Condition Monitoring Project

Transcription

1 Condition Monitoring Project Final Report Supported by: The Danish Maritime Fund Period: 1 st December th November 2015 Partners: C.C.JENSEN A/S, Maersk Supply Service, SDU, SIMAC

2 Content Condition Monitoring Project ABOUT THE CONDITION MONITORING PROJECT... 3 Vision... 3 Goals... 3 Milestones... 4 Danish Maritime Fund... 4 PROJECT GROUP MEMBERS... 5 TASK OVERVIEW... 7 Phase 2 Development of models and reporting procedures... 7 SUMMARY... 8 TASK 2.1 BUSINESS PLAN FOR PRODUCTS AFTER PHASE TASK 2.2 SCREENING OF POSSIBLE CORRELATIONS E.G. WITH MULTI VARIABLE ANALYSIS TASK 2.3 MODEL FOR TREATING MEASURED DATA COHERENCE WITH OPERATIONAL EXPERIENCES FROM E.G. OVERHAUL REPORTS AND OTHER COLLECTED DATA FROM VESSEL OWNER TASK 2.4 ASSESSMENT OF CONCLUSIONS TOGETHER WITH VESSEL OWNER TASK 2.5 MODEL FOR CONDITION BASED MONITORING OF LUBE OIL QUALITY TASK 2.7 DEVELOPMENT OF MODEL AND CRITERIA OF NORMAL CONDITION TASK 2.8 DEVELOPMENT OF MODEL AND CRITERIA OF ABNORMAL CONDITION TASK 2.9 DEVELOPMENT OF MODEL AND CRITERIA OF CRITICAL CONDITION TASK 2.10 DEVELOPMENT OF MODEL TO CALCULATE EXPECTED REMAINING USEFUL LIFETIME OF COMPONENTS TASK 2.11 DEVELOPMENT OF MODEL TO CALCULATE EXPECTED REMAINING USEFUL LIFETIME OF LUBE OIL TASK 2.12 PRODUCT MATURATION OF PRODUCTS AFTER PHASE TASK 2.16 CREW EDUCATION Page 2 of 21

3 About the Condition Monitoring Project This project was initiated together with Maersk Supply Service autumn 2011 with the overall goal to develop an innovative tool to use as basis for decisions of those who have the responsibility of vessel operation. The tool should provide the possibility to extend and predict docking intervals, overhaul intervals and increase the time between failures of equipment. Products developed in this project would be used as tools to collect, treat and compare measured data from equipment and operational data from vessels. Vision To offer vessel owners and crew a safe and sound tool to condition based maintenance of vital oil systems onboard vessels. Goals Phase 1 Phase 2 To develop equipment to conditioning of the oil before particle sensor, so water droplets are not counted as particles To get enough measurements to establish a baseline for normal operation To examine and determine which data from the operation of the vessel is relevant to combine with the measurements To collect practical operation experience and the relevance to vessel owner and classification society To produce an oil-wetted specification list in order to relate particle content to a single component, as for example a bearing in the equipment To develop a system to collect and forward data from vessel to database system To develop a database structure to handle the data Development of mathematical models for screening of coherence for example by multi variable analysis Development of model to link practical operational experience with measurements, for example an overhaul rapport Development of model for normal and abnormal operation Development of advice and recommended actions by detection of abnormal operation Development of reporting system, where advanced special knowledge and data treatment is translated to a foundation for decisions for a vessel responsible without specific knowledge about tribology and data treatment Page 3 of 21

4 Milestones Phase 1 A product that: Maintains oil clean and dry Measures particles in wet oils Collects data locally on vessel Measures the contamination level of the oil Collects relevant operational data from vessel o Data transfer o Ability to store data in a database Phase 2 A product that can combine: A method and process, which contains algorithms or process descriptions, which makes it possible to: o Link and treat measured data and operational data using automatic and advanced methods o Link data with practical operational data Establish valid limits for normal/abnormal operation Identify possible root causes to abnormal operation based on single components as for example bearings Danish Maritime Fund The Danish Maritime Fund provides financial support to initiatives and undertakings, which may serve to develop and promote Danish shipping- and shipbuilding industries. The Fund has financed the project with 50% of the total budget, and the support was paid continuously and in connection with the actual expenses. Since the support is a loan it has to be paid back if the project is giving profit within 5 years after the support was last paid out. Page 4 of 21

5 Project Group Members Condition Monitoring Project C.C.JENSEN A/S Carl Aage Jensen Kim Kjær Jens Fich Birgit M. Dabney Senior Management Senior Management Senior Management Project Assistant Morten Henneberg Lars N. Jensen Jesper Hoppe Ruben Hensen M. Sc. Marine Engineer Marine Engineer M. Sc. Svend Erik Lem Marine Engineer Henning Buch M. Sc. Maersk Supply Service Ivan Seistrup Peter Kragh Jacobsen Poul Visby Vaibhav Chavate Senior Management Senior Management Senior Management Senior Management Page 5 of 21

6 Vessels Condition Monitoring Project AHTS Maersk Achiever AHTS Maersk Assister AHTS Maersk Attender SDU René L. Eriksen Associate Professor Marie Lützen Associate Professor SIMAC Thorkil Hansen Associate Professor Consultants Henrik G. Mølbak Hydraulics Shamba Jumaine Tribology Page 6 of 21

7 Task overview Condition Monitoring Project Phase 2 Development of models and reporting procedures Task no. Task description Status 2.1 Business plan for products after phase 1. Closed 2.2 Screening of possible correlations e.g. with multi variable analysis. Closed 2.3 Model for treating measured data coherence with operational experiences from e.g. overhaul reports and other collected data from vessel owner. Closed 2.4 Assessment of conclusions together with vessel owner. Closed 2.5 Model for condition based monitoring of lube oil quality. Closed 2.6 Evaluation and reporting. Closed 2.7 Development of model and criteria of normal condition. Closed 2.8 Development of model and criteria of abnormal condition. Closed 2.9 Development of model and criteria of critical condition. Closed 2.10 Development of model to calculate expected remaining useful lifetime of components. Closed 2.11 Development of model to calculate expected remaining useful lifetime of lube oil. Closed 2.12 Product maturation of products after phase 2. Closed 2.13 Reference group meetings. Closed 2.14 End evaluation and reporting. Closed 2.15 Steering committee meetings. Closed 2.16 Crew education. Closed Page 7 of 21

8 Summary The goals and milestones in phase 2 have been met. Reporting system A software tool (CJC Trender Tool) has been developed which allows the user to monitor his equipment online. This software is developed with a strong focus on the needs for the crew and technical organisation, and is able to give a quick visual overview of the status of the equipment and measured values. Trender Tool will also give the possibility to plot detailed data from the sensors if the user wants to investegate further details. The important features can be highlighted as: Easy intuitive user interface Quick overview of the status of the equipment from ship level down to individual equipment Report generation Export of raw data from sensors Automatic actions in case of alarms (automated s) Platform for collecting input from crews (pictures, comments, oil analysis reports) Platform for communication between user and specialist Coupling between sensor data and root cause An important focus has also been to link measurements from the sensors to actual wear on components. Initially, the idea was to physically investigate worn components such as bearings and gears and compare the wear mechanisms to the sensor data from the operation of the equipment when in operation. During the project, it turned out to be logistically impossible, and it was therefore decided to build a tribological test bench which could actually simulate the wear mechanisms and have the sensors monitor the oil in parallel. Theoretically, the different types of wear should give different distributions of particles, but no research was available to prove this hypothesis. However, the results from the test bench proved that this theory was correct. It is very valuable to be able to distinguish between different wear mechanisms in the sensor signals. This means that for instance fatigue gives an unique fingerprint in the particle distribution, and it will thereby be possible to determine that a serious defect is under development. Advanced data treatment An important part of phase 2 has been to develop and verify statistical models that can treat big amounts of data. The current sensor packages measure approximately 70 values every 5 minutes per equipment. A method has been developed which allows these values to be transformed into one significant parameter via a statistical T2 model. This method has been verified in cooperation with MSS and is essential for creating an automated monitoring system. The methodology enables the monitoring system to give an alarm if the system deviates from normal operation. Furthermore, the methodology allows to identify which type of sensor signal that is deviating from its normal span. The statistical model is a learning model, which makes it simple to adapt to a given new system. Page 8 of 21

9 Task 2.1 Business plan for products after phase 1 Business plan for products after phase 1. Task 2.1, Business plan for products after phase 1, was removed in the revised application form for phase 2 as it was done in phase 1. Conclusions can be seen in the final report for phase 1 task Page 9 of 21

10 Task 2.2 Screening of possible correlations e.g. with multi variable analysis Screening of possible correlations e.g. with multi variable analysis. Keywords - Analysis of signals using autocorrelation function (ACF) - Analysis of signals using partial autocorrelation function (PACF) - Signal dependence and independence study using autoregressive integrating moving average (ARIMA) models - Correlations to temperature Normalised data has been investigated using ACF and PACF. Correlation dependence of 4 were observed meaning that noise correlates within 20 minutes. From the ACF and PACF were ARIMA model identified as MA(4) which were fitted to normalised data. Correlation analysis on the MA(4) fitted data showed that no significant correlation exists. Conclusion is then that data more than 20 minutes apart is mutually independent. This knowledge was used in the further data analysis to increase validity and robustness of models. Correlations to temperature were observed and taken into account in the further models using multivariate linear regression (MLR) methods. Statistical models constructed are thereby made less sensitive to different temperatures in the oil system as well as ambient. Page 10 of 21

11 Task 2.3 Model for treating measured data coherence with operational experiences from e.g. overhaul reports and other collected data from vessel owner Model for treating measured data coherence with operational experiences from e.g. overhaul reports and other collected data from vessel owner The purpose of this task was to find a way to integrate observations, oil analysis, overhaul reports, etc., into the data treatment being performed on the data from the measurements from the sensors. The situation is typically that when measured data are treated, they usually only represent a part of the whole picture. The solution to this was to implement a method into Trender Tool where it is possible to upload all kinds of information. The uploads can be commented on and will be placed on a timeline, where it can be seen when the observation was made. Examples of this kind of information, which can be uploaded are: Comments from the crew about the equipment and operation Oil laboratory reports Pictures taken on board of components This functionality will enable the user of Trender Tool to collect and evaluate information from various sources when looking into the data from the ship. Page 11 of 21

12 Task 2.4 Assessment of conclusions together with vessel owner Assessment of conclusions together with vessel owner. Keywords - Value of Trender Tool as data viewer - Value of T 2 models when implemented in Trender Tool - Crane Assesment at abnormal operation The web viewer Trender Tool is currently showing live data, filtered from invalid measurements, from the vessel. This enables crew, superintendents and technical management to receive much more reliable data of current contamination, than previously, when done manually with samples onboard the vessel. As a result, the manual sampling procedure can be decreased drastically in volume or perhaps completely. This is equivalent to one working day per week per vessel. Trender Tool in the current version can alert users of abnormal wear or operation by and users can then access the viewer to assess the data and perform analysis of the reason. A further value is, that the alarm can also trigger manual sampling for onboard and laboratory analysis of the oil, that will show more precisely evidence of the root cause of abnormal operation than samples taken by regular intervals. The tool enables users to interact with each other and upload everything from pictures to vibration monitoring reports to the system. The T 2 models have been programmed for use in Trender Tool, but not yet implemented. The model itself has, however, been presented and validated by Maersk Supply Service. The value of the T 2 model for the owner is that it treats all data into one value, which can detect abnormal operation more precisely than single measuring points. The AHC crane is representing equipment onboard the vessel which complexity makes it very difficult to react on abnormal operation seen in Trender Tool without a proven method. Together with Maersk Supply Service we have developed a method to narrow down the root cause component in a complicated hydraulic system. The method consists of a procedure of sampling, adjusting sampling points, a method for verifying data and analysis equipment. Based on the assessment of Trender Tool with Maersk Supply Service, further 2 tools have been developed and implemented. A tool to export data from Trender Tool into other systems used by the user and a reporting tool, which can automatically generate report of data from a given period and actual status in a form to be presented to management or for example classification societies. Page 12 of 21

13 Task 2.5 Model for condition based monitoring of lube oil quality Model for condition based monitoring of lube oil quality Keywords - T 2 control model including sensor-sensor correlations - Upper control limit (UCL) - Warning structure - MLR of temperature included in the T 2 models A model for oil condition has been investigated using Hotelling T 2 statistics. The model has proven robust and valid in monitoring different oil types and can from the sensor outputs estimate the lube oil s quality. From the phase I period (normal condition period) an upper control limit (UCL) has been estimated. A regression approach on a time window of 7 days has been proposed in order to construct a warning structure. The regression is based on orthogonal polynomial fitting. The warning structure is based on estimating UCL and two parameters from the polynomial fitting. MLR has been implemented in the T 2 models to increase robustness as the ambient conditions may change around the ships. Page 13 of 21

14 Task 2.7 Development of model and criteria of normal condition Development of model and criteria of normal condition Keywords - T 2 control model including sensor-sensor correlations - Distribution of data - Upper control limit (UCL) - MLR of temperature included in the T 2 models - Estimation of normal wear modes using test rig A model for normal operating condition has been investigated using Hotelling T 2 statistics. The model has proven robust and valid in monitoring different oil types, and can from the sensor outputs estimate the lube oil s quality. Distribution of data has shown that a learning period of 1 to 12 month is optimal. The learning period can be updated recursively, meaning that the system and model can be operational after a short period of time (1 week). Estimation of UCL is based on statistical analysis but further implementations can take run observations into account whereby warning differentiation is more distinct. MLR has been implemented in the T 2 models to increase robustness as the ambient conditions may change around the ships. The MLR on temperature is less needed/significant for the wear debris estimation than for the oil condition estimation. It has, however, proven to improve the model. Test rig Hephaestus has been successfully used to emulate wear debris from mild abrasion and mild fatigue. The wear debris size and distribution has been mapped out so references to on-board sensors are available for normal conditions. Particle distribution has been analysed for normal (abrasion) wear condition. Page 14 of 21

15 Task 2.8 Development of model and criteria of abnormal condition Development of model and criteria of abnormal condition Keywords - T 2 control model including sensor-sensor correlations - Upper control limit (UCL) - MLR of temperature included in the T 2 models - Warning model - Estimation of normal/abnormal wear modes using test rig A model for condition monitoring has been investigated using Hotelling T 2 statistics. The model has proven robust and valid in monitoring different oil types and can from the sensor outputs estimate the lube oil s quality. From the phase I period (normal condition period) an upper control limit (UCL) has been estimated. A regression approach on a time window of 7 days has been proposed in order to construct a warning structure. The regression is based on orthogonal polynomial fitting. The warning structure is based on estimating UCL, and two parameters from the polynomial fitting. MLR has been implemented in the T 2 models to increase robustness as the ambient conditions may change around the ships. Combining MLR, the T 2 model and orthogonal polynomial regression has made clear distinction of normal and abnormal conditions for the equipment monitored possible. Test rig Hephaestus has been successfully used to emulate wear debris from mild to severe abrasion, adhesion and mild fatigue. The wear debris size and distribution has been mapped out so references to on-board sensors are available for normal and abnormal conditions. This could enable interpretation of wear mode and thereby most likely damaged part. Page 15 of 21

16 Task 2.9 Development of model and criteria of critical condition Development of model and criteria of critical condition Keywords - T 2 control model including sensor-sensor correlations - Upper control limit (UCL) - MLR of temperature included in the T 2 models - Warning model - Radar plots - Estimation of critical wear modes using test rig A model for condition monitoring has been investigated using Hotelling T 2 statistics. The model has proven robust and valid in monitoring different oil types and can from the sensor outputs estimate the lube oil s quality. From the phase I period (normal condition period) an upper control limit (UCL) has been estimated. A regression approach on a time window of 7 days has been proposed in order to construct a warning structure. The regression is based on orthogonal polynomial fitting. The warning structure is based on estimating UCL, and two parameters from the polynomial fitting. MLR has been implemented in the T 2 models to increase robustness as the ambient conditions may change around the ships. Combining MLR, the T 2 model and orthogonal polynomial regression has made distinction of critical condition for the equipment monitored possible. Test rig Hephaestus has been successfully used to emulate wear debris from severe abrasion, adhesion and fatigue. The wear debris size and distribution has been mapped out so references to on-board sensors are available if critical condition occurs. Page 16 of 21

17 Task 2.10 Development of model to calculate expected remaining useful lifetime of components Development of model to calculate expected remaining useful lifetime (RUL) of components. Keywords - Expected life time of wear modules from test rig - Equipment condition estimation from statistical models - Statistical model for normal and out of normal operation Statistical models for condition monitoring have been developed and logging of operation load and load hours has been implemented. Test rig Hephaestus has conducted 4 tests of each major wear mode (16 tests in total) with different configurations. This can enables some interpretation between accelerated wear (abnormal/critical condition and remaining useful lifetime). Statistical model for RUL cannot be performed due to lack of data. No breakdown has been observed so no indications of abnormal/critical wear has been identified. With no empirical data can no RUL model can be constructed and validated. From the test rig the following major results in relation to RUL can be concluded: - Abnormal/critical wear will initially (often) transform back to normal wear if it is adhesion (will typically move back to abrasion/severe abrasion) - Fatigue will accelerate exponentially. - Both can be somewhat controlled if detected in due time by reduction of load The statistical platform for calculating RUL from the developed models has been prepared. When empirical data is available, they can be tested and validated. Page 17 of 21

18 Task 2.11 Development of model to calculate expected remaining useful lifetime of lube oil Development of model to calculate expected remaining useful lifetime of lube oil. Keywords - Laboratory aged oil - Impedance sensor analysis - Oil breakdown parameters - Parameters affecting the measurement stability of oil quality Typical breakdown parameters of oil have been identified. Artificial ageing of oil has been conducted in a laboratory test where several sensor techniques have been applied. Onboard sensors have also been tested on the artificial aged oil whereby quantitatively ageing parameters (sensor output) has been identified. Hereby, a relative reference frame to oil quality has been established. Impedance analysis of different oil types, aged oil and oil at different temperatures have identified parameters affecting the oil quality measurement: Relative moisture (RH) content, temperature, wear debris and additive package. RH, temperature and wear debris have been taken into account in the oil condition model. A specific model for remaining useful lifetime is made possible. However, data from oil that is depleted of additives, heavy oxidised and worn out has not been established due to lack of data (no oil change has yet been performed onboard the ships and no sensors indicate significant oil parameter changes). Page 18 of 21

19 Task 2.12 Product maturation of products after phase 2 Product maturation of products after phase 2. Keywords - Trender Tool as is - Products and services - Trender Tool in the future Product maturation of Trender Tool has been much more complicated than expected as the product and business model require a completely different set-up with regards to especially support. An investigation of the need for support per equipment sold is seen below: 50 units 100 units 250 units 500 units 1000 units Weekly tasks Hours Hours Hours Hours Hours Connection check 1,0 1,5 2,0 3,0 4,0 Ad-hoc connection check 1,0 2,0 4,0 8,0 16,0 Ad-hoc issue solving 1,0 2,0 4,0 8,0 16,0 SIM-card maintenance 0,5 0,8 1,5 3,0 6,0 Total hours and costs weekly 3,5 6,3 11,5 22,0 42,0 Total hours and costs monthly Total hours yearly A list of products has been matured and is ready for sales, but will be launched limited and specifically to individual customers to ensure that we develop the right system behind it to support it. Products launched are: CJC Datalogger with LAN connection CJC Datalogger with GSM/GRPS connection Licence CJC Trender Tool Viewer incl. 1 year subscription Licence CJC Trender Tool Viewer incl. 1 year subscription incl. reporting tool incl. data export function Yearly subscription - CJC Trender Tool Viewer Page 19 of 21

20 Yearly subscription - CJC Trender Tool Viewer incl. reporting tool incl. data export function CJC Trender Tool support CJC Trender Tool Webinar Further products in the future include: Trender Tool including the mathematical treatment, retrofit solutions for existing filter units, as well as complete new filter solutions including sensor packages and Trender Tool. Page 20 of 21

21 Task 2.16 Crew education Crew education Keywords - Weekly reports - Daily correspondence via telephone, mail and Trender Tool - Education In phase 1 we developed weekly reports to crew and superintendents to interact with them on all issues on a weekly basis and report findings until the Trender Tool was available. This has been continued until everybody had access to Trender Tool and then discontinued. Trender Tool has proven a great tool to secure that the basis of every discussion is the same for both crew, superintendents and C.C.JENSEN whenever discussing issues seen in the data or from other incidents seen on the vessel. It has also proven a great tool to communicate with and upload data, pictures, report laboratory oil analysis, so external data is stored and can be shared for root cause analysis. Several separate seminars have been held for crew members to ensure that they are comfortable with Trender Tool and trained to use it. Also, separate seminars have been conducted for all superintendents in the MSS HQ in Lyngby. Page 21 of 21