Predicting Asset Failure Using Similarity Based Modeling -Part II
The advantages of using SBM are as follows:
1) Physically linked signal behaviors can be modeled together
2) No regression or other parametric analysis needs to be done.
3) The parameters all move relatively together in identifiable patterns of behavior which are recognizable using SBM and are superior to an engineer using monitoring disciplines and techniques.
4) Simple design, robust estimate of each signals behavior and easy interpretation of results.
5) Provides sufficient time for diagnosis and recommendation for the problem in hand and the maintenance activity can be done well in time to avoid shutdown.
The Similarity Based Model can be better understood with the help of an example. In steam turbine, sensors are placed to monitor the oil temperature, bearing temperature and vibrations. The empirical thresholds in normal operating conditions have been defined by the OEMs for the above operating parameters. The data from all the sensors are obtained in real time and compared with the historical data to identify a pattern in the behavior of the signals from the sensors. The pattern thus obtained is then used for comparison with the actual real time value. The software applies a predictive analytics model to the sensor data and returns the information on the behavior of individual sensors. In a case where under normal operating conditions the differential bearing temperature is above the empirical threshold defined by the OEMs, then SBM model in the software highlights only the signal deviating from the normal expected pattern. This is the Detection stage in the incident life-cycle.
Fig.1 The top graph shows the differential bearing temperature in both normal operating condition and in alert condition while the bottom graph shows how the SBM generates the alert signal at an initial error stage.
As soon as the signal highlights an abnormal behavior in bearing temperature, the plant engineer can focus all his attention to identify the cause of the problem and make recommendations to rectify the problem before it causes the equipment failure. In this case of steam turbine, the temperature difference occurred between the two surfaces and the OEM recommendation has been exceeded. A shutdown of the turbine would have been necessitated had this small anomaly gone unnoticed.
Fig. 2 Incident Life Cycle of an asset
The Diagnosis stage of the incident life cycle focuses on the causal analysis and the Communication stage focuses on the recommendations of the plant engineer to the maintenance personnel. The Action stage in the incident life cycle focusses on implementing the recommendations of the plant engineer and ensures the breakdown is averted by proper maintenance. The maintenance personnel will take actions to mitigate the temperature difference. The early detection using the Similarity Based Modeling allows maintenance department and plant engineers the time to devise and enact a mitigation plan.
How is Similarity Based Modeling using Predictive Analytics is a Smart Solution?
Predictive Analytics Solution helps asset intensive industries achieve and exceed asset performance, improve reliability and availability by providing prior warnings and notifications on upcoming failures, resulting in ample time for planning, scheduling and executing maintenance jobs. Overall SBM optimizes maintenance activities and reduce maintenance cost expenditures helping organizations in gaining competitive advantage .Organizations will be able to focus on proactive maintenance rather on reactive maintenance. Plant engineers need to ensure that the critical equipment maintain the required levels of reliability, availability and performance. This requires frequent and accurate assessment of the equipment operating conditions to minimize the operational risks of unscheduled interruptions. Effectively analyzing the realms of data collected gives an idea not only about the overall health of the plant but also about production system elements like turbine, compressor etc. The data comes from periodic and real time systems. Functional failure prone elements in critical equipment's can be identified using periodic methods, but it can be time consuming and intermittent. Model based solutions are being increasingly used to obtain real time understanding of the equipment health. Periodic inspection methods will be much more efficient if the actionable intelligence created from large amount of data helps in detecting the problems automatically and provide a basis for diagnosis and prioritizing the problems.
Fig 3. Few Benefits of Predictive Analytics Based Asset Management
The foundation of next-generation asset management will be the ability to see, understand and respond to not just what is happening now in the asset base, but also to what will happen in the future. Similarity Based Modeling using Predictive analytics will provide utilities with the strong foundation and investment confidence they need to operate and excel with minimized risk. The imperative of using the wealth of asset data available in utilities' back offices and operations, along with expanding the future use and availability of real-time asset intelligence, will provide a means to planning and investing in a secure and dependable infrastructure while promoting efficient and effective operations.