Naval capabilities
The USS McCampbell is part of a US Navy pilot project that started in 2022 to test proof of concept for predictive maintenace.
By June 2016, it could no longer conduct normal operations and subsequently remained idle pier-side for over two years, waiting to enter a shipyard.
Embrace the algorithm Much of the US Navy’s shortcomings in this area can historically be put down to bad management and bad planning. However, the DoD has recently grasped the nettle – given the need to improve maintenance turnarounds – by employing condition-monitoring technology such as sensors, data analytics, algorithms and AI to schedule maintenance based on evidence of need. In short, it has finally chosen to embrace predictive maintenance. The argument is that, if implemented correctly, predictive maintenance can reform a military’s approach to weapon systems readiness by reducing unplanned and unneeded maintenance, cutting maintenance delays and shaving sustainment costs, according to officials. Already, the US Navy has predictive maintenance pilot projects in place for surface ships – including USS Mason, which began in 2020, and USS McCampbell and USS Bulkeley, which both started in 2022. These pilot projects aim to test proof of concept and further develop analysis techniques, as well as determining how best to provide predictive maintenance prompts to uniformed maintenance personnel aboard US ships. In 2022, in the same vein, the US Navy issued guidance requiring the use of predictive maintenance on all new and existing surface ships, submarines and aircraft carriers, where technically feasible, cost- effective and beneficial. For Tiedo Tinga, professor of life-cycle management at the Netherlands Defence Academy and professor of dynamics-based maintenance at the University of Twente, defining ‘predictive maintenance’ is a relatively simple affair. “My definition of predictive maintenance is [that it’s] a form of preventive maintenance, for which the future optimal moment of repair or replacement of the system or component is based on a combination of the present condition and a prediction or calculation of the remaining time to failure.”
Predictive maintenance can be considered an extension of condition-based maintenance, which bases decisions on the present condition of the system under review. Condition-based maintenance disregards any future degradation, and as a result frequently requires immediate action when a condition threshold is exceeded. Predictive maintenance provides additional benefits in adding response time, allowing for proper organisation of spares, facilities and workforce for upcoming maintenance tasks. Within predictive maintenance solutions, AI and machine learning (ML) constitute one approach to obtaining the remaining useful life (RUL) out of a system. Alternatively, applying physics-based degradation models – covering, for example, wear, fatigue or corrosion – is another approach that can be followed. The advantage of using AI and ML is that only limited system or domain knowledge is required, as the algorithms simply analyse the data to obtain the required relations and predictions.
The limited data sets challenge That’s not to say that predictive maintenance doesn’t face obstacles. “The disadvantage is the large amount of high-quality data that is required to train these algorithms,” says Tinga. “Specifically, [data related to] a considerable number of actual failures is required. That is a big problem in maintenance in general, and in military sustainment in particular, as maintenance of critical systems aims to prevent failures.” Typically, the number of examples for training AI is limited, or the examples are associated with a completely different operating profile of a ship or system, which makes it less useful, according to Tinga. Yet, there are successful examples of AI in maintenance – today, these are mainly related to fault or anomaly detection or diagnostics. However, AI shouldn’t necessarily be viewed as a magic bullet. Indeed, the largely variable and unpredictable operation of military systems makes it more difficult to effectively make use of. AI is far easier to train with known, repeatable processes, but it works less well in uncertain settings
Defence & Security Systems International /
www.defence-and-security.com 13
US Navy / AWS2 Jack Ryan
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45
