January 2014, Vol. 241 No. 1
Web Exclusive
Enhanced SCADA Alarm Management Using Machine Learning
The SCADA system is the first line of defense for protection and oversight of any pipeline. Monitoring the various alarms and alerts in a pipeline operation’s SCADA system, however, can be like trying to drink from a fire hose. Thousands of alarm points throughout a system, all reporting 24/7, require a significant investment in alarm management, which has become the next area for major developments in SCADA design.
Outgrowing standard control room concept
All SCADA systems are rule-based. When liquids in a tank reach a preset point, an alarm notification is triggered in the Human-Machine Interface (HMI) within the control room. All alarm levels – high, high-high, low, low-low – must be determined during set up and configuration of the installation system, along with desired actions (notification, control action or shut-in). This process is very time- consuming and never stops. Design conditions constantly change, as data points are added and alarms are adjusted, set and re-set.
That aggregated information is typically fed through trending tools to compare events to what occurred previously. This analysis provides the basis for making control decisions and dispatching personnel. All this information must be configured based on trending and reaction to alarm notification. Greater accuracy and greater scale within the system means more data points, more configuration and more monitoring are required in order to define the next step – as well as more alarms that must be investigated.
As a result, SCADA systems currently only tell the control room what has just occurred, rather than the conditions leading up to those events. An advance in alarm management can now close this gap. Rather than operating solely on events that have already transpired, this new approach moves decision-making closer to real-time by giving the control room the ability to recognize unexpected behaviors within data as they evolve. Solutions based on this technology extend data collection and rule-based alarm management so that control room operators can immediately identify data patterns that are different from what have previously been observed and make operational decisions based on that improved level of situational awareness.
Machine-learning and behavior recognition
The answer to better alarm management comes through machine learning and behavioral recognition algorithms originally developed by BRS Labs for the video surveillance industry. GlobaLogix and BRS Labs have partnered to bring this technology to SCADA systems, allowing operators to know where they should direct their attention. In other words, the system itself gains the ability to identify areas within a very long pipeline which may not have triggered an alarm yet, but may be worth noting because an alarm may be coming.
The system works at a metadata level. When brought online, it immediately begins recognizing normal data patterns and outliers within unusual or abnormal data patterns. Since the system teaches itself to recognize these patterns, rules or complex custom programming are unnecessary. This enhanced level of insight gives operators a view of the pipeline that is very similar to a weather map, showing which trends are coming down the line to indicate events with the greatest nodal consequences.
The nodal aspect is significant in that this new system is quite similar to the best available leak detection technologies. It analyzes data from SCADA systems and, in turn, provides data back into those systems for displays in the SCADA control room without adding instrumentation. Instead, modifications occur in server racks and in development of additional screens that the control room can utilize to make the most sensible responses for specific operations.
Alarm notification improves through an overall system view that shows trends in abnormal behavior, which a user can investigate to identify where in the system these abnormal readings are occurring. Subsequently, these areas of concern are brought to the control room operator’s attention so that existing screens can be monitored to identify unusual events as they occur.
This heightened awareness is a major improvement over the current state of the art, for example, as used in a compressor station’s bank of compressors. When one of compressors experiences higher discharge pressure – or higher discharge temperatures – than others, it may have not reached a critical alarm point. However, unless the rule-based SCADA system is configured to alarm on multiple points of temperature, the operator would not know that its temperature was rising unless he is standing directly in front of the temperature indicator.
With this new tool providing a broader view of the entire system, the operator can see trends in abnormal behavior and decide what actions, if any, should be taken. The final decision is the operator’s, but more timely information should lead to better, timelier decisions.
Example in-progress: Arkansas pipeline system
GlobaLogix has worked with BRS Labs to apply the behavioral recognition technology to data pulled from an existing gas gathering/gas delivery system in Arkansas, proving the ability to process very large data sets over long periods of time within the current SCADA system. Historical versus point-in-time comparisons have revealed insights that control room operators would have found very interesting and useful in a full-scale deployment.
A key element in this approach is that the in-progress evaluation is non-intrusive to existing systems – no new instrumentation or databases are required and there is little or no impact on normal operations. The application of the system in this case has proven itself to be intuitive, requiring no new training for an operator to be more effective in driving actions that could avoid major incidents, leaks and/or safety issues.
Multiple benefits
The collective advantage of this approach should be greater success and efficiency of control room management (CRM), through strengthening operators’ ability to use data from their systems to anticipate problems instead of reacting after the fact. It will reduce negative impacts such as downtime through early intervention, deliver reduced costs by not requiring a major overhaul of a company’s SCADA system (as an add-on or modular approach) and provide these benefits without impacting existing systems.
Unlike a rule-based system requiring considerable upfront configuration and rules establishment, the artificial intelligence behind this machine-learning system can be installed and fully functional quickly and teaches itself to recognize unexpected activity without human intervention. Finally, the solution supplements the intuition and experience level of the control room operator while leaving experienced human operators firmly in control of actions and response.
A new paradigm
Until now, there has been no development in alarm management in any SCADA product that makes use of machine learning or behavioral recognition technology. In effect, this solution adds as many eyes on events transpiring in an operation as there are data points.
A component using behavioral recognition improves situational awareness and supplements alarm management tools in most SCADA systems. The result is a new perspective on operations in the control room that reveals not just what happened, but what is happening right now.
Author
Jim Fererro, with 35 years’ experience in natural gas production and gas compression, is Senior Vice President of Houston-based GlobaLogix (www.globlx.com). He can be reached at jim.fererro@globlx.com or 713-987-7634.
Comments