The maritime industry has been flooded with promises around predictive maintenance for years. Software vendors promise cost reductions of 30–50%, conference presentations show dashboards full of green indicators, and pilot projects are celebrated as revolutions. The reality looks different.
Value depends heavily on three factors: data quality, technical assessment competence and functioning decision pathways. If any of these is missing, the result is at best an expensive dashboard that nobody uses. At worst, false confidence in unreliable data leads to poor decisions – such as postponing a necessary repair because the system shows “all green.”
The overselling is particularly visible with vendors who promise “AI-driven predictions” without explaining on which data basis these models were trained. An algorithm trained on data from land-based machinery is worthless for maritime application. Specific operating profiles, fuel qualities and environmental conditions at sea require maritime training data – and there is considerably less of that available than ashore.
Predictive maintenance works reliably today for equipment with clearly measurable condition parameters and high failure impact. These are primarily rotating machines: main engine, auxiliary diesels, turbochargers, large pumps and generators. Here, vibration measurements, temperature trends and oil analyses deliver robust early warning.
Trend-monitored lube oil analysis for combustion engines has also proven its worth. Regular sampling at fixed intervals – every 250 to 500 running hours – allows conclusions about cylinder liners, piston rings and injector nozzles without disassembly. The cost of an analysis is EUR 50–150 per sample – a negligible amount compared to the cost of a liner crack.
Predictive maintenance is currently less reliable for complex systems with many influencing variables, such as ballast water treatment systems, scrubber systems or control logic. Reference data is often lacking here, and the failure modes are too varied for simple trend models.
Overly broad projects without connection to the PMS and without operational planning rarely deliver substance. A classic example is the fleet-wide monitoring system where data is collected but never evaluated in a structured manner. The dashboards exist, but nobody has the responsibility or the time to derive actions from them.
Equally lacking in substance are approaches that rely solely on automatic alarms without incorporating technical context. A temperature threshold applied regardless of load and ambient conditions either produces too many false alarms (and gets ignored) or reacts too late. Thresholds must be calibrated load-dependently and machine-specifically – this requires technical understanding, not just IT competence.
Particularly concerning are vendors selling “Predictive Maintenance as a Service” without verifying the condition of onboard sensors. If the temperature sensor on the turbocharger bearing has not been calibrated for two years, or the vibration pickup is loosely mounted, all analyses built upon it are worthless. The foundation must be sound before analysis methods are applied.
Real impact comes from focus: a few critical assets, reliable measurements and clear escalation pathways. In concrete terms, this means the ship manager defines a maximum of five to eight assets per vessel for which predictive maintenance is introduced. For each asset, the relevant parameters, thresholds and response protocols are established.
The Chief Engineer on board produces a short trend report weekly – not an elaborate analysis, but a structured overview: which parameters have changed? Is there a trend? Is there a known cause? This report goes to the superintendent, who must respond within defined timeframes: acknowledgement, follow-up question, or action planning.
This simple cycle – measurement, assessment, decision, feedback – is more effective than any expensive platform without a defined process. IMO MSC.1/Circ.1378 (formerly the Guidelines on Maintenance) underscores the importance of systematic condition assessment as a complement to the PMS. Anyone who implements this approach consistently has already achieved more than most “digital” pilot projects.
A robust predictive maintenance system in the maritime environment must fulfil four core functions. First: reliable data acquisition with calibrated sensors and defined measurement points. Second: contextualisation of data – every reading must be linked to operating condition, load and ambient conditions. Third: trend analysis with defined thresholds and escalation levels. Fourth: feedback into the maintenance process – detected deviations must translate into PMS jobs and spare parts orders.
Most systems fail on points two and four. Data is collected but not contextualised. Deviations are detected but not converted into concrete maintenance tasks. This is the difference between a monitoring system and a genuine predictive maintenance process.
For sensor calibration, an annual cycle coordinated with the Annual Survey is recommended. IACS UR Z10.2 provides relevant requirements for maintaining class-relevant sensing equipment. Temperature probes should be checked against reference instruments, vibration sensors against reference standards with known amplitude and frequency.
For data storage, a hybrid approach has proven effective: raw data remains on board for detailed post-analysis, while aggregated trend values (daily averages, weekly averages, min/max) are transmitted ashore daily via satellite. This keeps communication costs low whilst still securing the trend overview ashore.
A bulk carrier with two ABB A175-L turbochargers showed a gradual change in the turbocharger-to-engine speed ratio on the port side. Over six weeks, the ratio dropped by 3% at comparable load. Simultaneously, scavenge air pressure decreased slightly. Individually, both values were within the acceptable range. In combination, they indicated progressive fouling on the turbine side that could no longer be compensated by online washing alone.
Systematic trend evaluation led to an offline wash at the next port, followed by a borescope inspection. This revealed incipient deposits on the turbine blades that, with continued operation, could have led to imbalance and bearing damage. The cost of the planned intervention: approximately EUR 8,000. The cost of a turbocharger failure at sea: EUR 80,000–150,000 plus off-hire.
Operators can quickly check whether their predictive maintenance approach has substance using five questions: Are the measurement data calibrated and contextualised? Are there defined escalation pathways from alarm to action? Do the results flow into PMS jobs and spare parts planning? Does the crew receive feedback? Can the superintendent demonstrate that at least one action per quarter was initiated on a data-driven basis?
Anyone who can answer all five with yes has a functioning process. Anyone who hesitates on more than two is probably running monitoring without impact – an expensive facade.
Free initial consultation – we analyze your situation and find the best path forward.
Request Consulting