The high demands placed on ship’s engines, propulsion systems and other critical equipment can lead to premature wear and tear, which can cause equipment to fail unexpectedly. Identifying such shortcomings has to occur at the earliest possible moment. Regular inspections, along with a rigorous programme of servicing and repairs (preventive, scheduled maintenance) and condition monitoring, are a vital part of that process.
Condition monitoring (CM) has long been used to improve plant performance and reduce costs, by helping to diagnose faults and optimise maintenance schedules. “There is a well understood kit bag of CM tools and techniques, of which the leading method is vibration analysis, with thermography, ultrasound, oil analysis, motion amplification and others also widely used,” says Geoff Walker, operations director of Faraday Predictive. “However, none of these tools provides a universal solution to every situation. Rather, like any kit of tools, the skilled practitioner understands the capabilities of each tool and selects the most appropriate tool, or combination of tools, for the particular task in hand.”
Faraday Predictive’s approach has been to develop new mathematical algorithms, in conjunction with the maths department at Cambridge University, resulting in the emergence of a new suite of solutions. “These give us a much more powerful set of diagnostics, with prediction of future condition, identification of the nature of the developing fault and much greater control of the set-up,” states Walker. “Outputs these systems provide include both present and predicted condition of the equipment at an overall level, along with the present and predicted status of individual faults up to three months into the future, allowing specific maintenance plans to be drawn up with the right materials, for the right work, at the right time.”
REPEATED FAILURES, COSTLY OUTCOMES
By way of illustrating these solutions in action, Walker cites an LNG tanker vessel that had been suffering repeated failures of seawater pumps, with the damage on each occasion resulting in the need to replace the pump completely at a cost of a few thousand dollars each time. “MBVI [Model-based voltage and current analysis] permanent monitoring units were installed and, after several months, one unit started showing some changes in behaviour. The two main features were a drop-off in power factor and the development of a hump in the spectrum around the vane pass rate of the pump.”
When the power factor had dropped off to 80%, repair was advised. It was discovered that internal corrosion of the pump had destroyed the flow straighteners on the pump inlet, leading to poorer efficiency. Wear was found at the replaceable wear rings, which had been leading to reduced pumping – and hence the drop off in power. “And, crucially, corrosion had created a hole right through the internal section of the pump casing from the high-pressure to the low-pressure zones,” recalls Walker. “Each time the impeller vane passed this point, the pressure wave created by the vane was able to partially escape through this hole, leading to a pressure pulse at vane-pass frequency. Due to flow turbulence, this was not at a single frequency exactly matching vane-pass rate, but was in a broad spread of frequencies around this frequency.”
The hole was repaired by cold-weld resin (Belzona) and the wear rings were replaced as a routine. “The total cost of the repair was a few hundred dollars – less than one tenth of the typical pump replacement cost,” he adds. Most importantly, detection had avoided the prospect of sudden, unexpected failure, and the hassle and disruption this might well have caused to a tanker out at sea.
With the increasing emphasis on sustainability and reducing carbon footprint, there is now a new generation of condition monitoring solutions that can also provide energy monitoring and energy optimisation capabilities, based on electrical measurements, to give information on electrical, mechanical and operational problems all in one go, he points out. “A variety of electrical condition monitoring techniques have been widely used for years, including partial discharge (PD) to detect early deterioration of insulation, motor current signature analysis (MCSA), which has mainly been used to identify specific motor faults, and motor start-up current analysis.”
WHAT PREDICTIONS FOR ‘PREDICTIVE AI?’
AI has often been touted as the next great breakthrough for predictive maintenance, so what role might it play within the marine industry? Not as significant as might be expected, according to Walker. “A few years back, when ‘IIoT’ and ‘Industry 4.0’ were buzzwords – and AI tools were first becoming more readily available – they were a bit of a solution in search of a problem and people hit on condition monitoring as a good use case for the technology,” he says. “The concept was the simple idea of easily grabbing data from plant with a few sensors, communicating the data to the cloud via low-cost ‘IoT’ comms and analysing them with AI, which would give an answer.”
While there are a small number of companies pushing hard in this area and investing a lot of cash trying to make it work, a lot of businesses discovered it’s not as easy as that, believes Walker. “Identifying a change in a signal is relatively easy, using AI, but identifying the cause and significance is much harder. In the world of serious condition monitoring, neural networks have always been viewed warily, because they were notoriously opaque – you can get an alert, but you don’t know how the neural network reached its conclusion.
“Since the benefit of condition monitoring is only realised when you use the information from it to make better maintenance and operations decisions, you generally want to understand why you are making those decisions,” he adds. “The higher the criticality of the plant in question, the more certain you need to be that you will be making the right decision before you either proactively shut down the plant to do maintenance or, conversely, keep it running, because you can’t afford to stop. So, a technique that doesn’t explain why it is giving you an alert is not as valuable as you might at first think.”
An additional challenge to AI-based solutions for serious condition monitoring situations is that the greatest benefit comes in the most critical areas – and yet those areas have the least available data on failures, simply because failure can’t be tolerated. “Aircraft can’t be allowed to fall out of the sky, and oil and gas production platforms can’t afford to lose production worth millions of pounds per day,” adds Walker. “So, the areas where AI systems are most likely to gain traction are the lower criticality areas, with relatively simple machines, where the maintenance manager just wants a simple indication that something different is going on with a machine and doesn’t need to know in advance the exact nature of the problem, and therefore the specific spare parts that will be required.”