Predictive maintenance gains traction with AI tipped as the future

Maintenance of ships’ equipment is still for the most part performed either under a planned maintenance regime following OEM’s service recommendations or as reactive maintenance when a sudden breakdown or failure occurs. These were almost the only two options available to operators until around 15 years ago when the idea of condition-based maintenance (CBM) began to be spoke of in maritime circles.

Engine makers were at the forefront of this movement, and most could draw on some years of experience garnered on shore-based equipment. At a time when a shortage of experienced sea staff and a rapidly expanding world fleet were coinciding, the idea of handing over monitoring of engines might have seemed an attractive proposition.

An attractive proposition

It is not difficult to see the attractions of a CBM strategy. Planned maintenance involved having a stock of components onboard that could be fitted whenever the requisite number of service hours had been completed. Furthermore, having disassembled the equipment, planned maintenance required components to replaced even if inspection showed the old part to have more useful life in it. Since the maintenance schedules followed the OEM’s manuals as regards time or running hours, often the actual operating conditions which may have involved long periods of high loads or idling, did not influence when maintenance was done.

CBM on the other hand makes use of sensors to measure and record vibrations, pressure, temperature, humidity, liquid flow rates and other vital parameters to give an indication of how close to optimum running conditions the machine is performing. From this it is possible to say that planned maintenance could be deferred until future data suggested that component change was desirable.

In most cases, it is beyond the skill of onboard engineers to determine this, which is why the data has to be transmitted or sent ashore for analysis by the maker’s expert service technicians. Ideally this would be done over the ship’s communication systems on a regular basis although where this is not possible, the recorded data can be despatched by other means, but this will mean a delay in analysing it. An intriguing option that originated in the shore power generation sector was that remote management of the engine might be possible allowing for a reduction in engine room staff.

However, a limited take up at the time can be put down to several reasons. Most of the engines in service were not the new generation of electronically controlled units and were not equipped with the multitude of sensors needed for remote monitoring. And at the same time, communication costs were high in the days before VSAT installations were the norm. Other calls on owners finances were also pressing – ECDIS was soon to become mandatory, the AFS Convention had only recently come into force and new anti-fouling coatings had to be applied, the NOx Tier II rules were coming into force, and the SOx emissions were being tightened in SECAs and in the open seas.

Expanding the field

Given their specialist knowledge, it is not surprising that OEM’s were the in the vanguard of CBM. But they were, and are, not without competition. Software specialists in the ship management and procurement arena and class societies were also on board. For ships with an operational history, a big task was migrating years of handwritten or printed data on engine parameters into a new electronic system that could make use of that data going forward. Software specialists can often make this task easier.

In some cases, there were tie ins with vessel performance software tools that combined trim optimisation and weather routing along with engine management software. Companies such as ABB, Kongsberg and SpecTec are all active in the arena today along with many others.

New electronic engines come with sensors already fitted and the sheer number, 500-600 on some large engines, recording data at intervals as short as one second gives an idea of the vast amount of data that needs to be analysed.

All the main engine manufacturers have their own service divisions and offer CBM and predictive services under their various maintenance contract offerings. All of these have digital analysis solutions and while some refer to algorithms used to interpret the data, others have often referred to the process as AI or machine learning. However it is termed, the result is the data is interpreted and used to determine if any planned maintenance needs to be done immediately or might be deferred.

An OEM’s perspective

Motorship asked Everllence as representative of the engine manufacturer community and accepted as the dominant player in the two-stroke market for its methodology and how it sees CBM morphing into predictive maintenance in the future. Stig Holm, Head of Marine & Power Digital & Academies Denmark, a part of Everllence, said that there are around 1,000 MAN B&W ME engines making use of the company’s CBM services, but the solution is not yet for all of its customers, there being around 15,000 older MC type engines still in operation.

He also said that from the outset, the service was designed to be a deep analysis of engine operation and as yet does not take account of outside factors such as sea state or weather, but this may happen in the future. In some areas it does co-operate with other organisations to get more value from data with lube and fuel oil analysis being an example.

Holm added that for Everllence, there are two future steps beyond CBM with Prediction being one of them and Forecasting being the holy grail. Explaining the subtle difference, he described prediction as being a statement about something that will happen, while Forecasting is estimating future trends or values over time, often in a time series context and is more about time patterns than individual outcomes. He adds that forecasting can reduce the impact of inappropriate engine operation (Engine specific or human driven).

Holm says that customers want a service that will prevent breakdowns whilst getting the most service life out of components. To get an early warning of component failure data is crucial but more important is the analysis. Even with the 550 plus sensors on a typical engine, there are still critical gaps in data. For example, there is as yet no sensor that can measure the conditions in the combustion chamber.

Whether called algorithms or AI, the analysis process is in constant development. With so much data available across many engines, forecasting can involve comparing outliers below alarm limits that in certain combinations can give clues to problems developing. The situation can be exacerbated in dual-fuel engines of in engines fitted with NOx reduction systems if a fuel type such as LNG or the SCR or EGR is not used for some time. This sort of information is covered by advisory letters sent out by the company, but it will remind chief engineers and superintendents when data indicates that a system has been unused beyond the recommended time that some corrective action is needed.

Holm believes that the CBM, prediction and forecasting services can only become more in demand in future, not least because there are now so many dual-fuel engines being installed and experience of the different fuel types and combinations is lacking or non-existent across the engine crew sector.

A new twist on old skills

It used to be said that the best Chief Engineers could walk into an engine room and tell by means of listening and sense of smell if something was amiss. That may or may not be true, but it does open up new avenues for AI.

A novel approach has been added to the ship maintenance army by a newcomer to the scene, Singapore-based Groundup,ai in what it calls Cognitive Maintenance. The technology has been deployed on vessels belonging to the Republic of Singapore Navy and uses sound sensing to alert for impending problems.

Over a period of weeks, the sound sensors recorded data from a range of equipment including starting air systems, compressors, pumps and other systems. Having ‘learned’ the normal operating sounds the AI system picked up on subtle changes in different pieces of equipment and was able to detect an imminent failure in the pinion gear of a starting air motor, a choking issue in an air compressor and a generator coolant pump with mild resistance that transpired to be non-critical but which increased the sound database of the system.

The above example gives an insight into the intriguing possibilities that AI may have in the area of ship maintenance and equipment monitoring. It only requires enough pioneering owners to take the initial steps and if shown to be successful others will follow.