RESEARCH INITIATIVES FOR IMPROVING THE SAFETY OF
OFFSHORE HELICOPTER OPERATIONS[*]
David Andrew Howson ()
Research Project Manager
UK Civil Aviation Authority, London, UK
16
Abstract
Since the late 1980’s, the UK Civil Aviation Authority (CAA) has been leading a programme of research aimed at improving the safety of offshore helicopter operations. The motivation for this initiative came from a major joint CAA/Industry review of helicopter airworthiness, commissioned in 1982. This study led to a number of research projects and other reviews which, in turn, led to further research projects. A total of over 20 projects have been undertaken covering airworthiness and operational issues, and covering helicopters and helidecks. This programme of work has been jointly funded and monitored by the UK CAA-run Helicopter Safety Research Management Committee (HSRMC). This paper provides a top-level summary of current activities on the seven main ‘live’ research projects.
16
Introduction
In the 1970’s and early 80’s the disappointing safety record of helicopters transporting people to work on oil rigs in the North Sea led to the formation of the Helicopter Airworthiness Review Panel (HARP). This group reported its findings in the HARP Report (CAP 491) [1] in 1984, which contained recommendations for research into helicopter health and usage monitoring, crashworthiness and ditching. The HARP Report also called for an investigation of human factors-related accidents which led to the formation of the Helicopter Human Factors Working Group. This group reported its findings in CAA Paper 87007 [2] in 1987, which included recommendations for research into a further seven, mainly operational areas.
In addition to these two initiatives, a major review of offshore safety and survival was commissioned in 1993 in response to an AAIB recommendation following the fatal accident at the Cormorant A platform in 1992. This study was conducted by the Review of Helicopter Offshore Safety and Survival (RHOSS) working group, which reported its findings in CAP 641 [3] in 1995. The overall effect of this exercise on the helicopter safety research programme was to add impetus to the crashworthiness (water impact) and ditching projects.
The resulting programme of helicopter safety research has been funded and monitored by the UK CAA-run Helicopter Safety Research Management Committee (HSRMC). The HSRMC was originally set up by UK CAA in the late 1980's to manage a joint UK CAA/UK Government/UK oil industry (UKOOA) research fund that was created to progress the recommendations of [1] following the loss of 45 lives in the Chinook accident in 1986. The committee is still thriving and has evolved over time expanding its membership to include the UK MoD, the UK helicopter operators (BHAB), the new European Aviation Safety Agency (EASA), the Norwegian CAA, the Norwegian oil industry (OLF), and the European Helicopter Association (EHA).
To date, the committee has overseen over £8M of research funding spread over a wide range of helicopter safety issues, the majority of which have their ultimate origins in the HARP Report. The remainder of this paper provides a top-level summary of current activities on the following seven main ‘live’ projects overseen by the HSRMC:
· Helicopter Health & Usage Monitoring Systems (HUMS) - Advanced Analysis Techniques.
· Helicopter Emergency Flotation.
· Helideck Environmental Issues - Turbulence Criterion.
· Operations to Moving Helidecks.
· Helideck Lighting.
· Helicopter Flight Data Monitoring - Extension to Low Airspeed Regime.
· Use of GPS for Offshore Helicopter Operations - Low Visibility Approaches.
Helicopter Health & Usage Monitoring Systems (HUMS) - Advanced Analysis Techniques
Background
The first vibration health monitoring (VHM) systems (known as ‘HUMS’) were installed on the North Sea helicopter fleet in the early 1990s which, the CAA believes, contributed significantly to a reduction in the airworthiness-related accident rate. Although in-service experience continues to demonstrate significant safety benefits, it is generally acknowledged that there is room for improvement in the diagnostic performance of HUMS. One study has suggested that for every four ‘successes’ there is one where a propagating defect was subsequently judged to be ‘evident’ but no warning was given as no warning threshold was exceeded.
In addition to the in-service experience, the ongoing review of the results of the two HSRMC funded helicopter main rotor gearbox (MRGB) seeded defect test programmes has indicated that there is scope for improving the effectiveness of HUMS data analysis. The main issues identified as requiring attention are:
· improvement of warning time (i.e. the time between warning and component failure) - when conducting retrospective analyses, the presence of defects is nearly always apparent to analysts in the data in advance of any indicator thresholds having being exceeded, and hence any warnings being generated. It should be borne in mind, however, that if a warning and the associated indicator histories are not judged conclusive, it is common practice to fly-on while ‘close monitoring’ for a defect. Hence improvement in warning time must not be at the expense of the warning’s ‘quality’ for maintenance decision making.
· detection of build defects - many warning thresholds are tailored on installation of the component/assembly using a simple ‘learning’ process. This improves sensitivity without increasing the false alarm rate. However, in the event of a build anomaly or defect these thresholds are set too high, effectively de-sensitising the analysis to the subsequent propagation of defects. Additionally, during the ‘learning’ period the threshold will, at best, be at a higher fleet average based level, further reducing the protection against defects introduced by build/maintenance errors. Hence a system that can provide increased sensitivity without increasing the false alarm rate and without requiring a ‘learning’ period after each maintenance action would represent a significant improvement.
· accommodation of unexpected gear indicator reactions - the identification of defects in a timely manner can be compromised by the rigid application preconceived ideas on how defects will manifest themselves in the vibration data. Experience has demonstrated that a wide range of reactions is possible, both in terms of which indicators react and how they respond. A more robust and capable analysis technique is therefore required if effectiveness is to be improved.
· accommodation of reducing gear indicator trends - certain types of defect can manifest themselves as reducing indicator trends. A technique is required that can detect these.
Experience has shown that the above issues can be mitigated through the use of well-trained and experienced human analysts. In the in-service environment, however, it is impractical for human analysts to examine all data in sufficient depth due to the large quantities generated on a daily basis. Hence a crucial factor in improving the effectiveness of HUMS is the establishment of a more sophisticated means of identifying the sections of data of interest. If this can be achieved it will result in a reduction of the quantity of data requiring detailed analysis, enabling human analysts to focus their efforts where their skills are still essential.
Earlier Research
As part of CAA’s helicopter MRGB seeded defect test programme, a number of alternative health monitoring techniques were evaluated. These included alternative sensors (e.g. stresswave sensors, acoustic sensors), and a number of alternative analysis techniques. In general, these alternatives were not able to demonstrate any significant improvement over the VHM techniques in-service at the time they were evaluated.
One notable exception was the programme of work aimed at demonstrating the benefits of supervised and unsupervised machine learning techniques. The scope of the project included the blind analysis of data from two of the S61 MRGB seeded defect tests. The work was completed in 1998 and the final report was published as CAA Paper 99006 [4].
The analysis of the seeded defect test data is covered in Study II of [4], and it is this section of the work that is of direct relevance to the current research. Although success was achieved with supervised machine learning, the absence of large numbers of examples of all possible failure conditions is expected to limit this technique to a retrospective response to in-service incidents. Conversely, given the vast quantities of data available to characterise serviceable components and/or systems and in view of the results of the blind analysis, unsupervised machine learning is considered to have great potential as a proactive tool.
This work was presented to both the CAA/Industry Helicopter Health Monitoring Advisory Group (HHMAG) in 1999 and to a Royal Aeronautical Society conference in March 2000, and CAA encouraged industry to exploit the results and develop the techniques for use in-service on civil helicopters.
Current Research
No further development of advanced HUMS data analysis techniques had taken place by 2003, however, when a significant HUMS missed detection involving a Super Puma bevel pinion occurred. The seriousness of this incident persuaded CAA to take the lead in developing the technology and, following a competitive tendering process, Smiths Aerospace Electronic Systems - Southampton, UK, were commissioned to conduct the research required. The overall objective of the project is to demonstrably improve the effectiveness of HUMS through the enhancement of VHM data analysis, and comprises the following tasks:
· a review of existing literature judged to be relevant to the project, CAA Paper 99006 in particular;
· development of the advanced HUMS data analysis techniques based on historical data;
· the design and production of analysis software;
· an off-line demonstration of the system and any consequent refinement;
· an in-service demonstration of the system.
The in-service demonstration is to be performed by Bristow Helicopters Ltd (BHL) at their Aberdeen, UK, base. The system will be installed/implemented in parallel with the existing HUMS ground station such that incoming HUMS data is analysed concurrently. This will be accomplished in a manner that does not affect the integrity of the existing ground station or analysis, i.e. all warnings provided by the existing analysis will be acted upon as usual, regardless of the output of the improved analysis.
Differences between the reactions of the existing and new analyses will be noted and compared. Feedback from any inspections or other maintenance actions or strip reports will be collected and catalogued with the associated results from both the new and existing analysis techniques.
An important aspect of the in-service trial is the evaluation of the system in terms of ease of use and workload. BHL are tasked with reporting on this aspect of the work from a user’s perspective.
Progress on Current Research
As at end June 2005 the literature survey had been completed and a large number of references on anomaly detection identified. Although many focussed on ‘intrusion detection’ on computer networks, most had general applicability. A few papers were related to aircraft health monitoring, the most relevant being related to novelty detection in jet engines. The survey confirmed that the data mining tool proposed for the research has the algorithms necessary to evaluate the most promising ideas cited in the literature.
The extensive archive of BHL HUMS data downloads has been decoded, ‘cleaned’ and catalogued in a SQL Server database in preparation for system development and testing. This was a significant but necessary task and included data for Super Puma main rotor, intermediate, tail and left and right accessory gearboxes.
Analysis of this data so far has shown it to be ‘noisy’ and, as expected, revealed that gearboxes tend to exhibit individual behaviour. Another challenge has been the inability to identify a ‘healthy’ data set for the development of anomaly detection models. All data has been found to contain some anomalies and, because of the lack of feedback from overhauls, the status of all data not containing documented faults can only be classed as ‘unknown’. A further challenge is the fact that the data contains many step changes, which are assumed to be due to unidentified maintenance actions on other parts of the helicopter. This has resulted in the work becoming more of a research effort than was originally envisaged. Furthermore, whilst data normalization techniques should be utilized wherever possible, for the VHM data pre-processing options are limited to filtering, and relatively simple techniques for characterising indicator trends. This places more emphasis on modelling and the analysis of model information. However, as a result of an intensive data analysis effort, there is now a high confidence that cluster modelling can reveal anomalous behaviour and can be used to characterise the significance of anomalies. The key on-going task is to determine the optimum structure for these models and define the best anomaly metrics.
Figure 1 - Cluster plot of 2 VHM indicators from AS332L LH Accessory Gearbox.
By way of an illustration, the plot in Figure 1 presents data trend information for two HUMS parameters in a multi-parameter cluster model. The light areas show outlying regions in the parameter space being modelled by a particular ‘anomaly cluster’. Any data trends moving into these regions would be classed as anomalous.
Subject to a satisfactory off-line demonstration of the system, the in-service trial is scheduled to start late 2005 and will last for six months. The project includes an option to extend the trial for a further six months should this be judged necessary.
Helicopter Emergency Flotation
Forced landing on the water (‘ditching’)
For extended over water flights (being in the UK beyond autorotation distance from land for a single engine helicopter, and more than 10 minutes flying time from a suitable forced landing site for a multi-engine helicopter), emergency flotation systems (EFS) have been mandated on UK offshore helicopters since the 1970s. However, it is difficult or impossible to design practical flotation systems that will keep a helicopter afloat and stable in the severest weather conditions. In [5] it was shown that, on average in the North Sea, a helicopter making a controlled landing on the water, and fitted with an emergency flotation system compliant with the guidance [6], might expect to be capsized by the waves in about 30% of cases.
Research into the design of EFS was undertaken in an attempt to improve the odds. Model tests conducted by British Hovercraft Corporation in the mid 1980s had investigated raising the float attachment positions in order to float the helicopter lower into the water, and the addition of water scoops to the emergency flotation bags (as routinely used on inflatable liferafts in order to improve stability). The former was found to give variable and inconclusive results, depending primarily on helicopter type and loading condition. The latter was seen to provide a uniform benefit, however, increasing the helicopter capsize threshold by about one sea-state. The benefits of float scoops, and their relatively low cost were described in [7].