Using sensors to cross check each other

Kyungrok Kim, Vessel Performance Manager, Bestship

Fuel flow, engine torque and engine RPM are correlated, but not directly.

But you can map their data together to identify faulty data, and then go on to see if a problem is due to engine or hull. Bestship explained

When ships fit sensors for vessel performance data, they typically fit a fuel flowmeter, an engine torque meter, and a sensor to measure revs per minute (RPM) of the engine.

The readings from the sensors should be correlated, because the vessel’s speed can be estimated from both the fuel flow and the torque, if you know the RPM.  

But the relationship is not direct, because changes in fuel flow will not immediately change RPM, for example.

Faulty data can be detected by looking for data which is outside an expected range, based on readings from the other sensors.

If all the sensors are working correctly, then you can determine if any loss of performance is due to the engine or the hull.

Kyungrok Kim, Vessel Performance Manager with Bestship, explained how to do it, speaking at People Tech Maritime event in Hamburg in May.

Bestship is a Hamburg-based company providing vessel performance management and regulatory compliance solutions, jointly owned by Wilhelmsen Group and MPC Capital.

Mr Kim is a former project engineer in the vessel performance centre of NSB Group, Hamburg, and a former marine engineer for Hanaro Shipping Company in Seoul. He has an MSc in maritime energy management from the World Maritime University.

Faulty sensor data

If you have faulty sensor data, you can end up with your vessel appearing to be in the wrong CII band, with commercial implications.

You may take rectification actions you do not need, such as unnecessary hull cleaning. You may also find you cannot respond easily to claims about fuel consumption and speed from the charterer, Mr Kim said.

Even if you have sensors which have passed ISO stability tests, ISO stability tests validate sensor repeatability in isolation, not cross-instrument agreement. You can have sensors which comply with regulations but are still providing incorrect data.

With better methods to identify faulty sensor data, you can build trust in the data, so seafarers no longer feel they should make separate calculations about fuel consumption for their noon day reports, he said.

It means you are only doing analysis on data which is good enough to be used in defence against potential technical and commercial claims.

Having high quality performance data would also make a better basis for AI analysis, he said.

Bestship’s method has four stages. The data is first validated, then given a stability analysis, visualised and then used to make a decision.

In a typical example from an actual vessel, Bestship had 166,000 raw high frequency data samples (RPM, flowmeter, torquemeter), of which only 38,000 passed the sensor cross check for validation. Of those 38,000 samples, only 13,000 data samples passed the stability check. So 92% of the data was rejected.

Bestship analyses sensor data at a granularity level of 60 second blocks, which gives higher resolution results than the 10 minute blocks recommended in ISO 19030.

Data validation

The first step of Bestship’s method is data validation. Data is taken from the three sensors, flowmeter, torquemeter and RPM. You make a calculation of the vessel’s power and speed through water, based on the flowmeter data and separately with the torque meter data, both based on the RPM.

The calculation method looks for the coefficient of variation (CV) between the calculation of power based on the flowmeter and torquemeter readings, assuming the RPM is correct.

If the CV is under 0.2 on 90 per cent of samples, that is an indication that both torque and flowmeter data can be trusted.

In another example, the CV was over 0.6, which is too high. It was clear from looking at the data that there is high noise in the fuel flowmeter reading. This suggested the right approach here was to reject the fuel flow meter reading and use the torque meter.

Bestship suggests that a CV above 0.2 means sensors should be taken as being in disagreement, and some recalibration is needed. It means your data about speed through water is affected.

Data stability analysis

The second stage is to analyse the data to see if it is stable. There is no value of using performance data from times the vessel operations are not in steady state, such as in manoeuvring or drifting.

Two methods to do this are ISO 19030 and a method from Dalheim and Steen.

Bestship advises to use both methods, and to reject any data where either method shows the ship was not in steady state at that time.

Data visualisation

With data which is both validated and stable, you can create a data visualisation, which helps identify the cause of any loss of performance.

For example if the engine is consuming more fuel than you expect for a given RPM, but you get the speed through water you expect for that RPM, that indicates engine wear, he said.

The parameter could be that the consumption is more than 5 per cent than at the shop test stage.

If it is giving higher power than expected for a given speed, but at the expected RPM, that indicates hull fouling. You should do something if the added resistance is calculated to be over 30 per cent, he said.

Decision

If you know you have engine wear or hull fouling, a next step is to do something about it.

For example, owners can do an underwater inspection.

Mr Kim showed data from an underwater inspection of a container ship, after the system identified high levels of hull fouling. The bow area, bow thruster, and sea chest gratings were 70 per cent covered in barnacles. The flat bottom, bilge keels, vertical sides and propeller were 80 per cent covered in barnacles. The propeller was also covered with slime and algae.