CEO Thomas JJ Meyer describes how Machine Prognostics AS is introducing health and usage monitoring systems to the shipping industry.
First developed as a result of the 1986 Chinook crash killing 44 passengers, health and usage monitoring systems are nowadays commonly used to monitor the health of helicopter components, such as gearboxes, bearings, shafts, and rotors during flight.
Ships can also largely benefit from the latest health and usage monitoring systems technologies as they are autonomous – that is, the decision support for maintenance is performed by algorithms and not by humans. The health of the machinery is systematically quantified and the data compressed to a few kilobytes per day, allowing seamless communication by satellite. This approach is a step ahead of the current shipping machine monitoring norms approved by class. This article discusses the work performed by Machine Prognostics AS, a small Norwegian technological start-up established in late 2016 taking on the challenge to bridge the technological gap in between health and usage monitoring systems in the aerospace and shipping sectors.
The situation today
Nowadays, in 98% of cases,1 maintenance procedures onboard ships are organised around a time-based strategy. In other words, machinery health inspection and maintenance tasks are performed at regular intervals regardless of the health condition of the machinery; the whole process is orchestrated around pre-defined calendar intervals.
The key advantage with time-based maintenance strategies is the ease of implementation. A human crew can easily construct a master calendar based on maintenance recommendations from equipment manufacturers and follow it. In addition, having visibility of the upcoming maintenance tasks makes human resources planning easier.
The drawbacks of time-based maintenance strategies are the:
- Costs – inspecting machinery or replacing parts regardless of their health condition engenders a waste of time and materials; and
- Dependencies on maintenance personnel’s diagnostics capabilities, experience and training.
Indeed, once a task, e.g. a routine machinery inspection, is finished, a report is logged for archive purposes. Maintenance logbooks, digitalised or not, are text-based entry systems – that is, the logged information about the health status of the inspected machinery is described by means of adjectives, i.e. the machinery is in good condition.
The drawbacks with such an approach are:
- Adjectives are by definition qualitative and cannot be used as a reliable quantitative source of information;
- Human-based diagnostics/judgements vary widely among experts;2,3
- A written log is not a natural digital source of information; and
- Human machinery inspection is neither a permanent solution nor a cost-effective one.
As long as humans are required to diagnose and prognose machinery health, the maritime industry will not be able to truly capitalise on the opportunities linked to the digitalisation of maintenance data, as the information provided will not be repeatable and fast enough to create truly automated health and usage monitoring systems enabling subsequent business opportunities.
Our solution: Foresight
To extract value out of the challenging case of machine monitoring, Foresight (Fig. 1), the health and usage monitoring systems technology developed for the shipping industry through a collaboration by Machine Prognostics and GPMS, Inc., addresses the business case around eight technical key points:
- Practicality – Our sensors (e.g. vibration, acoustics, temperature etc.) are bolted or glued on the asset shell. One single cable powers and transmits data for up to 100 sensors;
- Data quality – Foresight sensors have a unique design, with the signal condition and data conversion occurring within the sensor package itself, which is a Faraday cage. This ensures that there is no external electromagnetic interference when taking measurements. In addition, Foresight sensors have an automated built-in test; if a sensor is damaged and not sending the right data, that data will be automatically discarded;
- Embedded analytics and data compression – Component diagnostics is performed at the sensor level using condition indicator (CI) algorithms. CIs are the acquired vibration signals converted by algorithms sensitive to a given fault (e.g. unbalance, broken gear tooth etc.). Foresight uses a battery of CIs working in parallel to diagnose mechanical faults and to compress the data, allowing seamless communication in between IT systems. Typically, data sampled at 100,000 samples per second are compressed to a few kilobytes directly at the sensor level;
- Sensor fusion and multisensorial approach – Embedded analytics can make use of sensor fusion and interpret, for example, acoustic, vibration, temperature, oil etc. signals. Advanced mathematical techniques are used to automatically fuse information from CIs into health indicators (HIs);
- Simplicity – HIs provide a quantified machinery health status that can be shared and understood by all – a health index of a machine between 0 and 0.8 corresponds to a healthy machine; in between 0.8 and 1.0 maintenance actions shall be planned; above one maintenance is needed;
- Autonomy and reliability – Maintenance alerts and notifications are automatically set by the system. We use the HIs to enable alarm threshold setting strategies based upon combinations of probability density functions, thereby controlling the probability of false alarms and minimising missed alarms. Statistically, Foresight will report a false alarm one time out of a million;
- Actionability – Our remaining useful life (RUL) prognostics models target warnings at approximately 250 hours lifetime and have confidence intervals associated with them. A text message or an email is automatically sent to the operator when a machinery component is out of tolerances; and
- Integration capabilities – The team behind the technology has developed the hardware and software. We have the complete knowledge and can integrate it into various IT systems.
Foresight: a step ahead
At the end of the day, Foresight is especially well-matched for the marine business case. Indeed, as the system automatically diagnoses and predicts equipment status, and subsequently returns HI data from ship to shore, the communication, a few kilobytes of data stream, is therefore seamless; CI and HI data can be transferred from the Inmarsat-4 satellites with a symmetrical bandwidth of up to 432kbps or a Vsat system. This data transfer – which was once the bottleneck, expensive and inadequate for an onshore operation centre that analyses large volumes of machine data – is not an issue anymore.
The CIs and HIs are used to indicate the status of a monitored component and its trending towards a maintenance event. The analyst reviewing the signals can immediately see when the component should receive attention and if any imminent or planned maintenance is necessary. This trending reflects the system’s ability to predict (with high levels of accuracy) the RUL of the component and allows operational personnel the luxury of advance notification for asset location and scheduled servicing.
The Foresight approach is a step ahead of the current shipping machine monitoring norm approved by class. Shipping standards DNVGL-CG-0052 (2015) and DNVGL-RU-SHIP (2017) related to survey arrangement for machinery monitoring can only approve the process of humans generating diagnostics reports and do not cover autonomous maintenance decision support.
References
- ‘Beyond condition monitoring in the maritime industry’. DNV GL strategic research & innovation position paper. 6-2014
- Measuring Intrinsic Quality of Human Decisions, ‘The evidence that machines are better decision-makers is compelling’. Algorithmic Decision Theory, 567-57, ADT 2015
- ‘Human bias is everywhere’ – Judgment under Uncertainty: Heuristics and Biases Amos Tversky; Daniel Kahneman. Science, New Series, Vol. 185, No. 4157. (27 Sep 1974), 1124-11313
Thomas JJ Meyer
CEO
Machine Prognostics AS
+47 46442695
tjjm@machineprognostics.no
www.machineprognostics.no
