Advanced Machine Learning Techniques for Fault Diagnostics and Prognostics: From Modern Complex Manufacturing Systems to Healthcare

在当今全球化、互联互通和竞争激烈的市场中，现代复杂制造和服务（MCMS）系统（包括但不限于航空航天、交通运输和智能电网）以最高可实现的可靠性充分发挥其潜力是至关重要的。这类MCMS系统容易因退化和低质量部件而发生随机故障，这可能导致从破坏基础设施到危及人类生命的各种严重后果。为了避免代价高昂的失败，必须在质量和维护概念上投入巨大的努力，这是所提议的研究计划的目标领域。这项名为“用于故障诊断和预测的先进机器学习技术：从现代复杂制造系统到医疗保健”的研究计划的目标是在设计、开发和应用最先进的人工智能（AI）和机器学习（ML）技术方面调查和实施有前途的研究思路，这些技术有助于在过程监控和维护中推进故障诊断/预测。

In today's globalized, interconnected, and competitive market, it is critical and of paramount importance that the Modern Complex Manufacturing and Service (MCMS) systems, including but not limited to aerospace, transportation and smart power grid, operate at their full potential with highest achievable reliability. Such MCMS systems are subject to random failures due to degradation and low-quality parts, which could lead to a variety of severe consequences ranging from destruction of infrastructures to endangering human lives. To avoid costly failures, tremendous efforts have to be invested in both Quality and Maintenance concepts, which are the target areas of the proposed research program. The objective of this research program, entitled "Advanced Machine Learning Techniques for Fault Diagnostics and Prognostic: From Modern Complex Manufacturing Systems to Healthcare," is to investigate and implement promising research ideas in the design, development, and application of state-of-the-art Artificial Intelligence (AI) and Machine Learning (ML) techniques that contribute to advancement of fault diagnostics/prognostics in process monitoring and maintenance. Recently, advancements in sensing technologies with progressive advancements in computation and communication technologies have resulted in exponential growth of high-dimensional and multi-modal streaming (HDMS) condition monitoring (CM) data. Efficient utilization of HDMS data leads to highly accurate prediction results in process/system health diagnostics/prognostics. To achieve the aforementioned goal (i.e., to efficiently utilize these ever growing sources of CM data), recently, there has been a great surge of interest in AI/ML based data-driven methodologies. Similar to digitalization which transformed end-to-end business models, the AI and ML are positioning themselves as the transformative technologies of the century leaving industries with two options: "Embrace the process/system monitoring via AI/ML solutions or get left behind''. The "Common Theme'' of this research program is to apply advanced, hybrid (i.e., coupled with state-of-the-art statistical methods), and deep AI/ML techniques for process quality control, maintenance management, and survival analysis. In particular, the proposed research program focuses to address the following rigid research challenges: (i) How to monitor and control manufacturing processes with HDMS data? (ii) How to design "deep'' representations to monitor/control manufacturing processes/systems with HDMS data? (iii) How to develop optimal maintenance policy for a MCMS system subject to degradation via HDMS data? How to combine event data and CM data? In conclusion, the proposed research program is believed to be timely and of significant importance for development of future quality control and CBM programs in Canada. Should the new knowledge prove as valuable as I hope, it is expected they will have a significant impact on our country.