Explore chapters and articles related to this topic
The Industry 4.0 Architecture and Cyber-Physical Systems
Published in Diego Galar Pascual, Pasquale Daponte, Uday Kumar, Handbook of Industry 4.0 and SMART Systems, 2019
Diego Galar Pascual, Pasquale Daponte, Uday Kumar
Meaningful information must be derived from the data. Several tools and methodologies have been developed to convert data to information, with algorithms designed specifically for prognostics and asset health management applications. By calculating health, the remaining useful life (RUL) and so on, the second level of CPS architecture brings self-awareness to machines (see Figure 3.5) (Lee et al., 2014).
Ref: EATJ-D-19-00148 - prediction of remaining useful life of naval structures using a covariate-base hazard model
Published in Australian Journal of Structural Engineering, 2020
Nima Gorjian, Rameez Rameezdeen, Neda Gorjian Jolfaei, Chris Chow, Bo Jin
As critical structures become older and their total life cycle costs become higher, organisations in charge of them are facing challenges regarding maintenance and capital management based on the Remaining Useful Life (RUL) of those structures. Hence, they attempt to build an effective asset health management system that could use all available and relative asset data for future repair, replacement, renewal planning and budgeting. Despite a broader acknowledgement of the role of data in forecasting asset life, little is known about their true application in a real-life context. Though asset management professionals have been increasingly vocal on the need for data-driven decision makings, little is demonstrated in the academic literature or professional practice on how they should be dealing with the numerous data that is available at their disposal.
Prognostic modelling for industrial asset health management
Published in Safety and Reliability, 2022
Neda Gorjian Jolfaei, Raufdeen Rameezdeen, Nima Gorjian, Bo Jin, Christopher W. K. Chow
In an increasingly competitive global economy that enforces the maximising of cost savings with subsequent profit increases, successful organisations have identified prognostics and asset health management to be crucial for optimal asset life cycle management. Unexpected breakdowns would interrupt production and significantly impact on safety and total cost of ownership. There is always a requirement to reduce unexpected equipment downtime by a precise and effective asset life prediction method that is based on a robust CBM program. Though there are knowledge-rich books that introduce the concepts of CBM, they fail to critically analyse the use and applicability of the techniques for failure prognostics. In addition, research papers dealing with failure prognostics of physical assets cover disparate topics and numerous techniques developed over the last seven decades. They are mostly theoretical, hence, fail to relate those to real-life situations in organisations. Due to not having a practical orientation, contributions from these studies did not help asset managers and reliability engineers to solve their specific problems. Therefore, outcomes from most of the past studies have not been used or implemented in real-life applications and remain just as theoretical contributions. There was an acute need to capture the scattered knowledge developed over a very long period and generate new insights based on their usefulness to real-life applications. In addition, it should be presented to asset managers and reliability engineers in a way that is succinct and useful for their requirements. Hence, this review is centred on the recent research and development of CBM to explore potential approaches that could be applied in real-life situations and industry applications.
Active learning-assisted semi-supervised learning for fault detection and diagnostics with imbalanced dataset
Published in IISE Transactions, 2023
Xiaomeng Peng, Xiaoning Jin, Shiming Duan, Chaitanya Sankavaram
Asset health management of a large fleet of engineering assets is essential to ensure the performance and condition of assets continue to meet business expectations. The success of fleet health management in various industrial scenarios (e.g., fleets of ground vehicles, aerospace vehicles, and industrial robots) depends on accurate Fault Detection and Diagnostics (FDD) which are enabled by sensor data analytics, physics of the failure, machine learning, and statistical inference. Specifically, machine learning has emerged as one of the most prominent approaches for FDD given its superior performance over traditional methods during the past decade. Although the existing machine learning approaches have demonstrated ability in fault detection and classification, they still face several critical challenges in real-world FDD problems. The major driving force of the progress in machine learning-based FDD methods has been focused on hand-crafted features and classifications with presumably sufficient or complete knowledge of fault modes (Figure 1(a)), which may not be effective and applicable for a large fleet of assets that presents much higher uncertainty and unprecedented fault modes, due to various operating conditions. Such uncertainty can cause a potentially unbounded number of fault modes, known as an open-set problem (Scheirer et al., 2012) as shown in Figure 1(b). To address this problem, recent class-incremental learning methods attempt to detect and learn the new classes of fault incrementally (Liu et al., 2018; Yu and Zhao, 2019). However, these approaches require a large number of labeled samples that can infer good decision boundaries of known classes. In large fleet FDD problems, label scarcity and data imbalance are two major problems that can lead to poor performance of FDD methods and algorithms. Industrial assets are designed to work properly throughout a warranted lifetime, and faulty events or failures are extremely rare. For example, more than 99% of the data collected from an aircraft inspection system are categorized as (healthy) majority class, whereas rare failure events (i.e., ) are in one or multiple minority classes. Thus, the data collected from the fleet is naturally imbalanced. Limited by the label scarcity issue, it is almost impossible for a human to manually annotate all samples. Therefore, the number of failure modes and the types of faults are not known in advance. All these common issues make the fleet FDD problem extremely challenging, as depicted in Figure 1(c). To devise an efficient and reliable FDD method for fleet assets, we established a novel framework to address these theoretical and practical challenges.