Conquering the "Small-Sample, Deep-Structure" Data Challenge — AI Cube Lab Team Wins Global Championship at PHM 2025 Data Challenge

Conquering the “Small-Sample, Deep-Structure” Data Challenge!

AI Cube Lab Team Wins Global Championship at PHM 2025 Data Challenge

The SAM-IPA-1 team from AI Cube Lab — led by Prof. Jianshe Feng and composed of undergraduate, master’s, and doctoral students Peng Gao, Fanyu Qi, Yizhang Zhu, Jianyu Zhang, and Wenfei Li — leveraged outstanding technical expertise and an innovative algorithmic architecture to outshine numerous international competitors and claim first place at the prestigious 2025 Prognostics and Health Management (PHM) Data Challenge.

About the Competition

PHM is a core technology underpinning competitiveness in high-end manufacturing and aerospace. The PHM Data Challenge is the centerpiece of the annual PHM Society conference — the premier international event in the PHM community — and has generated far-reaching influence since its inaugural edition in 2008. Over the years, the competition has produced widely recognized benchmark datasets, such as the 2008 aero-engine remaining useful life dataset and the 2016 semiconductor CMP virtual metrology dataset. Thousands of research publications have been built on these datasets, making enormous contributions to the development and deployment of intelligent algorithms for advanced manufacturing. The competition is dedicated to tackling the most cutting-edge and demanding real-world industrial challenges, and is recognized as one of the most influential international data-science competitions in the field of predictive maintenance and intelligent manufacturing. Past editions have attracted top university teams from Georgia Tech, the University of Cincinnati IMS Center, the University of Wisconsin–Madison, the University of Maryland, NTNU, Tsinghua University, Shanghai Jiao Tong University, and Harbin Institute of Technology, as well as research teams from prominent organizations such as MathWorks, ByteDance, Hyundai, and Hitachi High-Tech.

This year’s challenge was exceptionally demanding: participants were required to achieve high-accuracy and highly generalizable predictions of the remaining useful life (RUL) of multiple aero-engines under the combined constraints of scarce fault samples and complex spatiotemporal data structures. The problem posed a high technical bar and challenged participants’ comprehensive abilities in data processing, information mining, AI modeling, and engineering domain understanding.

Student Highlights

In this top-tier global data competition, the AI Cube Lab team delivered an outstanding performance. Under the guidance of lab director Prof. Jianshe Feng, the SAM-IPA-1 team — comprising 2024 and 2025 cohort students Peng Gao, Fanyu Qi, Yizhang Zhu, Jianyu Zhang, and Wenfei Li — competed fiercely against dozens of teams from around the world.

Facing the formidable challenges of noisy data, scarce samples, and multi-task objectives, the AI Cube Lab team drew on a deep theoretical foundation and developed a highly innovative two-stage technical framework after thoroughly analyzing the problem setting. In the first stage, a multi-level data abstraction algorithm was designed to distill raw high-dimensional, multi-spatiotemporal, dynamic data streams into standardized core health indicators, resolving the tension between data redundancy and information scarcity. In the second stage, an attention-based multi-task collaborative learning architecture was constructed to achieve cross-spatiotemporal generalized learning of physical degradation patterns. This approach not only enabled accurate prediction under small-sample conditions but also demonstrated exceptionally strong generalizability, ultimately winning the PHM 2025 Data Challenge by a decisive margin — demonstrating enormous potential for industrial applications.

As the team representative, Wenfei Li, a 2024 cohort undergraduate student, presented the award-winning solution to global experts and scholars at the PHM Society Annual Conference held in Seattle, USA, excelling in the oral presentation, Q&A defense, and poster session, and displaying the professional competence and spirit of Sun Yat-sen University students. The team’s solid theoretical grounding, innovative algorithmic design, excellent engineering capability, and outstanding on-site performance together secured an undisputed championship victory. This achievement fully validates the technical advancement and practical value of the proposed framework. The methodology not only provides an innovative solution for predictive maintenance in high-value industrial sectors such as aviation and energy, but also demonstrates AI Cube Lab’s exceptional research strength and deep technical reserves in industrial big data and AI-empowered advanced manufacturing.

The ocean of knowledge is boundless, and exploration never ceases. This championship experience will undoubtedly inspire more AI Cube Lab members to devote themselves to frontier scientific research and to courageously participate in high-level discipline competitions — honing their skills and unleashing innovative potential in the pursuit of addressing national strategic needs and tackling key industrial challenges. Looking ahead, AI Cube Lab will continue to leverage high-level discipline competitions as a driving force, deepening the “research-driven, industry-academia integration” model and working tirelessly to cultivate more engineering leaders with international vision and outstanding innovative capabilities.