leakage inspection, and image processing techniques. Although these techniques have been extensively explored recently, there are still some practical limitations. In recent years, image processing techniques within AI and machine learning frameworks, are proposed with various levels of success. However, machine vision technologies are highly sensitive to image quality, which is affected greatly by variations of illumination and the speed of the rail inspection vehicles.

This presentation discusses an automated system for RSSI that combines ABA sensors, edge computing, and digital communications in one, low-cost, fast, accurate, and integrated system that allows for continuous rail surface health monitoring.  At the core of the system is an advanced hybrid algorithm developed by the authors. The proposed hybrid algorithm integrates Wavelet Packet Analysis (WPA) and Hilbert-Huang Transform (HHT) to effectively handle non-stationary data in noisy conditions. This proposed technique capitalizes on the strengths of each method, mitigating their individual limitations and resulting in improved overall performance.  Verification and validation studies through computer simulations and field measurements demonstrate the accuracy effectiveness and robustness of the proposed system and are discussed in detail along with general guidelines for implementation in practice.