DP Control System Pt.3 – Errors and Error Handling

Summary: Previous articles covered DP gain/PID controllers, Kalman filters, how the “Kalman filter” above isn’t really a Kalman filter but uses them (Pt.1), that neither the calibrated physical model nor the measurements can be fully trusted (Pt.1), how adaptive control loops correct for but also hide problems (DP current is actually DP error current, Pt.1), and assumptions that are sometimes false (unfairly described as “lies” to emphasize the point in Pt.2). DP control systems have come a long way, but they need to go further and incorporate modern technology, if they are going to become more reliable. There are limits to the adaptability of the current designs, and the DPOs & DPEs are still a vital part of error detection and correction.

Introduction: In my first draft of the first article, I cluttered up and doubled the length of the longest paragraph with examples of all the things that could go wrong, because I wanted operators to have concrete examples of why the adaption created by the feedback loops and model was so important. Instead, I cut it back to general descriptions, but the errors are still work knowing, so we’ll look at some examples.

Calibrated Physical Model: This is the physics based model used to keep the measurements grounded in physical reality. Before the vessel is even made, we know its hull form, thruster type/curves/power, and estimated current and wind drag (force coefficients) from different directions. Dockside and sea trial scaling is used to confirm or correct these values based on real vessel response, and the final results demonstrated by the DP trials. This doesn’t always go right - I have been on vessels where the scaling clearly did not match the vessel. You might be surprised to hear that I have even lost some fights where all the evidence was on my side, because the big DP company had to listen to big customers, but a small customer could be ignored. On one vessel, the yard was forced to physically change the ship to match the DP model, rather than adapting the software to match the ship. Unsurprisingly, subsequent ships from that shipyard tended to use alternate DP control system providers.

Calibrated? Assuming the physical model matches the vessel at the end of DP system and thruster scaling, doesn’t mean that it will match the vessel during operation. Ignoring the unknowns and assumptions covered in the previous article’s “lies”, the key word is calibration. Calibration doesn’t stick. It drifts off as the vessel changes, so it needs renewed. Calibration is a clock that winds down. When the ship left the shipyard, it had a clean hull, but after a year or two of operation, it has some growth on it and has more hull drag for current, even if the exact same hull profile is true. The hull current and wind profile, may not be true. Was the scaling performed at two or more drafts, and was it done at a draft that the vessel will work at? Scaling is usually done in light ship conditions, as there isn’t enough ballast to simulate load conditions without dirtying specialized tanks. I’ve seen two ships that had difficulty keeping their bow tunnels underwater, and a colleague saw one that couldn’t even get the bow tunnel into the water. These examples obviously make real scaling impossible, but the difference between scaling and operating drafts can sometimes be a hidden error, as the model may be wrong. Permanent pitch and roll can also have effects on wind and current drag. Is this scaled? Yes, for large ships and semis, but usually not. This is corrected and hidden in DP error current, but large error currents are dangerous. Having the wrong draft has caused loss of position in the IMCA DP incident reports, and DP guidelines mention the danger of it causing oscillation.

Physical: Even the placement of the equipment can be off. If the thruster is 1m away from what the DP control system manufacturer was told, they aren’t going to use laser measurement to find out. I’ve tested a ship where this has had a significant effect. Another example is position reference sensors whose positions aren’t correct in the DP control system. They used to be common but are easily found by doing a turn test (if the calibration is wrong the sensors diverge) and are rare today. A more common problem is a lack of thruster calibration. I’ve had some funny results doing box tests. In one case, a main azimuth thruster was off by 30-40 degrees. Quite a few thrusters have had much less thrust than shown in the capability plots. Thruster calibration and capability needs regularly checked. Small spot checks are effective (speed/pitch vs load, direction vs wash or local indication) and thruster calibration should be formally checked annually. Wrong information causes the DP control system to make bad actions, so these faults need avoided by the crew and corrected before loss of position. Again, these can be hidden in the DP error current and the crew needs to watch for them, because the DP control system does what works right now, but what works right now might not work in the future, if it is based on bad information.

Calibrated Measurements: The physical model doesn’t include everything important and neither do the measurements, but the measurements do provide vital control status information that the physical model lacks, and this is used to adjust to the hidden forces acting on the vessel. In a court room, it is preferred to have an expert eye witness give accurate testimony, but courts usually have to settle for inattentive people who were there, and even cameras can be wrong as they only show things from one angle, and an alternate view could change everything. Ideally, DP control systems would accurately measure everything important for DP control, but we don’t and we sometimes measure the wrong thing, the wrong way, because that’s how we used to do it. Speed and pitch feedback is only partially related to thrust, and errors in thruster feedback can be vital and require DPO detection and correction. We know that we need three mostly independent heading and motion reference units, and wish that we had three independent position references, so we could detect the faulty one rather than have them all go bad together.

Error Handling: Sensors can be expert witnesses, but they can also be unreliable narrators (e.g. The Usual Suspects) that lead the DP control system in the wrong direction. Calibration is vital but not enough, as the same equipment or principles can have common faults that may not be easily detected. The DP control system has a number of checks that it performs, but ultimately depends on the DPO to detect when it has failed to do so and make correction before position keeping is endangered. The DP control system tends to do the low level checks itself and depends on operators with better information and experience to cover more complex failures. When possible, the DP control system checks for valid information (not lost, garbage, or noise) in an acceptable range (not too high or low) that is consistent (not wildly varying or jumping) but not frozen (a constant breeze is possible but a lack of variation is often a sign of a sensor fault), and physically possible (200km/h is wrong and so is teleporting), from a healthy sensor (ready and measurement health within acceptable boundaries) that agrees with other sensors (mismatch or median check) measuring the same thing (hopefully independently).

Errors in Error Checking: Not all DP systems do all these checks and not for all signals. Some systems use physical limits that are sometimes wrong. E.g. a wave swinging heading faster than considered, or a lean limit to pass an MRU test that ignores that ships sometime pitch or roll that far. Storms are bad times to find these shortcuts. Some systems use automatic sensor weighting that can cause loss of redundancy by following consistent errors instead of noisy truth. It’s often helpful, but sometimes causes automatic errors that need prevented by the DPO. Some faults can be designed away by having independent sensors, but some environments and functions can’t support that, such as deep water drilling only having two position references, or some shallow water track follow operations only having DGPS (until Sonardyne Sprint-Nav is proven as a DP position reference). When the DP control system has a conflict, DPOs can sometimes choose the wrong sensor, and time needs spent getting familiar with each system and its limitations to avoid this. Design logic is sometimes faulty, and the DP control system can glitch. The combined DP-DPO system depends on designers and operators avoiding and correcting faults, but sometimes basic recommendations, like sensor variety and fault finding training, are ignored.

Conclusion: Both the physical model and the measurement are incomplete and subject to error, but the slow adaption of the errors and unknowns into the adjusted DP model (scaled model plus DP error current) allows better control than either could provide on their own, so long as the information in both the base model and measurements are mostly correct. When one or the other is off, the risk of being unable to respond to a fault or change increases, so it is vital to track the difference between real and error current, and to find & correct or manage the sources of the difference (e.g. T3). If this isn’t tracked and detected, unexpected failures of position control will occasionally occur when they don’t have too. Some unavoidable ones will occur anyway, due to some of the assumptions in Pt.2 (e.g. big wave, quick current change), but detecting and minimizing known errors supports the designed predictable operation of the system and is a primary function of the DPOs & DPEs.