Philip McCusker, Air Products and Chemicals, Inc, UK, explains how advanced data analytics has helped gas processing plants improve reliability and maximise operating efficiency.
In the gas processing sector, reliability is king, but process interruptions are hard to predict. The problem is that early warning of equipment failures is difficult to achieve with conventional process alarm management techniques, as the alarm settings are fixed and need to be suitable for all operating conditions. As such, alarms are set broadly so when they are eventually triggered, the equipment is already on the way to failure.
Figure 1. The winding temperatures of a fan’s motors jump up to an out-of-bounds condition.
Whether it is cryogenic processing or natural gas reforming, all gas processing plants can have equipment issues that cause costly downtime and expensive maintenance. Although alarms are set to notify operating teams of any problems as single measurement data points, they are often too wide in their measurement setting to provide meaningful early warning of a condition to avoid an impending equipment failure or process interruption. As such, problems can go completely undetected until it is too late, causing unplanned interruptions, which are costly to both plant operators and their customers, and require maintenance intervention to remedy the equipment deterioration.
What is changing?
In order to provide earlier warning of an abnormal condition, one needs to use more functionality than is offered by single variable alarm points. By using multi-variable data points fed into a condition algorithm, the abnormal or alarming points can be ‘conditioned’ to reflect the natural variability that is associated with the plant’s specific operating condition. Operating conditions can change on most plants due to factors such as operating mode, loading, and ambient extremes, which all serve to mask the start of an abnormal operating condition. Think of it like this: the minute detail of a snooker table, the smoothness of the cue ball, the speed and angle of the shot, the roughness of the felt, the room temperature, and the air flow are variables that all affect the outcome of the game. It is only by measuring these interdependent relationships that one can theoretically predict gameplay with accuracy. The same can be said of preventing interruptions to gas processing – and it is on this principle that Air Products’ Plant Performance Advisor service is based.
Figure 2. Excessively dirty air filters cause a lack of air cooling flow.
How does it work?
Huge datasets are collected from sensors on all manner of different equipment in use across fleets of plants including compressors, pumps, motors, expansion turbines, blowers, cooling towers and heat exchangers. The sensors measure vibrations, bearing temperatures, winding temperatures, pressures, pressure ratios, temperatures, approach temperatures, flows, performance efficiencies, and every other conceivable monitoring point in the key operating equipment.
Through careful data analytics, these enormous datasets are then transformed into composite variables that take into account the backdrop of the current operating conditions, creating a ‘fingerprint’ of the equipment’s operation. Rather than equipment being monitored by broad fixed alarm points, it has an ongoing ‘health check’ through condition multi-variable algorithms that can identify equipment deterioration before conventional process monitoring sensors can.
Figure 3. The filters are replaced causing motor winding and bearing temperatures to improve.
As such, the software will hone in on any deviation from the expected operating conditions.
The early warnings of a failure condition that can be identified through such a system are complex and varied, ranging from slightly raised lubrication oil temperatures for a nitrogen recycle compressor, through to mildly increased vibration of a high pressure boiler feed water pump. Often these issues are subtle and would not be picked up by simple human observation, and occur well before the equipment reaches its alarm point. What they have in common, however, is their ability – if left undetected – to negatively impact plant processing efficiency, costs, downtime, and, crucially, customer experience.
One example of this is the slight increase in valve temperatures of a multi-service reciprocating H2 compressor that occurred on a particular plant just before a planned outage at a sister plant feeding the same pipeline. The Plant Performance Advisor software was able to detect this deviation and predict valve failure. This, in turn, allowed reliability engineers to control and manage the situation, pro-actively shutting down the compressor and replacing the valves ahead of the sister plant’s planned outage. This avoided machine damage that could have cost in excess of US$100 000, as well as significant supply chain costs.
The ability to detect issues and predict and ‘manage out’ problems before they occur is invaluable. At another plant, for example, an axial position alert was received directly after a newly-installed pump was started up for the first time. Investigations revealed that the operating status of the pump was stable and that, in fact, this was an issue with the set-up of the axial position probe rather than the pump itself. Left undetected and unaddressed, this would have left a pump unprotected, which again, if it failed, would result in thousands of dollars worth of equipment damage and process interruptions. As it was, the engineers were able to correct the set-up of the probe, ensuring full protection was in place.
Figure 4. The vibration of a high pressure boiler feed water pump increases rapidly.
Examples like this underline the critical importance of understanding how the wealth of operating data available to a modern process plant can be used to improve the intelligence of equipment condition and facilitate an intervention at a more appropriate time. Real-time monitoring, combined with a pro-active detection of abnormalities, is key to ensuring processing efficiency is at its optimum.
In all of this, security of data remains paramount. Air Products' data processing is stored in a secure digital cloud with high dimensional data matrices, low signal-to-noise ratio and a high degree of correlation, all serving to give an accurate representation of overall asset performance.
The performance fingerprint is monitored by reliability engineers who receive alerts as well as proactively looking for abnormal operating conditions. This allows them to rule out likely causes and escalate alerts so that any issues can be resolved at a time that suits the customer, and before potential failures occur. Any action taken is then analysed for its effectiveness in comparison with previous operating conditions.
This process, which can be summarised in four key stages – monitor, detect, alert, diagnose – allows plant management to predict issues far sooner than they would normally be able to and, therefore, plan maintenance at a convenient time to coincide with pre-planned outages.
A pro-active approach
As well as helping to reduce process interruptions, the software measures the efficiency of each sub-system of the plant and estimates the conditions that would denote maximum efficiency. This is then used to provide monthly reports that contain recommended actions for cost and efficiency savings, ultimately optimising plant performance. This, in turn, gives the plant operator the confidence to extend the intervals between regular maintenance servicing, meaning less gross downtime and reduced overall costs.
The ability to compare a broad dataset is critical in assessing the performance of entire plants. At Air Products, for example, more than 400 industrial gas facilities worldwide are monitored and maintained. This generates hundreds of thousands of data points for analysis in proprietary algorithms and against a set of key performance indicators, allowing plant performances to be benchmarked and each specific site to run at the most cost effective level.
Figure 5. Due to a swift response, the repair is limited to the bearings, the balance disc and throttling sleeve.
Example: a fan motor
The software's dashboard indicates that the winding temperatures of a fan’s motors have jumped up to an out-of-bounds condition. Reliability engineers receive an alert and notify the site maintenance team to perform an online inspection (Figure 1).
The inspection reveals excessively dirty air filters that are causing a lack of air cooling flow (Figure 2). The filters are replaced, improving the motor winding and bearing temperatures, and avoiding overheating of the motor components, long-term abnormal behaviour and any collateral damage (Figure 3).
Example: high pressure boiler feed water pump
The vibration of a high pressure boiler feed water pump starts to increase rapidly, triggering an alert with the reliability engineers who promptly escalate the issue to the machinery engineer (Figure 4). The plant shuts the pump down and swaps operations to a secondary pump. A spare pump is installed while the original pump is sent to the workshop for an expedited rebuild.
Due to the swift response from the machinery engineer, plant operations and maintenance, the repair is limited to just replacing the bearings, the balance disc and throttling sleeve, thus avoiding impeller, bearing housing and casing damage, which could have cost over US$200 000 in extra maintenance expenses (Figure 5).
Figure 6. The lubrication oil temperature of a nitrogen recycle compressor, before and after the replacement of a faulty temperature control valve.
Example: nitrogen recycle compressor
The lubrication oil temperature of a nitrogen recycle compressor was on an upwards trend for several months. This caused the software to generate an alert – high lubrication oil temperature dramatically increases the rate of oil degradation, while negatively impacting vibrations and gear integrity.
The reliability engineers surveyed the lubrication oil temperature, identified a faulty temperature control valve and added the replacement of the valve to the plant’s planned maintenance shutdown. Left undetected, failure of this valve could have resulted in extended plant downtime, and thousands of dollars worth of damage (Figure 6).
When it comes to gas processing, the ‘gift of foresight’ is front and centre. As the sector grows and develops, big data and, more importantly, intelligent data analysis and processing will be key to preventing unplanned interruptions and reducing costly equipment damage.
This article was first published in Hydrocarbon Engineering. To receive your free copy, click here.
Read the article online at: https://www.hydrocarbonengineering.com/special-reports/10072017/reliability-is-king/